Deliverable D2.2. Measurement Results and Final mmmagic Channel Models

Transcription

1 Document Number: H2020-ICT mmMAGIC/D2.2 Project Name: Millimetre-Wave Based Mobile Radio Access Network for Fifth Generation Integrated Communications (mmmagic) Deliverable D2.2 Measurement Results and Final mmmagic Channel Models Date of delivery: 12/05/2017 Version: 2 Start date of Project: 01/07/2015 Duration: 24 months

2 Deliverable D2.2 Measurement Results and Final mmmagic Channel Models Project Number: Project Name: ICT Millimetre-Wave Based Mobile Radio Access Network for Fifth Generation Integrated Communications Document Number: Document Title: H2020-ICT mmMAGIC/D2.2 Measurement Results and Final Channel Models for Preferred Suitable Frequency Ranges Editor(s): Authors: Dissemination Level: Michael Peter (HHI) Katsuyuki Haneda (Aalto), Sinh L. H. Nguyen (Aalto), Aki Karttunen (Aalto), Jan Järveläinen (Aalto), Aliou Bamba (CEA), Raffaele D Errico (CEA), Jonas Medbo (EAB), Fabian Undi (HHI), Stephan Jaeckel (HHI), Naveed Iqbal (HWDU), Jian Luo (HWDU), Marcin Rybakowski (NOKIA), Cheikh Diakhate (Orange), Jean-Marc Conrat (Orange), Alexander Naehring (R&S), Shangbin Wu (SRUK), Angelos Goulianos (UniBris), Evangelos Mellios (UniBris) PU Contractual Date of Delivery: 12/05/2017 * Security: Status: Version: 2 ** File Name: Public Final mmmagic_d2-2.docx *The delivery date was postponed from 31/03/2017 to 12/05/2017 with the approval of the European Commission. **This version provides an updated Table 4.1, which reflects the final status of the channel model implementation. Its features go beyond the functionality reported in Version 1. mmmagic Public ii

3 Abstract This deliverable describes the extensive multi-frequency channel measurement and simulation campaigns conducted in the mmmagic project, covering mm-wave 5G propagation scenarios. The data is evaluated to characterize environment-specific propagation effects. Based on the findings, refined modelling approaches are developed and embedded in a geometry-based stochastic channel model (GSCM). The proposed mmmagic channel model incorporates mmwave specific features and major advancements concerning ground reflection and blockage effects, the support of large bandwidths and large antenna arrays, the provision of spatial consistency, and outdoor-to-indoor penetration loss modelling. Keywords 5G channel model, mm-wave propagation, geometry-based stochastic channel model, GSCM, channel measurement plan, QuaDRiGa, ray tracing, blockage, multi-frequency channel sounding, channel sounder validation, diffuse scattering, diffraction Acknowledgements We would like to acknowledge the following people for the valuable reviews to the deliverable: Henrik Asplund, Mythri Hunukumbure, Pekka Kyösti and Miurel Tercero. mmmagic Public iii

4

5 Document: H2020- ICT mmMAGIC/D2.2 Executive Summary The availability of reliable and accurate channel models is crucial for the design, deployment, and standardization of 5G radio access technology (RAT), and related performance evaluation. However, the elaboration and parameterization of such models is extremely challenging due to the wide range of targeted frequency bands, including mm-wave frequencies, the diverse propagation conditions in a large number of different scenarios, and the limitations of reported measurement results. One of the key ambitions of mmmagic (mm-wave based Mobile radio Access network for fifth Generation Integrated Communications) has been to develop advanced channel models for the entire frequency range from 6 to 100 GHz, based on joint measurement and simulation efforts. In this context, more than 20 channel measurement campaigns were conducted in a variety of propagation environments across eight frequency bands between 6 and 100 GHz. In the initial phase of the project, the requirements associated with 5G channel modelling have been identified. They are described in deliverable D2.1 [mmmagic-d2.1] along with a review of preceding state-of-the art channel models and methods for channel sounder validation. This deliverable provides a description of all conducted measurement campaigns and map-based simulations in the prioritised propagation environments, namely urban micro (UMi) street canyon and open square, indoor office and airport, and outdoor-to-indoor (O2I). A summary of each activity is provided in conjunction with the main findings and the impact on modelling. The measurement and simulation data have been analysed to extract environment-specific channel characteristics. In direct connection with the identified model requirements, special focus of the presented evaluations is on the frequency dependency of the delay spread, the impact of the ground reflection at mm-wave frequencies, improved clusters and subpaths modelling, smallscale fading, blockage, building penetration loss, spatial consistency and scattering behaviour. Substantiated by the findings, modelling approaches are derived for these aspects. The proposed mmmagic channel model has been developed in parallel with the 3GPP [3GPP38.900, 3GPP38.901], ITU-R [IMT-2020.EVAL], ITU-R building entry loss [ITU-R P-series] and QuaDRiGa [JRB+14] models. Consequently, there have been active contributions from mmmagic partners to 3GPP and ITU-R where several approaches developed in mmmagic have been adopted during the course of the project. The mmmagic channel model is a geometry based stochastic model (GSCM) comprised of baseline model components and additional features that extend its accuracy and applicability. The model features major advancements regarding the incorporation of ground reflection and blockage effects, building penetration loss, the support of large bandwidths and large antenna arrays, and the provision of spatial consistency. A subset of the proposed mmmagic features is implemented in QuaDRiGa v2.0, which is available as open-source software. mmmagic Public v

6 Document: H2020- ICT mmMAGIC/D2.2 Contents 1 Introduction Channel model requirements mmmagic propagation scenarios Summary of measurement and simulation campaigns Indoor scenarios Indoor measurement campaigns in the V and E band Indoor measurement campaigns at 2.4, 6, 15 & 60 GHz Indoor high resolution campaign at 60 GHz Airport check-in area measurements at 15, 28, 60, and 86 GHz bands Airport check-in area simulations at 15, 28, 60, and 86 GHz bands Ultra-wide band measurements at 7, 34 and 60 GHz in a small lecture room Measurement on spatial consistency Urban micro outdoor scenarios Street canyon measurements at 15, 28, 60 and 86 GHz bands Street canyon simulations at 15, 28, 60 and 86 GHz bands Street canyon measurements at 2, 15 and 60 GHz bands Open square measurements at 28 and 86 GHz bands Open square simulation at 28 GHz band Multi-frequency measurements in Berlin at 10, 28, 41 and 82 GHz Measurement and simulation campaign in Daejeon Ray-tracing simulation in Madrid-grid and Aalborg city layouts Open square measurements at 17 GHz Outdoor measurements at 3, 17 and 60 GHz Outdoor-to-indoor Scenarios O2I measurement campaigns at 3, 10, 17 and 60 GHz O2I measurement campaigns at 2, 6, 15, and 60 GHz O2I measurement campaign at 2, 6, and 60 GHz O2I measurement campaign at 17 GHz Scattering, ground reflection and blockage Specular reflections and diffused scattering Ground reflection in a courtyard Ground reflection in a street canyon Blockage measurements in indoor office with phantom Characterization of modelling components and related investigations Frequency dependency of large-scale parameters Literature review mmmagic results Ground reflection Background and motivation Fresnel reflection for ground materials Validation with mmmagic results Conclusion Clusters and subpaths Recent progress on the topic in 3GPP Kmeans++ cluster algorithm Intra-cluster characteristics in outdoor and indoor environments Clusters modelling Summary of studies of intra-cluster characteristics Small scale fading Effect of bandwidth at K-factor and fading depth Effect of bandwidth on the cross polarization ratio (XPR) Effect of antenna directivity Blockage mmmagic Public vi

7 Document: H2020- ICT mmMAGIC/D mmmagic blockage model Spatial consistency Geometric stochastic approach Specular reflection, diffused scattering and depolarization Characterisation of specular reflections Characterisation of diffused scattering Depolarisation Surface scattering: modelling of specular and diffused components Building penetration loss Channel model Coordinate system Antenna model Propagation-scenario description Building penetration loss Large-scale fading model Autocorrelation model Inter-parameter correlation model Small-scale fading model Initial delays and normalized path powers Departure and arrival angles Mapping of paths to sub-paths Polarization model Additional features Ground reflection Spatial consistency Blockage Conclusion References A Details on measurement and simulation campaigns A.1 Indoor scenarios A.1.1 Indoor campaign in the V and E band A.1.2 Indoor measurement campaigns at 2.4, 6, 15 & 60 GHz A.1.3 Measurements in airport check-in area at 15, 28, 60 and 86 GHz A.1.4 Simulations in airport check-in area at 15, 28, 60 and 86 GHz A.1.5 Ultra-wide band measurements at 7, 34 and 60 GHz A.1.6 Measurement on spatial consistency A.2 Urban micro outdoor scenarios A.2.1 Street canyon measurements at 15, 28, 60 and 86 GHz A.2.2 Street canyon simulations at 15, 28, 60 and 86 GHz A.2.3 Open square measurements at 28 and 86 GHz A.2.4 Multi-frequency measurements in Berlin at 10, 28, 41 and 82 GHz A.2.5 Measurement and simulation campaign in Daejeon A.2.6 Ray-tracing simulation in Madrid-grid and Aalborg city layouts A.2.7 Open square measurements at 17 GHz A.2.8 Outdoor measurements at 3, 17 and 60 GHz A.3 Outdoor-to-indoor Scenarios A.3.1 O2I measurement campaigns at 3, 10, 17 and 60 GHz A.3.2 O2I measurement campaign at 2.44, 5.8, and 58.7 GHz A.3.3 O2I measurement campaign at 17 GHz A.4 Reflection, scattering and blockage A.4.1 Specular reflections A.4.2 Non-specular scattering A.4.3 Ground reflection A.4.4 Blockage Measurements A.5 LSP frequency correlations for multi-frequency simulations mmmagic Public vii

8 Document: H2020- ICT mmMAGIC/D2.2 A.6 Cross-polarization ratio in indoor and outdoor environments A.6.1 Measurement campaigns A.6.2 Multipath detection A.6.3 MPC XPR Model A.6.4 MPC XPR model parameters B Subpath angle mappings C Results from previous measurements on frequency dependence of LSPs D Individual Parameter Tables mmmagic Public viii

9 Document: H2020- ICT mmMAGIC/D2.2 List of Figures Figure 2.1: Overview of measurement (black) and simulation (blue) campaigns Figure 3.1: Delay spread versus frequency (logarithmic units) in different indoor (left) and O2I (right) multi-frequency measurement campaigns. The solid lines show the linear fits Figure 3.2: Delay spread versus frequency (logarithmic units) in different outdoor LOS (left) and NLOS (right) multi-frequency measurement campaigns. The solid lines show the linear fits.. 35 Figure 3.3: Fitted α and corresponding confidence intervals (95%) for the different measurement campaigns Figure 3.4: Reflection coefficient for different ground conditions as a function of the angle of incidence at 28 GHz for vertical polarization (left) and horizontal polarization (right) Figure 3.5: Measurement results in courtyard (flagstone/ concrete) with variation of Tx-Rx distance for vertical polarization at 28 GHz (left) and 39 GHz (right) and model output Figure 3.6: Measurement results in courtyard (grass) with variation of Tx-Rx distance for vertical polarization at 28 GHz (left) and 39 GHz (right) and model output Figure 3.7: Measurement results on tarmac runway with variation of Tx-Rx distance for vertical (left) and horizontal polarization (right) and model output Figure 3.8: Measurement results on the tarmac runway at 220 m distance with variation of Rx height for vertical (left) and horizontal polarization (right) and model output Figure 3.9: Measurement results on grassland with variation of Rx height at 220 m distance (left) and variation of Tx-Rx distance (right), and model output Figure 3.10: Measurement results in the street canyon (flagstone/concrete ground) with variation of Tx-Rx distance for vertical (left) and horizontal polarization (right) and model output Figure 3.11: Example of Kmeans++ algorithm for subpath clustering Figure 3.12: Fading behaviour comparison for different channel bandwidths, LOS, V-V polarization Figure 3.13: Effect of bandwidth on the fading statistics and K-factor scaling, setup # 1 (LOS case) Figure 3.14: Effect of polarization on the K-factor scaling, BW=2 GHz, LOS Figure 3.15: Effect of polarization on the K-factor scaling, BW=2 GHz, Dual reflections from black board and wall Figure 3.16: K-factor analysis for different channel bandwidths, V-V polarization setup Figure 3.17: Fade depth scaling for different channel bandwidths, V-V polarization setup, s = Figure 3.18: Average XPR for (a) horizontal to vertical XPRH (b) vertical to horizontal XPRV versus channel bandwidth Figure 3.19: XPR behaviour comparison for different channel bandwidths Figure 3.20: Coefficient of variance for (a) horizontal to vertical CVH (b) vertical to horizontal CVV versus channel bandwidth Figure 3.21: Fade depth for different bandwidths and directive antennas for s = Figure 3.22: Blockage simulation for a 4m wide quadratic screen (two left hand graphs). The radio frequency is 6 GHz, the TX distance to screen 100m and RX distance to screen 1m. In a) the RX pathway is parallel and in b) 45 degrees relative to the horizontal screen edges mmmagic Public ix

10 Document: H2020- ICT mmMAGIC/D2.2 Figure 3.23: Measured signal strength when a garbage truck temporarily blocks the LOS condition of the 15 GHz transmitter at 70m distance Figure 3.24: METIS (upper) and improved (lower) blockage model output for scenario shown in Figure Figure 3.25: Same scenario as Figure 3.24 but with different placement of cab Figure 3.26: Modelled blocking scenario (left) and corresponding output for 62 GHz (middle) and 83.5 GHz (right) Figure 3.27: LSPs generation procedure Figure 3.28: AoA vs. Route Figure 3.29: UE trajectory Figure 3.30: cluster birth and death in excess delay domain Figure 3-31: Clusters are translated into geometric positions Figure 3.32: The problem of independent clusters of nearby users (current GSCM) Figure 3.33: Shared clusters (necessary improvement) Figure 3.34: Dropping of users Figure 3.35: Grid model (GGSCM): Calculate new cluster information at each grid point. Interpolate clusters between the four grid points Figure 3.36: Location of a cluster Figure 3.37: Drifting of angles and delays Figure 3.38: Measured profile for dressed stone wall and window scattering Figure 3.39: Power angular profiles at 45 0 transmit angle, for the concrete pillar measurements Figure 3.40: Power angular profiles at 45 0 transmit angle, for the red stone wall measurements Figure 3.41: Received Power versus angle for both vertical and horizontal polarizations at 6 meter distance and 45 0 transmit angle Concrete pillar Figure 3.42: Received Power versus angle for both vertical and horizontal polarizations at 6 meter distance and 45 0 transmit angle Red stone wall Figure 3.43: Modelling of amplitude distribution for (a) The window surface, (b) Bath stone wall and (c) The dressed stone wall surface Figure 3.44: Modelling of exponential power decay with respect to angle Figure 3.45: Modelling of variations (S) around the mean exponential decay model Figure 3.46: Cumulative distribution functions (CDFs) of loss in excesses of free space loss (building entry loss) for the calibrated model compared with measurement data Figure 3.47: Material penetration losses Figure 3.48: O2I penetration losses for low-loss model Figure 3.49: O2I penetration losses for high-loss model Figure 4.1: Channel model components Figure 4.2: Illustration of the angles and vectors used for the calculations Figure 4.3: Values of the reflection coefficients for ε = Figure 4.4: Effective path loss at 28 GHz and 6 m BS height mmmagic Public x

11 Document: H2020- ICT mmMAGIC/D2.2 Figure 4.5: Geometry of the improved blockage model Figure A.1: Channel measurement setup at CEA-LETI (a). Antenna steering in the V-band (b) and E band (c) in the office Figure A.2: Indoor environments: office (first line), conference room (second line) Figure A.3: Measured PDPs in the office (Rx1) and the detected multipath Figure A.4: Multipath extraction results for few positions Figure A.5: Bandwidth dependency of the rms delay spread. These delay spread values have been obtained from the full PDP (all delay bins considered) so that the comparison in both bands would not be impacted by the considered noise floor, nor by an applied algorithm to extract the MPCs Figure A.6: Path loss model in D2D scenario: office (left), conference room (right) Figure A.7: Path loss model in BS scenario Figure A.8: Delay spread: office (left), conference room (right) Figure A.9: Delay spread in BS scenario: office Figure A.10: Example of multipath clustering in the office room Figure A.11: Example of multipath clustering in the conference room Figure A.12: Cluster occurrence in indoor environments Figure A.13: Cumulative distribution function of the number of sub-paths per cluster Figure A.14: First cluster contribution in the total power Figure A.15: cdf of intra cluster rms angular spread Figure A.16: cdf of intra cluster rms delay spread Figure A.17: Clusters and rays arrival rate in the office room Figure A.18: Clusters and rays decay constant in the office room Figure A.19 Relative arrival angle of the rays in the office Figure A.20: Relative arrival angle of the clusters Figure A.21: Scatter plot of arrival times versus arrival angles for clusters Figure A.22: Scatter plot of arrival times versus arrival angles for rays Figure A.23: Measurement scenario LOS Figure A.24: Measurement scenario NLOS Figure A.25: Power angle distributions for full space angle for the LOS and NLOS scenarios for 5.8, 14.8 and 58.7 GHz Figure A.26: RMS power delay and angle spreads (left) and power delay profiles (right) for LOS and NLOS for the different frequencies Figure A.27: The window blocking effect at 5.8 GHz and transparency at 14.8 GHz is illustrated in the graphs above Figure A.28: Indoor channel impulse response measurement scenario Figure A.29: Power delay profiles for TX positions 1-4, and, 13, Figure A.30: Median RMS delay spread as a function of frequency Figure A.31: Virtual antenna array set-up mmmagic Public xi

12 Document: H2020- ICT mmMAGIC/D2.2 Figure A.32: Measurement scenario. The virtual antenna array is located at TX. The four RX antenna locations marked RX1-RX Figure A.33: Power delay profiles for the different RX locations. The expected free space loss for versus distance is marked with a dotted line Figure A.34: Directional power distributions on panoramic photo from the TX array location Figure A.35: Identified MPCs in elevation vs. azimuth (left) and propagation distance vs. azimuth (right) scatter plots for RX1 (upper) and RX2 (lower) Figure A.36: Cumulative distributions of fraction received power versus number of MPCs for the rectangular very high resolution array (upper) and the cubic array described previously on page 124 (lower graph) Figure A.37: Measurement environment in Helsinki airport check-in area: main terminal hall (left) and side corridor (right) Figure A.38: (Left) Floor plan of the environment with the measured locations in the airport, and (right) the exemplary PDP of one LOS link in different frequency bands Figure A.39: Large scale parameters for measured (black) and optimized simulated (green) channels at 15 GHz Figure A.40: Large scale parameters for measured (black) and optimized simulated (green) channels at 60 GHz Figure A.41: Measured and simulated PDPs at 15 GHz (left) and 60 GHz (right) Figure A.42: Simulated BS (Rx) and MS (Tx) positions in airport Figure A.43: 360 panoramic view of measurement scenario Figure A.44: Overview of the measured TX-RX locations Figure A.45: Overview of the channel sounding setup Figure A.46: Overview of the measurement setups Figure A.47: Picture of the measurement set-up and the Top-view schematic of the measurement set-up Figure A.48: CIR for various bandwidths in LOS case, H-H polarization setup Figure A.49: Average directional delay spread [ns] vs. bandwidth, dynamic threshold = 25 db, standard deviation [ns] values are shown at each average rms DS value along the plot Figure A.50: Average directional DS with and without beam selection, bandwidth = 4 GHz. 142 Figure A.51: Channel Sounder Transmitter Description Figure A.52: Channel Sounder Receiver Description Figure A.53: Channel Sounder Keysight Based Figure A.54: Transmitter Setup used for Indoor Large Scale Measurements Figure A.55: Receiver Setup used for Indoor Large Scale Measurements Figure A.56: Plane of the MVB Atrium, and location of measurement points and Tx Figure A.57: Points of measurement clustered in 3 types of scenarios Figure A.58: Route of the Outdoor Measurement Campaign Figure A.59: Transmitter for Outdoor Large Scale Measurement Campaign Figure A.60: Receiver for Outdoor Large Scale Measurement Campaign mmmagic Public xii

13 Document: H2020- ICT mmMAGIC/D2.2 Figure A.61: Correlation Coefficient of recorded impulse responses for all points in LOS Figure A.62: Correlation Coefficient for all points in LOS-NLOS scenario Figure A.63: Correlation Coefficient for all points in NLOS scenario Figure A.64. 3D Power Angular Profiles in Outdoor Scenario Figure A.65: Measurement site (left) and the point cloud model of the environment with 11 Tx and 1 Rx locations overlaid (right) Figure A.66: PDPs of the 15, 28, 60 and 86 GHz measurements at Tx3 (left) and Tx5 (right) Figure A.67: Large scale parameters for measured (black) and optimized simulated (green) channels at 15 GHz Figure A.68: Large scale parameters for measured (black) and optimized simulated (green) channels at 60 GHz Figure A.69: Measured and simulated PDPs at 15 GHz Figure A.70: BS and MS positions in street canyon simulation Figure A.71: Top-view of the measurement location (left) and the floorplan (right) Figure A.72: Comparison on PDP (left) and PAS (right) of link Tx10-Rx1 between 28 and 83 GHz measurements Figure A.73: DS versus link distance (left), and azimuth AS versus link distance (right). The dashed lines show the average values Figure A.74: Empirical CDFs of RMS delay spread Figure A.75: Bird s views of environment for measurement and ray tracing simulation Figure A.76: Channel sounders for measurement campaign in Daejeon Figure A.77: CDFs of number of subpaths in measurement and ray tracing results Figure A.78: CDFs of intra-cluster RMS delay spreads of measurement and ray tracing Figure A.79: CDFs of intra-cluster azimuth AoA spreads of measurement and ray tracing Figure A.80: CDFs of intra-cluster zenith AoA spreads of measurement and ray tracing Figure A.81: 2D and 3D layout of Madrid-grid with locations of 5 transmitters Figure A.82: 2D and 3D layouts of Aalborg city for 5 transmitters Figure A.83: The number of 5 strongest interaction types in Madrid-grid layout Figure A.84: The number of 5 strongest interaction types in Aalborg-city layout Figure A.85: Pathloss exponents for LOS and NLOS for both layouts Figure A.86: Shadow Factor for LOS and NLOS for both layouts Figure A.87: RMS delay spread for both layouts Figure A.88: RMS azimuth angle spread of departure for both layouts Figure A.89: RMS azimuth angle spread of arrival for both layouts Figure A.90: RMS zenith angle spread of departure for both layouts Figure A.91: RMS zenith angle spread of arrival for both layouts Figure A.92: Mean cross-polarization ratio for both layouts Figure A.93: Aerial view for open square measurement at R&S. Google, GeoBasis DE/BKG mmmagic Public xiii

14 Document: H2020- ICT mmMAGIC/D2.2 Figure A.94: Averaged PDPs for OS LOS scenarios RX 1 and RX Figure A.95: Averaged PDP for OS OLOS scenario RX Figure A.96: Averaged PDP for OS NLOS scenario RX Figure A.97: Measured path loss without antenna gain for different receiver locations in OS scenario Figure A.98: Aerial map view of the measurement environment Figure A.99: Tx and Rx measurement equipment Figure A.100: LoS CDFs of DS values Figure A.101: NLoS CDFs of DS values Figure A.102: PDPs with TX at TxA and RX at A Figure A.103: PDPs with TX at TxC and RX at C Figure A.104: LoS DS average values with different thresholds Figure A.105: NLoS DS average values with different thresholds Figure A.106: 3GPP TR DS model above 6 GHz Vs measurement results Figure A.107: O2I scenario and building layout. a) Co: corridor; b) and c) Fl: residential flat; d) BR: break room; e) LO1 and LO2: large office Figure A.108: Rx equipment. a) 3, 10 and 17 GHz Omni. antennas; b) 90 HPBW antenna at 17 GHz; c) 20 and 90 HPBW antennas at 60 GHz Figure A.109: Building entry losses Figure A.110: PDPs at 3, 10, 17 and 60 GHz at Rx position Figure A.111: Outdoor to indoor measurement scenario (left). The indoor Rx locations at floors 4, 7 and 8 are marked with blue-filled circles and the Tx locations at floor 4 are marked with a red-filled circles. On the right hand side, a photograph of the building is shown Figure A.112: Measured building entry loss (loss in excess of free space loss) versus elevation angle for the different frequencies together with the fitted linear model. The top graph corresponds to minimum loss (2.5% level); the middle graph corresponds to median loss; and the bottom graph to maximum loss (97.5% level) Figure A.113: Cumulative distribution functions of measured building entry loss for the different elevation angles and radio frequencies. The left hand graphs show distributions where Tx locations p1, p8 and p11 are excluded. The right hand graphs show distributions for all Tx locations Figure A.114: Aerial view for outdoor to indoor measurement at R&S. Google, GeoBasis DE/BKG Figure A.115: Single snapshot (left) and averaged (right) PDPs for O2I LOS TX1-RX Figure A.116: Averaged PDPs for O2I NLOS RX Figure A.117: 60 GHz Keysight channel sounder Figure A.118: Trolley used to conduct the scattering measurements Figure A.119: Graphical illustration of the trolley measurements for (a) 30 0 and (b) Figure A.120: Wall materials used in the non-specular study Figure A.121: Graphical description of the arc diffuse scattering measurement Figure A.122: Transmitter and Receiver Setup for an arc measurement following the chalk line mmmagic Public xiv

15 Document: H2020- ICT mmMAGIC/D2.2 Figure A.123: Diffused scattering measurement snapshot Concrete Pillar Figure A.124: Diffused scattering measurement snapshot Redstone wall Figure A.125: Setup Block Diagram Figure A.126: Measurement Locations (grassland) Figure A.127: Received signal strength over distance on flagstone/concrete: measured power (blue), FSPL (red), TRGR model (black) Figure A.128: Two-ray ground-reflection model for different frequencies with measured values on flagstone/concrete: measured power (blue), FSPL (red), TRGR model upper, lower and centre frequency (black, green, light blue Figure A.129: Received signal strength over distance on grassland: measured power (blue), FSPL (red), TRGR model (black) Figure A.130: Measurement Locations (concrete surface) Figure A.131: PDP for grassland (blue) and flagstone (red) Figure A.132: Measurement environment and scenario Figure A.133: Tx antenna at 17 GHz (left) and Tx measurement van (right) Figure A.134: Rx antennas at 17 GHz (left) and trolley with Rx antenna at 60 GHz (right) Figure A.135: Three typical impulse responses separated by 5 cm at 17 GHz with interferences in the main component Figure A.136: Channel gain vs distance at 3 GHz, Rx omni antenna Figure A.137: Channel gain vs distance at 17 GHz Figure A.138: Channel gain vs distance at 60 GHz Figure A.139: Blockage measurement in the office room. In the left hand side picture, red (resp. black) rectangles indicate desks (resp. storage cabinets and bookshelves) Figure A.140: Example of power delay profile with(out) the phantom Figure A.141: Correlations between 15 GHz ASD and 83.5 GHz ASD in airport (lhs.) and 61 GHz K-factor and 83.5 GHz K-factor in airport (rhs.) Figure A.142: Correlations between LSPs in airport (lhs.) and in street canyon (rhs.) Figure A.143: Main polarization PADP and its power delay profile in the cafeteria at centre frequency of 63 GHz. Detected MPCs are shown with markers and the noise threshold level is shown with black dash line [KJN+17] Figure A.144: MPC XPR as a function of the multipath excess loss E in the airport at 28.5 GHz. Markers show measured XPRs XPR= M i-c i and samples with XPR> M i-p th and XPR=< P th -C i). Lines show fitted model mean value μ and μ±2σ-limits [KJN+17] Figure A.145: MPC XPR parameter values for the model (A-33)-(A-35) in all the 25 measurement campaigns [KJN+17] Figure B.1: Comparison between the subpath angle mapping in 3GPP 3D-SCM and the MEA Figure B.2: Comparison between the subpath angle mapping of the new 3GPP channel model and the MEA mmmagic Public xv

16 Document: H2020- ICT mmMAGIC/D2.2 List of Tables Table 1.1: Definition of propagation scenarios... 3 Table 3.1: Defined requirements in order to ensure comparability across different frequencies Table 3.2: Indoor and O2I multi-frequency measurement campaigns in mmmagic Table 3.3: Outdoor multi-frequency measurement campaigns in mmmagic Table 3.4: Frequency-dependent linear regression model for DS in each campaign Table 3.5: Frequency-dependent linear regression model for DS in each combined scenario 36 Table 3.6: Measurement data used for investigation of ground reflection and model validation Table 3.7: Summary of recent progress on characteristics of clusters and subpaths Table 3.8: Decay rate and arrival rate of clusters and rays Table 3.9: Summary of parameters of intra-cluster characteristics Table 3.10: K-factor at different bandwidths and with different antenna combinations Table 3.11 Shadowing loss values. All values are expressed in decibels (db). The prediction error is the difference between measured and simulated shadowing loss values Table 3.12: 90% Power concentration over all transmit angles and distances Table 3.13: RMSE among various distributions to describe the variations of the specular reflections Table 3.14: Modelling parameters for the diffused scattering path for the case of concrete pillar and red stone wall at Table 3.15: 3GPP Material penetration losses Table 3.16: 3GPP O2I penetration losses Table 4.1: Model components Table 4.2: Combined mmmagic model parameter table Table 4.3: Comparison between the subpath generation method of mmmagic and the path/subpath generation method 3GPP channel models Table 4.4: Electrical properties of the environment, GHz Table 4.5: Shadow Fading for different frequencies in [db] Table A.1: LSP in indoor environments for D2D scenario Table A.2 Average rms angular and delay spread per cluster and Rx location for both frequencies (bandwidth of 6 GHz) in the office room. [-] indicates that the related cluster does not exist Table A.3: Number of MPCs needed to account for 95% of power Table A.4: Mean spread values in Helsinki airport Table A.5. Average measured and simulated large scale parameters Table A.6: Material parameter values Table A.7: Measurement campaign in view of requirements defined in Section Table A.8: Antennas and bandwidths used during the experiments Table A.9: Outdoors Channel Sounder Measurements Campaign mmmagic Public xvi

17 Document: H2020- ICT mmMAGIC/D2.2 Table A.10: Distances of the analysed points Table A.11: Spatial Correlation values for the atrium indoor measurements Table A.12: Spatial Correlation values for outdoor measurements Table A.13: Mean spread values for measured LOS links in street canyon Table A.14: Average measured and simulated large scale parameters Table A.15: Material parameter values in street canyon simulations Table A.16: Summary of the 28 and 83 GHz measurement campaign in Helsinki open square Table A.17: Mean DS and mean azimuth AS of 28 and 86 GHz measurements in Open Square Table A.18: Multi-frequency measurement campaign in Berlin view of requirements defined in Section Table A.19: Median and 95% Quantile of the RMS delay spread Table A.20: Model parameters of HiRes evaluation Table A.21: Comparison measurement and model results Table A.22: Channel Measurement Settings Table A.23: Ray Tracing Simulation Settings Table A.24: Measurement details for open square measurements in Munich Table A.25: Evaluation of measured and expected path loss for Vivaldi antenna Table A.26: Evaluation of measured and expected path loss for Dipole antenna Table A.27: Measurement setup parameters Table A.28: Computed DS values Table A.29: Measurement setup parameters Table A.30: Building penetration loss values Table A.31: Channel delay spread values Table A.32: Basic data of measurement setup Table A.33: Summary of the 17 GHz measurement campaign in Munich outdoor to indoor 195 Table A.34: Measured wall types for analysing diffused scattering effects at 60 GHz Table A.35: Positions for Measurement Campaign Table A.36: Path power for different measurements Table A.37: Measurement Setup Parameters Table A.38: Shadowing loss values. All values are expressed in decibels (db) Table A.39: Measurement environments and frequency bands [KJN+17] Table C.1: Previous measurements and results on frequency dependence of LSPs Table D.1: UMi outdoor parameter tables (SRUK + HHI results) Table D.2: UMi outdoor parameter tables (Aalto simulation results) Table D.3: UMi outdoor parameter tables (Orange + EAB results) Table D.4: Indoor parameter tables (Aalto simulation results for Airport) mmmagic Public xvii

18 Document: H2020- ICT mmMAGIC/D2.2 Table D.5: Indoor parameter tables (CEA results) Table D.6: Indoor parameter tables (HWDU + EAB results) Table D.7: O2I parameter tables (Orange + EAB results) mmmagic Public xviii

19 Document: H2020- ICT mmMAGIC/D2.2 List of Abbreviations 2D 3D 3GPP 2 Dimensions / 2-Dimensional 3 Dimensions / 3-Dimensional 3 rd Generation Partnership Project GSCM GTD HBS Geometry-based Stochastic Channel Model Geometrical Theory of Diffraction Human Body Shadowing 5GCM 5G Channel Model HPBW Half-power Beamwidth ACF autocorrelation function IF Intermediate Frequency AoA AoD ASA ASD AWG B2B BS CDF, cdf CIR COST Angle of Arrival Angle of Departure Azimuth of Arrival Angle Spread Azimuth of Departure Angle Spread Arbitrary Waveform Generator Back-to-Back Base Station cumulative distribution function Channel Impulse Response The Cooperation in Science and Technology IMT InH ITU ITU-R KED KF KS LO LOS, LoS LSF International Mobile Telecommunication Indoor International Telecommunication Union ITU-Radiocommunication Knife Edge Diffraction K-Factor Kolmogorov-Smirnov Local Oscillator Line-of-Sight Large Scale Fading CTIA CW D2D DDS DMC DoD Cellular Telecommunication Industry Association Continuous Wave Device-to-Device Direct Digital Synthesizer Dense Multipath Component Direction of Departure LSP LTE MATLAB MCD METIS Large-Scale Parameter Long Term Evolution MATLAB software of MathWorks Multipath Component Distance The Mobile and wireless communications Enablers for the Twenty-twenty Information Society DS Delay Spread MIMO Multi-Input Multi-Output E-field EIRP Electric Field effective isotropic radiated power MiWEBA Millimetre Wave Evolution of Backhaul and Access EMW EoA Electromagnetic Waves Elevation Angle of Arrival mmmagic mm-wave Based Mobile radio Access network for Fifth Generation Integrated Communications EoD Elevation Angle of Departure mm-wave Millimetre Wave ER Effective Roughness MS Mobile Station ESD Elevation Spread of Departure MT Mobile Terminal GB GF GO Gigabyte Geometry Factor Geometrical Optics NLOS, NLoS O2I PADP Non-Line-of-Sight Outdoor-to-Indoor Power Angular Delay Profile mmmagic Public xix

20 PAS Power Angular Spectrum SRT Standard Ray Tracing PDP Power Delay Profile SSF Small-Scale Fading PLE Path Loss Exponent Tx Transmitter plos Pseudo Line-of-Sight UE User Equipment QD Quasi-Deterministic UMa Urban Macro QuaDRiGa RAN RAT RF RMS, rms RMSE RSRP The QUAsi Deterministic RadIo channel GenerAtor Radio Access Network Radio Access Technology Radio Frequency Root Mean Square Root Mean Square Error Reference Signal Received Power UMi UTD UWB V2V VNA V-pol VR VSA Urban Micro Uniform Theory of Diffraction Ultra-wideband Vehicle-to-Vehicle Vector Network Analyser Vertical Polarization Visibility Region Vector signal analyser RSS Radio Signal Strength WiGig Wireless Gigabit RT Rx SAGE SCM SCM-E SF Ray Tracing Receiver Space Alternating Generalized Expectation Spatial Channel Model SCM-Extended Shadow Fading WIMAX WINNER WLAN WP WRC Worldwide Interoperability for the Microwave Access the Wireless World Initiative for New Radio Wireless Local Area Network Work Package World Radiocommunication Conference SINR SISO SLS SMa SNR Signal-to-Interference plus Noise Ratio Single-Input Single-Output System Level Simulation Suburban Macro Signal-to-Noise Ratio XP XPD XPR ZoD ZSA ZSD Cross Polarization Cross Polar Discrimination Cross-Polarization Ratio Zenith of Departure Zenith of Arrival Spread Zenith of Departure Spread mmmagic Public xx

21 1 Introduction Wireless communication is an integral part of our lives today. Driven by the proliferation of smartphones and tablets, mobile data traffic has grown by 4000-fold over the past ten years and the uptrend continues. It is expected that there will be 11.6 billion mobile-connected devices by 2020 exceeding the world's projected population [Cis17]. The continuously growing number of devices in conjunction with the increasing data rates has made spectrum in the traditional cellular frequency bands to be an insufficient resource. Utilizing spectrum at millimetre-wave (mm-wave) frequencies, where partially up to several Gigahertz of contiguous bandwidth is available, is therefore seen as one of the key elements to meet the requirements in the next decade. In the latest World Radiocommunication Conference in 2015 (WRC-15), the ITU-R selected candidate bands between GHz and 86 GHz for IMT2020 (5G) to be allocated at WRC-19 (agenda item 1.13) including possible additional allocations to the mobile service on a primary basis [ITUR R-809]. Standardization and regulations activities for 5G in ITU-R [ITUR ] and 3GPP [3GPP ] have already been kicked off in 2015 and are accompanied with huge involvement. The main objective of the mmmagic (mm-wave based Mobile radio Access network for fifth Generation Integrated Communications) project [mmm15] is to develop and design new concepts for mobile Radio Access Technology (RAT) for deployment in the frequency range of GHz. One of the key ambition of mmmagic is to develop advanced channel models, based on statistically significant measurement and simulation data, as these models are crucial for supporting related developments by enabling a performance analysis of different approaches. The propagation channel related investigations of mmmagic includes the identification of model requirements, the review and assessment of relevant literature results, the conduction of extensive channel measurement and simulation campaigns, the analysis of the acquired data, and the development of high frequency (6-100 GHz) specific model components in view of extending the current geometry-based stochastic channel models (GSCMs). The key achievements of the mmmagic channel modelling investigations are closely related to: Identifying the requirements for 5G channel models, covering a frequency range from 6 to 100 GHz. Conducting numerous multi-frequency channel measurement campaigns and accompanying ray tracing simulations between 2 and 100 GHz for a variety of 5G propagation scenarios. Evaluating measured and simulated data with a specific focus on the modelling of frequency-dependent large-scale parameters, ground reflection effects, clusters and subpaths, small-scale fading, blockage, building penetration, spatial consistency, surface reflections and diffused scattering. Proposing novel channel modelling approaches for mm-wave system deployments. Delivering a GSCM-based mmmagic channel model, incorporating approaches from the latest 3GPP channel model and QuaDRiGa. The proposed mmmagic model is comprised of baseline model components and additional features, both derived from the post-processing analysis of the extensive measurements carried out in the project. Providing an open source implementation of the channel model as part of the new QuaDRiGa release (v2.0), including the mmmagic baseline model and selected additional features. This document is built upon the work of one of the mmmagic work packages, and is comprised of the description of all the conducted measurement and simulation campaigns, the results on channel characterization, large frequency range (including mm-wave) modelling approaches mmmagic Public 1

22 and the mmmagic channel model, which can directly be employed for the performance evaluation of mm-wave systems. A review of the state-of-the art is documented in deliverable D2.1 [mmmagic-d2.1], which also describes the identified modelling requirements and the channel sounder validation methods. Based on the findings of D2.1, substantial progress on propagation characterization and channel model enhancements is provided in this deliverable. The remainder of this document is structured as follows: Section 2 provides an overview of all measurement and simulation campaigns. It covers both outdoor and indoor environments, as well as specific propagation scenarios that are mostly focusing on scattering, ground reflection, blockage and building penetration loss. Each campaign is briefly described in a coherent manner along with the main research findings and their impact on modelling. In Section 3, specific evaluations and modelling approaches are presented with the focus on the frequency dependency of large-scale parameters, the impact of ground reflection, the modelling of clusters and subpaths, the investigation of the small-scale fading and blockage, building penetration loss, as well as the study on spatial consistency and scattering behaviour. The topics are directly related to the open points identified in D2.1, and are further reflected in the mmmagic channel model, presented in Section 4. It describes the model components, incorporating new modelling features and improvements. Finally, Section 5 concludes the main part of the deliverable. This document is also enhanced with additional information presented in the form of annexes. Annex A provides further details on the measurement and simulation campaigns. In Annex B, the new subpath angle mapping proposed in the mmmagic channel model is thoroughly presented. Finally, Annex C summarizes measurements reported in the literature on the frequency dependency of the large-scale parameters. In conjunction with the QuaDRiGa implementation v2.0, the material provided in this deliverable aims to provide a basis for the design and evaluation of 5G radio communication standards. Many evaluations and model enhancements elaborated in mmmagic have been injected into standardization by 14 channel-related submissions to 3GPP [3GPP160846] [3GPP ] and six submissions to ITU-R [ITUR-3J22E] [ITUR-5D338E]. The improvements are therefore reflected in the respective latest channel models. 1.1 Channel model requirements The requirements for a comprehensive 5G channel model have been identified in the initial phase of the project [mmmagic-d2.1] and have been an important incentive for all associated investigations in the project. These requirements are summarized as follows: Support of large frequency range GHz as addressed in mmmagic; with potential applicability to the full range from 0.5 to 100 GHz. Support of large antenna arrays and large channel bandwidths. Provision of frequency and spatial consistency. Support of mobility and dynamic scatterers; incorporation of blockage effects. Incorporation of ground reflection effect. 1.2 mmmagic propagation scenarios Relevant propagation scenarios for channel modelling have been identified at the beginning of the project. They mainly target the study of urban micro-cellular (UMi) scenarios, indoor environments and outdoor-to-indoor (O2I) propagation. These scenarios are summarized and specified in Table 1.1. It is worth mentioning that the urban macro-cellular (UMa) is not taken into account since frequency spectrum above 6 GHz is expected to initially be used for small cell base stations (BS). mmmagic Public 2

23 Table 1.1: Definition of propagation scenarios Scenario Definition/Specifications BS/UE deployment UMi Street canyon UMi Open square Indoor Office Indoor Shopping mall Indoor Airport O2I Stadium Urban environment with high user density. Users are pedestrians or slow vehicular users. Buildings of 4-7 storeys on both sides. The length of the street is at least 100 m. The street canyon typically includes street furniture (lamp posts and/or traffic signs) and may include trees. Urban environment with high user density. Users are pedestrians or slow vehicular users. (semi-) flat areas in the shape of a square or circle usually surrounded by buildings of 4-7 storeys. The main elements inside an open square are lamp posts, vegetation (e.g. trees), etc. The width of open squares is usually between 50 and 100 meters. Traditional office (private office): An enclosed work space for (usually) one to four people. These offices are enclosed by doors and walls. Cubicle: A semi-enclosed work space for one person. Typically, these working spaces are partitioned rows with dividers which may or may not reach the ceiling. Rows are usually separated by aisles. Open-office: An open work space for more than ten people. There are no walls or cubicles to separate staff s working space. The entire office, including traditional office, cubicle, and open-office, can typically have the dimension from 25 to 100 meters, while the ceiling height varies from 3 to 5 m. A large enclosed shopping area in form of multi-storey buildings, usually with open ceiling (atrium) in the middle of the building Shops are arranged next to the outer walls of the building; corridors provide a walking area for the pedestrians. The shopping mall including shops with typical dimensions from 50 to 200 m. Ceiling height in the corridors and individual shops: 3 5 m (typically 3 m). Ceiling above open space (atrium): 9 18 m. High user density depending on the time of day. Gate area: A gate, or gatehouse, is an area of an airport that provides a waiting area for passengers before boarding their flight. Most gates consist of seating, a counter, an aircraft entry or exit doorway, and a jet bridge. Ceiling height is usually between 4 to 6 meters. The gate is connected to a corridor with or without walls. Typical width of the gate area can vary from 20 m to 50 m, while its length along the corridor can stretch up to 100 m. Check-in area: The area before the security check-point where checkin counters are placed. The ceiling height is about 4 9 m. Airline and service offices are located in the multi-storey walls. The user density varies depending on the time and airport. Typical dimensions can vary from 50 to 200 m. UE is located in an indoor environment, either following office, shopping mall or airport, and the building is surrounded by UMi street canyon or open square scenarios. Dimensions and antenna heights of the outdoor and indoor environments are similar to the street canyon, open square, office, shopping mall and airport environments. A place or venue for sports, concerts, or other events. Consists of a field or stage either partly or completely surrounded by a tiered structure designed to allow spectators to stand or sit and view the event. Can be roofed or unroofed. Typical width can vary from 100 to 150 m. BS is installed on the lamp posts, walls or below the rooftop of buildings. 1.5 h UE 3 m, 3 h BS 10 m BS is placed below the rooftop level of the surrounding buildings. 1.5 h UE 3 m, 3 h BS 10 m BS can be placed at the ceiling or on the walls. 1 h UE 1.5 m, 1 h BS 5 m (from floor level) BS is positioned at the ceiling. h UE = 1.5 m, 3 h BS 5 m (from floor level) BS should be installed near the ceiling. h UE = 1.5 m, 4 h BS 9 m (from floor level) BS is positioned at neighbouring building exterior wall, a lamp post or similar. h UE = 1.5 m, 3 h BS 10 m (from ground level) BS is put on higher levels of spectators area. h UE = 1.5 m, 3 h BS 10 m mmmagic Public 3

24 Scenario Definition/Specifications BS/UE deployment Metro station Very high user density. A railway station for rapid transit system. The station can be above the ground level or underground. There are entrances/exits at the ground level. The metro station can be a junction of two or more metro lines in different floors. These floors are connected to each other and to the exits/entrances via stairs, escalators and lifts. Typical dimensions of underground stations are: width: approx m, length: approx m, ceiling height: approx m. High to very high user density depending on the time of day. (from floor level in spectators area) BS can be put near the ceiling of the station or on the walls. h UE = 1.5 m, 3 h BS 10 m (from floor level) In the course of the project, the following scenarios were further prioritised: UMi street canyon and open square Indoor office and airport O2I The motivation for this prioritization was to concentrate on some widespread key scenarios that were expected to have different characteristics due to largely differing propagation environments. It enabled the collection of statistically significant data for these scenarios and the comparison of the results from multiple measurement campaigns carried out in similar propagation environments. Based on the results, parameter sets for the consolidated scenarios UMi Outdoor, Indoor and O2I are given in Section 4.3 for usage with the mmmagic channel model. In addition, the parameter tables for individual measurement and simulation scenarios are summarized in Annex D. Except from the aforementioned scenario-related investigations, considerable effort was undertaken for the evaluation of surface reflection characteristics at mm-wave frequencies. mmmagic Public 4

25 2 Summary of measurement and simulation campaigns Channel measurements play a crucial role in the process of characterizing the radio channel and develop a unified statistical channel model for link- and system-level simulations that is valid over the entire frequency range from 6 to 100 GHz. For this purpose, various channel measurements have been conducted by different partners, yielding in measurement data for a variety of indoor, outdoor, and O2I scenarios at multiple frequencies. In addition, there were simulation campaigns performed in selected popular environments and frequencies to provide a large data set of propgation channels for the purpose of channel modelling. Overall, 54 single-frequency equivalent campaigns have been conducted. An overview of these measurements and simulations is depicted in Figure 2.1. Figure 2.1: Overview of measurement (black) and simulation (blue) campaigns. Since one goal of these campaigns is to provide data which can be used to analyze the frequency dependency of the various channel characteristics, the comparability across different frequencies has to be ensured. As measurements at different frequencies require specialized hardware which has a huge impact on the measurement configuartion parameters such as measurement bandwidth, dynamic range, antenna patterns, etc further processing effort might be necessary to harmonize these parameters. Several requirements have been identified (see details in Table 3.1) that define the parameters which need to be aligned in order to make a reasonable comparison among the different channel sounders that have been employed for this study. In the following section, a description of the conducted measurement campaigns is listed in detail. This list can serve as a reference in highlighting whether the conducted measurements fulfill these requirements. mmmagic Public 5

26 2.1 Indoor scenarios Indoor measurement campaigns in the V and E band Name Environment Indoor channel measurement campaigns Indoor environments: Office and conference room Frequency 62 GHz (V band) 83.5 GHz (E band) Bandwidth 6 GHz 6 GHz Objective These measurement campaigns aim to determine and compare the propagation parameters such as path loss, rms delay and angular spread, in indoor environments. Moreover, the clustering patterns of the multipath and the intra-clusters parameters have been derived. Brief description A 4-port Vector Network Analyser (VNA) has been used to probe the radio channel. Directive antennas have been used during the measurements in both bands. For the V band, a 0 dbi gain standard omnidirectional antenna (resp. 20 dbi gain and approximately 20 HPBW) was used on the Tx (resp. Rx) side. A vertically polarized horn antenna with 10 dbi gain and a HPBW of 50 (resp. 20 dbi and a HPBW of 15 ) was used on the Tx (resp. Rx) side in the E band. Two positioner devices have been used to rotate the antennas and perform mechanical steering. Only the LOS and OLOS cases have been considered. Office Evaluations From directional measurements, we derive the synthetic omni-directional channel function. This results in power delay profiles being free of the antennas effects. Furthermore, a multipath detection algorithm is applied for paths identification (see Data evaluation in Section A.1.1). Next, the K-means algorithm is applied to group the estimated paths into clusters. Several parameters of interest such as the rms delay spread, rms azimuthal angle of arrival spread, path loss model, and so on have been compared for both bands. Conference room Main findings and impact on modelling Path loss exponents of 1.36 and 1.34 are obtained in the office room for measurements in GHz and GHz, respectively. For the conference room, path loss exponents of 1.33 and 1.44 are obtained respectively in GHz and GHz. In general, the path loss exponents are very similar for both bands and environments. Average rms delay spread values of approximately 8.30 ns and 6.10 ns is obtained in the office at 62 GHz and 83.5 GHz, respectively. Average delay spreads of 5.60 ns and 5.20 ns are obtained in the conference room at 62 GHz and 83.5 GHz, respectively. It is observed that the rms delay spread tends to decrease with increasing frequency. It turns out that the number of clusters varies from 2 to 5, regardless of the environment or frequency. In general, we find that the modelling framework is likely to be valid in both bands, at least in indoor environments. References Details on the measurements and evaluations can be found in Section A.1.1 and D2.1 [mmmagic-d2.1]. Results can be found in references [BMD16, BMD17]. mmmagic Public 6

27 2.1.2 Indoor measurement campaigns at 2.4, 6, 15 & 60 GHz Name Environment Multi frequency indoor campaign in Stockholm Office Frequency 2.44 GHz 5.8 GHz 14.8 GHz GHz Bandwidth 80 MHz 150 MHz 200 MHz 2 GHz Objective The objective is to characterize the indoor propagation channel over a very wide frequency range. Brief description The measurements were based on a VNA using frequency sweep. Delay domain data was obtained by FFT. Matched vertical dipole antennas (at 1.5 m height), with very similar patterns, were used for all frequencies. Evaluations Careful Line of Sight (LoS) calibration at m distance was performed. For comparisons over frequency the bandwidth was set to 80 MHz for all frequencies. Propagation loss (loss in excess of free space loss) was determined as well as delay spread. To provide high resolution estimation of the direction of wave propagation, the channel was massively sampled in the spatial domain, applying subsequent Fourier transformation. Floorplan of indoor office measurement scenario. Main findings and impact on modelling The propagation loss results agree very well with the 3GPP model [3GPP38.900]. There is clear frequency trend as the propagation loss (i.e. loss in excess of free space loss) increases about 5 db per decade of frequency increment. The RMS delay spread for NLOS (~25 ns) does not show any clear frequency dependence which is in disagreement with the 3GPP model. However, the 3GPP model delay spread is in a similar range (15-30 ns). Furthermore, the angle spread does not show any clear frequency trend which is in contrast to 3GPP which models substantial decrease of angle spread with increase of frequency. The measured RMS azimuth angle spread is in the range degrees for NLOS and degrees for LOS. Furthermore, the measured RMS elevation angle spread is in the range degrees for both NLOS and LOS conditions. These measured angle spread values are in general larger than the values of the 3GPP model. Analysis of the delay spread dependence on beamforming shows that a few of the highest power directions have the lowest values. However, the major part of directions shows a nearly uniform distribution of delay spread. Consequently, the usage of narrow steered beams may result in both increase and decrease of delay spread as compared with using omni antennas. References Details on the measurements and evaluations can be found in A.1.2 and D2.1 [mmmagic-d2.1]. mmmagic Public 7

28 2.1.3 Indoor high resolution campaign at 60 GHz Name Environment Frequency Bandwidth Objective High resolution indoor measurements Office GHz 2 GHz The objective is to characterize the indoor propagation channel with very high resolution in direction and delay. Brief description The measurements were based on a VNA using frequency sweep over 2 GHz and a robot antenna positioner for spatial sampling, forming a virtual antenna array of elements. Evaluations Delay and direction distributions are provided by FFT and Hanning windowing to supress side lobes. An angle resolution better than one degree is obtained with this technique. Panoramic view of office measurement scenario with overlayed directional power distributions. Main findings and impact on modelling Due to the high spatial and delay resolution, between 1000 and multipath components have been resolved within 95% of the received power, depending on the specific scenario. The least multipath richness is observed for LOS and the highest richness for the most obstructed NLOS scenario. Significant scatterers, such as curtains, walls, and metal structures, are identified. For the most severely obstructed scenario, a few significant scatterers are identified as clusters, which are very distinct in the angular domain, while substantially spread in the delay domain. These are attributed to scattering objects, which generate diffuse scattering components (e.g. the wavy curtain). The other scenarios demonstrate substantially faster decaying clusters, which are attributed to scattering by specular reflections. One important finding is that the number of identified multipath components depends strongly on the resolution of the measurement. This was found by comparing with a previous measurement campaign [MSA16] with about one order of magnitude lower resolution in both angle and delay giving 10 to 30 times fewer estimated multipath components. This result demonstrates the importance of providing channel modelling which scales with both bandwidth and antenna array size. References Details on the measurements and evaluations can be found in Section A.1.2. mmmagic Public 8

29 2.1.4 Airport check-in area measurements at 15, 28, 60, and 86 GHz bands Name Environment Centre Frequency Multi-frequency measurement campaign in Helsinki airport Check-in area in terminal 2 of the airport 15 GHz 28.5 GHz 61 GHz 83.5 GHz Bandwidth 2 GHz 3 GHz 4 GHz 4 GHz Objective The objective of the measurements was to provide multi-frequency data for access channel in the airport, to derive large-scale parameters for the channel models, investigate their frequency dependence, and calibrate the ray-tracing simulation. Brief description The measurements were conducted at night when almost no or just a few passengers were present. The Rx antenna was placed on the second floor at the terminal s check-in area providing an additional 3-m height and the Tx antenna was moved on the check-in area level at 1.5 m height. The Rx antenna was rotated in the azimuth plane from 20 degrees to 160 degrees in steps of 5, and in the elevation plane with steps of 0 and -20 degrees. Both V/V and H/V Tx/Rx polarizations were measured for each link. In total, 11 links were measured, consisting of 5 LOS and 6 NLOS links. Evaluations Measurement scenario: check-in area in Helsinki airport Based on the directional measured data, PADPs and PDPs were derived. From these, LSP including path loss, DS, azimuth and elevation spreads, and XPR were derived for each of the measured bands. Specular and diffuse components as well as decay factors have also been extracted and compared among the measured frequencies. Main findings and impact on modelling The reported mean DS, ASA and ESA are fairly close to each other at different frequencies, but show a slight decreasing trend with increasing frequency. When considering specular propagation paths only, the difference between frequencies is less noteworthy. The specular decay factor [VJN+16] decreases with frequency. For the diffuse power decay, the slope does not change with frequency, but the power level decreases with increasing frequency [VJN+16]. The XPR values do not change in the measured frequency band. References Details on the measurements and evaluations can be found in Section A.1.3, [VJN+16]. mmmagic Public 9

30 2.1.5 Airport check-in area simulations at 15, 28, 60, and 86 GHz bands Name Environment Center Frequency Multi-frequency simulation in Helsinki airport Check-in area in terminal 2 of the airport 15 GHz 28 GHz 61 GHz 83.5 GHz Bandwidth 2 GHz 2 GHz 2 GHz 2 GHz Objective The objective of the simulation was to obtain a large number of realistic channel data in the airport access scenario to derive complete 3GPP-type channel models, and investigate their frequency dependence. Brief description Simulation scenario: Helsinki airport Based on channel measurements at 4 frequencies, permittivity and penetration loss values for different materials were modelled as a function of frequency (Section A.1.4). The considered propagation mechanisms included first and second order specular reflections. 4 BSs were placed at a height of 5.7 m, and MSs were placed around the terminal hall at a height of 1.6 m with 0.6 m spacing. In total, 5937 LOS links were computed. Evaluations Based on the simulated data at 4 different frequencies, large-scale parameters were derived. By fitting the mean LSP values over frequency, the complete 3GPP-type channel model tables were obtained with frequency-dependent large-scale parameters. The data from different base station locations was then combined. Main findings and impact on modelling The same dominant paths appear in all frequency bands. All spread values show a slightly decreasing trend with the frequency while the K-factor shows a clear increasing trend. The spreads also vary as a function of distance, but this trend is quite specific to the environment. For, e.g., DS, the values first increase as a function of distance, but after roughly 30 m, there is a decreasing trend. No apparent frequency dependency is observed in correlation distance and cross-correlation of LSPs, except for the correlation distances of ZSA and ZSD, which decrease with frequency. The difference between trends with different BS position is only marginal. References Details on the simulations and evaluations can be found in Section A.1.4. mmmagic Public 10

31 2.1.6 Ultra-wide band measurements at 7, 34 and 60 GHz in a small lecture room Name Environment Small scale fading measurements in indoor scenarios Indoor small lecture room Frequency 34 GHz 60 GHz Bandwidth 4 GHz 4 GHz Objective Mm-wave systems are supposed to operate with large absolute channel bandwidth and high antenna directivity. A high Rician K-factor increases the contribution of deterministic channel components, thereby reducing the significance of stochastic parts of a channel. These measurements focus on the analysis of stochastic or deterministic channel fading behaviour, as function of channel bandwidth and antenna directivity. Brief description The 34 GHz measurements focus on the 4 different propagation setups i.e. a LOS, reflections from black board and wall are considered due to their different surface roughness and double bounce reflections from both surfaces. 60 GHz measurements focus on the specular reflections from the wall with different directive antennas, thus demonstrating different beamforming gains. Evaluations The small-scale fading measurements are used to analyze the fading depth, K-factor and XPD scaling as a function of bandwidth and antenna directivity. Measurement scenario: A small lecture room at Ilmenau university of technology,germany Main findings and impact on modelling 3GPP-SCM, WINNER and COST2100 channel models are based on the uncorrelated scattering assumption and the amplitude fading envelop of the taps follow Rayleigh amplitude envelopes in NLOS environments. In contrast to the 3GPP-LTE systems, which consider narrow band digital pre- and post-processing techniques, the mm wave systems proposed so far consider hybrid (digital and analogue) wide band pre- and post-processing techniques. Uncorrelated scattering assumption does not remain valid for wideband systems and the amplitude fading envelop may not follow Rayleigh fading distribution. Additionally, in order to mitigate higher path loss effects, mm-wave systems are supposed to operate with higher beamforming gains. Higher bandwidth and directivity may significantly affect the channel modelling methodology, which is the primary focus of the investigation of these measurements. It has been observed that the small-scale fading depth asymptotically converges towards zero db whereby K-factor increases with bandwidth and antenna directivity. Coefficient of variation of XPD around its mean value also reduces exponentially with an increase in channel bandwidth. In contrast to 3GPP-SCM, WINNER and COST2100 GBSCM channel models, these results demonstrate that mm-wave channel fading behaviour becomes quasi-deterministic due to large available bandwidth and high directive antennas. References Details on the measurements and evaluations can be found in Section A.1.5 and 3.4 and [DNC17, NCJ17, NJS+17]. mmmagic Public 11

32 2.1.7 Measurement on spatial consistency Name Environment Frequency Bandwidth Objective Spatial Consistency measurement campaigns in Bristol Street canyon and Indoor measurements 60 GHz 2 GHz The objective of the measurement campaigns was to provide data to investigate the spatial evolution of the channel by using large-scale parameters in UMi and Indoor scenarios. The spatial consistency of the channel was investigated by performing correlation of some parameters. Brief description Measurement campaigns have been carried out using a Keysight Technologies wideband channel sounder. In both campaigns, Tx was fixed at a specific location and the mobile Rx was located at different test points along the chosen scenario. The Rx performed a 360 rotation in the azimuth plane for the Indoor scenario, and 360 and 90 rotation in the azimuth and elevation planes respectively, for UMi scenario. CIRs were collected every 1 in the azimuth plane and every 5 in the elevation plane. Three different sub-scenarios were investigated in each scenario: LOS, transition LOS NLOS and NLOS. (a) Evaluations To investigate spatial consistency, the PDP was computed at all measured angles for each test point. PDPs that contain the strongest multipath were chosen for comparison in each test point. In each subscenario, correlation coefficient of chosen PDPs for every test point was computed. Power angular spectrum was also analysed in every scenario and sub-scenario. (b) (a) Cantocks Close opposite Faculty of Engineering (MVB) of Uni of Bristol (b) Merchant Venturer s Building (MVB) University of Bristol Main findings and impact on modelling Cross-polarised received signal in horizontal polarisation has better performance than Cross-polarised received signal in vertical polarisation. Cross-correlation was extremely high for both indoor and outdoor measurement campaigns, for all sub-scenarios. Power angular spectrum gives better description of the environment measured. Indoor measurements are very reflective due to metal materials; strong reflections can be seen even in the NLOS sub-scenarios. Outdoor measurements do not show any significant reflective multipath components, but ground reflection can be seen clearly in LOS sub-scenarios. References Details on the measurements and evaluations can be found in Section A.2.4 and D2.1 [mmmagic-d2.1]. mmmagic Public 12

33 2.2 Urban micro outdoor scenarios Street canyon measurements at 15, 28, 60 and 86 GHz bands Name Environment Multi-frequency measurement campaign in Espoo (Greater Helsinki) Street canyon near Aalto university s campus Frequency GHz GHz 61 GHz 83.5 GHz Bandwidth 0.5 GHz 0.9 GHz 4 GHz 4 GHz Objective The objective of the presented measurements is to provide multi-frequency channel data in the street canyon, derive large-scale parameters, investigate their frequency dependence, and calibrate the ray-tracing simulation. Brief description The measurements were conducted in the street canyon between Otakaari 5 and Otakaari 7 without pedestrians. Both Tx and Rx antennas were placed at a height of 2.6 m. The Rx was fixed in one location, and rotated in the azimuth plane from 0 degrees to 360 degrees with 5-degree steps. The Tx antenna was moved along the street and around the corner. V/V and H/V Tx/Rx polarizations were measured for each link. In total, 11 links were measured (7 LOS and 4 NLOS). Evaluations Based on the directional measured data, LSP including path loss, DS, ASA, and XPR were derived for each of the measured bands and compared between the frequency bands. Measurement scenario: street canyon in Espoo Main findings and impact on modelling Generally, the PDPs at different frequencies show strong correlation with peaks appearing at the same delay and angles. The results obtained support the trend that DS and AS decrease with frequency. However, the dependency appears significantly weak, and further measurements are required in order to clarify this observation. Also, due to the waveguiding effect the delay and angular spreads are very small, which makes it more difficult to observe clear differences between frequencies. No clear frequency dependency trend has been observed in the azimuth spread. NLOS links were not available with good enough dynamic range for the analysis. References Details on the measurements and evaluations can be found in Section A.2.1, D2.1 [mmmagic-d2.1]. mmmagic Public 13

34 2.2.2 Street canyon simulations at 15, 28, 60 and 86 GHz bands Name Environment Centre Frequency Multi-frequency simulations in Espoo (Greater Helsinki) Street canyon near Aalto university s campus GHz GHz 61 GHz 83.5 GHz Bandwidth 2 GHz 2 GHz 2 GHz 2 GHz Objective The objective of the simulation was to obtain a large number of realistic channel data in the street canyon access scenario, to derive complete 3GPP-type channel models, and investigate their frequency dependence. Brief description Based on channel measurements at 4 frequencies (Section 2.2.1), permittivity and penetration loss values for different materials were optimized and modelled as a function of frequency (see Section A.2.2 for details). The considered propagation mechanisms included first and second order specular reflections. 1 BS was placed at the measured Rx position with a height of 5 m, and MSs were placed along the street at a height of 1.5 m with a 0.5 m spacing. In total, 860 LOS links were computed. Evaluations Simulation scenario: street canyon in Espoo Based on the multi-dimensional simulated data, the complete 3GPP-type channel model tables were obtained with frequency-dependent large-scale parameters. Main findings and impact on modelling The simulated channels are consistent with the same paths observed at different frequencies. The results show that both delay and angular spreads decrease slightly with frequency, whereas the standard deviation of the spread does not depend on frequency. Also the K-factor shows a clear increasing trend. Generally, cross-correlation parameters and correlation distances show no frequency dependency with the exception of the DS correlation distance, which is slightly decreasing. References Details on the simulations and evaluations can be found in Section A.2.2. mmmagic Public 14

35 2.2.3 Street canyon measurements at 2, 15 and 60 GHz bands Name Environment Outdoor street canyon multi-frequency propagation measurements Office Frequency 2.44 GHz 14.8 GHz GHz Bandwidth 80 MHz 200 MHz 2 GHz Objective The objective was to accurately characterize path loss and delay spread over a large frequency range. Brief description Outdoor measurements are made in an urban area consisting of mainly modern office building blocks of approximately 100 m length, 25 m height and 20 m street width. The measurements are based on using a VNA sampling the radio channel response in frequency domain. Delay domain data was obtained by FFT. Matched vertical dipole antennas (at 1.5 m height), with very similar patterns, have been used for all frequencies. Evaluations Careful LoS calibration at m distance was performed. For comparisons over frequency the bandwidth was equalized to 80 MHz. Propagation loss (loss in excess of free space loss), as well as delay spread were determined. Street canyon measurement scenario. Main findings and impact on modelling In LOS scenarios, a multipath gain of up to 5 db is observed which is similar at all frequencies. This gain is due to additional paths from reflections off the ground and exterior walls. In the NLOS region behind the corner of the building, a substantial increase in the excess loss is observed. This loss is substantially lower than what would be expected by knife-edge diffraction only. Moreover, it is found that the oxygen compensation at 60 GHz is substantial, up to 4 db, for the NLOS data, which is more than what is expected from the link distance only. However, this is explained by the fact that the lengths of significant reflected propagation paths are substantially larger than the link distance. In order to get some further insight into the propagation mechanisms, manual ray tracing has been performed for a measurement location at 60 GHz. At the delay corresponding to the diffraction path around the corner, no signal above the noise floor is observed. The first cluster of weak paths is observed at substantially longer propagation distances than the diffraction path length. This cluster is likely to be caused by scatterers and/or rough surfaces in the area of the street corner. The strongest peak stands out having around 20 db higher power level than the rest of the power delay profile. It was possible to match the delay of this path with a ray-traced path using four specular reflections off exterior building walls. This result shows that specular paths may be important even far into the NLOS region. However, for most of the NLOS locations such pronounced peaks were not observed. The measured propagation path loss in NLOS increases about 3 db per decade increase of frequency, which is about two times higher than the 3GPP model [3GPP38.900]. The measured delay spread is around 150 ns in NLOS without any clear frequency dependence. This contrasts with the 3GPP model, which gives substantially smaller delay spread values which decreases substantially with increasing frequency. References Details on the measurements and evaluations can be found in D2.1 [mmmagic-d2.1]. mmmagic Public 15

36 2.2.4 Open square measurements at 28 and 86 GHz bands Name Environment Centre Frequency Multi-frequency measurement campaign in Narinkkatori Open square in Centre of Helsinki GHz 83.5 GHz Bandwidth 0.9 GHz 4 GHz Objective The objective of the measurements was to provide multi-frequency channel data in the city open square, to derive large-scale parameters for the channel models, to investigate their frequency dependence, and calibrate the ray-tracing simulation. Brief description The measurements were conducted in the open square Narinkkatori during daytime. The Rx side was mounted at a height of 5 m, and the Tx antennas were placed at a height of 1.6 m. The Rx was fixed in one location, with the Rx antenna rotated in the azimuth plane from 0 degrees to 360 degrees with 5-degree steps. The Tx antenna was moved along the square for LOS and around corners for NLOS scenarios. V/V and H/V Tx/Rx polarizations were measured for each link. In total, 10 links were measured (5 LOS and 5 NLOS). Measurement scenario: open square in Helsinki Evaluations Based on the directional measured data, frequency dependency was analysed in terms of DS and ASA. Main findings and impact on modelling Comparing channel in the two bands, the peaks appear relatively at the same delays and angles in the PADP for the same Tx-Rx link, however the normalized individual path amplitudes appear to be lower at 83 GHz, resulting in lower DS and ASA as compared to those in 28 GHz. The average DSs are 46 ns and 37 ns in 28 and 86 GHz respectively. The average ASAs are 27 and 23 in 28 and 86 GHz, respectively. References Details on the measurements and evaluations can be found in Section A.2.1 and D2.1 [mmmagic-d2.1]. mmmagic Public 16

37 2.2.5 Open square simulation at 28 GHz band Name Environment Frequency 28-GHz simulation campaign in Narinkkatori Open square in Centre of Helsinki GHz Bandwidth - Objective The objective of the simulation was to provide 28-GHz channel data in the city open square and derive 3GPP-type large-scale parameters for the mmmagic channel models. Brief description The simulation was performed for an open city square. Based on channel measurements at 28-GHz band (Section 2.2.3), permittivity and penetration loss values for different materials and objects were estimated. The considered propagation mechanisms included first and second order specular reflections. The BS was placed on a lamppost with 5m-height. There were 1457 MSs placed in the square at 1.6m-height with 0.5m-separation, resulting in 709 links with clear LOS condition. Evaluations Based on the multi-dimensional simulated data, statistical analysis was performed for extracting path loss, shadow fading, DS, RMS angular spreads (ASD, ASA, ZSD, ZSA), XPR, the Ricean K-Factor, cross-correlation and correlation distances of the parameters for the LOS condition in 28-GHz band. Simulation scenario: open square in Helsinki Main findings and impact on modelling Both measurement and simulation show that the propagation channel in the open square is dominated by strong specular paths, i.e., sharp peaks in the power delay profile. The delay spread in this open environment is greater than that in the street canyon (Sections and 2.2.2) due to the presence of many strong single- and double-bounce reflections. Distance dependency observed in almost all LSPs. References Details on the simulation can be found in D2.1 [mmmagic-d2.1]. mmmagic Public 17

38 2.2.6 Multi-frequency measurements in Berlin at 10, 28, 41 and 82 GHz Name Environment Multi-frequency measurement campaign in Berlin Street canyon in the city centre of Berlin, Germany Frequency GHz 28.5 GHz 41.5 GHz 82.5 GHz Bandwidth 0.5 GHz 1.5 GHz 1.5 GHz 1.5 GHz Objective The objective of the measurements was to provide multi-frequency data for UMi access channel in the street canyon to derive large-scale parameters for the channel model and investigate their frequency dependence. Brief description The measurements were carried out with a modular multi-frequency wideband channel sounder setup. The four frequency bands were measured simultaneously. The Tx (height: 5 m) was fixed and the mobile Rx was moved along the sidewalk with a constant speed of 0.5 m/s. CIRs were collected every 2 ms, corresponding to 1 mm spacing, from 10 to 300 m distance on three LOS tracks. Additional measurements were performed with a static Rx and for NLOS. Evaluations Based on line-of-sight data, a statistical analysis was made for the RMS delay spread (DS), the path loss, shadow fading and the Ricean K-Factor. Measurement scenario: Friedrichstr. In Berlin Main findings and impact on modelling In an initial analysis, the data for the higher bands was reduced to the same bandwidth as used for GHz, namely 500 MHz. Taking into account data for a distance range from 10 to 110 m, it reveals that the distribution of the DS for the lower bands (10, 28, 41 GHz) is very similar for evaluation thresholds of 15 and 20 db. For a threshold of 25 db, there is a trend to slightly higher values for 10 GHz. A statistical analysis based on a reduced data set for the distance range m shows that the DS is tendentially lower at 82 GHz in comparison to the other frequencies. In a second analysis step, the multi-path components were extracted from the raw data. This reduced the influence of measurement noise on the evaluated parameters and allowed using the entire distance range for the data analysis. The trend towards a slightly higher delay spread at 10 GHz could be confirmed. At the same time, there was a clear trend towards higher K- Factors at the higher frequencies. When comparing the measurement results with the 3GPP model [3GPP38.900], path-loss and delay spread results agree reasonably well. However, the strong and clear dependence of the K-Factor on the frequency is not included in the model. References Details on the measurements and evaluations can be found in Section A.2.4 and D2.1 [mmmagic-d2.1]. mmmagic Public 18

39 2.2.7 Measurement and simulation campaign in Daejeon Name Environment Frequency Bandwidth Objective Measurement and simulation campaign in Daejeon UMi Street Canyon 28 GHz 250 MHZ The objectives of the measurement campaign were to study intra-cluster characteristics in millimetre-wave channels and extract intra-cluster parameters for the final mmmagic channel model. Brief description An experiment including both ray tracing simulation and measurement was performed in Daejeon, Korea. Streets and buildings are the main components in the environment. The distance between the transmitter and receiver is up to 200 m. The experiment environment is categorized as urban micro (UMi) street canyon at 28 GHz. The length of a single street is in the order of 100 m. In addition, street furniture such as lampposts, traffic signs, and trees are typically seen in the environment. It should be noticed that the ray tracing simulation only considers reflections, penetrations, and diffractions. Both line-of-sight (LOS) and non-los (NLOS) were simulated via ray tracing. However, only NLOS scenario was measured. UMi street canyon scenario: Daejeon, Korea Evaluations The channel samples were clustered using the Kmeans++ algorithm and then analysed using standard procedures of calculating intra-cluster delay spreads and angular spreads. Results of measurement and ray tracing were compared mutually. Main findings and impact on modelling This campaign extracted intra-cluster parameters for a NLOS UMi street canyon scenario. It was found that the number of rays within a cluster follows a negative binomial distribution. Furthermore, single slope exponential intra-cluster delay spread should be sufficient. Lognormal distributions can be used for modelling arrival angle spreads. The distributions as well as their parameters can be applied to intra-cluster modelling in the final mmmagic channel model. References Details on the measurements and evaluations can be found in Section A.2.5. mmmagic Public 19

40 2.2.8 Ray-tracing simulation in Madrid-grid and Aalborg city layouts Name Environment Ray-tracing simulation in Madrid-grid layout and Aalborg city layout UMi Street Canyon Frequency 2GHz 5.6GHz 10GHz 28GHz 39GHz 73GHz Bandwidth Objective The objectives of the simulation were to study frequency dependency of LSPs for the same type of environment (UMi) but with different layout (cities) to verify environment impact for LSPs. Brief description The ray tracing simulation was performed in Urban Micro environment in two layouts: Madrid-grid, which is simple regular environment and Aalborg city layout which is realistic environment. WinProp ver.13 (Altair) was used for simulation with Standard Ray Tracing model. The same material types were used in both layouts (concrete material for walls and ground, parameters from ITU-R P.2040). The reflection and diffractions were considered only. Diffuse Scattering and penetrations (O2I) were disabled. The same isotropic antennas with the same heights were used (TX 5m and RX 1.5m). All ray-tracing parameters were identical in both scenarios. Five transmitters were located in different places in both scenarios. Madrid-grid layout Evaluations Based on simulation results statistical analysis was made for the path loss, shadow fading, RMS delay spread (DS), RMS angular spreads (ASD, ASA, ZSD, ZSA), XPR and the Ricean K-Factor for LOS and NLOS conditions. The part of Aalborg city layout Main findings and impact on modelling In the results we did not observe clear environment type (regular, not regular-realistic) impacts on LSPs. A certain frequency dependence is observed in the NLOS case for the number of interaction types, Shadow Factor, RMS Delay Spread, RMS Azimuth Angular Spread of Departure, RMS Azimuth Angular Spread of Arrival, RMS Zenith Angular Spread of Departure, RMS Zenith Angular Spread of Arrival. References Details on the simulations and evaluations can be found in Section A.2.6. mmmagic Public 20

41 2.2.9 Open square measurements at 17 GHz Name Environment Frequency Bandwidth Objective 17 GHz Open Square Measurement in Munich by Rohde & Schwarz Street Canyon and Open Square in Munich 17 GHz 160 MHz The objective of the measurement was to provide measurement data for channel modelling at 17 GHz for the UMi access channel. Brief description The measurements were carried out with the Rohde & Schwarz Channel Sounder Setup. Generator and Receiver are synchronized using GPS with additional time-alignment in post processing. The transmitter was positioned outside the fifth floor of an office building with Rx at 2 m height at different locations. LOS, OLOS and NLOS scenarios were measured. Evaluations Based on the measurements and reference data, the channel sounder setup was evaluated. A statistical analysis for the path loss and RMS delay spread was made. Vivaldi antenna at the receiver Main findings and impact on modelling The measurement evaluation focused on path loss and angular spread of individual paths that are received with omnidirectional and directional antennas. Besides this, the measurements posed as evaluation for further measurements to be conducted with this channel sounder setup. The RMS DS is found to be higher for measurements with an omnidirectional antenna, as more paths from different angles are received. References Details on the measurements and evaluations can be found in A.2.7. mmmagic Public 21

42 Outdoor measurements at 3, 17 and 60 GHz Name Environment Millimetre-Wave Outdoor measurements by Orange in Belfort O2O scenario in the industrial activity area Techn hom of Belfort Frequency 3 GHz 17 GHz 60 GHz Bandwidth 125 MHz 125 MHz 125 MHz Objective The objective of this measurement campaign was to evaluate the frequency dependence of large scale parameters (LSPs), especially the channel delay spread (DS). Brief description The measurement campaign was conducted using a wideband channel sounder. Three transmission points (TPs), i.e., TxA, TxB and TxC, were considered and a set of LoS and NLoS measurements were performed in each case. The distances between the TX antenna, placed on top a van at 2.5 m above ground, and the 35 considered RX positions ranged from 16 to 200 m. The RX antenna was kept at mobile user level (1.5 m high) and the RX positions included A1 to A16, B1 to B11 and C1 to C8 with the TX placed at the TPs TxA, TxB and TxC respectively. Both directional and omnidirectional antennas were used at each side to compute, and synthesize when necessary, omnidirectional channel characteristics such as PDPs. Evaluations The channel DS values were determined for each frequency band at the different RX positions. The values were computed by setting a 20 db threshold on the dynamic range of the computed PDPs. O2O Measurement in the industrial activity area Techn hom in Belfort. Main findings and impact on modelling It appears that the DS is statistically hardly dependent on the frequency. A 20% decrease on the average values is recorded from 3 and 17 to 60 GHz when the TX is placed at TxA or TxB (open square environment). An increasing behaviour is observed when the TX is placed at TxC (street canyon scenario). Overall, the values remain quite low and less correlated with the frequency in comparison to the 3GPP model results. This can be explained by the much less urban environment in which the measurement campaign was carried out compared to the 3GPP UMi scenario. References Details on the measurements and evaluations can be found in Section A.2.8. mmmagic Public 22

43 2.3 Outdoor-to-indoor Scenarios O2I measurement campaigns at 3, 10, 17 and 60 GHz Name Environment Millimetre-Wave measurements in Belfort O2I scenario in Orange Labs premises Frequency 3 GHz 10 GHz 17 GHz 60 GHz Bandwidth 125 MHz 125 MHz 125 MHz 125 MHz Objective The objectives of the measurement campaign were to give an estimation of the building penetration losses and the channel delay spreads in the different frequency bands for both coated and non-coated glass windows and also to assess the frequency dependence of these two parameters. Brief description The measurement campaign was conducted using a wideband channel sounder. The distance between the TX, placed outside at 2.5 m above ground level, and the building exterior walls was about 10 m. The RX antenna was kept at mobile user level (1.5 m high) and various rooms inside the building, with coated or non-coated glass windows, were considered for the RX positions. The TX-RX separation distances ranged from 10 to 25 m. Both directional and omnidirectional antennas were used at each side to compute, and synthesize when necessary, omnidirectional channel PDPs. Evaluations The building entry losses for the different window glass material as well as the penetration losses further into the building were derived for each frequency band. Their corresponding delay spread values were also provided. O2I measurement scenario: Orange labs, Belfort. Main findings and impact on modelling It was found that the penetration losses are strongly dependent on the widow material composition. For non-coated glass windows, the attenuation values are relatively low (less than 10 db) while windows with silver coated glass cause significantly higher attenuation (around 30 db). Regarding the frequency dependence, an increasing tendency of the attenuation with this parameter is observed for coated glass windows. However, for standard windows without silver coating, the attenuation values are almost constant irrespective of the frequency. The delay spread values, computed with a minimum of 20 db dynamic range, were found to be quite low (below 30 ns) and more or less uniformly distributed across the different frequency bands with a very slight decrease at 60 GHz. References Details on the measurements and evaluations can be found in Section A.3.1. mmmagic Public 23

44 2.3.2 O2I measurement campaigns at 2, 6, 15, and 60 GHz Name Environment Multi frequency outdoor to indoor measurements in Stockholm O2I scenario in Ericsson Research Office premises Frequency 2.44 GHz 5.8 GHz 14.8 GHz GHz Bandwidth 80 MHz 150 MHz 200 MHz 2 GHz Objective The objectives of the measurement campaign were to give an estimation of the building penetration losses and the channel delay spreads in the different frequency bands for non-coated glass windows and also to assess the frequencydependence of these two parameters. Brief description An outdoor-to-indoor multi-frequency measurement campaign has been performed in an eight stores tall office building in Kista, Stockholm. The measurements are based on a vector network analyser using frequency sweep. Delay domain data was obtained by FFT. Matched vertical dipole antennas (at 1.5 m height), with very similar patterns, were used for all frequencies. Evaluations Careful line of sight calibration at m distance was performed. For comparisons over frequency the bandwidth was equalized to 80 MHz. Propagation loss (loss in excess of free space loss) was determined as well as delay spread. O2I measurement Scenario. Main findings and impact on modelling Between 2.44 GHz and 14.8 GHz, the building penetration loss ranges from around 0 db up to 30 db. The lower end of penetration loss around 0 db is similar for all frequencies while the highest losses around 45 db occurs only at GHz. The minimum loss, due to penetration of the exterior wall/window only, is in the range 0-5 db with the highest values for 5.8 GHz and GHz. This non-monotonic dependence on frequency may be explained assuming that the three layers of glass, in the window frames, cause constructive or destructive interference, as an effect of multiple reflections, resulting in periodic varying attenuation as function of frequency. Subsequent measurements show that the window loss is about 2, 10, 0, and, 6 db at 2.44, 5.8, 14.8 and GHz, respectively, which confirms this effect and explains the measured minimum penetration loss. Moreover, it is clear that the spread of penetration loss is substantially larger for the higher frequencies. This may partly be the result of the venetian blinds, in some of the windows, which block the vertically polarized waves at the higher frequencies but are transparent at the lower frequencies. The measured building penetration loss agrees very well with corresponding 3GPP model. The RMS delay spread, however, increases with frequency from 50 ns to 100 ns, which is in contrast to the 3GPP model showing constant delay spread over frequency around 250 ns. References Details on the measurements and evaluations can be found in D2.1 [mmmagic-d2.1]. mmmagic Public 24

45 2.3.3 O2I measurement campaign at 2, 6, and 60 GHz Name Environment Multi frequency outdoor to indoor measurements in Stockholm O2I scenario in Ericsson Research Office premises Frequency 2.44 GHz 5.8 GHz GHz Bandwidth 80 MHz 150 MHz 2 GHz Objective The objectives of the measurement campaign were to give an estimation of the elevation angle dependence of building penetration losses. Brief description An outdoor-to-indoor multi-frequency measurement campaign has been performed in an eight stores tall office building in Kista, Stockholm. The measurements are based on using a VNA sampling the radio channel response in frequency domain. Delay domain data is obtained by FFT. Matched vertical and horizontal dipole antennas (at 1.5 m height), with very similar patterns, were used for all frequencies. Evaluations Careful line of sight calibration at m distance was performed. Propagation loss (loss in excess of free space loss) is determined for all measurement locations. O2I Measurement Scenario Main findings and impact on modelling The measurements are calibrated by free space measurements accounting for any antenna pattern effects at Rx due to the elevation angle. The antenna gain elevation angle variation was found to be within 1 db. It was found that the elevation dependent loss L el may be modelled with a simple linear expression L el = k abs (θ) (db) where k (db/ 90º) is a model parameter and θ is the elevation angle. This function has been fitted to the measurement data assuming that k is frequency independent. The model parameter k ranges from 24 db/90º to 31 db/90º between 2.5% to 97.5% of the CDFs showing that the suggested model has only little sensitivity to the high or low loss end of the distribution. The standard deviation between model and measurements is in the range 1.35 to 2.22 db. References Details on the measurements and evaluations can be found in Section A.3.2. mmmagic Public 25

46 2.3.4 O2I measurement campaign at 17 GHz Name Environment Frequency Bandwidth Objective 17 GHz Outdoor to Indoor Measurement in Munich by Rohde & Schwarz Outdoor to Indoor in Street Canyon in Munich 17 GHz 160 MHz The objective of the measurement was to provide measurement data for channel modelling at 17 GHz for Outdoor-to-Indoor environments. Brief description The measurements were carried out with the Rohde & Schwarz Channel Sounder Setup. Generator and Receiver are synchronized using GPS with additional time-alignment in post processing. The transmitter was positioned outside of the fifth floor of an office building with the receiver placed inside a ground floor office room. Both the LOS and NLOS O2I scenarios were measured. The measurements were conducted in the same campaign as the OS measurements as described in Section Evaluations Based on the measurements and reference data, the channel sounder setup was evaluated. A statistical analysis for the path loss and RMS delay spread was made. Aerial view of measurement scenario with Tx position and Rx positions. Main findings and impact on modelling Outdoor to indoor transmissions at high frequencies feature extreme path loss depending on the material. With metallised glass as used in many modern buildings, it must be expected that the path loss easily exceeds the available link budget. With normal glass windows, LOS transmissions can be established only with minimal additional loss. NLOS scenarios experience bigger RMS DS with very high overall path loss. References Details on the measurements and evaluations can be found in A.3.2. mmmagic Public 26

47 2.4 Scattering, ground reflection and blockage Specular reflections and diffused scattering Name Environment Frequency Bandwidth Objective Brief description Specular wall reflections and diffused scattering measurements Outdoor Environment 60 GHz 2 GHz To investigate the distribution of specular and diffused multipath components, resulting from wall scattering measurements The measurement campaign was conducted using a wideband time-domain channel sounder, with the capability to measure both the vertical and horizontal signal components at the receiver side. Towards this end, a circular horn antenna has been employed at the RX end, when the TX was equipped with a high-directional vertical polarized antenna. For the evaluation of the specular components, the distance between the TX and the receiver was at 1 m, whereas for the diffused scattering various distances have been investigated, ranging from 2 to 6 m. Evaluations Based on the experimental data, well-known probability distributions have been employed to characterize the distribution of the specular components for both smooth and rough wall surfaces. Furthermore, for the case of diffused scattering the power angular scattering profiles have been characterized with respect to the power concentration around the specular direction. Finally, depolarization properties have been studied for both specular and diffused wave components. Main findings and impact on modelling References (a) TX 45 RX 20 RX 10 (b) Investigation of Specular (a), and diffused (b) scattering resulting from wall reflections. Angular spread of power reduces with angle (for similar separation) and distance (for similar angles), for both smooth and rough walls. It was found that the smoother the surface the less the angular spread and furthermore, at large distances, diffused scattering is insignificant for smooth surfaces. Depolarization (XPD value) was found higher in rough surfaces; however no significant angle dependency has been observed. The Nakagami-m distribution can adequately describe the distribution of specular components for both rough and smooth wall surfaces. Details on the measurement equipment and scenarios can be found in Section A.4.1 and Section A.4.2. RX 30 RX 40 RX 50 RX 80 RX 70 RX 60 mmmagic Public 27

48 2.4.2 Ground reflection in a courtyard Name Environment Ground Reflection Evaluation Measurement in Munich Street Canyon, Open Square Frequency 28 GHz 39 GHz Bandwidth 2 GHz 2 GHz Objective The objectives of the measurement campaign were to evaluate the influence of ground reflection to small changes in distance between transmitter and receiver. Especially, the two-ray interference at distances below the crossover distance was investigated. Brief description The measurements are based on signal generator and spectrum analyser using a high sounding signal bandwidth. The same antennas were used for both frequencies resulting in similar antenna patterns. Evaluations The setup is calibrated before starting the measurements. The distances between transmitter and receiver are measured using laser ranging. All measurements use the same bandwidth and measurement points to the best knowledge. Measurement scenario grassland. Main findings and impact on modelling Ground reflection plays a significant role in small scale fading at the studied frequencies. The effects of two-ray reflection, most commonly ground reflection, have to be modelled accordingly. However, the ground material has a strong impact on the severity of ground reflections. It is derived from the measurements, that ground reflection has a strong impact on a concrete surface while the impact is negligible on grassland. Therefore, the modelling should take the predominant surface material into account and apply a ground reflection extension depending on the environment. References Details on the measurements and evaluations can be found in A.4.3. mmmagic Public 28

49 2.4.3 Ground reflection in a street canyon Name Environment Street Guided LOS measurements in Belfort Pedestrian area in an industrial area Frequency GHz GHz GHz Bandwidth 125 MHz 125 MHz 125 MHz Objective The objective of this outdoor measurement campaign, carried at 3, 17 and 60 GHz, was to evaluate the impact of the ground/wall reflection on the received signal strength in LOS scenario The transmission point is denoted BS and was placed on top of a van at 2.5 m above ground level. The RX antenna is kept at mobile user level at 1.5 m height and is set in a trolley pushed by an operator along a linear trajectory. The operator is either behind or in front of the Rx antenna leading to a LOS or obstructed measurement scenario. Omnidirectional antennas or sectorial antennas with vertical polarization were used at each end link. The distance between Rx and Tx ranged from 30 m to 160 m. An impulse response is measured every 200 ms corresponding to a distance around 5 cm between two successive impulse responses Evaluations The path loss is calculated from the impulse response, plotted as function of Tx-Rx distance and compared to the free space loss. Environment measurement in the industrial area Techn hom in Belfort. Main findings and impact on modelling A LOS street guided environment with a non-obstructed Tx-RX line and regular ground floor cannot be modelled as a free space environment. The channel gain difference with the free space loss ranged roughly between -10 db and +5 db. These variations are due to the interferences between the direct path, the ground reflection and other reflection on the vertical building walls. References Details on the measurements and evaluations can be found in Section A.4.3. mmmagic Public 29

50 2.4.4 Blockage measurements in indoor office with phantom Name Environment Indoor channel measurement campaigns: blockage measurements Office room Frequency 62 GHz (V band) 83.5 GHz (E band) Bandwidth 6 GHz 6 GHz Objective These measurements aim to determine the shadowing loss due to the presence of a phantom between the transceivers. Brief description The measurement campaign has been carried out at CEA- LETI, Grenoble, in a typical indoor office room. The measurement setup is the same as that described in Section 2.1.1, except that a phantom emulating a human body was located between transceivers in OLOS scenario in order to determine the blockage. At each frequency, the measurements are carried out with and without the phantom. Evaluations From directional antennas measurements data, we derive the synthetic omnidirectional channel function. A grid of 6 measurement points is considered in GHz (black dots in the illustration on the right) whereas 15 measurement points are considered in GHz (black and red dots). Only the 6 positions (indicated with black dots) are considered for the sake of comparison in the considered bands. The received power is computed with and without the phantom (between the transceivers) and the shadowing loss is determined. Main findings and impact on modelling The presence of a human phantom located between transceivers in the office causes path gain losses varying from db to (resp db to db) at 62 GHz (resp GHz), depending on the exact location of the phantom. Details on the result for each position can be found in Section A.4.4. Moreover, validation of the blockage measurements is provided in Section References Details on the measurements and evaluations can be found in Section A.1.1 and D2.1 [mmmagic-d2.1]. mmmagic Public 30

51 3 Characterization of modelling components and related investigations 3.1 Frequency dependency of large-scale parameters Several measurement and simulation campaigns have been conducted to study the frequency dependency of large scale parameters (LSPs). The LSPs are delay spread, shadow fading, K- factor, and azimuth and elevation spreads of departure and arrival. In this section we summarize findings on this topic based on literature survey and measurements and simulations done in this project. The work has focused on the frequency dependency of the mean and standard deviations of the LSPs. The comparability across different frequencies of one measurement campaign has to be ensured. As measurements at different frequencies require specialized hardware which in turn have a huge impact on measurement parameters such as measurement bandwidth, dynamic range, antenna patterns, etc further processing effort might be necessary to harmonize these measurement parameters. The list of requirements for a reasonable comparison is given in Table 3.1. It has been found that if any of the above given requirements is not fulfilled, fictitious frequency trends may be observed. Table 3.1: Defined requirements in order to ensure comparability across different frequencies Requirements (must be fulfilled) Equal measurement bandwidth Equal antenna pattern, either physical or synthesized Equal dynamic range for analysis both in delay and angle domains Equal angle resolution (e.g., array size equal in terms of number of lambda) Same environment and same antenna locations Other requirements (relatively small effects or not applicable to all results) Compensation of atmospheric absorption at the 60 GHz-band Sufficient sample size Static environment (when measurements are made successively) Same path estimation algorithms Same area of spatial averaging Literature review Frequency dependency of LSPs has been investigated back in 1990s. Table C.1 in Annex C summaries the results from previous works on frequency dependence of LSPs (mostly of the delay spread) for different indoor and outdoor environments, and for frequency bands ranging from 900 MHz to up to 70 GHz. The results from Table C.1 are mixed: the frequency dependency of the delay spread (DS) can be seen in many measurements [BMS89] [DBB+91] [JSB96] [NH97] [YSH05] [YYD+16] [P05] [RBM+12] [WPK+15] while the others: [D95] [DK95] [AZC99] [LR03] [AR04] [JHK+06] [BSK+07] [HJK+15] [KPS+15] [RCS16] [PLL+16] [PWU+16] support frequency independence. If the frequency dependence of the DS is observed in an environment, mmmagic Public 31

52 then the trend is that the DS is smaller at the higher frequency band (except result in [KPS+15]). The average difference in the DS between frequency bands are small in many cases, and is only pronounced when the difference between two compared centre frequencies are large, e.g GHz and 58 GHz. The DS, and hence the frequency dependence of the DS, is very sensitive to the threshold applied in the calculation (e.g., delay spreads decrease with a decrease of the threshold) [AZC99]. It should be noted that most of these results do not fulfil all the requirements for comparable results listed in Table 3.1. The studies that do fulfil the main requirements, i.e., [BMS89] [AZC99] [JHK+06] [KPS+15] [PLL+16] [PWU+16], show no clear frequency dependency or only slightly decreasing trend. The relatively small number of previous works that fulfil the list of requirements clearly demonstrates the need for more measurements. In the recent 3GPP NR channel model [3GPP38.900], first and second-order statistics of most LSPs for above 6-GHz bands are modelled as functions of carrier frequency. For example, the mean of DS in LOS condition is modelled as for UMi Street Canyon and for Indoor Office. μ lgds = 0.24log 10 (1 + f c ) 7.14 (3-1) μ lgds = 0.01log 10 (1 + f c ) 7.79 (3-2) It is noted that depending on the parameter and the type of the environment (outdoor or indoor) in [3GPP38.901], the frequency dependency is strong in some LSPs, while in other LSPs is rather weak or not modelled at all (i.e., the mean and variance values of LSPs are constant over the considered frequency range) mmmagic results The summary of the multi-frequency measurement campaigns conducted within the project to study the frequency dependency of channel s LSPs and the obtained results are shown in Table 3.2 for indoor and O2I scenarios, and in Table 3.3 for outdoor scenarios. Table 3.2: Indoor and O2I multi-frequency measurement campaigns in mmmagic Company/ Institution Environment Frequencies Reference Requirements Result CEA Indoor office and conference room 62 and 83.5 GHz Sec Different antennas used but omnidirectional patterns were synthesized in post-processing No O 2 compensation at 60 GHz Slight variation of DS with frequency. DS almost constant for bandwidth larger than 500 MHz. Intra-clusters rms delay and angular spreads comparable in both bands. EAB Indoor 5.8, 14.8 and 58.7 GHz Sec in D2.1 Fulfilled, Sample size=2 No trend in angle spread or DS EAB Indoor 2.4, 5.8, 14.8 and 58.7 GHz Sec All fulfilled No dependence of DS on frequency for NLOS. DS decreases mmmagic Public 32

53 slightly with increasing frequency for LOS. EAB Indoor 2.4, 5.8, 14.8 and 58.7 GHz Aalto Airport 15, 28, 60, and 86 GHz HWDU Small lecture room 7 and 34 GHz EAB O2I 2.4, 5.8, 14.8 and 58.7 GHz Orange O2I 3, 10, 17 and 60 GHz Sec in D2.1 Sec All fulfilled Fulfilled Different bandwidths in the measurements, but 2 GHz was used in post-processing. Sample size < 10 Loss increases with frequency. The reported mean parameters are fairly close to each other. The overall trend is that DS in LOS decreases with frequency. Sec. 0 Fulfilled no clear trend in rms DS Sec Sec All fulfilled Different RX antennas were used in the measurements but omnidirectional antenna patterns were synthesized each time in order to compare the results. Loss & DS increases with frequency DS weakly correlated with the frequency but a very slight decrease is observed at 60 GHz Table 3.3: Outdoor multi-frequency measurement campaigns in mmmagic Company/ Institution Environment Frequencies Reference Requirements Result HHI Street canyon 10.25, 28.5, 41.5 and 82.5 GHz Sec All fulfilled The DS tends to decrease with frequency, thought the trend is not very distinct and depends on the evaluation procedure. The K-factor increases with frequency. EAB Street canyon 2.4, 14.8 and 58.7 GHz Sec All fulfilled Loss increases with frequency. No trend in DS mmmagic Public 33

54 Aalto Street canyon 15, 28, 60 and 86 GHz Sec Fulfilled, Sample size < 10, no compensation of atmospheric absorption at 60 GHz DS in LOS decreases with frequency Aalto Open square 28 and 83 GHz Sec Fulfilled, sample size < 10 Both DS and ASA in 28 GHz are lower as compared to those in in 86 GHz Orange Street canyon 3, 17 and 60 GHz Sec Differences in antenna beamwidths. The DS displays an increasing behaviour with the frequency, especially from 3 to 17 GHz. Open square 3, 17 and 60 GHz Sec Different RX antennas were used in the measurements but omnidirectional antenna patterns were synthesized each time in order to compare the results. The DS slightly decreases with the frequency The measured DS (in log10 scale) values in different frequencies from all campaigns are combined and plotted in Figure 3.1 and Figure 3.2 for indoor, O2I, and outdoor scenarios, respectively. The solid lines in the figures show the linear fits of the measured data to the 3GPPP-like frequency-dependent model μ lgds = αlog 10 (1 + f c ) + β. (3-3) In each campaign, the same dynamic range (20 or 25 db, depending on the campaign) was used to calculate the DS values for different measured frequencies to have a reasonable comparison (see Annex A for the details of the measurement campaigns and the evaluation of the measured data). Table 3.4 presents the model parameters for each campaign from the linear regression and the accompanied 95% confidence bounds of the slopes and the p-values for the hypothesis that there is a frequency trend (α 0). The rows highlighted in red are the ones where the frequency trend is statistically significant with 95% confidence level, i.e., the p-value is small enough (smaller than 0.05) to reject the null hypothesis that The DS has no increasing or decreasing relationship with the frequency. mmmagic Public 34

55 Figure 3.1: Delay spread versus frequency (logarithmic units) in different indoor (left) and O2I (right) multi-frequency measurement campaigns. The solid lines show the linear fits. Figure 3.2: Delay spread versus frequency (logarithmic units) in different outdoor LOS (left) and NLOS (right) multi-frequency measurement campaigns. The solid lines show the linear fits. Indoor O2I Outdoor Table 3.4: Frequency-dependent linear regression model for DS in each campaign LOS Model: μ lgds = α log 10 (1 + f c ) + β Scenario α (95% confident bounds) β p-value CEA Office (-1.17, -0.78) CEA Conference Room (-0.46, -0.03) HWDU Lecture Room (-0.08, -0.03) Aalto Airport LOS (-0.47, 0.11) EAB Office (-0.27, 0.12) NLOS EAB Office (-0.06, 0.04) Orange Low-loss (-0.29, 0.07) Orange High-loss (-0.14, 0.04) EAB Traditional building 0.27 (0.16, 0.39) HHI Street Canyon ( ) Aalto Open Square (-2.00, 1.53) LOS Aalto Street Canyon (-0.50, 0.26) Orange Open Square (-0.32, 0.22) Orange Street Canyon 0.10 (-0.28, 0.49) Orange Open Square (-0.20, -0.02) NLOS Orange Street Canyon 0.13 (-0.20, 0.47) EAB Street Canyon 0.01 (-0.09, 0.12) mmmagic Public 35

56 As can be seen from Table 3.4, different campaigns show different trends of frequency dependency with different levels of statistical significance. There is no obvious and clear trend amongst the different campaigns. It should be noted that though the measurement data is extensive, only a few measurement environments were investigated. The corresponding observed frequency trends may very well be due to the specific characteristics of each scenario and environment. To assess more general trends similar scenarios (indoor office, street canyon, O2I) have been combined separately for LOS or NLOS conditions. The corresponding model parameters are obtained by weighted averaging which for the slope α is given by: α = N n=1 N i=1 w i w iα i, (3-4) 2 where w i = 1 σ i, σ 2 i is the variance of the estimated α i in the linear regression for the i-th campaign. The 95% confident bounds of the combined slope are calculated as [α 1.96σ, α σ], (3-5) where the variance of the combined model is: σ 2 N 1 = ( i=1 ) 1. (3-6) The same calculation is used for weighted intercept point β. The model parameters for each scenario are presented in Table 3.5. Table 3.5: Frequency-dependent linear regression model for DS in each combined scenario Scenario σ i 2 Model: μ lgds = α log 10 (1 + f c ) + β Weighted α (95% confident bounds) Weighted β Indoor office LOS* (-0.07, ) NLOS (-0.06, 0.04) Indoor Airport** LOS (-0.47, 0.11) O2I 0.05 (-0.02, 0.11) Street Canyon LOS (-0.19, -0.02) NLOS 0.02 (-0.07, 0.11) Open Square** LOS (-0.34, 0.23) NLOS (-0.20, -0.02) * The CEA measurements are not included in the combined fit due to the limited relative frequency range. ** This scenario is not included in the current 3GPP model. Table 3.5 shows that the overall trend, with 95% statistical confidence level, is that the DS decreases slightly for Indoor LOS, Street canyon LOS, and Open Square NLOS. For the other scenarios, any frequency trend is upper bounded by the confidence range. Corresponding maximum possible slope values are -0.06, -0.47, 0.11, 0.11, for Indoor Office NLOS, Indoor Airport LOS, O2I, Street Canyon NLOS and Open Square LOS, respectively. It may thus be concluded that only small frequency trends are allowed by the confidence intervals except for mmmagic Public 36

57 the Indoor Airport LOS and Open Square LOS scenario. The main conclusion is therefore that any frequency trend of the mmmagic measurement campaigns is small in general. In Figure 3.3 the corresponding slopes α are compared with the 3GPP model [3GPP38.901]. There is clear disagreement except for the O2I scenario. Generally, 3GPP overestimates the slope with a negative value. Figure 3.3: Fitted α and corresponding confidence intervals (95%) for the different measurement campaigns. To study frequency dependency not only of DS but also of other LSPs including angular spreads, shadow fading, K-factor and their cross-correlation and correlation distances, multifrequency multi-dimensional channel data were obtained from simulations, which were calibrated using the data from measurements, in various campaigns. The results indicate some frequency trends in delay and angle spreads, shadow fading and K-factor in both airport [Section A.1.4] and street canyon [Sections A.2.2 and A.2.6] scenarios. However, for time reasons an analysis assessing the statistical significance of the trends could not be performed. Therefore, any frequency dependency of angle spreads, shadow fading and K-factor are only indicative, which motivates further analyses in the future. The unified model parameters are given in Section 4, where channel data from all measurement and simulation campaigns have been combined for indoor, outdoor, and O2I environments. 3.2 Ground reflection Background and motivation In [mmmagic-d2.1], modelling of the ground reflection has been identified as being important to properly reflect the propagation characteristics at mm-wave frequencies in UMi scenarios. As the name implies, the ground reflection (GR) is a deterministic multipath component (MPC) that can be received by a mobile terminal (MT) which is in direct line of sight (LOS) to the base station (BS). The electromagnetic properties of the ground and the small angle of incidence usually lead to a significant part of the energy being reflected. Due to the different lengths of the two paths, they interfere with each other, which results in a deterministic fading pattern. At frequencies below 6 GHz, this fading occurs only close to the BS and the distance between successive fades can be up to several dozen meters. Hence, GR fading was counted as shadow mmmagic Public 37

58 fading (SF) in past measurements and models. However, at mm-wave frequencies this effect is critical. The distance between fades can fall below 1 m and the signal strength might suddenly drop by up to 20 db. Communication systems operating at these frequencies will also likely use narrow beams directed towards the user, but the delay and angle differences between the two components become very small for low BS heights. For a BS height of 5 m and an MT height of 1.5 m, the separation is only around 0.5 ns and 1.7 given a BS-MT distance of 100 m, and the values further decrease when the BS-MT distance is increased. Therefore, the fades may affect several GHz of bandwidth, and in general, it is hardly possible to suppress them by beamforming. This has been confirmed by measurements at 60 GHz, where severe GR fading occurred in all measured scenarios [WPK+15a], [PWK+15]. The current approach is to model this effect by a dual-slope path loss (PL) model [3GPP38.900]. At close distances between the BS and the MT, the PL is similar to the free-space loss and GR fading is approximated by SF. However, after a certain distance, the GR interferes destructively with the direct path, which leads to an increased PL. The transition point between the two slopes, the so-called break point (BP), depends on the BS and MT heights and on the carrier frequency. The higher the carrier frequency is, the further away is the BP. At mm-wave frequencies, the BP distance is several hundred meters from the BS which means that urban-microcell (UMi) deployments with typical cell sizes below 200 m will have to cope with the fading effects caused by the GR. The reason why this has not been addressed so far might be two-fold. First, many mm-wave measurements were done in static conditions where the MT did not move. However, GR fading depends on the receiver position and can therefore only be identified with mobile devices. Secondly, most GSCMs do not consider mobility. MTs are merely dropped into a propagation scenario and their instantaneous channels are used to predict the performance. However, 5G mm-wave communication systems will have to adjust very quickly to changing channels including the sometimes severe GR fading. Hence, mobility can no longer be neglected Fresnel reflection for ground materials The reflected amount of energy depends on the material properties, the incident angle and the polarization of the wave and frequency. For a plane wave that propagates in air and impinges on a plane surface, it is given by the Fresnel reflection coefficients as defined by (4-54). Figure 3.4 shows the reflection behaviour for different ground conditions at 28 GHz, using the permittivity values from [ITUR-P527] as summarized in Table 4.4. In UMi propagation scenarios with low BS heights, the angle of incidence with respect to the normal of the surface is typically between 70 and 90. In this angle range, the angle dependence is stronger for vertical polarization, as can be seen in Figure 3.4. Moreover, very dry ground yields the strongest reflections and wet ground the lowest. For horizontal polarization, the order is reversed. In the relevant angular range, however, the ground condition does not have a massive influence on the reflection behaviour. We therefore use the permittivity values of medium dry ground in the following evaluations. mmmagic Public 38

59 Figure 3.4: Reflection coefficient for different ground conditions as a function of the angle of incidence at 28 GHz for vertical polarization (left) and horizontal polarization (right) Validation with mmmagic results In this section, the impact of the ground reflection is investigated for different scenarios, based on uniform analyses of measurement data from three independent measurement campaigns. An overview of the measurement conditions is given in Table 3.6 together with references providing details for each measurement campaign. Table 3.6: Measurement data used for investigation of ground reflection and model validation Environment Courtyard Airport area UMi street canyon Ground material Flagstone/concrete and grass Tarmac runway and grassland Flagstone/concrete Frequency 28 GHz and 39 GHz 60 GHz 60 GHz Bandwidth 2 GHz 250 MHz 250 MHz Tx antenna horn, 10 dbi, height: 86 cm horn, 20 dbi, height: 4 m omnidirectional, height: 3.5 m Rx antenna horn with 10 dbi, height: 83.5 cm horn, 20 dbi, height: 3 5 m omnidirectional, height: 1.5 m Polarization vertical vertical and horizontal vertical Tx-Rx distance m with 25 cm spacing m with 1 mm spacing, here: results for m and 260 m 0 50 m with 0.4 mm spacing, here: m Reference Sections 2.4.2, A.4.3 [WPK+15a] [KPW+14] Courtyard In the project, dedicated measurements on ground reflection effects were conducted at 28 and 39 GHz in a courtyard. They are described in Sections and A.4.3. Since the excess travel distance of the GR path was below 10 cm, it cannot be separated with 2 GHz measurement bandwidth and superimposes with the LOS path. Figure 3.5 shows the normalized power of the first resolvable MPC as a function of the Tx-Rx distance for the flagstone/concrete ground. mmmagic Public 39

60 Figure 3.5: Measurement results in courtyard (flagstone/ concrete) with variation of Tx-Rx distance for vertical polarization at 28 GHz (left) and 39 GHz (right) and model output. The regular fading structure with maximums and minimums that result from constructive and destructive interference of both paths is clearly visible. In addition, the output of the GR model is plotted. Apart from slight deviations in the position of the extreme points, which are caused by ground irregularities, the model shows a very good agreement with the measurement curves at both frequencies. The results from measurements on grass are illustrated in Figure 3.6 and compared to the model output. As can be seen, in contrast to the flagstone/concrete surface, grass does not yield a significant specular ground reflection. The received power shows only small fluctuations around the free-space path loss and the GR model does not apply. Figure 3.6: Measurement results in courtyard (grass) with variation of Tx-Rx distance for vertical polarization at 28 GHz (left) and 39 GHz (right) and model output. Airport area Further measurement data was made available from a measurement campaign on an airport area [WPK+15a]. Compared to measurements in the courtyard, they were conducted at 60 GHz with 250 MHz bandwidth, and larger distances were considered. The spacing of two observations was 1 mm. Figure 3.7 shows the normalized received power as a function of distance between 100 and 160 m as measured on the tarmac runway for both vertical and horizontal polarization. Since no further multipath components besides the first one were observed on the open area, the received power is identical with the power of the first resolvable MPC. mmmagic Public 40

61 Figure 3.7: Measurement results on tarmac runway with variation of Tx-Rx distance for vertical (left) and horizontal polarization (right) and model output. In agreement with the results for flagstone/concrete in the courtyard, the fading indicates a strong GR. It is 1 10 db weaker than the LOS path. Though the strength of the GR slightly fluctuates over distance, e.g. the reflection is weaker between 115 and 125 m, the model is in very good agreement. This is confirmed by Figure 3.8, which illustrates the results for 220 m distance with variation of the Rx antenna height between 3 and 5 m. Figure 3.8: Measurement results on the tarmac runway at 220 m distance with variation of Rx height for vertical (left) and horizontal polarization (right) and model output. Figure 3.9 shows the corresponding results for grassland. In full accordance with the measurements in the courtyard, the received power only slightly fluctuates around the expected power of the LOS path, yielding the conclusion that the ground does not cause a specular reflection. This may be due to the grass acting as diffuse scatterer or the unevenness of the related ground surface. mmmagic Public 41

62 Figure 3.9: Measurement results on grassland with variation of Rx height at 220 m distance (left) and variation of Tx-Rx distance (right), and model output. UMi street canyon Finally, measurement data from a 60 GHz UMi street canyon measurement campaign is considered. The measurements were conducted on the sidewalk, paved with concrete flagstones. In contrast to the preceding GR measurements, omnidirectional antennas were used, which gives rise to a higher degree of multipath propagation. Furthermore, pedestrians could temporarily block the ground-reflected path. Figure 3.10 shows the normalized received power for a distance between 25 and 50 m. The fluctuations are less regular compared to the other scenarios. Nevertheless, the fading periodicity is in line with the GR model, which also well predicts the maximum and minimum expectable power. This confirms that the GR may have a dominant impact for busy urban propagation scenarios. Figure 3.10: Measurement results in the street canyon (flagstone/concrete ground) with variation of Tx-Rx distance for vertical (left) and horizontal polarization (right) and model output Conclusion In this section, we investigated the impact of the ground reflection on the LOS channel based on data from a dedicated mmmagic measurement campaign and additional data from former campaigns, which was made available in the project. All data was processed in a uniform manner. The measurement results for flagstone/concrete and tarmac ground are in very good agreement with the GR model and confirm its applicability for different mm-wave frequencies, different bandwidths, both polarizations and various distance ranges. The exemplary measurement data for the UMi street canyon scenario shows that the effect of the GR on the received power can even be dominant if omnidirectional antennas are used. mmmagic Public 42

63 Grass-covered ground does not yield a significant specular ground reflection at mm-wave frequencies. The received power shows only small fluctuations around the free-space path loss. If it is known that the ground is dominated by grass areas and unpaved paths, e.g. for an urban park area, the GR should therefore be switched off. In typical UMi scenarios, however, the ground is unusually determined by pavements and asphalt roads. They are relatively smooth and act as good reflectors. Assuming the existence of a ground reflection, we propose to choose one of the tree ground conditions randomly (very dry ground, medium dry ground and wet ground) for stochastic modelling. The GR fading causes both small and large-scale fading. Hence, it is not sufficient to include it in the PL model only. The reflected component must be explicitly included in the GSCM. This is not trivial because it changes the way MPCs are generated. The model in Section 4.6 describes how this can be done by adjusting the powers, delays and angles of the paths. Furthermore, we discuss the influence of the GR on the PL and propose corrections that need to be made in order to remain compatible with the existing models. 3.3 Clusters and subpaths Recent progress on the topic in 3GPP In the new 3GPP channel model [3GPP38.900], modelling of intra-cluster delay spreads and angular spreads has been enhanced in order to satisfy the requirements of large bandwidth and large antenna arrays. The number of subpaths per cluster has been proposed to be a function of the bandwidth, wavelength, and the size of antenna aperture. Increase in bandwidth or aperture size will linearly scale up the number of subpaths per cluster. The resolvable subpaths result in subpath-specific power, delay, and angles. Also, modelling of intra-cluster angular and delay spreads has been included. The assumption of the 3GPP channel model implied that intra-cluster delay and angle are independent. This assumption is fundamentally different from the cluster generation procedures in the 3GPP 3D-SCM [3GPP36.873]. The delays of subpaths (relative delays of rays) are generated by a sequence of uniformly distributed random variables (r.v.s) scaled by the intracluster delay spread. Intra-cluster offset angles of subpaths are also generated by scaled uniformly distributed r.v.s. Then, subpath powers are obtained by the product of power delay spectrum and power angular spectrum, after plugging in the subpath delays and angles into these spectrum functions. Table 3.7 lists recent progress on the distributions of characteristics of clusters and subpaths, by showing the distributions of cluster and intra-cluster characteristics. A full comparison of intra-cluster characteristics between the mmmagic proposed approach and the new 3GPP channel model [3GPP38.900] will be introduced in Section mmmagic Public 43

64 Table 3.7: Summary of recent progress on characteristics of clusters and subpaths How to generate [SR14] [WHW+16] [HJK+15] [MSA16] [3GPP38.900] #clusters Poisson Constant Poisson - Constant #subpaths/cluster Exponential Negative binomial Constant - Constant subpath delay Equally spaced - Equally spaced - Uniform distribution subpath power Exponential - Exponential Non-uniform A deterministic function Kmeans++ cluster algorithm After obtaining data of subpaths via measurements or raytracing simulations, clustering algorithm needs to be performed to group these subpaths into a number of clusters. The clustering of subpaths depends on their delays, powers, and AoAs. In principle, AoDs can be considered as well. However, in this report, we consider clustering from the UE s perspective. Therefore, the cluster algorithm does not use AoDs. The Kmeans++ algorithm [AV07], which groups sample points into a number of clusters according to a predefined metric, was developed based on the Kmeans algorithm with an additional simple randomized seeding technique. Here, the Kmeans++ algorithm was employed to analyse the number of clusters and other cluster properties. We determine the multipath component distance (MCD) in the azimuth of arrival and delay domains between two paths i and j as follows. ij MCD AoA = 1 2 a i a j, (3-7) where a k = [sin(θ k ) cos(φ k ), cos(θ k ) cos(φ k ), cos(φ k )] T, θ k and φ k are the elevation and azimuth of arrival of path k, respectively. MCD ij τ = ζ τ i τ j τ std Δτ2, (3-8) max where τ std is the standard deviation of the delays, Δτ max = max (τ i τ j ), and ζ is a suitable delay scaling factor setting the prominence of the delay domain. Here we fix ζ = 1. The resulting i,j metric to distinguish clusters: ij MCD ij = MCD AoA 2 + MCD ij τ 2. (3-9) To avoid detection of insignificant clusters, the power of an identified cluster should represent at least 0.5 % of the total power. Then, each centroid is updated by averaging subpaths inside and the Kmeans++ clustering algorithm continues iteratively until convergence. Finally, the optimum number of clusters N = K opt is determined by the Kim-Park (KP) index in [KPP07]. The KP index is computed by the mean intra-cluster distance and the inter-cluster minimum distance. It should be noticed that other clustering algorithms such as the agglomerative algorithm and time cluster and spatial lobe clustering algorithm approach were also reported in the literature. The intra-cluster properties depend on the clustering algorithm and parameters used. An example of subpath clustering is depicted in Figure 3.11, where points with the same color are subpaths in the same cluster. mmmagic Public 44

65 Figure 3.11: Example of Kmeans++ algorithm for subpath clustering Intra-cluster characteristics in outdoor and indoor environments Intra-cluster characteristics were studied in outdoor and indoor scenarios. The outdoor scenario was investigated via both measurement and raytracing simulation in an UMi street canyon scenario. Both measurement and raytracing were performed in the NLOS scenario while the LOS scenario was only simulated via raytracing as complementary. Details of this UMi street canyon campaign can be found in Section A.2.5, where intra-cluster characteristics such as cluster number, number of paths within a cluster, intra-cluster delay/angular spreads were studied. Indoor environments were investigated via measurements in obstructed-los (OLOS) and LOS scenarios. Details of this indoor campaign and the procedure to extract the MPCs from measurement can be found in Section A.1.1. We present here the clustering patterns and properties of indoor channel LOS measurements carried out in the two environments shown in Figure A.2, i.e., the office and conference room at 62 GHz and 83.5 GHz Clusters modelling The clusters are modelled based on the extended version of Saleh-Valenzuela (S-V) [SV87]. The original S-V model characterizes the cluster only in the delay domain. The extended model [SJJ+00] includes both the delay and spatial domain. The combined channel impulse response is: L K l h(t, φ) = β kl e jφ kl δ(t T l τ kl )δ(φ Θ l φ kl ), (3-10) l=0 k=0 with l and k the cluster and path index; L, and K l are the number of cluster and the number of MPCs in the cluster l; T l is the arrival time of the l th cluster; β kl, Φ kl, φ kl, and τ kl are the amplitude, phase, arrival azimuth, and arrival delay of the k th path in the l th cluster, respectively. The path gain is modelled as follows: β 2 kl = β 2 (T l, τ kl ) = β e 2 (0,0) T l Γ e τ kl γ, (3-11) where β 2 (0,0) is the average power of the first arrival of the first cluster. The procedure to calculate this average path gain can be found in [SV87, SJJ+00]. Γ and γ are the power-delay time constants for the clusters and rays, respectively. Next it is assumed that the clusters and rays arrival rate follow a Poisson distribution with fixed rate Λ and λ, respectively: p(t l T l 1 ) = Λe Λ(T l T l 1 ) l > 0 (3-12) mmmagic Public 45

66 p(τ kl τ (k 1)l ) = λe λ(τ kl τ (k 1)l ) k > 0. (3-13) Our aim is to determine these parameters from the measurements, along with the statistical properties of the angular parameters of the clusters and rays. Their assessment methodology is completely described in [SV87, SJJ+00], and the experimental results are shown in Section A.1.1. Table 3.8 summarizes the clusters and rays decay constant and arrival rate average values in in the two environments and frequency bands. In the office room, the clusters and rays decay faster at GHz than at GHz. This might be explained by the higher path loss at the higher frequencies. This trend is not observed in the case of the conference room due to, e.g., insufficient number of samples in the averaging. However, the cluster and ray parameters are similar in both frequency bands. Table 3.8: Decay rate and arrival rate of clusters and rays Γ (ns) γ (ns) 1 Λ (ns) 1 λ (ns) Office room GHz GHz Conference room GHz GHz We also determine the statistical description of the cluster and ray arrival angles. Details on the characterization of the rays and clusters arrival angles can be found in Section A Summary of studies of intra-cluster characteristics In summary, parameters of intra-cluster characteristics in the measurement and simulation campaigns in Section are listed in Table 3.9. To incorporate the abovementioned intra-cluster modelling to the mmmagic channel model, a number of steps need to be adapted. For example, in the extended SV channel model, the numbers of clusters and rays follow Poisson distributions. However, these numbers are assumed constants in the new 3GPP channel model [3GPP38.900] and the mmmagic baseline model in Section 4.1. Therefore, the Poisson distributed numbers of clusters and rays can be used optionally in the mmmagic channel generation procedure. In addition, as the number of intra-cluster subpaths is not always equal to 20, a procedure of intra-cluster subpath generation for the mmmagic channel model, which is adaptive to different number of subpaths per cluster and backward compatible to legacy channel models, will be introduced in Section Table 3.9: Summary of parameters of intra-cluster characteristics Outdoor LOS (Raytracing) Outdoor NLOS (Raytracing) Outdoor NLOS (Measurement) Indoor LOS (59-65 GHz) Indoor LOS ( GHz) Number of clusters N /5 4 Number of rays per cluster M Not available Not available 26 9/12 (3/4) 9/10 (3) mmmagic Public 46

67 Intra-cluster delay spread c DS Intra-cluster AoA spread c ASA Intra-cluster ZoA spread 16.0 ns 25.1 ns 23.5 ns 7.20 ns/14.20 ns 6.10 ns/8.0 ns / / Not available Not available c ZSA Note: For indoor results, X/Y indicates that X pertains to the conference room and Y to the office. A(B) indicates that A pertains to the first cluster and B to the following clusters. 3.4 Small scale fading Effect of bandwidth at K-factor and fading depth SCM and WINNER models are based on the uncorrelated scattering (US) assumption. To model this a narrow band Jake's [Jak94] like Sum-Of-Sinusoids (SOS) based approach is used where a large number (20/cluster in SCM/WINNER) of equal amplitude sinusoids are summed up to ensure Rayleigh fading of each resolvable MPC. At millimetre wave frequencies, however, a large amount of bandwidth is available and ultra-wide band (UWB) channel setups could easily be realized. Therefore, increased resolution in the delay domain increasingly leads to reduced number of non-resolvable MPCs. Hence, fading statistics of MPCs will no longer be Rayleigh and uncorrelated scattering assumption breaks down [MOL05]. Therefore, it is expected that with increased system bandwidth (reduced bin width), fading statistics of mm-wave channels will increasingly deviate from Rayleigh towards Rice distribution. Rician fading channels are modelled by a non zero-mean stochastic process which demonstrates the presence of either a LOS or fixed (a strong specular reflection) summed up with the diffuse MPCs. A high Rician K- factor demonstrates an increased contribution (in power) of LOS/fixed components, thereby reducing the significance of diffuse MPCs. This section focuses on the modelling of small scale fading distribution, the analysis of K-factor and fading depth as a function of bandwidth. A Rician K-factor is defined as K Ric = P 0 P d, (3-14) where, P 0 corresponds to the power of LOS/fixed paths and P d corresponds to the power of diffuse MPCs. In this work, four different propagation setups have been studied including a LOS, reflections from black board and wall are considered due to their different surface roughness and double bounce reflections from both surfaces. It has been observed that in all propagation cases, small scale fading depth asymptotically converges towards zero db whereby K-factor increases with bandwidth. Fading depth in our analysis is defined as F d = sσ d, (3-15) where, σ d is the standard deviation of total Rx signal power and s is a system parameter and could be defined as an arbitrary number. Figure 3.12 shows a snapshot of the normalized Rx power along the measurement track (setup 1) shown in Figure A.46. The amplitude fading reduces for higher bandwidths whereby power spread around the local mean (0 db) is significantly higher for lower bandwidths. For 4 GHz bandwidth, fading depth is almost negligible and amplitude decay due to path loss effect dominates in the channel. mmmagic Public 47

68 Figure 3.12: Fading behaviour comparison for different channel bandwidths, LOS, V-V polarization. (a) Co-pol (b) Cross-pol. Figure 3.13: Effect of bandwidth on the fading statistics and K-factor scaling, setup # 1 (LOS case). Figure 3.13 shows the cumulative distribution functions (CDF) of Rx signal P(x) for LOS channel setup under different bandwidths. Results show that the amplitude distribution of P(x) converges towards its local mean with increase in bandwidth and as a result K-factor increases in both polarization setups. The cross polarized setup in Figure 3.13(b) shows that standard deviation of P(x) is relatively higher and K-factor is lower as compared to the co-polarized case in Figure 3.13(a). As expected, lower bandwidths result in reduced K-factor showing the convergence of fading distribution towards Rayleigh fading. These results are in perfect agreement with FCC UWB channel investigations in [MAE08]. Ideally both transmitting and receiving antennas must be perfectly co-polarized to capture the signal energy efficiently. However, a channel rotates the polarization of a signal and increased scattering results in greater signal depolarization [Mal08]. This phenomena is more relevant in indoor scenarios where a signal is subjected to multiple reflections. Figure 3.13 compares P(x) and K-factor values of two extreme cases. Results in Figure 3.14 correspond to a clear LOS setup shown in Figure A.46 and signal depolarization is intentionally introduced by transmitting a signal either horizontally or vertically while receiving with same or opposite polarization. On mmmagic Public 48

69 the other hand, results in Figure 3.15 correspond to setup 4 shown in Figure A.46 where a signal is subjected to scatterers at both wall and black boards due to multiple reflections in the channel. In this case, channel itself may also depolarize a signal due to increased scattering. In LOS case shown in Figure 3.14, K-factor reduces up to 9.1 db for cross polarized setup which is still very high. Results in Figure 3.15 also demonstrate a high K-factor except for the V-H scenario. This proves that even after the depolarization by a channel, mm-wave UWB channels are still fading with a Rician distribution with a high K-factor. Figure 3.16 and Figure 3.17 show how the K-factor and small scale fading depth evolve with an increase in channel bandwidth. In LOS case amplitude of fixed path gain ρ is always higher than the scattered components which are present in the channel due to 30 wider beamwidth of the antenna. This results in higher K-factor in LOS case as compared to other setups, as shown in Figure The smooth surface of the black board turns out to be a good reflecting object resulting in more fixed/specular path amplitude than the scattered ones. This is apparent from relatively higher K-factor as compared to wall reflections. However, at lower bandwidths (wider bin width) around 500 MHz, summation of these MPCs results in Rayleigh fading amplitude distributions implying that energy of fixed path ρ 0, resulting in very low K-factor. Figure 3.14: Effect of polarization on the K- factor scaling, BW=2 GHz, LOS. Figure 3.15: Effect of polarization on the K- factor scaling, BW=2 GHz, Dual reflections from black board and wall. Figure 3.16: K-factor analysis for different channel bandwidths, V-V polarization setup. Figure 3.17: Fade depth scaling for different channel bandwidths, V-V polarization setup, s = 3. mmmagic Public 49

70 Rough surface of the wall is expected to result in relatively lower amplitude of specular reflections as compared to black board and LOS path. As a result, the small scale fading behaviour stays in the Rayleigh fading regime even for relatively larger bandwidth. The case for double reflections from both reflecting objects (i.e. wall and black board) is even worse than single wall reflections. However, in each measurement setup, it is clear that K-factor increases with an increase in channel bandwidth. It shows that channel fading behaviour evolves from Rayleigh towards Rician fading. As a result, fade depth also falls rapidly with increase in bandwidth up to 1 GHz. However, after this point an increase in K-factor is not so significant and fade depth is lower than 2-4 db of the received power. It shows that minuscule movements of Rx do not lead to a significant variation in the total received power. Hence, it is fair to conclude that with an increase in bandwidth, channel fading behaviour turns out to more deterministic than stochastic Effect of bandwidth on the cross polarization ratio (XPR) In addition to fade depth scaling as discussed earlier in Section 3.4.1, electromagnetic polarization behaviour is another related attribute to be studied under small scale movements of transmitter and/or receiver. In narrow-band channels, minuscule Tx/Rx movements change the delay and phase relationship of non-resolvable MPCs with in a resolvable MPC, causing a random amplitude fading of the delay bin. As a result, total channel power of co- and cross polarized propagation channels may vary significantly with small scale movements. To model this effect, XPR is usually assumed to be a log-normally distributed random variable [Sha06, KMH+07] which in turn causes randomness in the polarimetric scattering matrix as well. For a fully delayresolved UWB channel, amplitude fading of the MPCs is caused only by attenuation losses. Therefore, the channel power may not vary significantly with small scale movements. As a result, randomness in the XPR and polarimetric scattering matrix is expected to be reduced demonstrating a quasi-deterministic channel behaviour. Ideally, polarization of a received signal must exactly be the same as that of a transmitted signal to capture maximum of the signal energy. However, a wireless channel rotates and thus depolarizes a transmitted signal, which is an entropic effect. The degree of a signal depolarization is known to be a function of carrier frequency and reflection coefficient of the interacting surface [JKF99]. The objective of this work is to characterize the XPR of different reflecting surfaces and to track the degree of randomness of XPR as a function of channel bandwidth. To date, with the best of our knowledge, such a characterization study has not been done for mm-wave channels and the results reported so far [DNC17,NCJ17, SR+15] have focused only at the small scale fading. By definition, XPR refers to a situation where a signal's polarization is perfectly maintained, XPR 0 corresponds to a perfect signal depolarization and XPR < 0 implies a large coupling with orthogonal polarization. XPR power level in db for horizontal to vertical XPD H is defined as the channel power ratio between H-H (Tx horizontal-rx horizontal) and H-V polarized channels whereby vertical to horizontal XPR V is the power ratio between V-V and V-H setups. Figure 3.18 shows average values of XPR H and XPR V computed over a small scale area. mmmagic Public 50

71 (a) (b) Figure 3.18: Average XPR for (a) horizontal to vertical XPR H (b) vertical to horizontal XPR V versus channel bandwidth. As expected, XPR power is higher in LOS, demonstrating that polarization is maintained significantly. Theoretically, in the LOS measurements, if the channel is composed only of a single direct LOS path, XPR power should be infinite. However, in our measurements, MPCs probably reflected from tables/chairs are also present in LOS. Depolarization of these reflected MPCs cause a finite but relatively higher XPR power as compared to other propagation setups. Due to relatively smooth surface of the black board as compared to wall, XPR H and XPR V values are higher in case of reflection from black board. This demonstrates that depolarization of a transmitted signal is relatively lower for black board reflections as compared to wall which is theoretically expected. Interestingly, double reflections (reflected first from black board and then sound absorber wall) also result in higher XPR power as compared to reflections from wall. This is indeed possible, as polarization rotations by multiple reflections may end up in a signal orientation parallel to Rx antenna polarization. Results in Figure 3.18, do not show any conclusive relationship of average XPR values with bandwidth. However, in certain cases a 5-10 db difference in narrow-band and UWB channels is apparent. Figure 3.19: XPR behaviour comparison for different channel bandwidths. mmmagic Public 51

72 Now we compare the difference in the degree of randomness of XPR values for narrow-band and UWB channels. In Figure 3.19, XPR power levels are shown for the whole 30 cm measurement track. The values are normalized to have a unit mean in the linear scale. These results clearly show that an increase in bandwidth results in reduced variation in the XPR power with minuscule movements of the Rx. Results in Figure 3.20 show the coefficient of variation (in %) defined in (A-14) as a function of bandwidth. It can easily be followed that both CV H and CV V converge towards zero percent with an increase in channel bandwidth. This convergence is almost linear along log-bandwidth scale (i.e. exponential in linear bandwidth scale) in many cases. This clearly demonstrates that XPR behaviour becomes more and more deterministic with increase in the channel bandwidth. If only a single direct LOS path is present in the channel, standard deviation in the XPR power must be zero for each channel bandwidth. However, this is not the case in our measurements as described earlier. Figure 3.20 shows that coefficient of variation for LOS case is lowest in most of the bandwidth setups. However, in our measurements, no clear relationship is observed between the degree of randomness of XPR and different reflecting surfaces. (a) (b) Figure 3.20: Coefficient of variance for (a) horizontal to vertical CV H (b) vertical to horizontal CV V versus channel bandwidth Effect of antenna directivity Figure 3.21 shows the observed fade depth for the different antenna combinations and bandwidths. It can be observed that there is a reduction of the fade depth not only with the bandwidth as already shown in [MAE08, NCJ17], but also with an increment on the directivity of the antennas. It is interesting to notice that for the 15 HPBW at TX and omni RX, with shorter bandwidth the fade depth is increased considerably in comparison to the 15 HPBW RX antenna. Figure 3.21: Fade depth for different bandwidths and directive antennas for s = 3. mmmagic Public 52

73 Table 3.10 summarizes the estimated K-factors. It can be easily followed that an increase in bandwidth results in increased K-factor. It is interesting to note that under the same bandwidth more directive antennas result in higher K-factor in NLOS scenario as compared to the LOS setup with omni-omni setup. Table 3.10: K-factor at different bandwidths and with different antenna combinations Visibility 0.1 GHz 0.2 GHz 0.4 GHz 1 GHz 2 GHz 4 GHz omni/omni LOS 6.15 db 8.45 db db db db db 15 /omni NLOS 3.92 db 6.72 db 8.77 db db db db 15 /30 NLOS 7.24 db db db db db db 15 /15 NLOS 7.27 db db db db db db 3.5 Blockage Modelling of blocked pathways becomes more important as the carrier frequency increases. Particularly in the mm-wave frequency range, where high gain antennas and beamforming techniques will be used to mitigate the decrease of omni antenna aperture and resulting increase of path loss, possible blockage by vehicles, people, vegetation etc. may be severe. Consequently, it is crucial to model this effect realistically. Diffraction models are suitable to model blocking. The problem is, however, to provide a comprehensible and simple model. A simplistic approach is to assume that the basic blocking element is a rectangular totally-absorbing screen. It should be noted that a totally absorbing screen gives diffraction which is polarization invariant. This contrasts with the UTD wedge model which uses the actual electromagnetic material properties. This totally-absorbing screen problem is solved accurately by means of the Kirchhoff diffraction integral [BW99]. A major drawback is though that there is no closed solution of the two-dimensional integral. To obtain convergence with numerical integration, the screen has to be discretised into elements substantially smaller than half a wavelength. At 60 GHz, for example, the wavelength is 5mm resulting in a discretization at the order of 1 mm resulting in one million elements for a 1m by 1m screen size. For this reason, different simplified models have been proposed. Most of these models are formulated for the 2D case. A simplified model for the 3D case (METIS blockage model) was proposed in [METIS-D1.4]. It is formulated in such a way that it fulfils Babinet s principle, i.e. different shapes of blocking objects may be synthesised by combining multiple screens. Due to its simplicity and comprehensibility, it was agreed to be a component of the 3GPP NR (5G) channel model [3GPP38.901]. Further details on the METIS blockage model are provided in the mmmagic deliverable D2.1. Within mmmagic a substantially improved blockage model has been developed based on the METIS blockage model. This model accounts for the fast fading due to summing of the complex amplitude of paths from the four edges of a rectangular screen. The output of the mmmagic model is considerably more accurate than the output of the METIS model. Furthermore, the mmmagic blockage model is, as the METIS blockage model, conveniently expressed using standard analytic functions mmmagic blockage model A problem with the METIS blockage model is that it does not model the effect of the combination of different phases of the paths from the four edges of the screen. Furthermore, the 3D modelling has not been validated by comparison with accurate physical models. In this section, an improved version of the METIS blockage model is presented and validated based on both measurements and Kirchhoff s diffraction formula. The mathematical formulation of the model is provided in section A simulation example using a quadratic screen of 4 m size and radio frequency of 6 GHz is shown in Figure The transmitter (Tx) is located at 100 m distance mmmagic Public 53

74 from the screen and the receiver (Rx) at 1 m distance behind the screen. For this scenario the model output is shown for the case when the receiver (Rx) moves from a LOS location to the blocked location at the centre behind the screen. There are two different pathways for the Rx movement: a) parallel to, and, b) 45 degrees relative to the horizontal edges of the screen. The improved model output is very accurate as compared with the Kirchhoff integral output. It is also clear that the previous (METIS) model, which does not account for the phase differences of the four paths, only provides the minimum loss in the fading zone. Another example based on measurements at 15 GHz with a 5G radio access prototype [OSH16] is shown in Figure 3.23 and Figure It is clear that the improved model has better agreement with the fading immediately before blocking as well as in the shadow zone. Moreover, the METIS model provides underestimation of loss in the deep shadowing zone, whereas the improved model does not (two right hand graphs of Figure 3.23). Figure 3.22: Blockage simulation for a 4m wide quadratic screen (two left hand graphs). The radio frequency is 6 GHz, the TX distance to screen 100m and RX distance to screen 1m. In a) the RX pathway is parallel and in b) 45 degrees relative to the horizontal screen edges. mmmagic Public 54

75 Figure 3.23: Measured signal strength when a garbage truck temporarily blocks the LOS condition of the 15 GHz transmitter at 70m distance. Figure 3.24: METIS (upper) and improved (lower) blockage model output for scenario shown in Figure Figure 3.25 shows that the blockage model provides accurate output when nonrectangular screen shapes are synthesized by joining multiple rectangular screens together. In this case the scenario shown in Figure 3.23 and Figure 3.24 is modified so that the cab is lowered and joined with the container. mmmagic Public 55

76 Figure 3.25: Same scenario as Figure 3.24 but with different placement of cab. The model is further validated based on the measurements described in A.4.4, at 62.0 GHz and 83.5 GHz, where a phantom has been used for blockage of the LoS path. In Figure 3.26 the modelled scenario is shown. Two screens are used to model the head and torso of the phantom. In the same figure the model outputs are shown for 62 GHz and 83.5 GHz. Figure 3.26: Modelled blocking scenario (left) and corresponding output for 62 GHz (middle) and 83.5 GHz (right). The measured and modelled blocking loss for positions P1-P6 are given in Table 3.11, along with the prediction errors. There are some variations which are likely to be caused by the fluctuations in the shadowing zone which may not be precisely modelled. Furthermore, there might be some inaccuracies in the modelled geometry and/or the phantom positioning since the measurements in both bands were carried out at different dates. However, the overall agreement is good. mmmagic Public 56

77 Table 3.11 Shadowing loss values. All values are expressed in decibels (db). The prediction error is the difference between measured and simulated shadowing loss values. P1 P2 P3 P4 P5 P6 ΔP 0 (62 GHz) meas ΔP 0 (62 GHz) sim Prediction error (62 GHz) ΔP 0 (83.5 GHz) meas ΔP 0 (83.5 GHz) sim Prediction error (83.5 GHz) Spatial consistency A newly defined procedure, namely a spatial consistency procedure, has been initiated by 3GPP in release 14 [3GPP38.900] in order to update the large and small scale channel modelling parameters so that they are spatially consistent. A spatial consistency procedure and correlation distances are defined for: Cluster-based random variables Line of Sight (LoS) to non-los (NLOS) state transition 3GPP defines a procedure that is based on the parameter-specific correlation distance values for spatial consistency. The corresponding metrics for various location types defined in the 3GPP standardization can be found in [3GPP38.900]. There are two typical methods to model spatial consistency, one is the spatially correlated random variable based method, and the other is the geometric stochastic approach. The spatially correlated random variable based method has already been used in large-scale parameter generation in the mmmagic channel model. Its principle will not be repeated here. How the spatially correlated random variable based method is embedded in the procedure of mmmagic channel generation will be described in Section The rest of this section will focus on the description of the geometric stochastic approach Geometric stochastic approach The basics of this approach were originally been proposed in [5GCM16]. Spatial consistency means that channel realizations including large-scale parameters (LSPs) and small-scale parameters (SSPs) would need to vary in a continuous and realistic manner as a function of position in geometry. Two features of spatial consistency are important. Firstly, user equipments (UEs) sharing similar locations should have correlated LSPs, and the LSPs should be crucially dependent on UE s position instead of random allocation in each drop as done in 3GPP SCM. Moreover, the path loss including shadow fading should vary smoothly as UE moves in geometry. This is particularly important to the evaluation of multiuser MIMO or multiuser beam-forming techniques. Secondly, SSPs in a drop (e.g. angle, power, and delay) should be dynamically changing with position. The new model realizes time-variant angles and cluster death and birth as UE is moving which is important to evaluate mobility and beam tracking for 5G communications. Geometry position Geometry positions of UE, scatters, and BS are the fundamental information of SCM, and are fixed in a drop. Actually, the position of UE is time variant as UE is moving. Suppose the moving mmmagic Public 57

78 speed of UE is v and moving direction is v in global coordination system (GCS), the position of UE at time t is given by where d(t 0 )sin(θ ZoD )cos(φ AoD ) + v(t t 0 )cos(φ v ) X UE (t) = ( d(t 0 )sin(θ ZoD )sin(φ AoD ) + v(t t 0 )sin(φ v )), (3-16) h UE d t 0 is the distance between BS and UE at previous time t 0. Notice that the time interval t t 0 can be a sub-frame duration as used in 3GPP. Time-variant Path loss The path loss is dependent on the distance between BS and UE. Since BS s position is fixed at X BS = (0,0, h BS ) T, the distance between BS and UE at time t is d(t) = X BS X UE (t). (3-17) With the path loss model for above 6 GHz or 3GPP path model for sub-6ghz, the path loss at time t can be updated accordingly. The correlated shadow fading in different positions are discussed in section in [3GPP36.873]. The correlated shadow fading is given by F(d) = F 1 + F 2 1 α 2. (3-18) where α = exp( d/d cor ), d cor is the correlation distance of shadow fading, F 1 and F 2 are the shadow fading allocated in two neighbouring grids in the space. Position-based Large-scale Parameters (LSPs) 3GPP SCM allocate LSPs randomly for each UE. Two UEs may have much different LSPs although they are close in locations. In fact, the LSPs of the two close by UEs should be similar which leads to channel impulse response with high correlation. To circumvent the problem, we divide each cell under the BS s coverage into multiple grids. Each grid is spatially consistent in the sense of large-scale fading characteristics. Each grid is configured with a set of LSPs following the given probability density function. Grid centre is assumed as the location in calculating LSPs. In the step to generate the LSPs for a UE location, it firstly checks which grid the UE locates, and then take the LSPs of the corresponding grid to the UE channel. In this way, UE sharing the same grid will have the same LSPs. Grid-based LSPs Set scenario, network layout and Grid parameters Assign propagation condition (NLOS/ LOS) for every grid Generate correlated large scale parameters (DS, AS, SF, K) for every grid Save the Gridbased LSP table General parameters: Set scenario, network layout and antenna parameters Get propagation condition (NLOS/ LOS) Calculate pathloss Get correlated large scale parameters (DS, AS, SF, K) Small scale parameters: Figure 3.27: LSPs generation procedure. mmmagic Public 58

79 Figure 3.27 illustrates the procedure to generate LSPs where red texts are the new steps based on [3GPP38.900]. Notice that the grid-based LSPs are calculated only once and are saved as a table. Most LSPs of UEs are taken from the table. Thus, the computational complexity of LSPs is lower than 3GPP SCM. Time-variant Angle Variant angles are introduced for each ray including azimuth angle of departure and arrival (AoD, AoA) and zenith angle of departure and arrival (ZoD, ZoA) in [WHS15+]. Since UE s position at time t is available, the angles can be updated with transmitter and receiver information in the global coordination system (GCS). Linear approximation is an efficient way to reduce complexity with acceptable errors. The linear method for variant angles is formulated as [WHS15+] θ m,n,angle (t) = θ m,n,angle (t 0 ) + S m,n,angle (t t 0 ), (3-19) where the sub-index Angle represents AoA, AoD, ZoA, or ZoD. S m,n,angle is the slope which describes the changing ratio of time-varying angles. For LOS cluster, the expression of AoD and ZoD slopes are given by [WHS15+] vcos(φ v φ AOD (t 0 )) S ZOD = S ZOA = (h BS h UE )/cos(θ ZOD (t 0 )) (3-20) S AOD = S AOA = vsin(φ v φ AOD (t 0 )) (h BS h UE )tan(θ ZOD (t 0 )). (3-21) For NLOS cluster, the model can be simplified by introducing a virtual UE which is the mirror image of UE based on the reflection surface. The simplified slopes in NLOS channel are given by [WHS15+] S ZOD = S ZOA = vsin(φ v + φ AOD (t 0 ) φ RS ) (h BS h UE )/cos(θ ZOD (t 0 )), (3-22) S AOD = S AOA = vcos(φ v + φ AOD (t 0 ) φ RS ) (h BS h UE )tan(θ ZOD (t 0 )), (3-23) where φ RS is the angle of the reflection surface and it can be deduced from the initial φ AoD and φ AoA. Based on the time-variant angle, cluster AoA smoothly changes with UE s position. Figure 3.28 shows the AoA changing rate is about 1.8 degree to 2.2 degree per meter (Figure 3.28(a)), which is comparable to the measured results 2 degree to 5 degree per meter (Figure 3.28(b)) [WSH16]. (a) Simulation (b) measurement Figure 3.28: AoA vs. Route. mmmagic Public 59

80 Cluster Birth and Death Cluster birth and death are assumed to happen at the same time in order to keep a fixed number clusters as defined in 3GPP SCM. Scatters are assumed to be independent with each other. In this sense, cluster birth and death can be modelled with Poisson process if looking at the rate of cluster birth/death in time. The cluster birth/death will happen at time t with the probability Pr(t) = 1 exp( λ c (t t 0 )), (3-24) where t 0 is the previous time of cluster birth/death. The model has a single parameter λc, which represents the average number of cluster birth/death per second. The parameter λc depends on the number of birth/death in a spatial consistency distance and UE moving speed. For cluster death, the cluster selection can be based on the cluster power from weak to strong since weak cluster is easy to change [KMH+07]. For cluster birth, new cluster can copy the cluster (power, delay, and angles) from nearest grid. The priority of cluster selection is based on the cluster power from weak to strong. When UE is moving to the neighbouring grid, the clusters will be replaced by the new clusters of neighbouring grid gradually and hence keep spatial consistency. Figure 3.29 shows the UE trajectory and Figure 3.30 shows the cluster birth and death in delay domain [WSH16]. Figure 3.29: UE trajectory. Figure 3.30: cluster birth and death in excess delay domain. Method using geometric locations of clusters (Grid-based GSCM, GGSCM) According to the drop concept of the conventional GSCMs (SCM, WINNER, IMT-Advanced, 3GPP 3D, etc.), the UEs are located randomly and the propagation parameters are randomly drawn from the pre-defined probability distributions. The channel is assumed to be stationary along a short distance (segment), but this assumption does not hold for longer distances, therefore the parameters need to be re-calculated (new drop/segment). This approach is called as block-stationary modelling in which LSPs and SSPs are fixed during the segment and fully different between the segments. The transition from a segment to another provides a rapid change of channel model parameters thus the channel is discontinuous. To improve the realism of time evolution, it is possible to interpolate between the segments. However, it is difficult to ensure spatial consistency between nearby users in multi-user case. Therefore, a new method (partly based on [METIS-D1.2]) is proposed in [5GCM16] and drafted in the following. In this method, called Grid-based GSCM (GGSCM), cluster and path angles and delays are translated into geometrical positions (x, y) of the corresponding scatterers (see Figure 3-31). The benefit is that the cluster and path evolution in delay and angle domains can be naturally traced and will have very realistic variations. mmmagic Public 60

81 Figure 3-31: Clusters are translated into geometric positions. This method needs to be complemented with some birth/death process to maintain uniformity of clusters during movement. An example clarifies this discussion. Let us consider a case of three users: A, B, and C (Figure 3.32). Users A and C are far away from each other. They may assume independent clusters. However, the users A and B are located nearby. The current 3GPP-3D model [3GPP36.873] assumes independent small-scale parameters (SSPs), which lead to a non-physical situation, and too optimistic MU-MIMO throughput evaluations. Figure 3.33 illustrates the thinking of spatially consistent case in which all or some of the clusters are shared between nearby users. A B A B C C Figure 3.32: The problem of independent clusters of nearby users (current GSCM). Figure 3.33: Shared clusters (necessary improvement). Figure 3.34 depicts the situation in which a high number of users are dropped onto a 2-dimensional map. Each user has a ring around them, and the radius of that ring is equal to the correlation distance (or stationarity interval). If another user is located inside that ring, the spatial consistency must be taken into account. Otherwise, current method of random SSPs is acceptable. In the case of nearby users, the clusters should be interpolated between the users. User B takes the N strongest clusters (N is the number of clusters defined per scenario). mmmagic Public 61

82 These users share some clusters Independent User, independent clusters Correlation distance Figure 3.34: Dropping of users. Figure 3.35: Grid model (GGSCM): Calculate new cluster information at each grid point. Interpolate clusters between the four grid points. The interpolation can be done along a route based on a pre-defined grid (Grid based GSCM, GGSCM). In the GGSCM approach a discrete two-dimensional map of possible UE locations is defined as shown in Figure Instead of drawing LSPs and SSPs for the actual user locations, the cluster parameters are drawn for every grid point. Then the cluster parameters for the actual UE locations are interpolated between four nearest grid points. The grid can be intuitively understood as a drop in which the distance between two adjacent users is constant in x and y dimensions. The drops are independent between the grid points (GPs), i.e., LSPs and SSPs are randomly drawn from the pre-defined distributions (similar to the legacy GSCM). The locations of the clusters are defined in (x, y) or (x, y, z) coordinates. The maximum distance between UE (or BS) and the cluster location is determined from the geometry of UE and BS locations, AoA, AoD, and delay. This geometry is an ellipse with focal points at UE and BS locations, and the major axis equals to the delay of the MPC multiplied by the speed of light. In the case of single bounce, the cluster is located on the locus of an ellipse defined by AoA, AoD, and delay (see Figure 3.36, SBC, single bounce cluster). In the case of multi-bounce, the same locus defines upper bound of the distance of the cluster, i.e. the cluster can be anywhere in the segment between BS (or UE) and the locus (see Figure 3.36, FBC/LBC, first/last bounce cluster). A distribution for that cluster location could be uniform between the two ends of said segment. Because the AoA, AoD, and delay are randomly drawn in the GSCM, most likely the geometry of these three parameters does not fit to the ellipse. Thus the 50% of the cluster locations may be based on UE -side cluster parameters and another 50% based on BS-side cluster parameters. After fixing the physical locations, drifting of LSPs and SSPs are enabled for a short distance movement as illustrated in Figure Implementation of the drifting is straightforward and is fully based on the geometry (for each impulse response, phase, delay, and angle of arrival is recalculated). This supports dynamic channels and simulation of very large arrays. This approach also allows spatially consistent LOS (and specular reflection). Since the BS, UE and scatterers have physical coordinates, also a simple map for LOS (and specular reflection) can be created. The clusters may be calculated only for the grid points and on the need basis to avoid excessive use of memory. A smooth birth-death process of clusters can be realized by weighting the cluster powers in each grid point based on the distance from UE. All clusters of the four closest grid points are active and the clusters are selected by cutting the weakest clusters away. If the number of a cluster in the scenario of interest is N = 20, the total number of clusters in any position between the grid points is 4*N = 80, but only 20 strongest clusters are selected. The strength of the cluster is scaled by a path loss from the cluster location to the UE. This approach keeps the number of clusters constant, and allows smooth birth-death process. mmmagic Public 62

83 SBC d 1 2 d d c FBC AoD LBC d 1 d 2 AoA Tx Rx Figure 3.36: Location of a cluster. Figure 3.37: Drifting of angles and delays. 3.7 Specular reflection, diffused scattering and depolarization Characterisation of specular reflections One of the biggest issues in the mm-wave bands is the modelling of diffuse scattering. At lower frequencies (< 6 GHz) building and terrain surfaces have usually been assumed to be electrically smooth (i.e. their surface height variations were small compared to the carrier wavelength) and the reflection process is dominated by a strong specular path at an angle of reflection equal to the angle of incidence. On the contrary, in the mm-wave bands, surface height variations are significant compared to the carrier wavelength and the reflection process may need to be replaced with diffuse scatter. This work presents the results of a measurement campaign of the specular scattering process for a number of different smooth and rough building materials at 60 GHz. Analysis of the measured results which are presented in Section 4.2, indicate that a Nakagami-m distribution can adequately model the random fluctuations caused by surface roughness of the various building materials. This can be achieved by merely adjusting the m-parameter of the distribution. The modelling can be employed to generate the randomness caused when surface scattering occurs, by superimposing random samples of the Nakagami-m distribution on the deterministic specular component. In order to analyse the specular scatter process of mm-waves, and the depolarization that is caused by diffused scattering, a measurement campaign was carried out for a number of different outdoor and indoor smooth and rough wall surfaces within the University of Bristol premises. A detailed description of the measurement setup is provided in A.4.1. Figure 3.38 shows the measured received power of the first order scattered component for a travel distance of 4 m in parallel to the dressed stone wall (rough surface) and the glass window (smooth surface). mmmagic Public 63

84 Figure 3.38: Measured profile for dressed stone wall and window scattering. It is apparent from the presented figure that the power fluctuations of the specular component for the rough dressed-stone wall measurement are considerably larger compared to the signal fluctuations resulting from the smooth window reflections. The modelling of the specular as well as the diffused components for both smooth and rough surfaces will be described in detail in Section Characterisation of diffused scattering The small-scale variations instigated by the propagation mechanism of diffused scattering have been investigated through the process of wall reflection measurements at a central frequency of 60 GHz and a bandwidth of 2 GHz. To measure the power received at all directions, the transmitter was placed at a fixed angle with respect to the normal from the wall (~0 0 to 90 0 ) and the receiver was mounted on a trolley and moved along an arc of a predefined radius. More details on the experimental set-up are provided in paragraph A.4.2. For the purpose of determining the concentration of power within a specified range we introduce the term of power concentration which can be defined as the angular span corresponding to 90% of the power in the angular profile. Therefore, the higher the power concentration, the less the impact of the diffused scattering, since a low concentration implies that most of the received power is found within a small range around the incident angle. Figure 3.39 and Figure 3.40 depict the power angular profiles resulting from diffused scattering measurements at a representative transmit angle of 45 0, for the concrete pillar and the red stone wall scenarios. mmmagic Public 64

85 Figure 3.39: Power angular profiles at 45 0 transmit angle, for the concrete pillar measurements. Figure 3.40: Power angular profiles at 45 0 transmit angle, for the red stone wall measurements. The above figures illustrate the important differences observed when comparing the angular scattering profiles of rough and smooth surfaces. More specifically, it is apparent that for the case of reflections from the concrete pillar, there is a clear separation of the angle range where most of the power is concentrated. Clearly this angular span is centred at the angle of incidence. Furthermore, it can be observed that an increase in the transmitter-receiver separation decreases the angular spread for the same incident angles. In addition, it is shown that keeping the distance fixed, the power concentration is reduced for higher transmit angles. In the case of reflections from the red stone wall, the power angular profile appears sparse, making thus difficult to distinguish and visualize the angular span corresponding to the power concentration metric, which obviously is significantly bigger than the case of a smooth wall surface. After analysing the results, it was found that similar to the case of the concrete pillar, an increase in distance implies reduction of the angular spread. Furthermore, higher transmit angles lead to smaller values of angular spread when there is a reference at the same distance. mmmagic Public 65

86 However, those findings are not constant. In general, power concentration reduces with distance and transmitting angle. However, when considering reflections from rough surfaces, such as the red stone wall, the value of this metric appears quite large, implying that the effect of diffused scattering is quite significant and should be incorporated into electromagnetic-based prediction tools such as ray-tracing software/algorithms. Table 3.12, illustrates the power concentration values as for both wall types as obtained from the measured data. Table 3.12: 90% Power concentration over all transmit angles and distances Distance (meters) Angle (Degree) Concrete Pillar Red Stone Wall Depolarisation The effect of polarization has been investigated through means of the cross-polar discrimination ratio (XPD). The XPD is given by XPD = 10log 10 ( P V P H ), (3-25) where P V and P H correspond to the received power of the vertical and the horizontal component respectively. Due to the property of the channel sounder used in these measurements both the horizontal and vertical components of the received signal have been recorded. Therefore, the estimation of the XPD ratio is straight forward and it is calculated directly from the measured data. Some representative results on the depolarization of the vertically transmitted signal are shown in Figure 3.41 and Figure 3.42, for the case of concrete pillar and red stone wall respectively at a distance of 6 m and for a transmit angle of m, 45 0 Figure 3.41: Received Power versus angle for both vertical and horizontal polarizations at 6 meter distance and 45 0 transmit angle Concrete pillar. mmmagic Public 66

87 6m, 45 0 Figure 3.42: Received Power versus angle for both vertical and horizontal polarizations at 6 meter distance and 45 0 transmit angle Red stone wall. Measurement analysis suggests that as a general observation, XPD values are much lower in the case of the red stone wall. This finding implies that depolarization is more severe when considering reflections from rough wall surfaces. Another observation was that for the concrete pillar scenario, no XPD dependency on distance was observed, whereas it was shown that the cross-polar discrimination becomes bigger for higher incident angles Surface scattering: modelling of specular and diffused components The most appropriate distribution for analysing the small-scale signal strength fluctuations of the specular component over short travel distances is subsequently determined by comparing amongst typical well-known probability distributions that are commonly applied for the modelling of multipath fading channels. The mathematical formulation for modelling and generating signal amplitudes that corresponds to the fading of the specular components is quite simple. Assuming that the received power is Y, then Y can be expressed as Y = a p E(Y), (3-26) where E(Y) denotes the expected value of the variable Y and the parameter a p is the normalized power so that E(a p ) = 1. From (3-26), it is straightforward that the normalized amplitude a v is given by a v = Y E(Y) (3-27) Therefore, investigation of this parameter allows determining the amount of fading, a metric that depends on the roughness of the surface under test. Three probability distribution functions have been employed in order to characterize the amplitude fading of the specular component; The Rayleigh, the Rician and the Nakagami-m distributions. These probability functions have been extensively used in modelling the multipath fading profiles of mobile communication systems. The suitability of each of the above distributions to model the measured data was identified by means of Root Mean Squared Error (RMSE). RMSE is expressed as follows mmmagic Public 67

88 RMSE = E ((X X) 2 ), (3-28) where X and X represent the theoretical (predicted) and measured data respectively. Table 3.13: RMSE among various distributions to describe the variations of the specular reflections Material m- parameter K factor (db) RMSE (db) Nakagami Rician Rayleigh Window Bath stone wall Dressed stone wall Table 3.13 presents the parameters of the best-fit Nakagami-m distribution (i.e. m-parameter) and the best-fit K-factor (which is estimated by the theoretical fitting of Rician fading onto the experimental data). The RMSE is also included for the three considered distributions and the three surfaces under test. The fitting for the window, the bath stone, and the dressed stone are shown in Figure It is apparent that for all three surface types considered, the Nakagami-m and the Rician distribution provide good fits on the experimental data. The Rayleigh distribution fails to interpret the experimental data for the window (smooth surface) and for the Bath stone wall (surface of medium roughness). However, as it is shown in Figure 3.43 it provides adequate fit for the case of a very rough surface, (i.e. dressed stone wall). This is expected from theory since Rayleigh fading is mostly applicable at conditions of severe fading. The Nakagami-m distribution provides adequate interpretation of the measurements undertaken and the RMSE value obtained from comparison with the measured data is almost identical to the Rician distribution. Moreover, the Nakagami-m distribution provides the flexibility required to model a wide range of signal propagation conditions, ranging from very small to severe fluctuations on the received signal strength. The Nakagami-m distribution is given by f(x) = 2 ( μ ω ) μ 1 Γ(μ) x(2μ 1) e μ ω x2, (3-29) where x denotes the amplitude of the received signal, and μ, ω are the shape and the scale parameters of the distribution. In particular, the expected value of the power is given by ω, and this implies that in the case of the modelling process followed in this work, it should be always ω = 1. This facilitates the modelling process, and is an easy way to verify the validity of the normalized amplitude values (i.e. for the normalized amplitudes (a v 2 ) = 1 ). (a) (b) (c) Figure 3.43: Modelling of amplitude distribution for (a) The window surface, (b) Bath stone wall and (c) The dressed stone wall surface. mmmagic Public 68

89 Beyond the modelling of specular reflections, an important issue that arises is the prediction of the received power when the AoA at the receiver side, is different than the incident angle. According to ray optics, no signal is expected at a direction out of the specular component when the surface structure is perfectly smooth. However, it is shown in Section that especially in mm-wave frequencies this assumption is not realistic but depending on the surface type (i.e. smooth or rough), the percentage of scattered energy can vary for different directions, and antenna separation distances and transmit angles. This section attempts the modelling of the signal variations due to diffused scattering and presents a step-by-step procedure in order to model the received signal strength at various received angles. The generation of diffused components is based on the measurements described in Section A.4.2, where two different surface types are considered: the concrete pillar (smooth surface) and the red-stone wall (very rough surface). The steps for generating multipath components due to reflections from building materials are the following: 1) Select the surface type. Surface type 1 represents very rough surface (i.e. for this example the red-stone wall). Surface type 2 represents very smooth surface (i.e. in this study concrete pillar) 2) Select deterministic time based on the total distance Tx-wall-Rx. 3) Set the the angle of incidence θ i and receiving angle. 4) Identify θ, which is the difference between the specular angle and the receiving angle. 5) Evaluate the power of the specular component (i.e. θ =0) based on a superposition of a deterministic component based on the Fresnel reflection coefficient, and a Nakagami-m RV with m-parameter depending on the surface type. 6) To find the diffused component for a specific θ, we assume symmetric power decay on both sides of incident angle (specular angle). The respective power is modelled as follows: P( θ ) = Sαe β θ, (3-30) with θ > 0. S is given by a lognormal distribution with mean 1, so that the average power follows the above equation (i.e. S~Ln(1, σ 2 )). Parameter S describes the variations of data around the mean exponential decay. It is similar to shadowing in the standard path-loss model. Therefore, ln(s) should follow a normal distribution with parameters: ln(s) N~Ln(0, σ 2 ), as shown in Figure 3.45 below. The parameters σ and β depend on the type of surface, the angle of incidence θ i and the distance from the wall. Exemplary values for an angle of 45 0 and two separation distances are provided in Table Table 3.14: Modelling parameters for the diffused scattering path for the case of concrete pillar and red stone wall at Surface Type Distance : 4 meters Distance : 6 meters β σ β σ Concrete wall Red stone wall mmmagic Public 69

90 Figure 3.44: Modelling of exponential power decay with respect to angle. Figure 3.45: Modelling of variations (S) around the mean exponential decay model. Figure 3.44 depicts the received signal degradation for the red stone wall as a function of angle 45 0 transmitter angle 4 meters distance. As illustrated in this figure, the signal strength can be sufficiently presented with an exponential decrease model, as long as the variations around the mean are further modelled with the log-normal distribution. This trend is demonstrated in Figure 3.45, where both experimental and modelled values of the parameter S are shown. The presented modelling approach has been evaluated for two distances (4 and 6 m) for one particular angle at However, the methodology can be generalized for all angles and distances based on available measurements. As shown previously, diffused scattering depends on the transmit angle, the antenna separation and the surface type. Therefore, future work will focus on the investigation of the impact of the variables on the received signal strength. 3.8 Building penetration loss The mmmagic building entry loss model is based on the corresponding 3GPP model [3GPP38.901] with a couple of additions described in Section In order to provide a credible model, it has been calibrated based on the measurement data presented in [mmmagic mmmagic Public 70

91 deliverable D2.1] for the type 0 building ( old low loss building). The fitted parameter values for (4-2) are as follows: β = 0 L el = 20 abs (θ / 90º), elevation angle loss L az = 0 (db) as the angle of incidence of the measurements was close to perpendicular in the measurements θ = 0 (deg) as the angle of incidence was close to perpendicular in the measurements α = 1 (db/m), indoor loss coefficient d room = 2 (m) room size k σ = 0.08 (db/ghz) A comparison between the fitted model and measurement data is shown in Figure In general, the agreement with the measurement data and the model is good. However, a disagreement for 5.8 GHz may be observed. This is probably due to an effect of the specific multi-layer glass design of the windows, which results in the window loss fluctuating over frequency. At this point it is not known if the fluctuations might be coherent amongst most of the actual window types. If so, a systematic fluctuation may be accounted for in further future extensions of the proposed model. The results from Orange O2I measurement campaign at 3, 10, 17 and 60 GHz are compared to the 3GPP TR penetration loss model for frequency above 6 GHz. The parameters for such a model are summarized in Table 3.15 and Table Table 3.15: 3GPP Material penetration losses Materials Losses [db] Standard multi-pane glass IIR glass Concrete Note : f is in GHz L glass L IIRglass L concrete 5 4 Table 3.16: 3GPP O2I penetration losses f f f Penetration Loss Models Low-losss High-loss Loss through external wall [db] L 5 10 log glass L concrete L 5 10 log IIRglass L concrete Note : d2d-in is in m Indoor loss [db] Standard deviation [db] 0.5 d2d-in d2d-in 6.5 mmmagic Public 71

92 CDF CDF CDF Document: H2020-ICT mmMAGIC/D2.2 0m < d in < 4m 4m < d in < 7m Excess Loss [db] Excess Loss [db] 7m < d in < 15m Excess Loss [db] Figure 3.46: Cumulative distribution functions (CDFs) of loss in excesses of free space loss (building entry loss) for the calibrated model compared with measurement data. It can be noticed in the O2I penetration loss model that an additional 5 db is added to the external wall loss to account for non-perpendicular incidence. Since most of the measurements were performed with an incidence angle close to 90, the 3GPP model is shifted 5 db downward in the next figures to allow for a more consistent comparison. First, in Figure 3.47, the material penetration loss values provided by the 3GPP model from Table 3.15 is compared to Orange measurement (Meas.) results. For that purpose, the measurements performed at RX positions behind windows, where the attenuation is directly due to the window glass materials, are considered. R21 and R23, located in BR, are the RX positions chosen to estimate the attenuation caused by the non-coated windows while R14 and R17, located in Fl, are the ones used to evaluate the losses through the coated windows. mmmagic Public 72

93 Material Losses [db] GPP Standard multi-pane glass 3GPP IIR glass Meas. 2-layer non-coated windows Meas. 2-layer coated windows Frequency [GHz] Figure 3.47: Material penetration losses In Figure 3.47, a good agreement is observed between the measurements and the 3GPP model regarding the increasing behavior of the attenuation with the frequency for high-loss materials. However, for low-loss materials, the measurements exhibit an almost constant attenuation value of about 5 db regardless of the frequency. This disagrees with the increasing behavior of the attenuation values, up to 14 db at 60 GHz, provided by the 3GPP model. In order to compare the 3GPP O2I penetration loss model from Table 3.16 with Orange measurement (Meas.) results, the measurements performed in BR and Fl are considered for the lowloss and high-loss models respectively. The attenuation values provided by the model and the measurements are shown Figure 3.48 and Figure Penetration Losses [db] Low attenuation 3GPP model 10 GHz 3GPP model 17 GHz 3GPP model 60 GHz Meas. 10 GHz Meas. 17 GHz Meas. 60 GHz Indoor Distance [m] Figure 3.48: O2I penetration losses for low-loss model mmmagic Public 73

94 Penetration Losses [db] High attenuation 3GPP model 10 GHz 3GPP model 17 GHz 3GPP model 60 GHz Meas. 10 GHz Meas. 17 GHz Meas. 60 GHz Indoor Distance [m] Figure 3.49: O2I penetration losses for high-loss model In Fl where the windows are composed with high-loss materials, the measured attenuation values do not vary a lot (about 5 db) from one RX position to another. In fact, since the attenuation of these windows is close to the attenuation of the concrete, the building behaves as an attenuator whose value slightly depends on the distance. For the low-loss window materials in BR, the penetration loss is strongly dependent on the optical visibility condition between the TX and the RX. This explains the much greater variation range (up to 20 db) of the measured attenuation values. The measurements are more or less consistent with the 3GPP O2I penetration loss model and follow the same conclusions made upon Figure The largest mismatch is recorded at 60 GHz where the 3GPP attenuation values are significantly greater than the ones computed from the measurements. For instance, the mean values provided by the 3GPP low-loss model correspond with the upper values computed from the measurements in BR. This is partly due to the L glass parameter modelling. In the 3GPP model, it is considered to be increasing with the frequency whereas Orange measurements suggest a single constant attenuation of about 5 db for this parameter for all frequencies. mmmagic Public 74

95 4 Channel model This section describes the stochastic model, i.e. a geometry-based stochastic channel model (GSCM) developed in conjunction with the 3GPP [3GPP38.900, 3GPP38.901] and the QuaD- RiGa [JRB+14] model, including some of those models components and further enhancements. The model is focused on the frequency range GHz. However, since the measurements and evaluations were extended to lower frequencies, it is applicable down to 2 GHz. Based on the wideband measurements, it supports a bandwidth up to 10% of the centre frequency but no larger than 2 GHz. The mmmagic project was actively working on channel modelling concepts throughout the 3GPP and ITU standardization phases for 5G channel models and various new modelling concepts were implemented in the open-source QuaDRiGa channel model. Many ideas (namely the ground reflection model, the blockage model, the method to realize spatial consistency and the extended O2I penetration loss model) from mmmagic found their way into 3GPP, ITU and QuaDRiGa due to the fruitful discussions and collaborations, both in the project and the standardization groups. Overall, 20 channel-related contributions were made to 3GPP [3GPP160846] [3GPP ] and ITU-R [ITUR-3J22E] [ITUR-5D338E]. Therefore, references to these other models do not represent the state of the art that was found at the beginning of the project, but rather include the mmmagic achievements. An overview of the model components is given in Figure 4.1. The basic assumption is that building/scene models are not available and that the propagation paths are generated fully stochastically. It is possible to apply various antenna models, e.g. the antenna architectures developed in [mmmagic-d5.1]. The channel model is therefore independent of the antennas. The output of the channel model is a complex-valued MIMO channel impulse responses per user link depending on the user location. Major advancements have been made concerning specific propagation characteristics for mmwave channels. Those are the detailed characterizations of propagation scenarios based on measurements and simulations, a model for the ground reflection, extensions for large bandwidths and large array antennas, spatial consistency, blockage and improved O2I penetration loss modelling. Table 4.1 gives an overview of the model components, their realization in 3GPP and QuaDRiGa, and the proposed mmmagic approach. As mentioned above, several mmmagic concepts have been successfully adopted by 3GPP. Further components are essential for a complete model to work, but they have not been studied by mmmagic (e.g. the coordinate system, interparameter correlation model, polarization model, etc.). These components of the mmmagic model are either identical to the 3GPP proposal or are taken from QuaDRiGa. The model consists of 13 baseline components and 6 additional features. The latter are not absolutely necessary to generate the channel coefficients, but they offer decisive improvements in the modelling of mm-wave channel characteristics. mmmagic Public 75

96 Figure 4.1: Channel model components Table 4.1: Model components Model Component 3GPP (v14.0.0) QuaDRiGa (v2.0) mmmagic proposal Coordinate system Global coordinate system: Cartesian coordinates (in units of meters) with arbitrary origin. Local coordinate system: Spheric coordinates (elevation θ = 0 points to the zenith, θ = 90 points to the horizon) Global coordinate system: Same as 3GPP. Local coordinate system: Geographic coordinates (elevation θ = 90 points to the zenith, θ = 0 points to the horizon) Same as 3GPP (Reference implementation in QuaD- RiGa will use geographic coordinates) (see Sec. 4.1) Antenna model Uniform rectangular panel array Freely configurable array antenna support (including 3GPP antennas). Supports import of measured and simulated far-field antenna patterns. Same as QuaDRiGa (see Sec. 4.2) Propagation-scenario description (Section 4.3) mmmagic Public 76

97 Model Component Supported scenarios 3GPP (v14.0.0) QuaDRiGa (v2.0) mmmagic proposal UMa UMi street canyon Rural Macro (up to 7 GHz) Indoor office 3GPP and mmmagic parameters are supported. Final parameters are based on measurements in mmmagic and 3GPP results. UMi outdoor Indoor Pathloss model Provided for supported scenarios. 3GPP and mmmagic parameters are supported. Final PL models based on measurements in mmmagic and literature results. LOS probability model Provided for supported scenarios. Same as 3GPP (for 3GPP scenarios) Same as 3GPP O2I penetration loss model Provided for supported scenarios. 3GPP and mmmagic models are implemented. 3GPP approach with extensions (see Section 4.3.1) Large-scale fading model (Section 4.4) Autocorrelation model Only requirements are specified. Implementation details are left open. Support for 3D spatial consistency of LSPs (and SSF). Implementation is based on sum-of-sinusoids method. Same as 3GPP Only requirements are specified. Implementation details are left open. (see Sec ) Inter-parameter correlation model Ordered Cholesky decomposition (not validated) Matrix-Square root (validated, destroys decorrelation when high inter-parameter correlations are used) Same as QuaDRiGa (see Sec ) Small-scale fading model (Section 4.5) Path-powers and delays Randomized delays and clusters, single slope exponential power delay profile, scaled to match KF. Same as 3GPP Additional normalization step to enforce same DS as in the large-scale-fading (LSF) model. Same as QuaDRiGa (see Sec ) Departure and arrival angles Fitted to a wrapped Gaussian distribution based on path powers, scaled to match KF. Randomized angles (Gaussian distributed), normalization step to enforce same AS as in the LSF model. Same as QuaDRiGa (see Sec ) Sub-path mapping Baseline model: Mapping to 20 equal-power sub-paths with fixed angleoffset and identical delays. Large-bandwidth extension (optional): Individual delays for each sub-path, angles are drawn from a continuous uniform distribution, unequal subpath powers. 3GPP baseline model and mmmagic method are implemented. Additional spherical-wave extension to support larger antenna arrays. New method (not adopted by 3GPP) (see Sec ) mmmagic Public 77

98 Model Component 3GPP (v14.0.0) QuaDRiGa (v2.0) mmmagic proposal Intra-Cluster DS Defined for the two strongest clusters (baseline model only) 3GPP and mmmagic method are implemented. Part of the sub-path mapping (see Section 4.5.3) Polarization model Random phases scaled by XPR (ignores circular polarization) Successive linear transformations (leads to correct XPR also for circular polarization, validated) Same as QuaDRiGa (see Sec ) Additional features (Section 4.6) Oxygen absorption Optional for GHz band Can be included in customized parameter-tables for 60 GHz band Same as 3GPP (not investigated by mmmagic) Ground reflection Explicit reflection path added. No adjustment of LSPs. Same as mmmagic model More complex (complete) model compared to 3GPP (see Sec ) Spatial consistency Decorrelation distance model is used to spatially correlate all random variables in the model. Implementation details are left open. Spatial consistency is implemented based on the sum-ofsinusoids method. New proposal (see Section 4.6.2) UT mobility modelling Supports linear movement on horizontal plane. Initial delays are updated using movement direction. Powers and angles are updated using the 3GPP spatial consistency model. Short-term mobility (drifting): Based on explicit cluster positions (single or dual-bounce model) and geometric updates of the path-delays and angles. Long-term mobility: Same as QuaDRiGa short-term mobility Implements birth / death process of scattering clusters. Blockage (Option A) Stochastic method for capturing human and vehicular blocking. Defines blocking regions based on a MT-centric model. Impact on LSPs is not corrected. Spatial consistency model required. (Option B) Geometric method based on rectangular screens. Attenuation based on knife edge diffraction. No placement rules for blocking screens given. Not supported. New proposal (see Section 4.6.3) mmmagic Public 78

99 Model Component 3GPP (v14.0.0) QuaDRiGa (v2.0) mmmagic proposal Multi-frequency simulations An alternative channel generation method is proposed. Path delay and angles are generated for an anchor frequency. Path powers are scaled for different frequencies. 3GPP method is implemented. Same as 3GPP 4.1 Coordinate system The coordinate system is defined with respect to a spherical coordinate system where the zenith angle θ = 0 points to the zenith and θ = 90 points to the horizon. In contrary, QuaDRiGa [JRB+14] uses the geographic coordinate system where the elevation angle θ = 90 points to the zenith and θ = 0 points to the horizon. The conversion between the two is straight forward. To avoid confusion between the coordinate systems, 3GPP uses the term zenith, i.e. zenith angle of arrival (ZoA), zenith angle of departure (ZoD), zenith angle spread of arrival (ZSA), zenith angle spread of departure (ZSD), while QuaDRiGa uses the term elevation, i.e. elevation angle of arrival (EoA), elevation angle of departure (EoD), elevation angle spread of arrival (ESA), elevation angle spread of departure (ESD). 4.2 Antenna model The antenna is defined by its directional response, also known as the radiation pattern. The spherical antenna coordinate system has two angles and two poles. The elevation angle θ is measured relative to the pole axis. A complete circle will go through each of the two poles. The azimuth angle ϕ moves around the pole. The electric field is resolved onto three vectors that are aligned to each of the three spherical unit vectors êθ, êϕ and êr of the coordinate system. In this representation, êr is aligned with the propagation direction of a path. In the far field of an antenna, there is no field in this direction. Thus, the radiation pattern consists of two components, one is aligned with êθ and another is aligned with êϕ. It is usually described by a 2-element vector F(θ, φ) = ( F[θ] (θ, φ) F [φ] (θ, φ) ). (4-1) This description in the so-called polar spherical polarization basis (see [Lud73, GMP07]) can be generated by any common antenna modelling software and is also supported by all common antenna measurement tools. 4.3 Propagation-scenario description The parameters for the mmmagic channel model are summarized in Table 4.2. They include the combined propagation scenarios UMi Outdoor (Street Canyon and Open Square), Indoor (Office and Airport) and O2I. The values were derived by combining the results from all mmmagic measurement and simulation campaigns. The underlying parameter tables for individual measurement and simulation scenarios are given in Annex D. Furthermore, the delay spread results from the Aalto measurements have been incorporated. They can be found in from Table 3.4. For the fusion of the results, the methodology from Section 3.1 was adopted, but with the simplification of equal weighting factors w i = 1. Single frequency campaigns were taken into account in the average intercept point β by calculating their individual intercept point β i based on the average slope α derived from the multi-frequency campaigns. In cases where no values are available from mmmagic, the values from 3GPP [3GPP38.901] are included for the sake of completeness. These values are printed in italics. mmmagic Public 79

100 Path loss (PL) [db] Scenarios Delay spread (DS) lgds=log 10(DS/1s) AOD spread (ASD) lgasd=log 10(ASD/1 ) AOA spread (ASA) lgasa=log 10(ASA/1 ) ZOD spread (ZSD) lgzsd=log 10(ZSD/1 ) ZOA spread (ZSA) lgzsa=log 10(ZSA/1 ) Shadow fading (SF) [db] K-factor (K) [db] Cross-Correlations Cross-Correlations Table 4.2: Combined mmmagic model parameter table PL lgds lgds lgasd UMi Outdoor Indoor LOS NLOS LOS NLOS PL UMi-LOS = 19.2 log 10(d 3D) log 10(f c) 0.11 log 10(1+f c) log 10(1+f c) log 10(1+f c) PL UMi_NLOS = 45.0 log 10(d 3D) log 10(f c) 0.01 log 10(1+ f c) log 10(1+ f c) lgasd lgasa 0.18 log 10(1+ f c) lgasa lgzsd 0.16 log 10(1+ f c) lgzsd lgzsa lgzsa 0.12 log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) PL Ind--OS = 13.8 log 10(d 3D) log 10(f c) 0.13 log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) PL Ind-NLOS = max(pl Ind-LOS, PL' Ind-NLOS), with PL' Ind-NLOS = 36.9 log 10(d 3D) log 10(f c) 0.01 log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) mmmagic Public 80 O2I see 3GPP [3GPP38.901] with extension in Section log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) log 10(1+ f c) N/A 0.36 N/A log 10(1+f c) log 10(1+f c) SF K 4.60 log 10(1+ f c) N/A K 5.86 N/A 5.0 log 10(1+ f c) log 10(1+ f c) ASD vs DS ASA vs DS ASA vs SF ASD vs SF DS vs SF ASD vs ASA ASD vs 0.10 N/A -0.6 N/A N/A ASA vs 0.20 N/A -0.5 N/A N/A DS vs 0.33 N/A -0.5 N/A N/A SF vs 0.04 N/A 0.30 N/A N/A ZSD vs SF ZSA vs SF ZSD vs K 0.20 N/A 0.30 N/A N/A ZSA vs K 0.33 N/A 0.30 N/A N/A ZSD vs DS ZSA vs DS ZSD vs ASD ZSA vs ASD ZSD vs ASA N/A N/A N/A N/A

101 ZSA vs ASA ZSD vs ZSA Delay scaling parameter r XPR [db] XPR XPR Number of clusters N Number of rays per cluster M Cluster DS (c DS ) in [ns] N/A 11 Cluster ASD (c ASD ) in [deg] Cluster ASA (c ASA ) in [deg] Cluster ZSA (c ZSA ) in [deg] Cluster ZSD (c ZSD ) in [deg] modelled as c ZSD = μ lgzsd following [3GPP38.901] Per cluster shadowing std [db] Correlation distance in the horizontal plane [m] DS 8.20 log 10(1+ f c) ASD ASA SF N/A 5 N/A N/A ZSA ZSD log 10(1+ f c) log 10(1+ f c) O2I penetration loss N/A N/A N/A N/A fc is carrier frequency in GHz log 10(1+f c) Building penetration loss The mmmagic building penetration loss model is basically the same as the corresponding model agreed in 3GPP [3GPP38.901]. There are however a couple of refinements introduced: 1. A lognormal frequency dependent spread is introduced 2. A term for elevation angle dependence is introduced With these additions the model is formulated as follows: BEL (db) = L bpl + L el + L az + L in + N(0, σ db ) (4-2) where L bpl = 10 log 10 ( (0.3(1 β) + 0.7β) 10 L1/10 + (0.7(1 β) + 0.3β) 10 L2/10 ) L 1 = 20β + (2 + β)( f), window loss model L 2 = 5 + 4f, concrete/brick wall loss model f = frequency (GHz) β = {0, 1} representing building class (1=thermal efficient, 2=traditional) L el = 20 abs (θ / 90º), elevation angle loss (value aligned with ITU-R) L az = 5 db effective additional loss for azimuth angle relative to exterior wall 1 mmmagic Public 81

102 θ = Elevation angle relative to exterior wall (degrees). For NLOS θ is relative to the local clutter height. L in = α max (0, d in d room ) indoor loss or penetration depth loss α = [ , uniform dist.] (db/m), indoor loss coefficient d room = [2-4, uniform dist.] (m) room size d in = indoor distance (m) N(0, σ db ) = log-normal distribution and and L el = 20 abs (θ / 90º) (4-3) σ db = 4 + k σ f (db), (4-4) where k σ is a frequency dependent constant (db/ghz) with a fitted value of k σ = 0.08 building class 0 (see Section 3.8). for This model uses a distribution of glass and concrete or brick wall materials that is set to 30% standard two-pane glass and 70% concrete/bricks for building class 0, while for building class 1 the model is set to 70% standard three-pane IR reflective glass with metal coating and 30% concrete/bricks. For frequencies above 6 GHz the dominant component is the standard glass windows for building class Large-scale fading model The first part of the channel model deals with the large-scale parameters (LSPs). Hence, it can also be considered as the large-scale fading (LSF) model. A subsequent small-scale-fading (SSF) model then generates individual scattering clusters for each MT. The positions of the scattering clusters are based on seven LSPs: 1. RMS delay spread (DS) 2. Ricean K-factor (KF) 3. Shadow fading (SF) 4. Azimuth spread of departure (ASD) 5. Azimuth spread of arrival (ASA) 6. Elevation spread of departure (ESD) 7. Elevation spread of arrival (ESA) The granularity of each LSP can be described on three levels: the propagation scenario level, the link level, and the path level. Propagation scenario level Normally, LSPs are assumed to be log-normal distributed. For example, the median log-normal delay spread DSμ in an urban cellular scenario is 6.89 log 10(s) which corresponds to a DS of στ = 128 ns. With a standard deviation of DSσ = 0.5, typical values may vary in between 40 and 407 ns. The same principle applies for the other six LSPs. The decorrelation distance (e.g., 40 m) describes the distance-dependent correlation of the LSP. If, e.g., two mobile terminals in the above example are 40 m apart from each other, their DS is correlated with a correlation coefficient of e 1 = Additionally, all LSPs are cross-correlated. A typical example is the dependence of the ASA on the KF. With a large KF (e.g., 10 db), a significant amount of energy comes from a sin- mmmagic Public 82

103 gle direction. Thus, the ASA gets smaller which leads to a negative correlation between the ASA and the KF. Link level A MT at a specific position is assigned to a propagation scenario. Depending on the position and the scenario, it experiences a radio channel which is determined by the specific values of the seven LSPs. Due to the autocorrelation properties, small distances between users in the same scenario also lead to high correlations in the channel statistics, e.g., a second terminal right next to the current user will experience a similar DS. The second granularity level thus contains the specific values of the LSPs for each user position. Generating those values can be seen as going from the scenario-wide distribution of LSPs to virtual measurement -values for each MT. Path Level Finally, the individual components of the CIR are calculated. This procedure takes the values of the LSPs into account and calculates the path-powers and the path-delays of the MPCs. The detailed procedure for this is described in the following sections Autocorrelation model The log-normal fading in the logarithmic scale is used for all large-scale parameters, i.e., the shadow fading (SF) around the mean path loss PL (in db), the delay-spread (DS), the Ricean K-factor (KF) and the angular spreads (ASD, ASA, ESD, and ESA). Log-normal fading is characterized by a Gaussian distribution with zero mean and standard deviation. Due to the slow fading process versus distance d, adjacent fading values are correlated. Its normalized autocorrelation function R(d) can be described with sufficient accuracy by the exponential function [ITUR-P1816] R(d) = exp ( d ), (4-5) d corr with the correlation length d corr being dependent on the environment. In the spatial consistency procedure, the cluster specific random variables are also correlated following the exponential function with respect to correlation distances in the two dimensional horizontal plane Inter-parameter correlation model In order to account for the inter-lsp correlation, a 7 7 matrix X is assembled containing all cross-correlation values ρ between each two LSPs. 1 ρ DS,KF ρ DS,SF ρ DS,ASD ρ DS,ASA ρ DS,ESD ρ DS,ESA ρ KF,DS 1 ρ KF,SF ρ KF,ASD ρ KF,ASA ρ KF,ESD ρ KF,ESA ρ SF,DS ρ SF,KF 1 ρ SF,ASD ρ SF,ASA ρ SF,ESD ρ SF,ESA X = ρ ASD,DS ρ ASD,KF ρ ASD,SF 1 ρ ASD,ASA ρ ASD,ESD ρ ASD,ESA ρ ASA,DS ρ ASA,KF ρ ASA,SF ρ ASA,ASD 1 ρ ASA,ESD ρ ASA,ESA ρ ESD,DS ρ ESD,KF ρ ESD,SF ρ ESD,ASD ρ ESD,ASA 1 ρ ESD,ESA ( ρ ESA,DS ρ ESA,KF ρ ESA,SF ρ ESA,ASD ρ ESA,ASA ρ ESA,ESD 1 ) (4-6) Then, the square-root matrix X 1 2 is calculated from X. In order to calculate the matrix-squareroot, X must be positive definite to get a unique, real-valued solution. The matrix X 1 2 is then multiplied with the seven values obtained from the maps at the sample point (y,x). mmmagic Public 83

104 The procedure is repeated for all MTs. 4.5 Small-scale fading model Initial delays and normalized path powers B DS A DS ( ) = X 1/2 ( ) (4-7) B EsD A EsD Initial delays are drawn randomly from a scenario-dependent delay distribution as τ l [1] = r τ σ τ ln (X l ), (4-8) where the index l denotes the path number, X l = U(0,1) is an uniformly distributed random variable having values between 0 and 1, στ is the initial DS from large-scale-fading model, and rτ is a proportionality factor (see [KMH + 07]). The term rτ was introduced in [3GPP25.996] because στ is influenced by both the delays τl and the powers Pl; rτ is usually calculated from measurement data. Next, the delays are adjusted such that the first delay is set to zero and then they are sorted τ l [2] = sort{τ l [1] min (τ l [1] )}. (4-9) The NLOS path powers are drawn from a single slope exponential power delay profile (PDP) depending on the DS στ and a random component Z l = N(0, ξ 2 ) [KMH + 07]. The term ζ is a scenario-dependent coefficient emulating an additional shadowing process. It is obtained from measurements. P [1] l = exp ( τ [2] l r τ 1 σ r τ ) 10 Zl 10 (4-10) τ Here, the power values are given in units of Watts. The power of the first path is further scaled according to the initial KF from the map and the path powers are normalized so that their sum power is one Watt. L P [2] [1] 1 = K P l l=2 ; P [2] [1] 2 l = P 2 l L ; P l = P [2] [2] l / P l l=1 (4-11) The scaling of the powers by the KF changes the DS. Hence, in the last step, this is corrected by calculating the actual delay spread using the scaled powers and normalizing the delays in order to obtain the desired RMS delay spread in the PDP. The DS after applying (4-11) is L l=1 L l=1 2 σ [actual] 2 τ = P l τ l ( P l τ l ). (4-12) This value differs from the initial value στ that was provided by the parameter map. Hence, the delays (4-9) are scaled such that the correct delay spread can be calculated from the generated path-delay and path-powers. τ l = σ τ [actual] σ τ [2] l τ (4-13) The last steps are different from other models. The WINNER [KMH + 07] and 3GPP-3D model [3GPP36.873] scale the delays with an empiric formula that corrects the delays to reduce the effect of a high KF. However, due to the random variables in (4-8) and (4-10) the resulting DS is always different from the value in the map. The new method ensures that scattering clusters mmmagic Public 84

105 are distributed in a way that the DS calculated from the MPCs is exactly the same as the DS from the large scale fading model. In the following section, the departure and arrival directions of each MPC are generated. Those are the combined with the delays in order to calculate the 3D positions of the scattering clusters Departure and arrival angles The 2-D WINNER model [KMH + 07] introduced the AoD, ϕd, and the AoA, ϕa. In 3-D coordinates, the ZoD, θd, and the ZoA, θa are also needed. The angles share similar calculation methods but have different angular spreads (ASs) σϕ a, σϕ d, σθ a, and σθ d. The individual departure and arrival angles of the MPCs are generated by first assigning random angles to the already known path powers from the previous step. In order to obtain the correct ASs, a scaling operation is used to readjust the angles. This approach is different from the WINNER and 3GPP-3D model where the angles are mapped to the already known powers using a wrapped Gaussian distribution [PMF97]. As for the DS, the intention behind the new method is to achieve the best possible match between the ASs that can be calculated from the MPCs and the values from the LSF model. Azimuth angles A random list of angles is created for the NLOS paths from a Gaussian normal distribution with zero-mean and a variance corresponding to the given AS from the LSP maps. The LOS angle is defined to be zero. φ [1] [1] 1 = 0 and φ 2...L N(0, σ φ 2 ) (4-14) The so obtained angles need to be mapped to the interval [ π;π]. This is done by a modulo operation which wraps the angles around the unit circle φ l [2] = (φ l [1] + π mod 2π) π. (4-15) The relationship between path powers and angles is random. Hence, the resulting AS is undefined. In the next step, the actual AS is calculated. This requires calculating the power-weighted mean angle φ because the angles are distributed on a sphere and, therefore, the AS depends on the reference angle, i.e., the definition of where 0 is. The angle φ is subtracted from the angles ϕl [2] to map the mean angle to 0. Then, and the wrapping around the unit circle (modulo operation) is applied a second time. The AS then follows from φ = arg [ P l exp(jφ [2] l )], φ l [ ] σ φ [actual] L l=1 = (φ l [2] φ + π mod 2π) π, L = P l (φ [ ] l ) 2 l=1 L [ ] ( P l φ l ). Now, with σϕ being the initial AS from the map, the angles ϕl [2] are updated to l=1 2 (4-16) φ l [3] = σ φ [actual] σ φ [2] l. φ (4-17) If σϕ is larger than σϕ [actual], then (4-15) needs to be applied again in order to account for the periodicity of the angles. However, this could lower the AS instead of increasing it as intended by the scaling operation. A solution is to create new angles with a bias to the negative side of the circle. mmmagic Public 85

106 φ [3] l,if φ [3] l < π; φ [4] l = { N (π, π2 4 ), otherwise (4-18) φ l [5] = (φ l [4] + π mod 2π) π. (4-19) However, this changes the AS and the calculations (4-16) to (4-18) need to be repeated iteratively until the actual AS σϕ [actual] converges either to the given value σϕ or a maximum value. Finally, the LOS direction is applied. Zenith angles φ l = φ l [5] + φ LOS (4-20) Since the zenith angles can only have values in between 0 and π, the calculation method differs from the method used for the azimuth angles. As for the azimuth angles, a random list of angles is created for the NLOS paths from a Gaussian normal distribution with zero-mean and a variance corresponding to the given AS from the LSF model. θ 1 [1] = 0 and [1] θ 2...L N(0, σ θ 2 ) (4-21) θ l [2] = θ l [1] + θ LOS (4-22) The so obtained angles need to be mapped to the interval [ 0; π]. This is done by a modulo operation which wraps the angles around the unit circle and an additional operation that mirrors the angles at the poles of the unit sphere, e.g., a zenith angle of 181 is mapped to 179, 182 to 178, and so on. This ensures that the previously calculated azimuth angles are not changed. As for the azimuth angles, the resulting AS is undefined. Hence, the actual elevation spread σθ [actual] is calculated using (4-16). Then, with σθ being the initial elevation spread from the LSF model, the angles θl are updated. Since the angles should be distributed around the LOS direction, θ LOS needs to be subtracted before scaling the angles and added again after scaling them Mapping of paths to sub-paths The subpath generation procedure consists of the generation of delays, powers, and angles of subpaths. The mmmagic approach generates delays and powers of subpaths by generalizing the generation procedure of path delays and powers in the new 3GPP channel model [3GPP38.900]. The method of equal area (MEA) is used to generate subpath angles in the mmmagic channel model. A comparison between the subpath generation methods of the mmmagic and the 3GPP channel models is summarized in Table 4.3. mmmagic Public 86

107 Table 4.3: Comparison between the subpath generation method of mmmagic and the path/subpath generation method 3GPP channel models Delay Power AOA/AOD/ZOA Correlation between delay and angles Legacy 3GPP channel model [3GPP36.873] Constant (only in the first strongest path) Equal Laplacian PAS Sampling method unknown Subpath generation New 3GPP channel model (additional feature) [3GPP38.900] mmmagic approach Path generation Both new and legacy 3GPP channel model [3GPP38.900] [3GPP36.873] Uniform r.v.s Exponential r.v.s Exponential r.v.s Monte Carlo Sampling Uniformly distributed r.v.s Mapping from delay Laplacian PAS MEA Sampling Mapping from delay Wrapped Gaussian r.v.s No No No Yes Generate delays of subpaths Delay offset of rays are drawn randomly from the delay distribution. τ l,m = r μ c DS ln(x l,m ), (4-23) where c DS is subpath delay spread, r μ is the delay distribution proportionality factor, X l,m ~ uniform (0,1), and subpath index m = 1,, M. With uniform delay distribution the delay offset values τ l,m are drawn from the corresponding range. Normalize the delay offset by subtracting the minimum delay offset and sort the normalized delay offset to ascending order: The delays of subpath are given as Generate powers of subpath The received power levels of subpaths are determined by Subpath power needs to be normalized as τ l,m = sort (τ l,m min(τ l,m )). (4-24) τ l,m = τ l + τ l,m. (4-25) r μ 1 P l,m = exp ( τ l,m ). (4-26) r μ c DS P l,m = P l ( P l,m P l,m M m=1 ). (4-27) Generate AOAs/AODs of subpaths AOA of each subpath φ n,m,aoa is defined by the sum of the AOA of its associate path s and an azimuth angular offset, i.e., φ l,m,aoa = φ l,m,aoa + φ l,aoa. (4-28) Azimuth angular offsets (assuming Laplacian distribution) are computed using the MEA as mmmagic Public 87

108 φ l,m,aoa = c ASA { 2 c ASA 2 ln (2m M ), m < M/2 m) ln (2(M ), m M/2, M (4-29) where c ASA is path azimuth angle spread. The generation of AODs (φ l,m,aod ) follows a procedure identical to AOA as described above. Generate ZOAs/ZODs of subpaths ZOA of each subpath θ l,m,zoa is defined by the sum of the ZOA of its associate path s and a zenith angular offset, i.e., θ l,m,zoa = θ l,m,zoa + θ l,zoa. (4-30) Zenith angular offsets (assuming Laplacian distribution) are computed using the MEA as θ l,m,zoa = c ZSA { 2 c ZSA 2 0.5) ln (2(m ), m < M/2 M m + 0.5) ln (2(M ), m M/2 M (4-31) where c ZSA is path elevation angle spread. The generation of ZODs is (θ l,m,zod ) fully aligned with the new 3GPP channel model [3GPP38.900] Polarization model The complex-valued amplitude of a path between a transmit antenna and a receive antenna is g = P F r (θ a, φ a ) T M F t (θ d, φ d ) e j2π λ d (4-32) where F r and F t describe the polarimetric antenna responses at the receiver and the transmitter, respectively. P is the power of the path, λ is the wavelength, d is the length of the path determined by the path delay, (θ a, φ a ) are the arrival and (θ d, φ d ) the departure angles that were calculated in the previous step. M is the 2 2 polarization coupling matrix. LOS polarization 3GPP [3GPP36.873] proposes a model for the LOS and NLOS polarization, where the LOS polarization M LOS is M LOS = exp (jψ LOS ) ( ) (4-33) In this equation, the 2 2 matrix can be interpreted as a mirror operation that transforms the outgoing direction of a path at the transmitter into an incoming direction at the receiver. The phase of the LOS path is with λ being the carrier frequency wavelength. ψ LOS = 2π λ (h BS h MT ) d 2D (4-34) NLOS polarization The cross polarization ratio (XPR) is calculated from measurement data. A log-normal distribution is fitted to the measurement results having the average XPRμ and the standard deviation (STD) XPRσ 2. When those parameters are calculated from measured data, they are usually averaged over different propagations paths. Thus, the XPR value from the parameter tables is a LSP with a scenario-dependent distribution, i.e., it depends on the positions of the MT. However, mmmagic Public 88

109 [db] here, the values XPR l,m for individual MPCs are needed. Those are calculated in two steps. [LSP] First, a value μ XPR is drawn from [LSP] μ XPR = 2 N(μXPR, σ XPR ). (4-35) This value represents the average XPR over all MPCs at the receiver positions. Then, in a second step, the XPR for the individual MPCs is drawn using XPR μ [LSP] instead of XPRμ. This maintains the original spread XPRσ in the generated channel coefficients. [db] [LSP] XPR l,m = N (μxpr, 2 σxpr ) (4-36) The model for the NLOS polarization maps the XPR to two matrices. The first matrix adjusts the linear polarization, the second matrix scales the phase offset and therefore models the elliptic component. Linear component During scattering, the linear polarization of a MPC might change. For example, a transmit antenna sends a vertically polarized wave which only oscillates in the êθ direction. Then, a receiver might detect a wave that oscillates in both the êθ direction and êϕ direction because scattering changed the polarization angle while the phases of the êθ and êϕ components remain unchanged. In other words, a linear polarized wave stays linear polarized. In order to model this polarization change, the XPR of a path (4-36) is mapped to a rotation matrix. This was also suggested by [ZRP + 05]. [linear] = ( m θθ m φθ m ) = ( cos γ l,m sin γ l,m ). (4-37) φφ sin γ l,m cos γ l,m M l,m The rotation angle γ is calculated as m θφ XPR l,m = m θθ 2 m φθ 2 = m φφ 2 m θφ 2 = (cos γ l,m) 2 (sin γ l,m ) 2 = (cot γ l,m) 2, γ l,m = arccot( XPR l,m ). (4-38) Elliptical component When channel measurements are done with circular polarized antennas such as in land-mobile satellite scenarios, there is a very clear indication that scattering alters the phase between the two polarization components which are aligned with the spherical basis-vectors in phi and theta direction. In other words, a purely left hand circular polarized (LHCP) signal can be received with a right hand circular polarized (RHCP) antenna after scattering. There might also be a transformation from linear to elliptic polarization and vice versa. This is not covered well by the existing GSCMs. The commonly used approach by 3GPP creates a random phase difference between the polarization components. As a result, all paths have a (random) elliptic polarization and there is no way to adjust the XPR for scatting of circular polarized waves. This is addressed in the new model by adding elliptic polarization using an additional Jones matrix. The phase difference between the êθ and êϕ component is obtained by a scaling matrix M [elliptic] l,m = ( exp (jκ l,m) 0 0 exp ( jκ l,m ) ). (4-39) The phase shift κ is calculated using the XPR from (4-36). κ l,m = X l,m arccot( XPR l,m ) (4-40) Xl,m { 1,1} is the positive or negative sign. In this way, the same XPR can be calculated from the channel coefficients at the output of the model when using circular polarized antennas. mmmagic Public 89

110 The influence of the scattering cluster is modelled by a combination of three operations: a scaling operation that introduces a phase shift between the vertical and horizontal component to obtain a phase difference that matches the XPR, a reflection operation, and a rotation operation to obtain the desired (linear) XPR. 4.6 Additional features Ground reflection [NLOS] M r,t,l,m,s = [linear] Ml,m ( ) M [elliptic] l,m (4-41) It is possible to add a single ground reflection (GR) that dominates the multipath effect [Gol03]. If the height of the MT is small compared to the distance between BS and MT, it will be difficult to resolve the GR in the delay domain. For example, in a typical UMi scenario with a BS height of 10 m and a MT height of 1.5 m, there are only 3.2 ns between the direct path and the reflected path given a BS-MT distance of 30 m. It would require more than 300 MHz of bandwidth to resolve both paths. In our proposed model extension, the GR path is added to the model. Its power, delay, departure and arrival angles, and polarization can be explicitly calculated as described in the remainder of this section. Figure 4.2: Illustration of the angles and vectors used for the calculations Path powers and path delays In order to incorporate the ground reflection into this existing model (see Section 4.5.1), an additional path is added having the delay τ GR and power P GR. It is common to use relative instead of absolute delays. Hence, the delay of the GR is calculated by τ GR = (h BS + h MT ) 2 + d 2 2D (h BS h MT ) d 2D, c (4-42) where c is the speed of light. The power of the reflected path depends on the reflection coefficient R, which varies depending on the polarization of the incident wave and the electromagnetic properties of the ground. P GR = R2 2 P [2] (4-43) 1 In order to obtain the correct angle and delay spreads, the power of the LOS component must be adjusted to keep the normalization (4-11) of the path powers. P LOS = (1 R2 2 ) P [2] 1 (4-44) mmmagic Public 90

111 Since the reflected path has a later delay compared to the LOS path, the delay spread (DS) is altered. This can be corrected by multiplying the NLOS delays with a scaling coefficient S τ. This coefficient is calculated from σ 2 2 τ = P GR τ GR L + S 2 [2] [2] 2 τ P l (τl ) l=2 L [2] [2] (P GR τ GR + S τ P l τl mmmagic Public 91 l=2 ) 2, (4-45) by using the initial DS σ τ that was also used to calculate the delays in (4-8). Then, the final pathdelays and path-powers are τ l = [0 τ GR S τ τ 2 [2] P l = [P LOS P GR P 2 [2] S τ τ L [2] ] (4-46) P L [2] ]. (4-47) In the next step, the departure and arrival angles are updated in a similar way to account for the GR path. Departure and arrival zenith angles As for the DS, the additional GR path changes the zenith angular spread, both at the BS and the MT. Hence, the angles of the other NLOS paths need to be corrected as well in order to achieve the given AS values σ θ d and σ θ a. Since the azimuth angles of the GR path are identical with the LOS path, only zenith angles need to be considered here. The AS can be obtained by (see Section 4.5.2) L [ ] 2 σ θ = P l (φ l ) l=1 L l=1 2 [ ] ( P l φ l ). (4-48) Unfortunately, there is no closed form expression that can be used to calculate a scaling coefficient that corrects the angles of the NLOS paths. Hence, numerical methods must be used to determine S θ. The updated angles then are θ l = [θ LOS θ GR S θ θ 2 [2] + θlos S θ θ L [2] + θlos ]. (4-49) In the next step, the polarization state of the GR path is determined. This takes the dependence of the reflection coefficient on the polarization of the incident wave into account. Polarization Due to the power scaling of the LOS path, the LOS polarization differs from (4-33). M LOS = (1 R2 1 2 ) 2 exp (jψlos) ( ) (4-50) In this equation, the normalization is removed. For the reflected path, the generally complexvalued, reflection coefficients R and R are applied to the channel coefficients. M GR = 2 R exp (jψ GR) ( R 0 0 R ) (4-51) The additional factor 2/R reverses the power scaling of the reflected path in (4-43) that was needed to correctly adjust the delay and angular spreads. The phase of the GR path is ψ GR = 2π λ (h BS + h MT ) d 2D (4-52) Note that for dielectric materials (i.e., common ground materials), the reflection coefficients generally have negative values. Hence, there is a 180 shift between the phases of the direct path

112 and the GR path in most of the cases. In the next section, the values of the reflection coefficient are discussed. Reflection coefficient The reflection coefficient is a function of the electromagnetic properties of a material. The complex-valued relative permittivity is given by σ σ ε = ε r j ε 2π f c ε r j, (4-53) 0 f c [GHz] where ε r is the relative permittivity and σ is the conductivity of the material. The reflection coefficients for the two polarizations are then calculated to [Gol03], [ITUR-P2040]: R = ε sin θr Z ε sin θ r + Z, R = sin θr Z sin θ r + Z (4-54) Z = ε cos 2 θ r (4-55) R = 0.5 R R 2 (4-56) θ r = θ d GR = arctan ( h BS + h MT d 2D ). (4-57) θ r is the angle between the ground and the reflected path (see Figure 4.2). An illustration of the magnitude of the reflection coefficient for a value of ε = 5 is illustrated in Figure 4.3. The average coefficient R 2 is shown as a thick black line. This value was used in (4-43) and (4-44) to correct the influence of the GR path on the delay and angular spreads. The figure also shows that there is a point where only horizontally polarized waves are reflected. In optics, this corresponds to Brewster s law. Figure 4.3: Values of the reflection coefficients for ε = 5 In a typical radio-propagation scenario, the values of the relative permittivity and the conductivity are frequency-dependent. A general guideline on how to model this dependency has been provided by [ITUR-P2040], where ε r = A (f c [GHz] ) B ; σ = C (f c [GHz] ) D (4-58) [ITUR-P527] published curves for different ground materials. These curves have been fitted to the above model for the range from 6 to 100 GHz (see Table 4.4). We propose to randomly choose one of the three ground types (very dry, medium dry and wet) to determine the value of the reflection coefficient. mmmagic Public 92

113 Material Rel. permittivity Conductivity A B C D Very dry ground Medium dry ground Wet ground Table 4.4: Electrical properties of the environment, GHz Path loss and shadow fading At larger distances, the GR influences the PL. For example, Figure 4.3 the power of the reflected path is almost equal to the direct path at distances greater than 1,000 m. Due to the 180 phase shift, both paths (almost) cancel each other. Thus, the PL increases with the 4-th power of the distance. This effect has been taken into account by the 3GPP model [3GPP38.900] by using a dual-slope PL model for the LOS channels. In the first part, which ranges from the BS to a socalled break point (BP), the PL for the UMi street canyon scenario is close to the free-space PL. After the BP, the second slope accounts for the influence of the GR. PL 1 = 21 log 10 d 3D log 10 f c [GHz] (4-59) PL 2 = 40 log 10 d 3D log 10 f [GHz] c 9.5 log 10 (d 2 BP + (h BS h MT ) 2 ) (4-60) The break point distance d BP is defined as d BP = 4 (h BS 1) (h MT 1) f c c (4-61) A comparison of the 3GPP PL model and the new model is presented Figure 4.4. The thick black line corresponds to the 3GPP UMi street canyon model. The thin blue line shows the proposed model where the characteristic small-scale fading effects are included. However, at large distances, the asymptotic PL of this model is a combination of (4-60) and the 2-ray ground reflection PL. The resulting PL scales with the 6-th power of the distance. This is shown as a red dashed line. PL GR = 60 log 10 d 3D 20 log 10 (h BS h MT ) 9.5 log 10 (d 2 BP + (h BS h MT ) 2 ) (4-62) In order to compensate this increased PL, we propose to add a third slope to the 3GPP model. The first two parts are identical to (4-59) and (4-60). The third slope is modelled according to PL 3 = 20 log 10 d 3D log 10 f [GHz] c 9.5 log 10 (d 2 BP + (h BS h MT ) 2 ) log 10 (d BP2 + (h BS h MT ) 2 ) (4-63) d BP2 = 41.7 h BS h MT f c [GHz] (4-64) The second BP is at the intersection of (4-60) and (4-63). As a result, the PL at large distances is identical to the existing PL model that includes the ground reflection by a dual-slope model. In the same way, it is possible to extend all existing PL models with implicit ground-reflection models. mmmagic Public 93

114 Figure 4.4: Effective path loss at 28 GHz and 6 m BS height The GR fading oscillates around the average PL. Constructive interference increases the maximum power by up to 6 db. In the worst case, the power can drop by more than 20 db. However, the average power is the same as in the 3GPP PL model. Hence, the additional power of the GR in (4-47) does not need to be compensated in the path loss formulas. However, the GR fading influences the SF. The 3GPP model assumes a SF standard deviation (STD) of 4 db for all distances. Since the GR fading is now explicitly included, the SF values need to be adjusted. In order to determine the optimal SF settings for the new model, a parameter study was done. The results are listed in Table 4.5. The fluctuations are the strongest in between the BS and the first BP. We therefore propose reduce the SF STD at the model input to 1 db. This leads to reasonable SF values in between 3.9 and 4.8 db at the model output. In between the two break points, the SF STD needs to be reduced to 3 db. At large distances beyond the second BP, the 4 db SF STD from the 3GPP model remain unchanged. Table 4.5: Shadow Fading for different frequencies in [db] Distance Input New Model Output Distance 3GPP New Carrier Frequency [GHz] UMi Model m - d BP d BP - d BP d BP2-2d BP Spatial consistency This subsection describes two spatial consistency methods, one is the spatially correlated random variable based method, and the other is the geometric stochastic approach. The spatially correlated random variable based method provides a more accurate solution to spatial consistency modelling and is already used in large-scale parameter generation in the mmmagic channel model. This is extended to small-scale channel parameter generation and is by default the spatial consistency procedure in the mmmagic channel model. The spatially correlated channel parameters will be stored with respect to different positions and require large storage space. On the other hand, the geometric stochastic approach updates small-scale channel parameters online without any extra storage space, at the cost of less accurate descriptions of 2- mmmagic Public 94

115 D spatial consistency. Users of the mmmagic channel model can choose which procedure to use. Spatially correlated random variable based method This spatial consistency procedure in mmmagic is based on 2-D spatially correlated r.v.s. These spatially correlated r.v.s will be used during the generation of small-scale parameters. After this procedure, UEs located nearby will share correlated parameters such as delays and angles. From the mmmagic channel generation procedure described in Section 4.1, r.v.s are either Gaussian distributed or uniformly distributed. Gaussian r.v.s will be generated using the Gaussian random field method, which is a computationally efficient method to generate 2-D spatially correlated Gaussian r.v.s. Uniform r.v.s will be obtained by transforming Gaussian r.v.s. Consider any two points P (a,b) = (x a, y b ) and P (a,b ) = (x a, y b ) in a 2-D (W meter by W meter) plane. The distance metric can be computed by d (a,b),(a,b ) (4-65) = (min{ x a x a, W x a x a }) 2 + (min{ y b y b, W y b y b }) 2 where the minimum operator is used for wrapping around the region of interest. Then, the correlation between P (a,b) and P (a,b is expressed as ) R (a,b),(a,b ) = exp ( d (a,b),(a,b ) d corr ), (4-66) where d corr is the correlation distance depending on the type of the small scale parameter. Values of d corr for different parameters can be found in Section A.1.6 or in [3GPP38.900]. To reduce computational overhead, a granularity g should be defined within the region of interest. The ranges of a and b can be determined by 1 a, b W, where computes the minimum g integer no less than W. As a result, the correlation matrix of points within the region of interest g can be formed by computing the correlation values with respect to the first point (left upper corner point) R = [R (a,b),(1,1) ]. Generate spatially correlated Gaussian r.v.s After obtaining the correlation matrix R, compute its 2-D fast Fourier transform (FFT) by G = FFT2(R) (4-67) where FFT2 is the 2D FFT operator. Next, generate a zero-mean unit variance Gaussian matrix V iid with independently and identically distributed (i.i.d.) entries. The matrix V storing correlated Gaussian r.v.s can be calculated by V = real{fft2{sqrt(g)} H iid / W } (4-68) g where real{ } obtains the real part of a matrix, sqrt{ } computes the element-wise square root of a matrix, and is the Hadamard product which computes the element-wise product of two matrices. In this case, any two entries v (a,b) and v (a,b in V will have the target correlation value ) R (a,b),(a,b ). Generate spatially correlated uniform r.v.s According to the property of a Gaussian copula, spatially correlation uniform r.v.s with the same correlation matrix R can be obtained by mapping the spatially correlated Gaussian r.v.s to [0,1] via the cumulative distribution function (CDF) of a Gaussian distribution. Let matrix U = [u (a,b) ] be the matrix storing correlated uniform r.v.s, its entries can be calculated by mmmagic Public 95

116 u (a,b) = Φ(v (a,b) ) (4-69) where Φ( ) is the CDF of standard Gaussian distribution. In this case, any two entries u (a,b) and u (a,b ) in U will have the target correlation value R (a,b),(a,b ). Impact on mmmagic channel generation In order to implement spatially correlated random variable based spatial consistency in the mmmagic channel model, parameter-specific spatially correlated r.v.s should be calculated and stored before user dropping. After a user is dropped, the steps of channel generation involving random drawing Gaussian or uniform r.v.s will be replaced by reading stored spatially correlated r.v.s according to the user s position. Implementation procedure of geometric stochastic approach The geometric stochastic approach for spatial consistency discussed in Section is an optional feature in mmmagic as the baseline approach is based on [3GPP38.900] with small changes in sense of backward compatibility. We believe that this approach could be implemented in the mmmagic model as basic methodology of both [3GPP38.900] and QuDRiGa is nearly same and stems from state of the WINNER II models [KMH+07]. At the moment, a detailed comparison of spatially correlated random variable based method and the geometric stochastic approach described in Section is out of scope of this work, and we leave it as a future work. The implementation procedure described here is based on [3GPP38.900], and the step number here correspond to the fast fading model steps of [3GPP38.900, Section 7.5]. Step 1 and 2: Pre-compute the LSPs for each grid, grid shape can be rectangular with side length of spatial consistent distance. Step 3 and 4: Every UE takes the LSPs of the grid that the UE locates. See item-c. Calculate the path loss based on UE s position. Step 5: Add the decision of cluster birth and death. If yes, take the procedure of cluster birth and death. Step 11: Update the angles at the beginning of the step Blockage The blockage model is based on a simple formulation of two-dimensional knife-edge diffraction using standard mathematical functions. The amplitude of a received diffracted path,, relative to the amplitude of the same unblocked path, A, i.e. A A A rel diff is given by diff A diff where 2 2 rel 1 phij Adiff 1 sij Fij, (4-70) i 1 j 1 2 Ph ij cos tan 2 2 F ij ij (4-71) proj proj proj ij D1ij D2ij ri (4-72) mmmagic Public 96

117 The sign parameter is given by i2 ph exp D1 D2 ij ij ij (4-73) i2 Ph exp r. (4-74) if NLOS condition in projection i s 1, and if LOS condition in projection i s sgn D1 D2 D1 D2 where k mod( j,2) 1. ij ij ij ij ik ik, (4-75) The corresponding geometry is shown in Figure 4.5. For a NLoS case, Tx and Rx should be replaced with consecutive interaction points (excluding specular reflections) for which the blocking object is located in between. Figure 4.5: Geometry of the improved blockage model. This is an extension of the METIS blockage model, which is obtained by using the following parameterization: phij 1, Ph 1, and cos ij 1. Ph and ph ij account for the phase variations of the direct path and the paths diffracted by the four edges of the screen. As they are fixed to unity (real valued) in the METIS model, maximum constructive addition i.e. minimum diffraction loss is obtained. For the mmmagic blockage model, substantial fading variations are expected due to different phases of the different paths. Furthermore, the cos ij factor accounts for increase of diffraction loss in the shadow zone close behind the screen. When the relative distance to the screen is sufficiently large this factor may be neglected. mmmagic Public 97

118 5 Conclusion This deliverable D2.2 summarizes the work on channel measurements and modelling done in the mmmagic project. The document provides a description of all conducted measurement campaigns and simulations, the results on channel characterization, mm-wave specific modelling approaches and the final mmmagic channel model. It is to be seen in conjunction with D2.1, which contains a review of preceding state-of-the art channel models and measurement campaigns and gives details on the identified model requirements and channel sounder validation methods. During the project, more than 20 measurement campaigns were conducted in the targeted frequency range between 6 and 100 GHz for outdoor and indoor propagation scenarios. Special focus was on the following scenarios: UMi street canyon, UMi open square, office, airport checkin area and outdoor-to-indoor. Up to four frequencies were considered for the same propagation environment to enable the investigation of frequency-dependent channel characteristics. Frequency bands below 6 GHz were included to close the gap to the traditional cellular frequencies. As the measurement data is statistically limited, complementary channel data have been generated by ray tracing simulations, calibrated with the measurement results, enabling the derivation of additional parameters for stochastic modelling. Specific measurements were performed to study specular and diffused scattering and the impact of the ground reflection and blockage at mm-wave frequencies. The collected data was thoroughly evaluated by visual inspection, parameter estimation techniques and statistical analyses with special focus on the frequency dependency of LSPs, the impact of the ground reflection at mm-wave frequencies, cluster and sub-path characteristics, small-scale fading, blockage, building penetration loss, spatial consistency and scattering behaviour. Specific models were elaborated and validated by comparing the model output with the measurement results. The measurement parameters, such as measurement bandwidth, dynamic range and antenna pattern, may severely affect the results when analysing the frequency dependency of channel characteristics. Therefore, a set of requirements was defined that must be met to enable full comparison of measurement results in different frequency bands. The analysis of the mmmagic measurements shows with 95% statistical confidence level that the DS slightly decreases with frequency for some scenarios (Indoor LOS, Street canyon LOS, Open square NLOS). For the other scenarios, any frequency trend is upper bounded by the confidence range. Overall, any frequency trend was found to be small. The ground reflection can have a huge impact on the received signal power and cause frequency-selective fading. For UMi scenarios at mm-wave frequencies, these effects are not appropriately covered by a dual-slope PL model. Therefore, a model has been developed to incorporate the ground reflection explicitly. It enables to reproduce the small-scale as well as the large-scale effects. Investigations on multipath clustering in both indoor and outdoor environments showed potential deficiencies in the current 3GPP channel model. A constant number of 20 irresolvable subpaths may not sufficiently characterize above 6 GHz channels. Intra-cluster RMS delay spreads are found to be larger than the current assumption in the new 3GPP channel model. An improved version of the METIS blockage model (mmmagic blockage model) has been presented and validated based on both measurements and Kirchhoff s diffraction formula. The spatial consistency procedure, which has been initiated in by 3GPP in release 14 [3GPP38.900], mmmagic Public 98

119 has been investigated and, furthermore, a different model, called grid-based GSCM, (partly based on [METIS-D1.2]) has been proposed. The new model realizes time-variant angles and cluster death and birth as the UE is moving, which is an important feature to evaluate mobility and beam tracking for 5G communications. Investigations on reflections and scattering lead to the conclusion, that a Nakagami-m distribution can adequately model the random fluctuations caused by surface roughness of various building materials. The proposed mmmagic channel model has been developed in parallel with the 3GPP [3GPP38.900, 3GPP38.901], ITU-R [IMT-2020.EVAL], ITU-R building entry loss [ITU-R P-series] and QuaDRiGa [JRB+14] models. It incorporates the aforementioned findings and modelling approaches. The mmmagic channel model is composed of baseline components and additional features. It provides major improvements regarding the incorporation of ground reflection and blockage effects, the support of large bandwidths and large antenna arrays, the provision of spatial consistency, and elevation angle dependent O2I penetration loss modelling. By actively collaborating with 3GPP and ITU-R, various mmmagic results and modelling approaches have been adopted and are reflected in the latest respective channel models, e.g., measurement results of several propagation scenarios, the ground reflection model, and the extended O2I penetration loss model. An open source implementation of the mmmagic baseline model and selected additional features is provided as part of the new QuaDRiGa release (v2.0) on mmmagic Public 99

120 6 References [3GPP160846] 3GPP R , Ericsson, "Street microcell channel measurements at 2.44, 14.8, and GHz," 3GPP RAN1 #84, St Julian s, Malta, February [3GPP161688] [3GPP161691] [3GPP161700] [3GPP161707] [3GPP161708] [3GPP162720] [3GPP162721] [3GPP162722] [3GPP162872] [3GPP163254] [3GPP164808] 3GPP R , Ericsson, "Measurements of path and penetration losses at multiple carrier frequencies," 3GPP TSG RAN WG1 #AH Channel Model, Ljubljana, Slovenia, March GPP R , Ericsson, "Indoor and outdoor to indoor channel measurements," 3GPP TSG RAN WG1 #AH Channel Model, Ljubljana, Slovenia, March GPP R , Fraunhofer HHI, "mmmagic white paper," 3GPP TSG RAN WG1 #AH Channel Model, Ljubljana, Slovenia, March GPP R , Samsung, "Discussion on blockage modeling," 3GPP TSG RAN WG1 #AH Channel Model, Ljubljana, Slovenia, March GPP R , Samsung, "Discussion on spatial consistency," 3GPP TSG RAN WG1 #AH Channel Model, Ljubljana, Slovenia, March GPP R , Samsung, "Discussion on blockage modelling for above 6 GHz channel model," 3GPP TSG RAN WG1 #84b, Busan, Korea, April GPP R , Samsung, "Discussion on spatial consistency for above 6 GHz channel model," 3GPP TSG RAN WG1 #84b, Busan, Korea, April GPP R , Samsung, "Discussion on large antenna array modeling," 3GPP TSG RAN WG1 #84b, Busan, Korea, April GPP R , Keysight, "Initial 60 GHz channel sounding results from the mmmagic project for corner diffraction and surface scattering," 3GPP TSG RAN WG1 #84b, Busan, Korea, April GPP R , Ericsson, "On the frequency dependence of LSPs," 3GPP TSG RAN WG1 #84b, Busan, Korea, April GPP R , Samsung, "Remaining details on blockage modelling for above 6 GHz channel model," 3GPP TSG RAN WG1 #85, Nanjing, China, May [3GPP164809] 3GPP R , Samsung, "Remaining details on spatial consistency for above 6 GHz channel model," 3GPP TSG RAN WG1 #85, Nanjing, China, May [3GPP ] [3GPP ] [3GPP25.996] [3GPP36.873] [3GPP38.900] [3GPP38.901] [5GCM16] 3GPP R , Fraunhofer HHI, ZTE, Ericsson, "TR38.900_CR_Ground_Reflection_Model", Change Request, TSG-RAN WG1 NR Ad Hoc, Spokane, WA, USA, January GPP TR , Scenarios and Requirements for Next Generation Access Technologies, v0.2.0, Tech. Rep., Feb GPP TR , Spatial channel model for multiple input multiple output (MIMO) simulations, v6.1.0, Tech. Rep., GPP TR , 3rd generation partnership project; technical specification group radio access network; study on 3D channel model for LTE (release 12), v12.0.0, Tech. Rep., GPP TR , 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Study on channel model for frequency spectrum above 6 GHz (Release 14), v14.2.0, Tech. Rep., GPP TR , 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Study on channel model for frequencies from 0.5 to 100 GHz (Release 14), v14.0.0, Tech. Rep., G channel model for bands up to 100 GHz, revised version, Tech. Rep., Mar [Online]. Available: mmmagic Public 100

121 [METIS-D1.2] METIS D1.2, Initial channel models based on measurements, V1.0, ICT , METIS project, [ALS+14] M. R. Akdeniz, Y. Liu, M. K. Samimi, S. Sun, S. Rangan, T. S. Rappaport, and E. Erkip, Millimeter Wave Channel Modeling and Cellular Capacity Evaluation, IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp , June [AR04] [AV07] [AZC99] [BHL+14] [BMD16] [BMD17] [BMS89] [BSK+07] [BST11] [BW99] [CBY+05] [Cis17] [Cor94] [Czink07] C. R. Anderson and T. S. Rappaport, In-Building Wideband Partition Loss Measurements at 2.5 and 60 GHz, IEEE Transactions on Wireless Communications, vol. 3, no. 3, May D. Arthur and S. Vassilvitskii, K-means++: the advantages of careful seeding, in Proc. SODA 07., New Orleans, Louisiana, 2007, pp A. Affandi, G. El Zein and J. Citerne, "Investigation on frequency dependence of indoor radio propagation parameters," IEEE Vehicular Technology Conference, 1999 Fall, Amsterdam. F. Boccardi, R. Heath, A. Lozano, T. Marzetta, and P. Popovski, Five disruptive technology directions for 5G, Communications Magazine, IEEE, vol. 52, no. 2, pp , A. Bamba, F. Mani, and R. D Errico, E-Band Millimeter Wave Indoor Channel Characterization, IEEE 27 th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Valencia, Spain, 4-7 September A. Bamba, F. Mani, and R. D Errico, A Comparison of Indoor Channel Properties in V and E bands, 1 st European Conference on Antennas and Propagation (EuCAP 2017), Paris, France, March R. J. C. Bultitude, S. A. Mahmoud, and W. A. Sullivan, A Comparison of Indoor Radio Propagation Characteristics at 910 MHz and 1.75 GHz, IEEE Transactions on Selected Areas in Communications, vol. 7, no. 1, January R. J. C. Bultitude, T. C. W. Schenk, N. A. A. Op den Kamp and N. Adnani, "A Propagation-Measurement-Based Evaluation of Channel Characteristics and Models Pertinent to the Expansion of Mobile Radio Systems to Frequencies Beyond 2 GHz," in IEEE Transactions on Vehicular Technology, vol. 56, no. 2, pp , March A. Bottcher, C. Schneider, P. Vary, R.S. Thomae, Cross Correlation Characteristics of Large Scale Parameters in Urban Macro Cell, VTC2011-Fall, San Francisco, CA, USA, M. Born, E. (1999), Principles of optics: electromagnetic theory of propagation, interference and diffraction of light, Cambridge: Cambridge University Press. p ISBN N. Czink, E. Bonek, X. Yin, and B. Fleury, Cluster Angular Spreads in a MIMO Indoor Propagation Environment, IEEE 16 th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), pp , Berlin, Germany, September Cisco, Cisco visual networking index: global mobile data traffic forecast update, , White Paper, February [Online]. Available: L. M. Correia and P. O. Françês, Estimation of materials characteristics from power measurements at 60 GHz, IEEE International Symposium on Personal, Indoor and Mobile Radio Communication (PIMRC), pp , September, N. Czink, The Random Cluster Model A Stochastic MIMO Channel Model for Broadband Wireless Communication Systems of the 3 rd Generation and Beyond, PhD thesis, Technischen Universitat Wien, mmmagic Public 101

122 [D95] [DBB+91] [DM16] [DNC17] [DK95] [GME99] [GMP07] D. M. J. Devasirvathan, "Multi-Frequency Propagation Measurements And Models In A Large Metropolitan Commercial Building For Personal Communications," IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), pp , R. Davies, K. Bensebti, M. A. Beach, and J. P. McGeehan, Wireless propagation Measurement in Indoor multipath Environment at 1.7 GHz and 60 GHz, IEEE VTC Dmitrii Solomitckii et al, "Characterizing the Impact of Diffuse Scattering in Urban Millimeter-Wave Deployments", IEEE Wireless Comms Letters, Vol. 5, No. 4, August D. Dupleich, N. Iqbal, C. Schneider, S. Haefner, R. Mueller, S. Skoblikov, J. Luo, and R. S. Thomä, Investigations on fading scaling with bandwidth and directivity at 60 GHz, 11th European Conference on Antennas and Propagation (EuCAP 2017), Paris, France, Mar H. Droste and G. Kadel, "Measurement and analysis of wide band indoor propagation characteristics at 17 GHz and 60 GHz," 9th International Conference on Antennas and Propagation, pp vol. 2, Eindhoven, L. Greenstein, D. Michelson, and V. Erceg, Moment-method estimation of the ricean k-factor, IEEE Communications Letters, vol. 3, no. 6, pp , S. Gregson, J. McCormick, and C. Parini, Principles of Planar Near-Field Antenna Measurements, IET, [Gol03] A. Goldsmith, Wireless Communications, Cambridge University Press, [Hat80] [HCL+14] [HJK+14] [HJK+15] [HNJ+16] M. Hata, Empirical formula for propagation loss in land mobile radio services, IEEE Trans. Veh. Technol., vol. 29, no. 3, pp , S. Hur, Y. J. Cho, J. Lee, N. G. Kang, J. Park, and H. Benn, Synchronous Channel Sounder Using Horn Antenna and Indoor Measurements on 28 GHz, IEEE International Communications and Networking (BlackSeaCom), pp , May K. Haneda, J. Järveläinen, A. Karttunen, M. Kyrö and J. Putkonen, "Indoor shortrange radio propagation measurements at 60 and 70 GHz," 8th European Conference on Antennas and Propagation (EuCAP 2014), pp , The Hague, 2014 K. Haneda, J. Järveläinen, A. Karttunen, M. Kyrö, and J. Putkonen, A statistical spatio-temporal radio channel model for large indoor environments at 60 and 70 GHz, IEEE Trans. on Antennas and Propag., vol. 63, no. 6, pp , Jun K. Haneda, S. L. H. Nguyen, J. Järveläinen and J. Putkonen, "Estimating the omnidirectional pathloss from directional channel sounding," th European Conference on Antennas and Propagation (EuCAP), Davos, 2016, pp [IMT.2020.EVAL] ITU-R preliminary report for evaluations of IMT [ITUR-3J22E] [ITUR-3J37E] ITU-R Document 3J/22-E, Orange, "Building entry loss measurements at 3, 10 and 17 GHz frequencies", June ITU-R Document 3J/37-E, Ericsson, "Measurement results of building entry loss in the 2 GHz to 60 GHz frequency range", June [ITUR-3J82E] ITU-R Document 3J/82-E, Orange, "Building entry loss measurements at 3, 10, 17 and 60 GHz ", March [ITUR-3J114E] ITU-R Document 3J/114-E, Ericsson, "On Recommendation ITU-R P.[BEL]: Building entry loss measurement and modelling results on elevation angle dependence", March mmmagic Public 102

123 [ITUR-5D336E] [ITUR-5D338E] [ITUR-P1407] [ITUR-P1411] [ITUR-P1816] [ITUR-P2040] [ITUR-P527] [ITUR R-809] ITU-R Document 5D/336-E, Samsung, "Ground reflection modelling for IMT-2020 Channel model", September ITU-R Document 5D/338-E, Samsung, "Remaining issues on ground reflection modelling for IMT-2020 channel model", February ITU-R P , Multipath propagation and parameterization of its characteristics, Tech. Rep., ITU-R P , Propagation data and prediction methods for the planning of short range outdoor radiocommunication systems and radio local area networks in the frequency range 300 MHz to 100 GHz, Tech. Rep., ITU-R P.1816, The prediction of the time and the spatial profile for broadband land mobile services using UHF and SHF bands, Tech. Rep., ITU-R P , Effects of building materials and structures on radiowave propagation above about 100 MHz, Tech. Rep., 2015 ITU-R P.527-3, Electrical characteristics of the surface of the earth, Tech. Rep., ITU-R, Resolution 809 (WRC-15), World Radiocommunication Conference, Geneva, [ITU-R P-series] To appear in the ITU-R P-series propagation reccomendations, [ITUR ] [Jak94] [JGB03] [JKF99] [JHK+06] [JKH15] [JHK16] [JNH+16] [JRB+14] [JSB96] ITU-R M.2083, Framework and overall objectives of the future development of IMT for 2020 and beyond, Sep W. C. Jakes and D. C. Cox, Eds., Microwave Mobile Communications. Wiley-IEEE Press, E. J. Jaselskis, J. Grigas, and A. Brillingas, Dielectric properties of asphalt pavement, Journal of Materials in Civil Engineering, vol. 15, no. 5, pp , September/October J. D. Kraus, D. A. Fleisch, and S. H. Russ, Electromagnetics, 5th ed., ser. McGraw- Hill series in electrical and computer engineering, Boston: McGraw Hill, T. Jämsä, V. Hovinen, A. Karjalainen and J. Iinatti, "Frequency Dependency of Delay Spread and Path Loss in Indoor Ultra-Wideband Channels," The Institution of Engineering and Technology Seminar on Ultra Wideband Systems, Technologies and Applications, pp , London, J. Järveläinen, M. Kurkela, and K. Haneda, Impacts of room structure models on the accuracy of 60 GHz indoor radio propagation prediction, IEEE Antennas Wireless Propag. Lett., vol. 14, pp , J. Järveläinen, K. Haneda, and A. Karttunen, Indoor Propagation Channel Simulations at 60 GHz Using Point Cloud Data, IEEE Trans. Antennas Propag., vol. 64, no. 10, pp , Oct J. Järveläinen, S. L. H. Nguyen, K. Haneda, R. Naderpour, and U. T. Virk, Evaluation of millimeter-wave line-of-sight probability with point cloud data, IEEE Wireless Commun. Lett., vol. 5, no. 3, pp , Jun S. Jaeckel, L. Raschkowski, K. Borner, and L. Thiele, QuaDRiGa: A 3-D multi-cell channel model with time evolution for enabling virtual field rials, IEEE Transactions on Antennas and Propagation, vol. 62, no. 6, pp , G. J. M. Janssen, P. A. Stigter and R. Prasad, "Wideband indoor channel measurements and BER analysis of frequency selective multipath channels at 2.4, 4.75, and 11.5 GHz," in IEEE Transactions on Communications, vol. 44, no. 10, pp , Oct mmmagic Public 103

124 [KGM+16] [KMH+07] [KJN+17] [KLK+15] [KPP07] [KPS+15] [KPW+14] A. Karttunen, C. Gustafson, A. F. Molisch, J. Järveläinen, K. Haneda, "Censored Multipath Component Cross-Polarization Ratio Modeling," in IEEE Wireless Communications Letters, vol. PP, no. 99, pp. 1-1, Pekka Kyösti, Juha Meinilä, Lassi Hentilä, et al., IST WINNER II D1.1.2 v.1.1: WINNER II channel models, Technical report, A. Karttunen, J. Järveläinen, S. L. H. Nguyen, and K. Haneda, Above 6 GHz multipath cross-polarization ratio in indoor and outdoor scenarios, COST CA15104 TD(17)03052, Lisbon, Portugal, February 1-3, M.-D. Kim, J. Liang, H.-K. Kwon, and J. Lee, Directional delay spread characteristics based on indoor channel measurements at 28 GHz, in Proc. 26th IEEE Ann. Int. Symp. Pers. Indoor Mobile Radio Commun. (PIMRC), Hong Kong, pp , Aug. 30-Sept. 2, D. -J. Kim, Y. -W. Park, and D. -J. Park, A novel validity index for determination of the optimal number of clusters, IEICE Trans. Inf. & Syst., vol. E84-D, no. 2, pp , O. H. Koymen, A. Partyka, S. Subramanian and J. Li, "Indoor mm-wave Channel Measurements: Comparative Study of 2.9 GHz and 29 GHz," 2015 IEEE Global Communications Conference (GLOBECOM), pp. 1-6, San Diego, CA, W. Keusgen, R. J. Weiler, M. Peter, M. Wisotzki, Propagation measurements and simulations for millimeter-wave mobile access in a busy urban environment, 9th International Conference on Infrared, Millimeter, and Terahertz Waves (IRMMW- THz 2014), September [LR03] Dai Lu and D. Rutledge, "Investigation of indoor radio channels from 2.4 GHz to 24 GHz," IEEE Antennas and Propagation Society International Symposium, pp vol. 2, Columbus, OH, USA, [Lud73] [MAE08] [Mal08] [METIS-D1.4] A. Ludwig, The definition of cross-polarization, IEEE Trans. Antennas Propagation, vol. AP-21, pp , W. Q. Malik, B. Allen, and D. J. Edwards, Bandwidth-dependent modelling of small scale fade depth in wireless channels, IET Microwaves, Antennas Propagation, vol. 2, no. 6, pp , Sept W. Q. Malik, Polarimetric characterization of ultrawideband propagation channels, IEEE Transactions on Antennas and Propagation, vol. 56, no. 2, pp , Feb ICT METIS Deliverable D1.4 v3, METIS channel models, Tech. Rep., [MH16] R. Müller, S. Häfner, D. Dupleich, R. S. Thomä, G. Steinböck, J. Luo, E. Schulz, X. Lu and G. Wang, Simultaneous multi-band channel sounding at mm-wave frequencies, 10th European Conference on Antennas and Propagation (EuCAP), pp. 1 5, April [MHD+14] [mmm15] [mmmagic-d2.1] [mmmagic-d5.1] [MSA16] R. Mueller, R. Herrmann, D. A. Dupleich, C. Schneider, and R. S. Thomae, Ultrawideband multichannel sounding for mm-wave, 8th European Conference on Antennas and Propagation (EuCAP), pp , mmmagic website: ICT mmmagic Deliverable D2.1, Measurement campaigns and initial channel models for preferred suitable frequency ranges, Tech. Rep., ICT mmmagic Deliverable D5.1, Initial multi-node and antenna transmitter and receiver architectures and schemes, Tech. Rep., J. Medbo, N. Seifi, H. Asplund, Frequency dependency of measured highly resolved directional propagation channel characteristics, European Wireless, mmmagic Public 104

125 [NH97] [NSB08] [NJS+17] [NCJ17] [MOL05] P. Nobles and F. Halsall, "Delay spread and received power measurements within a building at 2 GHz, 5 GHz and 17 GHz," Tenth International Conference on Antennas and Propagation, pp vol. 22, Edinburgh, B. Neekzak, K. Sayrafian-Pour, J. S. Baras, Clustering Characteristics of Millimeter Wave Indoor Channels, IEEE Wireless Communications and Networking Conference (WCNC), 31 March - 3 April N. Iqbal, J. Luo, C. Schneider, D. Dupleich, R. Müller, S. Haefner, and R. S. Thomä, Stochastic/Deterministic behavior of cross polarization discrimination in mmwave channels, IEEE ICC Wireless Communications Symposium (ICC 17 WCS), Paris, France, May N. Iqbal, C. Schneider, J. Luo, D. Dupleich, R. Müller, S. Haefner, and R. S. Thomä, On the Stochastic and Deterministic Behavior of mmwave Channels, 11th European Conference on Antennas and Propagation (EuCAP 2017), Paris, France, A. F. Molisch, Ultrawideband propagation channels-theory, measurement,and modeling, IEEE Transactions on Vehicular Technology, vol. 54, no. 5, pp , Sept [OSH16] P. Okvist, N. Seifi, B. Halvarsson, A. Simonsson, M. Thurfjell, H. Asplund and J. Medbo, 15 GHz Street-Level Blocking Characteristics Assessed with 5G Radio Access Prototype, IEEE 83rd Vehicular Technology Conference (VTC Spring), May, [P05] [PLL+16] [PMF97] [PK09] [PWK+15] [PWU+16] [Rap02] [RBM+12] [RCS16] [Sato97] P. Papazian, "Basic transmission loss and delay spread measurements for frequencies between 430 and 5750 MHz," IEEE Transactions on Antennas and Propagation, vol. 53, no. 2, pp , Feb J. J. Park, J. Liang, J. Lee, H. K. Kwon, M. D. Kim and B. Park, "Millimeter-wave channel model parameters for urban microcellular environment based on 28 and 38 GHz measurements," IEEE 27th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), Valencia, K.I. Pedersen, P.E. Mogensen, and B.H. Fleury, Power azimuth spectrum in outdoor environments, Electronics Letters, 33(18): , M. Peter and W. Keusgen, Analysis and comparison of indoor wideband radio channels at 5 and 60 GHz, 3rd European Conference on Antennas and Propagation (EuCAP), March 2009, Berlin. M. Peter, R.J. Weiler, T. Kühne and others, Millimeter-Wave Small-Cell Backhaul Measurements and Considerations on Street-Level Deployment, IEEE Globecom Workshops, M. Peter et al., "Investigations on the frequency dependence of the delay spread in an UMi street canyon scenario," International Symposium on Antennas and Propagation (ISAP), pp , Okinawa, Japan, T. Rappaport, Wireless Communications. Principles and Practice, 2nd ed. Prentice Hall, 2002 T. S. Rappaport, E. Ben-Dor, J. N. Murdock and Y. Qiao, "38 GHz and 60 GHz angle-dependent propagation for cellular & peer-to-peer wireless communications," IEEE International Conference on Communications (ICC), pp , Ottawa, ON, X. Raimundo, S. Salous and A. A. Cheema, "Indoor radio propagation measurements in the V-band," Radio Propagation and Technologies for 5G (2016), pp. 1-5, Durham, K. Sato, et al., Measurements of reflection and transmission characteristics of interior structures of office building in the 60-GHz Band, IEEE Transactions on Antennas and Propagation, pp , December, mmmagic Public 105

126 [Sha06] [SJJ+00] [SR14] [SR15] [SR+15] [STJ10] [SV87] [VJN+16] [WHW+16] [WPK+15] [WPK+15a] [WSH16] [YLH05] [YSH05] [YYD+16] [WHS15+] [ZRP + 05] M. Shafi, M. Zhang, A. L. Moustakas, P. J. Smith, A. F. Molisch, F. Tufvesson, and S. H. Simon, Polarized MIMO channels in 3-D: models, measurements and mutual information, IEEE Journal on Selected Areas in Communications, vol. 24, no. 3, pp , March Q.H. Spencer, B. D. Jeffs, M. A. Jensen, and A. L. Swindlehurst, Modeling the Statistical Time and Angle of Arrival Characteristics of an Indoor Multipath Channel, IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp , March M. K. Samimi and T. S. Rappaport, Ultra-wideband statistical channel model for non line of sight millimeter wave urban channels, in Proc. GLOBECOM 14, pp , Austin, USA, Dec M. K. Samimi and T. S. Rappaport, Statistical channel model with multi-frequency and arbitrary antenna beamwidth for millimeter-wave outdoor communications, Proc. GLOBECOM 15., pp. 1 7, San Diego, USA, M. K. Samimi and T. S. Rappaport, 28 GHz millimeterwave ultrawideband smallscale fading models in wireless channels, CoRR, vol. abs/ , [Online]. Available: S. Jaeckel, L. Thiele and V. Jungnickel, Interference Limited MIMO Measurements, Proc. IEEE VTC '10 Spring, A. A. M. Saleh and R. A. Valenzuela, A statistical model for indoor multipath propagation, IEEE J. Sel. Areas Commun., vol. SAC-5, pp , February J. Vehmas, J. Järveläinen, S. L. H. Nguyen, R. Naderpour, and K. Haneda, Millimeter-wave channel characterization at Helsinki airport in the 15, 28, and 60 GHz bands, Proc. IEEE VTC Fall, Montreal, Canada, September, S. Wu, S. Hur, K. Whang, and M. Nekovee, Intra-cluster characteristics of 28 GHz wireless channel in urban micro street canyon, Proc. GLOBECOM 16, Washington, USA, Dec R. J. Weiler, M. Peter, T. Kühne, M. Wisotzki and W. Keusgen, "Simultaneous millimeter-wave multi-band channel sounding in an urban access scenario," 9th European Conference on Antennas and Propagation (EuCAP), Lisbon, R. J. Weiler, M. Peter, W. Keusgen, A. Kortke, and M. Wisotzki, Millimeter-wave channel sounding of outdoor ground reflections, 2015 IEEE Radio and Wireless Symposium, pp , Jan Y. Wang, Z. Shi, L. Huang, Z. Yu, and C. Cao, "An extension of spatial channel model with spatial consistency", Proc. IEEE VTC 2016-Fall, Montreal, Canada, Sep K. Yu, Q. Li, and M. Ho, Measurement investigation of tap and cluster angular spreads at 5.2 GHz, IEEE Transactions on Antennas and Propagation, vol. 53, no. 7, pp , July H. Yang, P. F. M. Smulders and M. H. A. J. Herben, "Indoor Channel Measurements and Analysis in the Frequency Bands 2 GHz and 60 GHz," 2005 IEEE 16th International Symposium on Personal, Indoor and Mobile Radio Communications, pp , Berlin, H. Yan et al., "Comparison of large scale parameters of mmwave wireless channel in 3 frequency bands," International Symposium on Antennas and Propagation (ISAP), Okinawa, Japan, 2016, pp Y. Wang, L. Huang, Z. Shi, K. Liu, and X. Zou, A millimeter wave channel model with variant angles under 3GPP SCM framework, in Proc. IEEE PIMRC 2015 Workshop, Hong Kong, China, Aug Y. Zhou, S. Rondineau, D. Popovic, A. Sayeed, and Z. Popovic, "Virtual channel space-time processing with dual-polarization discrete lens antenna arrays," IEEE Trans. Antennas Propag., vol. 53, pp , Aug mmmagic Public 106

127 Annex A mmmagic Public 107

128 A Details on measurement and simulation campaigns A.1 Indoor scenarios A.1.1 Indoor campaign in the V and E band Measurement campaign A 4-port Vector Network Analyser (VNA) has been used to record the two measurement campaigns. The first one was carried out at GHz whereas the second was at GHz. To be able to measure at these mm-wave frequencies, two pairs of frequency converters - one for each band - have been employed at both ends of the measurement setup. The low IF bandwidth permits us to have a noise floor of approximately -140 dbm. For the GHz measurements, we used a 0 dbi gain standard omni-directional antenna on the Tx side while we use a vertically polarized standard horn antenna with 20 dbi gain and approximately 20 half power bandwidth (HPBW) on the Rx side. For the GHz measurements, a vertically polarized horn antenna with 10 dbi gain and HPBW of 50 is used on the Tx side whereas a vertically polarized horn antenna with 20 dbi gain and HPBW of 15 is utilized at the Rx. In post-processing we synthetized the omnidirectional PADP in both cases. Moreover, the clusters are analysed in the angular domain only for AoA where the same kind of antenna is used. Hence, we believe that the comparison with omnidirectional measurements is valid. To fully investigate the channel in the azimuth domain, two positioner devices have been used to rotate the antennas and perform mechanical steering. Additionally, the rotating motor at the Rx side is collocated with a 2D positioner that is able to scan the grid of 80 cm 80 cm. Figure A.1 (a) presents the VNA-based measurement setup, while in Figure A.1 (b) and (c) we show the antenna scanning performed in the two measurement bands. The measurement campaigns have been carried out at CEA-Leti premises. The first environment was a classical office of approximately 50 m 2 with desks, storage cabinets and bookshelves (furniture was arranged at the sides of the room leaving a large empty part in the centre where there was possible to perform the measurements). The room dimensions L, W, and H are approximately 7 m, 7 m, and 2.6 m respectively. The second environment was a conference room with different seats. Figure A.2 shows the floor plans (first column) of the environments with the Rx locations, and pictures of the measurement environments (second column) with antennas connected to the converters and placed on the positioners. We consider one Tx location and four (resp. seven) Rx locations in the office (resp. conference) room. At each Rx location in the office, the 2D positioner moved the 3 5 (resp. 2 3) grid at GHz (resp GHz) and the channel is probed at each position. For the conference room measurement, the 2D positioner moved the 1 2 grid at each Rx locations in both bands. Two scenarios were considered. In the device-to-device (D2D) scenario, both the Rx and Tx were located at 1.20 m from the floor. For the base station (BS) scenario, the Tx was located at 2.10 m from the floor whereas the Rx had an height of 1.20 m. BS measurements were only performed in the office room in the GHz band. mmmagic Public 108

129 (a) (b) (c) Figure A.1: Channel measurement setup at CEA-LETI (a). Antenna steering in the V-band (b) and E band (c) in the office. Figure A.2: Indoor environments: office (first line), conference room (second line). mmmagic Public 109

130 Data evaluation At each pairing scanning direction of Tx and Rx (Ω S T, Ω S R ) the channel frequency response H p (f, Ω S T, Ω S R ) has been recorded and the corresponding channel impulse response h p (τ, Ω S T, Ω S R ) computed. The power angular delay profile (PADP) of the channel is as PADP p (τ, Ω S T, Ω S R ) = h p (τ, Ω S T, Ω S R ) 2. (A-1) Note that (A-1) is further corrected to remove the gains introduced by the antennas. The synthetic omnidirectional PDP is then obtained as N s T N s R PDP p (τ) = PADP p (τ, Ω S T, Ω S R ), T=1 R=1 (A-2) where N s T, N s R is the number of scanning directions of the Tx (resp. Rx). Additional details on the synthetization procedure can be found in [HJK+14]. The number (N k ) of detectable paths and their corresponding delay τ k (k = 1,, N k ) is determined from the synthetic PDP. Each delay τ k is mapped to a unique couple defining the direction of departure and arrival of the k-th detected path. In the following, only the azimuthal direction is considered as the transceivers have the same height. The path detection algorithm [HJK+14] is applied for paths identification. The algorithm searches for the local maxima in the PDP to find the delay by comparison with a threshold function [HJK+14]. Then the amplitude and angle is searched for in the PASP. Consider a path is found at τ k, the corresponding power is given by the following maximization function α k 2 = max Φ S T,Φ S R PADP p(τ, Ω S T, Ω S R ) (A-3) where T = {1,, N s T } R = {1,, N s R }. The (azimuth) angle of arrival (AoA) and departure (AoD) of the kth path is given by the couple (Ω S T, Ω S R ) maximizing (A-1). Figure A.3 shows an example of omnidirectional PDP along with the detected multipaths. The two PDPs have been measured at the same position, after antenna de-embedding. We can see from Figure A.3 that i) the power levels of multipaths are in general higher at 62 GHz than those at 83.5 GHz, and ii) the PDP drops faster to the noise floor at 83.5 GHz. These observations might be explained by the free space attenuation that is higher at higher frequencies. Figure A.3: Measured PDPs in the office (Rx1) and the detected multipath. mmmagic Public 110

131 Furthermore, Figure A.4 shows the results of the multipath extraction for few positions. The results of this measurement campaign have been exploited to assess the frequency dependency of LSPs and characteristics of clusters and sub-paths. Figure A.4: Multipath extraction results for few positions. Frequency dependency of LSPs (Delay spread as the example) As seen in the literature review (Section 0), the rms delay spread dependency on the frequency is somewhat controversial. Usually, the higher the operating frequency is, the lower is the rms delay spread value. The results comparison in terms of propagation parameters such as the rms delay spread is tricky for different studies as the bandwidths might be different. In order to investigate the frequency dependency of the rms delay spread, its values have been determined from 60 to 64 GHz (V band) and from 81 to 85 GHz (E band). In both bands, a frequency step of 1 GHz is considered. For a given frequency, the rms delay spread is determined for different bandwidths in order to assess a possible trend. Only the results obtained in the office room are exploited here because the number of measurements points was sufficiently large to obtain meaningful interpretations. Figure A.5 shows the bandwidth and frequency dependency of the rms delay spread for the two bands in the office room. We observe that the rms delay spread only varies slightly with the frequency. For the different bandwidths, the maximum variation of the rms delay spread in the GHz and GHz bands equals only 1.25 ns (7.80 %) and 1.70 ns (15 %), respectively. Furthermore, the rms delay spread dependency on the bandwidth is addressed. For each frequency in the considered bands, the rms delay spread is determined with a bandwidth of 0.1, mmmagic Public 111

132 0.2, 0.3, 0.5, 1, 1.5, 1.8, and 2 GHz. The results for both bands are depicted in Figure A.5 where it can be seen that the rms delay spread decreases with increasing bandwidth, regardless of the centre frequency. Moreover, it is observed that the rms delay spread value converges towards its actual value for increasing bandwidth. The actual rms delay spread value is that obtained with infinite bandwidth, i.e., all the multipath are resolved in the delay domain. On the average, the rms delay spread value decreases of about % and % for the aforementioned bandwidths in the GHz and GHz bands, respectively. The bandwidth dependency of the rms delay spread might explain variations in results from different studies since the bandwidth is not always the same. However, we observe from the experiments that the rms delay spread values are almost constant for bandwidths larger than 500 MHz. Figure A.5: Bandwidth dependency of the rms delay spread. These delay spread values have been obtained from the full PDP (all delay bins considered) so that the comparison in both bands would not be impacted by the considered noise floor, nor by an applied algorithm to extract the MPCs. Path loss model Here we present the results obtained from the measurement campaigns in indoor premises at CEA-LETI. The path loss model is obtained from the integration of the full PDP. To model the path loss, we adopt the floating intercept model: PL(d)[dB] = PL n log 10 ( d d 0 ) + X σ (A-4) where PL 0 is the intercept in db and n is the path loss exponent that characterizes the increase in the channel path loss as a function of the Tx-Rx distance d. Both PL 0 and n are obtained from the intercept and slope of the best-fit to (A-4) according to the minimum mean square error (MMSE) criteria. X σ represents the variation due to long-term fading effects. Figure A.6 shows the path loss model in the two frequency bands and environments for the D2D scenario. In the office we obtained n = 1.36 (resp ) and PL 0 equals to db (resp db) at GHz and GHz, respectively. We found that X σ = 0.66 db (resp db). Similar results were obtained in the conference room and reported in Table A.1. Table A.1: LSP in indoor environments for D2D scenario Office Conference room GHz GHz GHz GHz PL 0 [db] 67,90 69,60 70,20 71,10 n 1,36 1,33 1,33 1,44 mmmagic Public 112

133 τ rms (ns) 8,30 6,10 5,60 5,20 Figure A.6: Path loss model in D2D scenario: office (left), conference room (right). Figure A.7 shows the path loss model for the BS scenario. Compared to the D2D, a higher path loss exponent is obtained in the BS scenario as it amounts to 1.82 and PL 0 equals to about 63 db. Figure A.7: Path loss model in BS scenario. The higher path loss in the BS scenario indicates that the signal power decreases at a faster rate as compared to the D2D scenario. This is mainly explained by larger propagation distances of the MPCs, which is due to the tilt between the Tx and Rx. Delay spread After multipath component extraction, the delay spread τ rms values were obtained. Figure A.8 shows the delay spread in the two frequency bands and environments. We found that the measured delay spread throughout the office room vary from 4.60 ns to 8.40 ns for the first band (i.e., GHz) while it ranges from 6.15 ns up to 11.5 ns for the GHz band. There is clearly no trend between the τ rms values and the Tx-Rx distance. This observation might not be valid in larger indoor offices and/or outdoor environments where the Tx-Rx separations are mmmagic Public 113

134 much important. Therefore, the investigated office room can be characterized by the mean delay spread, which is about 6.10 ns and 8.27 ns at 83.5 GHz and 62 GHz, respectively. Variation around the mean value can be modelled by a normal random variable with standard deviation of 0.7 ns at 62 GHz and 0.83 ns at 83.5 GHz. Similar results were obtained in the conference room and reported in Table A.1. Figure A.8: Delay spread: office (left), conference room (right). Figure A.9 shows the delay spread values for the BS scenario. As it can be seen in the figure, an average rms delay spread of about 7.50 ns is obtained in the office room for the BS scenario. We observed that the obtained delay spread is slightly higher than that obtained in the D2D scenario. However, the average delay spread values in both scenarios are within the same order of magnitude as the relative difference is only about 9.60%. Figure A.9: Delay spread in BS scenario: office. Clustering of multipath, intra-clusters parameters, and cluster model Figure A.10 and Figure A.11 show the clustering results of the multipath in both environments and frequency bands. The red circles, blue squares, black triangles, green diamonds, and pink stars indicate the first, second, third, fourth, and fifth cluster, respectively. The yellow markers show the corresponding centroids. The concentric circles on the polar plots indicate the travelled distance in meters by the multipath. mmmagic Public 114

135 Figure A.10: Example of multipath clustering in the office room Figure A.11: Example of multipath clustering in the conference room. Moreover, the number of clusters in the environments has been considered as well. Figure A.12 shows the occurrence - or probability density function (pdf) - of the number of identified clusters in the office and conference rooms. mmmagic Public 115

136 Figure A.12: Cluster occurrence in indoor environments. Up to five clusters in both the office and conference rooms are observed, regardless of the frequency band. However, as it can be seen in Figure A.12, the number of clusters varies throughout the considered environment. As comparison, [NSB08] obtains the maximum of six clusters in a residential room at 60 GHz, [ALS+14] found up to four clusters in an outdoor micro-cell at 73 GHz, and two to five clusters have been obtained for an indoor channel measurements at 28 GHz [HCL+14]. However, it is worth mentioning that the number of clusters will depend on the measurements dynamic range, the used algorithm to assess the multipath, etc. In the office room, four clusters appear with the highest likelihood, independently of the frequency band. For the conference room, five and four clusters occur with highest likelihood in the band GHz, and GHz, respectively. In general, we observed that the number of clusters slightly decreases with increasing frequency. For instance, i) the probability of having five clusters is always higher at GHz than that at GHz and ii) we often assess two clusters at GHz, whereas the minimum number of clusters is three at GHz. This observation might be explained by the higher path loss at the higher frequencies. At GHz, we obtain 4 and 4-5 clusters for the office and conference room, respectively. For the E band, 4 clusters have been observed in both environments. Only the LOS conditions are considered here. Furthermore, we experimentally observed from the clustering results that the first arriving cluster differs from the others, i.e., second, third, fourth, and so on As shown in Figure A.14, the first cluster contributes mostly in the total received power as compared to the following clusters. Moreover, the first cluster is usually comprised of much more multipath than the following clusters. Figure A.13 shows the cdf of the number of sub-paths per cluster where we distinguish the first cluster from the following because of the aforementioned reason. We can see from the figure that the number of sub-paths of the first cluster differs notably from that of the following clusters. In the office room, the first cluster (resp. following clusters) - on the average - is (resp. are) comprised of 12 and 10 sub-paths (resp. 4 and 3 sub-paths) at GHz and GHz, respectively. For the conference room, the first cluster (resp. following clusters) - on the average - is (resp. are) comprised of 9 sub-paths (resp. 3 sub-paths) for both frequency bands. Therefore, we conclude that the number of cluster and sub-paths per cluster is very similar in both environments and bands. mmmagic Public 116

137 Figure A.13: Cumulative distribution function of the number of sub-paths per cluster. We now investigate the dispersion properties of the clusters in the delay and angular domains. The general definition of the root mean square (rms) angular spread provides values from 0 to 1, and overcome the cyclicity of the angles when calculating the directional spread. However, for small angular spread values as expected for clusters, we adopt the definition of the rms angular spread in [CBY+05]. Table A.2 shows the rms angular and delay spread values in the office room. For an Rx location, each parameter is calculated by averaging the values obtained from all the positions at that location, i.e., 60 positions at GHz and 24 positions at GHz. Table A.2 Average rms angular and delay spread per cluster and Rx location for both frequencies (bandwidth of 6 GHz) in the office room. [-] indicates that the related cluster does not exist GHz GHz 1 st cluster 2 nd cluster 3 rd cluster 4 th cluster 5 th cluster σ φ[ ] τ rms [ns] σ φ[ ] τ rms [ns] σ φ[ ] τ rms [ns] σ φ[ ] τ rms [ns] σ φ[ ] τ rms [ns] Rx Rx Rx Rx Rx Rx [-] [-] Rx Rx We performed the measurements over only 14 positions in the conference room, which is not enough to assess meaningful average results. Therefore, we only show the results for the office room. In addition, we investigate the power ratio of the first cluster in each of the considered scenarios and frequency. As it can be seen in Figure A.14, the first cluster carries most of the received power. Furthermore, it turns out from the experiments that the first cluster contribution in the total power is higher at GHz than that at GHz. This is in agreement with the aforementioned comments about the number of clusters at both bands. mmmagic Public 117

138 Figure A.14: First cluster contribution in the total power. The statistics of intra-clusters large scale parameters such as the angular spread (c ASA ) and rms delay spread (c DS ) have also been determined. Figure A.15 shows the cdf of the rms (azimuthal) angle of arrival spread values of the intra-clusters in both environments and bands. Similarly, Figure A.16 shows those of the rms delay spread. The Kolmogorov-Smirnov (KS) test shows that the values of c ASA at both bands follow the same probability distribution, regardless of the environment. Therefore, the statistics of c ASA (e.g., median value) are frequency independent, at least for the considered bands. The new 3GPP channel model [3GPP38.900] shows frequency dependency of c ASA, i.e., decrease with increasing frequency. However, the analytical frequency dependency of c ASA indicated in the new 3GPP channel model yields to a difference of less than 1 for the considered centre frequencies, which is negligible. Figure A.15: cdf of intra cluster rms angular spread. mmmagic Public 118

139 Figure A.16: cdf of intra cluster rms delay spread. The KS test shows that the values of c DS in the conference room follow the same probability function in both bands. In the office room, the values of c DS for the two bands do not follow the same distribution. Notwithstanding, it is observed that the values of c DS (e.g., the median) are higher in the GHz band than those at GHz. The rms delay spread slightly decreases with frequency and the frequency-dependent model is reported in Table 3.4. Furthermore, he median value of the intra-clusters parameters are summarized in Table 3.9. On one hand, the statistics show that the rms (azimuthal) angle of arrival spreads of intra-clusters follow the same probability distribution, regardless of the frequency. On the other hand, the cumulative distribution functions analysis shows that the intra-clusters rms delay spreads are higher at 62 GHz than those obtained at 83.5 GHz, regardless of the environment. Now, we determine from the measurements the parameters of the cluster model (i.e., extended S-V) defined in Section Moreover, the statistical properties of the angular parameters of the clusters and rays are derived as well. Figure A.17, and Figure A.18 show the assessment of the clusters and rays arrival rate, the clusters and rays decay constant, respectively, in the office room at both frequency bands from the measurements. Figure A.17: Clusters and rays arrival rate in the office room. mmmagic Public 119

140 Figure A.18: Clusters and rays decay constant in the office room. We only show the plots of the office room measurements for the sake of space and clarity. However, the clusters and rays decay constant and arrival rate average values in the two environments and frequency bands are summarized in Table 3.8 (Section 3.3.4). We also determine the statistical description of the ray arrival angles. The ray arrival angles are relative to the mean arrival angle of the considered cluster. The arrival angle of a cluster is the mean arrival angle of all paths comprised in that cluster. Figure A.19 shows the probability density function (pdf) of the relative arrival azimuth angle of the rays in the office room. We clearly see from the figure that the Laplacian distribution suits best to the measurements. The mean and standard deviation of the relative arrival angle are μ =4, σ = 29 and μ =7, σ = 29 at 62 GHz and 83.5 GHz, respectively, for the office room. Figure A.19 Relative arrival angle of the rays in the office. In theory, the mean value of the relative angle is expected to be zero. We obtain a shift of the mean value due to the angular resolution during the measurements, 10 step at the Rx in the two bands and 45 step at the Tx at 83.5 GHz. Following the occurrence of the ray arrival angles, we now address those of the cluster arrival angles. For this purpose, the cumulative distribution function (cdf) of the clusters relative arrival angles is derived and shown in Figure A.20. Except outliers occurring around 50 at 62 GHz (Office) and around for both frequencies (Conference room), the cdf plot can be roughly approximated with a straight line in each case. Therefore, this indicates that the cluster mmmagic Public 120

141 relative angle is quasi - uniformly distributed in [ ] at 83.5 GHz and in [ ] at 62 GHz in the office. For the conference room, the cluster relative angle is uniformly distributed between at 62 GHz and between at 83.5 GHz. This distribution is not spread over [0-360 ) because clusters appearing near the reference cluster are considered as part of the reference cluster. The same behaviour or trend has already been observed and reported [SJJ+00]. Notwithstanding, the results show that the clusters arrival angles are uniformly distributed over (approximately) all azimuthal angles. Figure A.20: Relative arrival angle of the clusters. Correlation of time and angle The arrival of clusters and rays are modelled with their probability density functions shown in (3-12) and (3-13), respectively. However, these distributions can be used only for time-only models (original S-V model), i.e., (3-10) without the second delta function. With the introduction of the angle of arrival into the model, the pdf of clusters and rays arrival become dependent of the corresponding arrival angles and are expressed as follows [SJJ+00]: p(t l, Θ l T l 1, Θ 0 ) = p(t l T l 1 )p(θ l Θ 0 ) p(τ kl, φ kl τ (k 1)l ) = p(τ kl τ (k 1)l )p(φ kl ) (A-5) (A-6) The above equations show that the distributions of the clusters and the rays are separable functions, which is only valid if the time and angle distributions are independent. Figure A.21 and Figure A.22 show the scatter plot of the arrival times versus arrival angle of azimuth for the clusters and rays, respectively, in both environments. In Figure A.21, we see that the time of arrival and the angle of arrival for the clusters are not correlated. One can notice that the higher time of arrival does not imply necessarily higher or lower angle of arrival, meaning thereby a decorrelation between these variables. Moreover, the low inter-correlation values (i.e., ρ) between the times and the angles of arrival of the clusters (displayed in the figures) clearly indicate decorrelation between the two variables. The same comments are valid for the rays where we can see from Figure A.22 that their times and angles of arrival are uncorrelated. Similarly, the low inter-correlation values shown in the figures indicate decorrelation between the two variables. The decorrelation between times and angles of arrival for both clusters and rays imply that their statistics are independent. Therefore, their conditional distributions can be respectively separable functions as shown in (A-5) for the clusters and (A-6) for the rays. mmmagic Public 121

142 Figure A.21: Scatter plot of arrival times versus arrival angles for clusters. mmmagic Public 122

143 Figure A.22: Scatter plot of arrival times versus arrival angles for rays. A.1.2 Indoor measurement campaigns at 2.4, 6, 15 & 60 GHz In this section channel measurement results from indoor scenarios are presented. The focus is to determine if there are any frequency trends. In order to provide a credible comparison between measurements at different radio frequencies, the requirements introduced in Section 3.1 have to be fulfilled, recapitulated here: Requirements (must be fulfilled) Equal measurement bandwidth Equal antenna pattern, either physical or synthesized Equal dynamic range for analysis both in delay and angle domains Equal angle resolution (e.g., array size equal in terms of number of lambda) Same environment and same antenna locations mmmagic Public 123

144 Other requirements (relatively small effects or not applicable to all results) Compensation of atmospheric absorption at the 60 GHz-band Sufficient sample size Static environment (when measurements are made successively) Same path estimation algorithms Same area of spatial averaging Only few of the results reported in literature fulfil these requirements. It has been found that if any of the above given requirements is not fulfilled, fictitious frequency trends may be observed. These requirements are however fulfilled for the results reported below. As a result of thorough analysis, no clear frequency trend has been observed as shown in the following two subsections. Frequency dependency of measured highly resolved directional propagation channel characteristics The result presented in this section is a summary of a comprehensive published paper [MSA16]. The paper focuses on determining the frequency dependence of directional as well as wideband indoor channel properties. A novel method is used for the analysis, which is based on using a cubic virtual array and has previously been proven to provide exceptional measurement accuracy at 58.7 GHz. It is shown that the measured channel power distributions over direction and delay are surprisingly similar over the full frequency range in both LOS and NLOS conditions. One exception is that the window transmission attenuation and reflectivity is substantially different at the two frequencies under test (i.e. 5.8 GHz and 14.8 GHz). This difference results in that one of two dominant paths at 14.8 GHz goes out of the building and is reflected off an adjacent building back to the receiver location. This does not occur at 5.8 GHz as the windows block penetration at this frequency. Both measurement scenarios are depicted in Figure A.23 and Figure A.24. The corresponding directional power distributions in the full space angle are shown in Figure A.25. Figure A.23: Measurement scenario LOS mmmagic Public 124

145 Figure A.24: Measurement scenario NLOS The distributions are strikingly similar for 5.8, 14.8 and 58.7 GHz showing no obvious frequency trend. In Figure A.26, the measured RMS angle and delay spreads are shown together with corresponding power delay profiles. There is no evident frequency trend except for the increase in delay spread for NLOS and 14.8 GHz. This is however not a monotonic frequency dependency. It was found that the additional peak in the NLOS power delay profile at 14.8 GHz is purely an effect of full transparency of windows at this frequency, whereas the windows are causing substantial attenuation at 5.8 GHz. This effect is attributed to the multiple reflections caused by the three glass layers of the window. Further, this effect is resonant for integer number of wavelengths, inside the different layers of glass and air space, resulting in an oscillating behaviour over considered frequency range. At 14.8 GHz there is a strong pathway out of the window reflected off an adjacent building back in again through another window as illustrated in Figure A.27. This pathway is heavily attenuated at 5.8 GHz. Figure A.25: Power angle distributions for full space angle for the LOS and NLOS scenarios for 5.8, 14.8 and 58.7 GHz. mmmagic Public 125

146 Figure A.26: RMS power delay and angle spreads (left) and power delay profiles (right) for LOS and NLOS for the different frequencies. Figure A.27: The window blocking effect at 5.8 GHz and transparency at 14.8 GHz is illustrated in the graphs above. The main takeaway of these measurements is that no clear dependence of angle and delay spreads on frequency is observed over the range 6-60 GHz. Indoor delay spread frequency dependence In this measurement campaign, the indoor delay spread frequency dependence is investigated. The measurement scenario is depicted in Figure A.28. The receiver (Rx) was placed at a fixed location and the transmitters (Tx) at 15 different locations mainly in NLOS. All aforementioned requirements for comparability over different frequencies were fulfilled. The channel was measured at 2.4, 5.8, 14.8 and 58.7 GHz. mmmagic Public 126

147 Figure A.28: Indoor channel impulse response measurement scenario. In Figure A.29, power delay profiles of six (exemplary) Tx locations are shown, and in Figure A.30 the median RMS delay spread as a function of frequency is shown. Moreover, spread bars indicating the 10% to 90% range are shown. Once more, no clear frequency dependence is observed. Figure A.29: Power delay profiles for TX positions 1-4, and, 13, 14. mmmagic Public 127

148 Figure A.30: Median RMS delay spread as a function of frequency. Ultra-high spatial resolution indoor measurements at 60 GHz The results presented in this section are based on an indoor channel measurement campaign with the goal to provide as high resolution in direction as possible. For this purpose, the same measurement set-up is used as presented in the previous section and is depicted in Figure A.31. A vertically oriented planar virtual antenna array, using a vertically polarized open waveguide, with 256 and 64 elements in the x and z directions, respectively, is used. The array element distance is 2 mm. This configuration provides a directional resolution better than 1 degree in azimuth. In order to maximize the resolution, the highest possible carrier frequency of the setup is used, namely GHz. For the same purpose, the measurement bandwidth is maximized, which in this case is 2 GHz. Figure A.31: Virtual antenna array set-up. The measurement scenario, in a typical indoor office environment, is depicted in Figure A.32. As the TX antenna array is large, the resolution is somewhat degraded for scattering objects that are located close. For this reason, the TX antenna-positioning robot is placed in an open space area. Power delay profiles (PDP) from the four RX locations are shown in Figure A.33. Note that the more obstructed scenarios (RX1 and RX3) have fewer clear spikes in their PDPs. By overlaying propagation path lengths and direction of arrivals with photographs and floorplans of the environment, it was observed that these spikes are specular reflections. Figure A.34 shows angular mmmagic Public 128

149 power spectrums superimposed on measurement photos. Another observation made by analysing the angular power spectra is that the channel is much richer for the NLOS scenarios compared to the LOS scenario. One striking observation is that the most significant scatterer in the deepest NLOS measurement (RX1) actually is a curtain. Though the curtain is rather thick, this effect was not expected. The corrugated shape makes it scatter waves in all directions. This result is of great value as it shows that standard ray tracing may be inadequate for channel modelling in many cases. Figure A.32: Measurement scenario. The virtual antenna array is located at TX. The four RX antenna locations marked RX1-RX4. Figure A.33: Power delay profiles for the different RX locations. The expected free space loss for versus distance is marked with a dotted line. mmmagic Public 129

150 Figure A.34: Directional power distributions on panoramic photo from the TX array location. The measurement data is further analysed to determine the distributions of multipath components (MPCs). For each peak of the three-dimensional measured channel responses (2D in space + 1D in delay) a corresponding multipath component is identified. It turns out that the number of MPCs may be very large due to the very high resolution in both direction and delay as shown in Figure A.35. Figure A.35: Identified MPCs in elevation vs. azimuth (left) and propagation distance vs. azimuth (right) scatter plots for RX1 (upper) and RX2 (lower). mmmagic Public 130

151 Figure A.36 shows cumulative distribution functions of the total received power vs. number of multipath components. Note that the MPCs are ordered in descending power. A key observation is that the more obstructed the scenario is (RX1 and RX3), the more MPCs are needed to account for a substantial fraction of the received power. It may also be noted that the MPCs are more spread out in angular domain in the NLOS scenarios than in the LOS scenario as shown in Figure A.35. Another interesting observation is that the LOS scenario indicate clustering in both angle and delay domain, whereas the heavily obstructed scenario RX1 is predominantly clustered in angle and spread out in delay. Cumulative Fraction of Total Power RX1 RX2 0.1 RX3 RX Power ordered MPC number Figure A.36: Cumulative distributions of fraction received power versus number of MPCs for the rectangular very high resolution array (upper) and the cubic array described previously on page 124 (lower graph). Figure A.36 shows that the number of identified MPCs depends strongly on the measurement resolution. The higher the resolution, the more MPCs are identified. This is quantified in Table A.3 below where the number of MPCs for both the medium resolution cube array measurement (page 128) and the very high resolution rectangular array measurement are provided. Table A.3: Number of MPCs needed to account for 95% of power. BW=2 GHz Array size=100 λ BW=150 GHz Array size=10 λ LOS NLOS A.1.3 Measurements in airport check-in area at 15, 28, 60 and 86 GHz Aalto University s airport channel sounding measurements were done at the Helsinki airport s check-in area at the Terminal 2 for the frequency bands of GHz, GHz, 59-63, and mmmagic Public 131

152 GHz. Figure A.37 shows the measurement environment including the main terminal hall and the side corridor. The floorplan of the environment with the measured points and exemplary PDP of one LOS link in different frequency bands are shown in Figure A.38. We refer to [mmmagic D2.1, Section 6.6.1] and [VJN+16] for the specific details of the measurement environment and setup. From the measurements, delay and azimuth spreads have been calculated as presented in Table A.4. Here, a bandwidth of 2 GHz was used for all frequencies. Figure A.37: Measurement environment in Helsinki airport check-in area: main terminal hall (left) and side corridor (right). Figure A.38: (Left) Floor plan of the environment with the measured locations in the airport, and (right) the exemplary PDP of one LOS link in different frequency bands. LOS NLOS Table A.4: Mean spread values in Helsinki airport 15 GHz 28 GHz 60 GHz 86 GHz DS [ns] ASA [ o ] DS [ns] 55 N/A ASA [ o ] 28 N/A A.1.4 Simulations in airport check-in area at 15, 28, 60 and 86 GHz The channel simulations performed by Aalto rely on a newly developed deterministic field prediction tool (ray tracer) which is based on point clouds [JHK16]. The accurate descriptions of the propagation environments, i.e., the point clouds, are obtained by laser scanning [JKH15]. The prediction method takes into account both specular reflections and scattering, and combines the paths to give the total field. The effect of shadowing caused by blocking objects in the environment such as lamp posts or people is taken into account by assuming that points found mmmagic Public 132

153 inside the Fresnel zone induces additional attenuation to a path [JNH+16]. The material parameters of the model, i.e., the permittivity and the shadowing loss, are tuned based on channel measurements. The point cloud-based simulation at the airport was carried out by omitting the scattering, and only calculating reflections up to the second order, as the higher order reflections were found negligible. The channel sounding results at 15 and 60 GHz were used to calibrate the permittivity ε r and attenuation loss L a of the shadowing objects in the environment. Paths propagating through walls were assigned with very high (> 70 db) attenuation losses, which means that these paths will not contribute to the received power. For 15 and 60 GHz, the optimization yielded ε r values of 4.2 and 3.6, respectively. A comparison between measured and simulated large scale parameters is shown in Figure A.39 and Figure A.40, and examples of measured and simulated PDPs are presented in Figure A.41. The average measured and simulated values in the two frequency bands are presented in Table A.5. The permittivity for 28 and 86 GHz was derived by linear fitting, and the attenuation loss L a was chosen heuristically. The final parameter values are presented in Table A.6. Figure A.39: Large scale parameters for measured (black) and optimized simulated (green) channels at 15 GHz. Figure A.40: Large scale parameters for measured (black) and optimized simulated (green) channels at 60 GHz. mmmagic Public 133

154 Figure A.41: Measured and simulated PDPs at 15 GHz (left) and 60 GHz (right). Table A.5. Average measured and simulated large scale parameters. Freq. ε r Path loss [db] Mean delay [ns] Delay spread [ns] Meas. Sim. Meas. Sim. Meas. Sim. 15 GHz GHz Table A.6: Material parameter values Frequency ε r L a 15 GHz GHz GHz GHz Next, a large set of channel data is generated in order to model the large scale behaviour of LOS channels. Channel data is calculated for 15, 28, 60 and 86 GHz with the parameters presented in Table A.6. Base stations (BSs) and mobile stations (MSs) were placed in the main terminal hall as depicted by Figure A.42. In total, 5937 LOS links with 4 BS locations were calculated. The MS spacing was 0.6 m and the link distance varied between 4 and 64 m. The BS and MS heights were 5.7 and 1.5 m, respectively. The parameter tables for the airport are presented in Table D.4. mmmagic Public 134

155 Figure A.42: Simulated BS (Rx) and MS (Tx) positions in airport. A.1.5 Ultra-wide band measurements at 7, 34 and 60 GHz Large scale channel fading measurements at 7 and 34 GHz Figure A.43 shows a 360 panoramic view of a small lecture room at the Technische Universität Ilmenau campus. The dimensions of the room are 7.36 m x 4.7 m x 3.43 m. The walls are made up of bricks, where one of the inner-walls is covered by metallic blackboards and the opposing wall has 5 cm thick sound absorbers. The outer wall is largely covered by the windows with metallic frames. Doors and windows are closed during the measurements, otherwise indicated. The distance to the building opposite to the seminar room is approximately 55 m. A perpendicular wing of the building next to the seminar room is approximately 19 m away. Measurements are conducted in a LOS environment, however during the azimuth scans, both TX/RX horn antennas may also illuminate a pseudo-los (plos) environment. Here, a plos environment corresponds to a situation when both TX and RX antennas are not pointed exactly towards each other during the azimuth scans. Figure A.44 and Figure A.45 show an overview of the measurement setup. The channel sounder used in this measurement is described in [MHD+14]. This channel sounder has a 3 db measurement bandwidth of 4 GHz after calibration. During the measurements, a linear positioning device is used for controlling the linear RX movements and two spherical positioning devices are used for fixing TX/RX antenna pointing angles. In the subsequent discussions, the linear positioning device is denoted as rail. One spherical positioning device carries the transmitter of the mm-wave band (30.4 to 37.1 GHz) and the receiver of the FCC-UWB band (3.4 to 10.1 GHz). This transceiver unit is denoted as TX MRX F and is kept at a fixed location throughout all measurements. The other spherical positioning device carries the receiver of the mm-wave band and the transmitter of the FCC-UWB band. It is denoted as TX FRX M. This unit is mobile and is moved to 6 positions with half a meter step size on the rail. Additionally, the rail was placed at 5 different locations inside the room to cover the entire room and to collect measurement data for various transceiver distances. In this way, overall 30 locations have been measured inside the room. For both frequency bands, dual polarized horn antennas with identical 3 db half power beam width of 30 with dbi gains have been used. As shown in Figure A.45, these antennas are mounted at the positioners in opposite directions, and hence they mmmagic Public 135

156 scan 180 opposite directions at the same time for both mm-wave and FCC UWB bands using TX MRX F and TX FRX M transceiver units. Both spherical positioning devices perform a 360 azimuth scan of lecture room in 12 steps at a particular position. In this way, for 30 locations inside the room, a total of 12 x 12 x 6 x 5=4320 directional power delay profiles (PDP) have been measured for deriving the temporal characteristics of the channel. Three PC s are used to control the positioning devices, switches for the polarization and record the data sent from the UWB sounding units. The height of both TXFRXM and TXMRXF units is 1.68 m, which is approximately at eye level. Table A.7 shows a list of requirements that have been fulfilled during the measurement in context of Table 3.1. Figure A.43: 360 panoramic view of measurement scenario. Figure A.44: Overview of the measured TX- RX locations. Figure A.45: Overview of the channel sounding setup. Table A.7: Measurement campaign in view of requirements defined in Section 3.1 # Requirement Centre frequency 7/34 GHz 1 Equal measurement bandwidth 4 GHz ( ) 2 Equal antenna pattern 3 db HPBW 30 ( ) 3 Equal dynamic range for analysis both in delay and angle domains 25 db below the max power peak ( ) mmmagic Public 136

157 5 Same environment and same antenna locations Yes ( ) Small scale fading measurements at 34 GHz In our experimental setup, UWB measurements are carried out at 34 GHz centre frequency. The Rx sounding unit is placed in the centre of the room at the rail with an initial 4.06 m distance between both Tx-Rx units. Then the Rx unit is moved in a certain direction according to the setups with a step distance of 2 mm. This step distance is approximately λ f 4, where λ f = 8.01 mm corresponds to the wavelength of the mm-wave-uwb upper edge frequency of 37.1 GHz. A total measured distance from first to the last Tx-Rx position is around 30 cm in each of the 4 experimental setups shown below. Figure A.46: Overview of the measurement setups. Small scale fading measurements at 60 GHz In order to study the fading characteristics at 60 GHz, we have performed measurements in a room as indicated in Figure A.47. The use of beamforming in a NLOS condition was emulated by using directive antennas pointing to a wall in order to induce a specular reflection. Furthermore, the RX was automatically moved along a track in λ steps to investigate the variations on 2 the signal strength during displacement. Besides the reflection in the wall, there are other scatterers as a monitor on the side, shown in Figure A.47, which influences the channel when using low directive antennas. However, these scatterers remain almost invisible with high directive antennas since they are in a different direction. The TX was mounted on a static tripod pointing to the wall with an incident angle of 30 relative to the normal of the wall. On the other side, the RX was located on a rail pointing to the same spot in the wall, with an incident angle of 30, as shown in Figure A.47. The position of the rail was adjusted with a laser pointer in such a manner that the RX keeps focus on the same point on the wall while moving along the rail. This is important to avoid fluctuations on the received signal of the main reflected path due to misalignment with the antenna's bore-sight. mmmagic Public 137

158 Figure A.47: Picture of the measurement set-up and the Top-view schematic of the measurement set-up. The RX was displaced automatically along 2.2 m in 2.5 mm steps ( λ at 60 GHz) giving a total 2 of 881 measurement positions. However, the spatial resolution of the channel sounder with 4 GHz bandwidth is about 5.88 cm, meaning that there are around 24 samples in space (positions on the rail) that correspond to the same delay tap of the measured channel impulse response. This work shows only the investigation results for TX-V and RX-V. Table A.8 summarizes the different bandwidths and combination of antennas that were used at the TX and RX. Table A.8: Antennas and bandwidths used during the experiments TX RX Visibility Bandwidth [GHz] Omni V Omni V LOS 0.1, 0.2, 0.4, 1, 2, 4 15 HPBW 15 HPBW NLOS 0.1, 0.2, 0.4, 1, 2, 4 15 HPBW 30 HPBW NLOS 0.1, 0.2, 0.4, 1, 2, 4 15 HPBW Omni V NLOS 0.1, 0.2, 0.4, 1, 2, 4 Analysis procedure Let H ab (x, f) correspond to a channel transfer function (CTF) when a signal is transmitted with a polarization a and received with a polarization b. Let x correspond to a particular measurement position on the rail. Then the CTF can be expressed as n f H ab (x, f) = A ab (x, n)e jφab (x,f) δ(f n f), n=1 (A-7) where, A ab and φ ab correspond to the magnitude and phase responses of the channel at the n th frequency bin of the CTF at position x. For bandwidth reduction, we adopt the same procedure as used in [MAE08]. At first, the frequency domain samples are extracted while maintaining the centre frequency f c, delay resolution τ and the length of the full-band frequency domain channel such that mmmagic Public 138

159 H ab W (x, f) = { Hab (x, f) f l f c f h 0 otherwise. (A-8) Now the channel bandwidth W is defined as W = f h f l, where f h and f l are the highest and lowest frequencies of the band, respectively. Time domain channel impulse response (CIR) vector h W ab (x, τ) is now obtained by inverse Fourier transform (IFFT) of H W ab (x, f). Figure A.48 shows an example of the CIR for various channel bandwidths. For comparison purpose, the power P(x) of the channel transfer function H W ab (x, f) corresponding to a particular bandwidth WW and a position xx on the rail is normalized such that f h H ab W (x, f) 2 df f P(x) = l, (A-9) 1 L f h H W ab (x, f) 2 df dx f l where, L corresponds to a total number of locations at the rail. Normalization in (A-9) removes the global pathloss (due to an initial Tx-Rx distance) of each measurement, however, local pathloss or path gain effect due to movement afterwards on the rail is maintained. As discussed earlier, that due to reduced bin width at higher bandwidths, a Rice distribution is most likely to fit the fading envelope. It is defined as p ζ (z) = z σ 0 2 e z 2 +ρ 2 2σ 0 2 I 0 ( zρ ), z 0, (A-10) σ 0 2 where, I 0 is a modified Bessel function of first kind with zeroth order. The K-factor is defined as the ratio of power of fixed ρ2 to the power of scattered MPCs σ By fixed path, we mean either a direct LOS path or strong specular reflection from a reflecting surface: K = ρ2 2 2σ. (A-11) 0 In Eq. (A-11), ρ and σ 0 are determined by fitting the data in P(x) to its best Rician distribution fit. The fade depth F is defined as s times the standard deviation of signal received power P(x) from its mean [MAE08], where s is a system parameter: F = sσ. Now the XPD channel power X a W (x) for a particular bandwidth W is computed as (A-12) X a W (x) = f h H W aa (x, f) 2 df f l H ab W (x, f) 2, (A-13) f h df f l where, H W aa (x, f) corresponds to a channel when a signal transmitted with polarization a is received with same polarization. The degree of randomness of the XPD is quantified by the coefficient of variance (CV), which is the ratio of standard deviation of X W a (x) at all positions to their ensemble value. CV a (W) = std x {X W a (x)} E x {X W a (x)} (A-14) mmmagic Public 139

160 Figure A.48: CIR for various bandwidths in LOS case, H-H polarization setup. Delay spread analysis for different bandwidths At mm-wave frequencies, a large amount of bandwidth is available and ultra-wide band (UWB) channel setups could easily be realized. Therefore, increased resolution in the delay domain increasingly leads to reduced number of non-resolvable MPCs. Hence, fading statistics of MPCs will no longer be Rayleigh and uncorrelated scattering assumption breaks down [MOL05]. Therefore, it is expected that with increased system bandwidth (reduced bin width), the fading statistics of mm-wave channels will increasingly deviate from Rayleigh towards Rice distribution as shown in Section 3.4. Channel characterization of FCC UWB band in [MAE08] and mm-wave UWB channels in [NCJ17] have provided evidences that increase in channel bandwidth results in an increase in Rician K-factor. K-factor is in general negatively correlated with rms DS [BST11,WINNER]. Therefore, it is quite intuitive that higher bandwidth results in reduced rms DS given that same dynamic range is applied to CIR obtained at a particular bandwidth. Note that absolute 3 db bandwidth of our measured channel is around 4 GHz after calibration; therefore for further bandwidth reduction in our analysis we have applied the procedure described in (A-7) and (A-8).Figure A.49 shows the average directional DS results for a bandwidth in the range of GHz. Average DS values have been calculated from 4320 (144 per position times 30 measured locations) directional PDPs. It can be observed that DS reduces significantly with the bandwidth and after a certain bandwidth beyond 2 GHz there is no significant change in the rms DS. However, one may note that there is no clear trend in rms DS values between both frequency bands for different Tx-Rx polarization cases. At 4 GHz bandwidth, a maximum of 22 ns difference is observed, which demonstrates a high degree of similarity between two frequency bands. mmmagic Public 140

161 Figure A.49: Average directional delay spread [ns] vs. bandwidth, dynamic threshold = 25 db, standard deviation [ns] values are shown at each average rms DS value along the plot. Dynamic threshold and strongest beam selection Now we consider the case where only a single maximum power TX-RX beam pair is selected for each of the 30 measured locations. For the results in Figure A.50, average rms DS is computed over 30 directional beams. It reveals that beam selection reduces the DS significantly in each polarization setup. However, higher dynamic thresholds may increase the DS. Results show that if the RX is horizontally polarized, the DS in the 7 GHz band increases more with the dynamic threshold as compared to the 34 GHz channel. This basically provides evidence that weak reflections at longer delays are more likely to occur at 7 GHz. Note that for the 34 GHz mmmagic Public 141

162 band, the DS with beam selection is always lower than 2.5 ns, whereas in the 7 GHz band it is always lower than 44 ns for different polarization setups and dynamic ranges. The above discussion leads to the conclusion that the channel bandwidth has a significant impact on the fading behaviour of the channel. Our results clearly demonstrate that the directional rms DS reduces with bandwidth, which results in increased coherence bandwidth. Beam selection further reduces the rms DS. However, our studies of different Tx-Rx polarization combinations show that there is no clear trend in the rms DS of two frequency bands. Depending on the polarization setup, the rms DS for two frequency bands may reverse. Figure A.50: Average directional DS with and without beam selection, bandwidth = 4 GHz. A.1.6 Measurement on spatial consistency Description of the measurement kit Transmitter description A 2 GHz wide baseband signal was generated using Keysight M9099 Waveform Creator software and used to configure a Keysight M8190A arbitrary waveform generator (ARB). This signal is taken from the direct output of the two channels in the ARB such as I and Q signals. In each of the two ARB a transmit signal voltage of 700 mvolts was set. Power measurements in the same outputs gave a value of dbm. This signal is driven to I and Q ports on a Sivers IMA up-converter through three 1-meter high performance coaxial cables (M17/60-RG142) with SMA connectors. The cable produced each a power loss of -1.7 db, so the input power in the Sivers device I and Q ports were dbm. mmmagic Public 142

163 M8190A Chassis M9505A Arbitrary Waveform Generator SMA SMA DIRECT DIRECT 0.7 VOUT OUT 0.7 V TXQ1 CH1 TXI1 CH GHz N5183B MXG Analog Signal Generator RF OUT TXI GHz GHz TXQ2 SMA -1.7 db -1.7 db TXLO TXI3 TXQ3 SIVERS-IMA TX FC1003V/01 Up-converter dbm GHz TXIF-I GHz TXLO CO1211A/00 Power Supply Up 1-meter cable (M17/60-RG142) MIL-DTL17 with SMA connectors in both sides SMA SMA db 9.64 GHz 1.5-meter Low phase noise cable with SMA connectors in both sides TX 15 60GHz +25 db 40 60GHz Figure A.51: Channel Sounder Transmitter Description. The 60 GHz passband signal is generated using a Keysight N5183B MXG Analog Signal Generator. The signal generator is configured to drive a 15 GHz signal to Local oscillator (LO) port on a Sivers IMA up-converter through a low phase noise cable. The power generated by the Signal Generator was dbm, and the power driven to the Sivers LO port was 9.64 dbm, which implies a cable loss of 2.83 db. The Sivers up-converter internally multiply the frequency times four (X4) in order to create a 60 GHz. In addition, the Sivers IMA up-converter modulated the 60 GHz signal with the 2 GHz baseband signal. The power signal at the antenna input was 15 dbm. The Keysight VSA & Waveform Creator channel sounding function operates by repeatedly transmitting a single carrier signal, bearing a modulated waveform. The waveform has excellent auto-correlation properties, and a low peak-to-average power ratio. The bandwidth and duration of the modulating waveform may be varied to suit the channel measurement required. Spectrum shaping can be applied to reduce out of band interference when transmitting the signal in a live environment. Receiver description The 60 GHz modulated signal is first captured by the receiving circular horn antenna, and thereafter the signal is directed to an orthomode transducer which can be used to split the signal into its vertical and horizontal component (Co and Cross Polar). The isolation provided by the orthomode has been measured approximately 20 db. The orthomode is connected to two Sivers IMA devices in order to down-convert the two 60GHz signals into I and Q IF signals which are then directed to the performance digital oscilloscope (DSO), MSOS804A. A back to back calibration was performed to determine the equivalence between the power measured in the oscilloscope and the actual power in the system. According to this test, the value at the input of the orthomode transducer is 11 db lower than the value measured for the scope within the dynamic range. This difference will be explained later. mmmagic Public 143

164 N5183B MXG Analog Signal Generator SMA GHz -5.5 GHz RF OUT -5.6 db ZX Power Splitter Combiner 13.9 dbm 12 dbm 25 GHz SMA SMA IN OUT OUT IN OUT SMA ZX60-183A+ Wideband Microwave Amplifier Power Supply for Amplifier SIVERS-IMA RX1 (TOP) CO1211A/00 Power Supply Up FC1003V/01 Up-converter SMA RXIF-Q RXIF-I SMA SMA RXLO 11 GHz Transducer Orthomode LX Port1 Digital oscilloscope, MSOS804A BNC CH4 BNC CH3 BNC CH2 CH1 BNC SIVERS-IMA RX2 (BOTTOM) CO1211A/00 Power Supply Up FC1003V/01 Up-converter SMA RXIF-Q RXIF-I SMA SMA RXLO 1.5-meter cable coax cable SLU18-SMSM-01.50M 10-cm cable coax with SMA connectors 3-meter cable (M17/60-RG142) MIL-DTL17 with SMA connectors Ethernet Cable with RJ45 connectors Figure A.52: Channel Sounder Receiver Description. The down-conversion process also requires a 15 GHz signal in the LO port of the Sivers IMA transceiver, which is generated in a way similar to the one used at the transmitter. However, there is a difference at this instance, as the LO signal must be driven into two Sivers devices and for this reason a power amplifier and a splitter have also been used. The output power in the signal generator was set to -5.5 dbm. The power driven into the power amplifier was dbm, with a 25 db gain and the output of the power amplifier was 13.9 dbm. In addition, for splitting the signal into two Sivers devices a splitter was used, whose loss is about 3 db including a small cable to the two LO ports. Finally, the input power in both Sivers devices LO ports is 11 dbm. The receiver was mounted on a trolley. The setup of the trolley will be explained in each individual measurement campaign, as it has been used in various configurations. mmmagic Public 144

165 Figure A.53: Channel Sounder Keysight Based Indoor measurement MVB (University of Bristol) atrium measurements Measurement campaign setup The transmitter was mounted on a large pole in a fixed position in the MVB Atrium (Figure A.54) at a height of 2.5 metres. Figure A.54: Transmitter Setup used for Indoor Large Scale Measurements The receiver was attached in a pole, whereas at the same time it was attached to a motor placed on the trolley to perform the rotation measurements as shown (Figure A.55). mmmagic Public 145

166 Figure A.55: Receiver Setup used for Indoor Large Scale Measurements The receiver was moved along 83 test points along the MVB Upper Atrium, as depicted in Figure A.56. The height of the receiver was 1.5 metres. The separation between the measurement points was 0.5 metres. Figure A.56: Plane of the MVB Atrium, and location of measurement points and Tx. At each of the test points, the receiver performed a 2D scan (360 spin) in the azimuth plane. The resolution of the rotation was 1. The closest point between the Tx and Rx was the point 34 as it is pointed out in the Figure A.56, with a distance of 9.6 metres between them. It is important to note; the transmitter is not pointing directly to the front of the receiver boresight. The receiver has to rotate 11 counter clock wise to point directly to the transmitter. Antenna setup The transmitting antenna was a vertically polarized rectangular horn with 20 dbi gain and 20 of beamwidth. The receiving antenna was a circular horn, which was connected through an orthomode transducer at the receiver allowing the simultaneous recording of both horizontal and vertical components of the incoming signal. The receiving antenna has 25 dbi gain and 14 of beamwidth. mmmagic Public 146

167 Analysis procedure In order to analyse the spatial evolution in the large scale, the 83 test points have been clustered in three different categories for the analysis; LOS, transition between LOS and NLOS (quasi Line of Sight) and NLOS. Once the points have been classified, a correlation based study was performed. Figure A.57 clearly illustrates the grouping of the various points. nlos quasilos LO S Figure A.57: Points of measurement clustered in 3 types of scenarios. The recorded CIRs have been used in order to obtain the PDP at each of the measured angles. Given the 1 0 measurement resolution, 361 PDPs were computed for each of the measured points. The correlation coefficient was computed by correlating the selected PDP at each point. Thereafter, the PDPs with the strongest power were chosen and post-processed for the evaluation of correlation among adjacent spatial points. Outdoor measurement campaign The outdoor channel measurement campaign was carried out at Cantock s Close, outside Merchant Venturer s Building (MVB), at the University of Bristol. The transmitter Tx was located in a fixed point at the third-floor-balcony of MVB building, and was mounted on a pole at 1.95 m. In addition, the balcony was 4.5 metres above the level of the street, and therefore the Tx was at 6.45 m height. The receiver antenna was mounted on a pole, and this in turn on a trolley. The antenna height was 1.75 metres above the level of the street. Sixteen test points have been considered overall. All sixteen test points are located in Cantock s Close north pavement (top pavement), as it is shown in Figure A.58. Figure A.58: Route of the Outdoor Measurement Campaign. mmmagic Public 147

168 These points provide three clear scenarios such as: LOS or points 1-8, transition LOS and NLOS or points 9-11, and NLOS or points Figure A.59: Transmitter for Outdoor Large Scale Measurement Campaign. The Tx and Rx antennas were manually aligned at each of the each test points. Similar to the indoor measurements, the alignment was performed such as the Tx and Rx antennas form a LOS link, and the Rx antenna captures the maximum power. In every test point, the receiver performed a quasi 3D scan (360 spin) in the azimuth plane and 90 (45 to -45 ) in the elevation plane. The signal is captured every 1 in the azimuth plane, and every 5 in elevation plane. The Tx and Rx antennas that have been used are dual polarised circular horn antennas. Both antennas have 25 dbi as nominal gain, and a Half Power Beam-Width (HPBW) of 13, based on antenna pattern measurements. Figure A.60: Receiver for Outdoor Large Scale Measurement Campaign. mmmagic Public 148

169 Although the Tx antenna can be configured for both horizontal and vertical transmissions, only one polarisation was transmitted at a time. On the other side, the Rx antenna was connected to an orthomode transducer, which allows for simultaneous recordings of both polarizations. Table A.9 summarises some important parameters used for the setup in the channel measurement campaign. Table A.9: Outdoors Channel Sounder Measurements Campaign Parameter Value (metres) Rx antenna height from ground 1.75 Tx antenna height from ground 6.45 Distance of Tx from Location Point 1 29 Distance of Tx from Location Point Distance of Tx from Location Point Distance between Location Points 3 In addition, the transmitted power for all points in both polarisations was 15 dbm. Table A.10 summarizes the distances respect to the transmitter location and the distances between the 16 points in which the receiver was measured. Table A.10: Distances of the analysed points Rx Location Distance from the Tx (metres) Distance respect to the previous point Point (metres) 0 Point Point Point Point Point Point Point Point Point Point Point Point Point Point Point Analysis procedure In order to analyse the spatial evolution of the large scale parameters, the 16 measurement points have been clustered in three different categories or scenarios, as mentioned above. Once the points have been classified, a correlation based study was performed. The correlation coefficient was computed by means of correlating the selected PDPs at every measurement point. For the correlation study, only 1 elevation plane has been evaluated, namely the plane at 0 0. It is important to notice that the transmitter and receiver antennas were manually aligned at all the sixteen measurement points. Based on the alignment process, the azimuthal angle with the maximum power has been used as the reference coordinate system for the determination of power angular characteristics of the recorded profiles. mmmagic Public 149

170 Indoor Results In this section, we present the correlation distance values of the recorded impulse responses as obtained from both indoor and outdoor measurement campaigns carried out in the premises of UoB. The correlation was measured by correlating the complex impulse response values of the profiles recorded in adjacent positions. Figure A.61 illustrates high correlation of adjacent points along the LOS measured points. In order to produce these results 35, points have been used and the correlation coefficient amongst the temporal profiles that have been captured, was evaluated. Figure A.61: Correlation Coefficient of recorded impulse responses for all points in LOS. Figure A.62 illustrates the correlation distance for the LoS-NLoS transition. As shown in the figure below, the correlation was found very high for this particular scenario; this could be attributed to the limited number of points that have been used in the transition analysis. Figure A.62: Correlation Coefficient for all points in LOS-NLOS scenario. mmmagic Public 150

171 For the NLoS scenario, a very high correlation was found among the points that have been analysed. Contrary to intuition, when comparing amongst all three different scenarios it was found that the highest correlation was observed for the NLoS case. A possible factor that contributes to the high correlation metrics can be attributed to the limited number of the measured points. Figure A.63: Correlation Coefficient for all points in NLOS scenario A summary of the correlation values found for 0.9 and 0.5 correlation threshold can be seen in the table below: Table A.11: Spatial Correlation values for the atrium indoor measurements Sub-Scenario Distance at 0.9 correlation Distances at 0.5 correlation Line of Sight 13 m Does not match (Much higher correlation) Quasi Line of Sight 2 m Does not match (Much higher correlation) No Line of Sight Does not match (Much higher correlation) Does not match (Much higher correlation) Outdoor Results Figure A.64, the LOS sub-scenario is illustrated by a three-dimensional representation of the received signal along a route dominated by LoS propagation, this point is the point 1. The transition region is only made up of 3 points. In Figure A.64, the point 9 was analysed for this region and it can be seen the power drops about 20 db respect adjacent points due to the partial blockage which is caused by the foliage along the route. Finally, for the NLOS scenario there is a very weak signal caused by wall reflection, which together with the loss caused by foliage results in a path-loss degradation of more than 30 db when compared to the LOS region. mmmagic Public 151

172 Figure A.64. 3D Power Angular Profiles in Outdoor Scenario A summary of the correlation values for the outdoor measurements are presented in Table A.12. Table A.12: Spatial Correlation values for outdoor measurements Sub-Scenario Distance at 0.9 correlation Distances at 0.5 correlation Line of Sight 9 m Does not match (Much higher correlation) Quasi Line of Sight Does not match (Much higher correlation) Does not match (Much higher correlation) No Line of Sight Does not match (Much higher correlation) Does not match (Much higher correlation) Similar to indoor environment, the empirical results indicate that for all three different scenarios, the correlation coefficient is extremely high. Furthermore, it is shown that foliage does not produce a substantial effect in the correlation according to the presented analysis. Finally, NLOS scenario also demonstrates the highest correlation among the PDPs. The correlation is no lower than This evident is quite unexpected since all NLoS measurements were blocked by foliage and building which impose. However, the high correlation values can be interpreted by the limited measurement points, as well as the high directional antennas that have been employed for the presented measurements. A.2 Urban micro outdoor scenarios A.2.1 Street canyon measurements at 15, 28, 60 and 86 GHz A multi-frequency measurement campaign in a street canyon environment was performed by Aalto University in its campus area in Espoo (Greater Helsinki), Finland. The measurement frequencies include 15, 28, 60 and 86 GHz. Figure A.65 shows the measurement site and the measured points with the overlaid point cloud model. We refer to [mmmagic D2.1, Section 6.1.1] for the specific details of the measurement environment and setup. Figure A.66 shows mmmagic Public 152

173 the overlaid PDPs in the measured frequencies for link Tx3-Rx1 and Tx5-Rx1. These measurements use horn and omnidirectional antennas with similar radiation patterns for all these frequencies. Figure A.65: Measurement site (left) and the point cloud model of the environment with 11 Tx and 1 Rx locations overlaid (right). Figure A.66: PDPs of the 15, 28, 60 and 86 GHz measurements at Tx3 (left) and Tx5 (right). For the frequency comparison of LSPs, delay and azimuth spreads were extracted using a 500- MHz bandwidth and a 30 db dynamic range. The measurements at different bands were carried out at different times. The results were not derived at 28-GHz since large deviations were found compared to the other bands due to the changes in the environment (parked cars). The mean spread values for LOS links, shown in Table A.13, show a very weak decreasing trend in DS and AS, and in general, quite low spread values due to the tunnelling effect of the narrow street canyon. Table A.13: Mean spread values for measured LOS links in street canyon. Mean DS [ns] Mean ASA [ o ] 15 GHz 28 GHz 60 GHz 86 GHz A.2.2 Street canyon simulations at 15, 28, 60 and 86 GHz The point cloud-based simulation at the street canyon was carried out by omitting the diffuse scattering, and only calculating specular reflections up to the second order, as the higher order reflections were found negligible. The channel sounding results at 15 and 60 GHz were used to calibrate the permittivity ε r. For 15 and 60 GHz, the optimization yielded ε r values of 5.5 and 4.0, respectively. A comparison between measured and simulated large scale parameters is shown mmmagic Public 153

174 in Figure A.67 and Figure A.68, and examples of measured and simulated PDPs are presented in Figure A.69. The average measured and simulated values in the two frequency bands are presented in Table A.14. The permittivity for 28 and 86 GHz was derived by linear fitting and the attenuation of shadowing objects L a were chosen heuristically. The final parameter values are presented in Table A.15. Figure A.67: Large scale parameters for measured (black) and optimized simulated (green) channels at 15 GHz. Figure A.68: Large scale parameters for measured (black) and optimized simulated (green) channels at 60 GHz. Figure A.69: Measured and simulated PDPs at 15 GHz. mmmagic Public 154

175 Freq. 15 GHz 60 GHz Table A.14: Average measured and simulated large scale parameters. ε r Path loss [db] Mean delay [ns] Delay spread [ns] Meas. Sim. Meas. Sim. Meas. Sim Table A.15: Material parameter values in street canyon simulations. Frequency ε r L a 15 GHz GHz GHz GHz Next, a large set of channel data is generated in order to model the large scale behaviour of LOS channels. Channel data is calculated for 15, 28, 60 and 86 GHz with the parameters presented in Figure A.70. One base station (BS) and 860 mobile stations (MSs) were placed in the main terminal hall as depicted by Figure A.71. The MS spacing was 0.5 m and the link distance varied between 6 and 104 m. The BS and MS heights were 5 and 1.5 m, respectively. The LSPs derived from the simulations are presented in Table D.2. Figure A.70: BS and MS positions in street canyon simulation. A.2.3 Open square measurements at 28 and 86 GHz The dual-band 28/86 GHz open square measurement campaign was conducted in Narinkkatori square in downtown Helsinki, Finland. The square is approximately m 2 and surrounded by buildings on all four sides. In addition to the buildings and the ground, the propagation environment is affected by lamp posts and statues, which cause the line-of-sight (LOS) to be obstructed, i.e., obstructed-line-of-sight (OLOS) in some Tx-Rx links. There were no vehicles in the area. Thanks to the highly elevated BS (5 m), the LOS paths were not obstructed by the presence of humans' movements in the square. For all links, Rx (BS) was placed near one lamp post at one corner of the square while the Tx (UE) were moved to different locations to cover the whole area of the square. Deep NLOS links were also measured where the Tx was placed mmmagic Public 155

176 behind buildings. The top-view of the measurement location and the floor plan of the measurements are shown in Figure A.71. Figure A.71: Top-view of the measurement location (left) and the floorplan (right). For each Tx-Rx link the horn antenna on the Rx side was scanned over the azimuth angles with 50 steps; the elevation angle was set to 10 downtilted direction. The measurement was repeated with horn antennas having vertical and horizontal polarizations to obtain vertical-tx/vertical-rx and vertical-tx/horizontal-rx polarization measurements; they are called co- and crosspolarization measurements, respectively, hereinafter. A back-to-back (B2B) calibration was performed prior to the measurement by connecting interfaces of Tx and Rx antennas through a 20- or 30-dB attenuator. In total five LOS, four NLOS and one OLOS measurements were conducted. A summary of the measurement campaign is shown in Table A.16. Table A.16: Summary of the 28 and 83 GHz measurement campaign in Helsinki open square Description 28 GHz 86 GHz Carrier frequency GHz 83 GHz Bandwidth 900 MHz 4 GHz No. of Tx positions 10 No. of Rx positions 1 No. of LOS links 6 No. of NLOS/OLOS links 4 Tx (MS) height 1.6 m Rx (BS) height 5 m Tx antenna Rx antenna bicone, 2-dBi gain, 60 o elevation HPBW, vertically polarized horn, 19-dBi gain, 10 o azimuth/40 o elevation HPBW, vertically and horizontally polarised mmmagic Public 156

177 Figure A.72: Comparison on PDP (left) and PAS (right) of link Tx10-Rx1 between 28 and 83 GHz measurements. Figure A.73: DS versus link distance (left), and azimuth AS versus link distance (right). The dashed lines show the average values. For the frequency comparison of LSPs, the results were extracted using 900 MHz bandwidth 20 db dynamic range. Figure A.72 shows the comparison on the normalized PDP and PAS of link Tx10-Rx1 between 28 and 86 GHz measurements, whereas Figure A.73 compares the omni-directional delay spread (DS) and azimuth angular spread at Rx (ASA) of the same link, which are calculated as in [HNJ+16]. Table A.17 compares the mean DS and mean azimuth AS of 28 and 86 GHz measurements. As can be seen from these results, both DS and ASA in 28 GHz measurements are lower as compared to those in 86 GHz measurements. Table A.17: Mean DS and mean azimuth AS of 28 and 86 GHz measurements in Open Square 28 GHz 86 GHz Mean DS [ns] Mean ASA [ o ] A.2.4 Multi-frequency measurements in Berlin at 10, 28, 41 and 82 GHz The channel measurement campaign was carried out in Berlin, Germany, aiming to provide statistically evaluable multi-frequency data for the urban microcellular (UMi) access channel in the street canyon. Four bands, namely 10.25, 28.5, 41.5 and 82.5 GHz, were measured simultaneously. In this section, statistical evaluations on the frequency dependence of the root-mean-square delay spread (DS), the path-loss (PL), and the Ricean K-Factor (KF) under line-of-sight (LOS) conditions are presented. By tendency, the largest DS and lowest KF values occur for GHz when a high relative evaluation threshold is used. DS values get smaller and KF values increase with increasing frequency. However, results also show that the estimated DS is very sensitive to data selection and processing. In the following sections, the measurement data is further processed and evaluated. In a first step, a high-resolution pre-processing algorithm is applied. This algorithm extracts the multipath components from the broadband data. In this way, the bandwidth limitation and the measurement noise can be reduced. In the second step, characteristic channel properties are calculated from the processed data. Those are the delay spread, path loss, shadow fading, and the Ricean K-Factor. mmmagic Public 157

178 Introduction Fifth generation (5G) mobile networks will need to make use of much higher frequencies to provide ultra-high data rates and capacity [BHL+14]. The targeted frequency bands comprise a huge range from 6 up to 100 GHz. Since the provision of an accurate reference channel model is essential for enabling link and system level simulations at these high frequencies, there are strong related activities in the research community and standardization bodies like 3GPP and ITU-R. Most of the work builds upon state-of-the-art three-dimensional (3D) geometry-based stochastic channel models (GSCMs) that have been developed in recent years for the lower frequency bands [3GPP36.873], [METIS-D1.4], [JRB+14]. The ultimate target is to elaborate a flexible and comprehensive frequency-agile channel model, which covers the entire range up to 100 GHz for targeted 5G scenarios rather than providing a model for specific bands only. With this respect, the question arises to what extent the large-scale parameters (LSPs), like delay spread and angle spread, depend on frequency. The results in the literature regarding the frequency dependence of LSPs are divergent. A trend to have lower delay spreads at higher frequencies is reported e.g. in [PK09], [5GCM16]. However, the results in [5GCM16] are not fully coherent. For several measurements, significant frequency dependence could not be observed. For the initial channel model proposed in [mmmagic-d2.1], frequency dependence has been taken into account based on literature values, but the investigations are ongoing. In this section, results of a multi-frequency measurement campaign performed in Berlin, Germany, are presented. The four frequencies (10.25, 28.5, 41.5 and 82.5 GHz) were measured simultaneously, and the data is particularly well suited for evaluations on the frequency dependence of channel characteristics. Table A.18 provides a comparison of the measurement conditions with the requirements defined in Section 3.1. Details on the channel sounder, the scenario and the measurement procedure can be found in [mmmagic-d2.1] (Sections and A.3). Table A.18: Multi-frequency measurement campaign in Berlin view of requirements defined in Section 3.1 # Requirement Comment 1 Equal measurement bandwidth The measurement bandwidth was 500 MHz at GHz, and 1.5 GHz at 28.5, 41.5 and 82.5 GHz; however, to assess the frequency dependence, the same evaluation bandwidth of 500 MHz is assured by post-processing 2 Equal antenna pattern, or antenna omnidirectional in azimuth pattern de-embed 3 Equal dynamic range for analysis Different evaluation thresholds are investigated: both in delay and angle domains 4 Equal angle resolution (e.g., array size equal in terms of number of lambda) 15, 20, 25 db below the strongest MPC. not relevant 5 Same environment and same antenna locations Yes, besides the small offset between different Tx/Rx antennas < 40 cm. 6 Compensation of atmospheric absorption not relevant at the 60 GHz-band 7 Sufficient sample size The measurement data used for the presented evaluations comprise several hundred thousand CIRs for each frequency. 8 Static environment (when measurements not relevant, since measurements were performed ( ) are made successively) simultaneously 9 Same path estimation algorithms For all comparative analyses, the same algorithm is used for all frequency bands. 10 Area of spatial averaging Yes ( ) ( ) mmmagic Public 158

179 Extraction of MPCs The basic pre-processing method is described in [STJ10]. It is assumed that the measurement data can be described as y n = h n + v n, (A-15) where y n is the measured channel coefficient at the n th pilot position. We further assume that the frequency response (FR) of the channel can be modelled as a sum of discrete reflections, i.e. the multipath components (MPCs), of the transmitted waveform at obstacles in the physical environment. The pre-processing decomposes the channel into L h n = α l exp (jφ l ) exp ( j2π τ l B n N ), (A-16) l=1 where α l is the amplitude, φ l is the phase, τ l is the delay of the l th MPC, B is the measurement bandwidth, n is the index of the sample point in frequency domain and N is the number of sample points in the frequency domain. Due to the pre-processing, the CIR has an approximately 6 db better signal-to-noise ratio (SNR) than the raw measurement data. Delay Spread of LOS channels The root-mean-square (RMS) delay spread (DS) measures how the multipath power is spread out over time. For calculating the DS, one needs a representation of the CIR in time domain where L multi-path components, i.e., their power P l and delay τ l can be identified. The DS is then calculated as [Rap02]. σ τ = 1 P P l(τ l ) 2 L L 2 ( 1 P P lτ l ) with L P = P l and P l = α l 2 (A-17) l=1 l=1 l=1 The index l indicates the path number and P is the total received power. Since the number of the detected MPCs depends on the SNR in the channel, it is common to remove weak paths having power values below a threshold Γ relative to the strongest path. In order to make the results comparable, the same threshold is used for low and high SNR channels and it needs to be ensured that there is always enough dynamic range in the CIR to support the chosen threshold. Influence of pre-processing and the threshold on the results The investigations in this section are based on the collected data from the LOS tracks. For each segment of 50 m length, 50,000 CIRs are available with a spatial separation of 1 mm. In a first processing step, the CIRs in different frequency bands were reduced to the same bandwidth, namely 500 MHz and a Kaiser window (parameter β = 6) was applied in the frequency domain to reduce the sidelobes in the delay domain. Low-pass-interpolated versions (upsampling with factor 4) of the CIRs were used for further processing. The DS was calculated based on averaged power delay profiles (APDPs). Each APDP was obtained by averaging over K successive CIRs according to K P(τ) = h k (τ) 2 k=1 (A-18) Where P(τ) denotes the APDP and h k (τ) is the k th CIR. An averaging factor of K = 100 was used, which corresponds to a spatial averaging over 10 cm. A relative threshold Γ with respect to the maximum in P(τ) was applied and contributions below the threshold were discarded [ITUR-P1411], [ITUR-P1407]. Great care has been taken to ensure that the dynamic range of the APDPs was always large enough to support the respective threshold in order to exclude mmmagic Public 159

180 distortions of the estimated values by noise. In cases where the evaluation threshold was not supported by a PDP of at least one band, the data of all bands was discarded. Figure A.74 (top left) shows the resulting empirical cumulative distribution functions (CDFs) of the DS for three different thresholds, namely 15, 20 and 25 db and a distance range (2D distance between Tx and Rx) from 10 to 110 m. They are based upon 3,000 data samples (APDPs) for each frequency. Since all frequency bands used dipole antennas, it can be assumed that the antennas do not alter the results. As expected, increasing the threshold yields higher values and the CDFs shift to the right. It can be seen that the results for the three frequencies are very similar for Γ = 15 db and Γ = 20 db. There is only a small trend towards larger spreads for GHz, which occurred with lower probability. However, for Γ = 25 db this trend becomes more distinct, whereas the curves for 28.5 and 41.5 GHz are still practically congruent. Figure A.74 (top right) shows the CDFs for Γ = 25 db, when the preceding bandwidth limitation to 500 MHz for the frequencies 28.5 and 41.5 GHz is omitted. The results with bandwidth limitation are included for comparison. It can be seen that the estimated DS values for 28.5 and 41.5 GHz are lower without bandwidth limitation. Accordingly, the difference to the GHz curve increases. When comparing results derived from data with different bandwidth, significant deviations may occur which, however, are not related to the propagation channel, but further dependent on the data processing and the estimation procedure. Results on the comparison of all four frequencies are illustrated in Figure A.74 (bottom left). In this case, the evaluated distance range was limited to m and a threshold of 15 db was chosen. Compared to the results based on the distance range m (CDFs included in Figure A.74 (bottom left) for comparison), lower delay spreads occur with higher probability. The CDFs are less smooth as only 600 APDPs for each frequency are available that exhibit the required dynamic range at the respective Rx position for each measured band (after bandwidth limitation of the upper bands), including 82.5 GHz. The DS for 82.5 GHz is clearly smaller than for the lower bands. However, for the lower bands a trend of a potential dependence on frequency cannot be identified based on this subset of data samples. Figure A.74 (bottom left) exhibits that the considered distance range can affect the results and that a large sample is beneficial to enable clear conclusions. Preliminary investigations on the dependence of the DS on Tx-Rx distance indicate that lower values and lower variations occur for very small distances (10 20 m). Both tend to increase with distance, but saturate at distances between 30 and 50 m. The influence of the high resolution processing, i.e., the extraction of the MPCs, is shown in Figure A.74 (bottom right) and is compared with the band-limited processing in Table A.19. When comparing the results for the distance range from m, the DS estimation after MPC detection shows a 30% increase compared to the estimation without MPC detection for the 10 GHz band. However, no increase is observed for the 28 and 41 GHz bands. Careful inspection of all the LOS measurement files showed that a threshold of 25 db is supported for the lower channels frequencies (10, 28, 41 GHz) for the full distance range from 10 to 250 m. Model parameters for the LOS channels The empirical distribution function of the DS for the three frequencies is shown in Figure A.74 (bottom right). It is usually fitted to a log-normal distribution which served as an input to channel models [3GPP38.900, JRB+14]. The median log-ds is fitted to DS = 0.44 log 10 (f c ) N(DS σ 2 ). DS σ = 0.18 log 10 (f c ) (A-19) (A-20) where f c is the carrier frequency in GHz. DS μ is measured in units of log 10 s. The fitted curves are also shown in Figure A.74 (bottom right) for comparison. mmmagic Public 160

181 A comparison of the DS results from the measurements (250 m distance, 25 db threshold and high-resolution processing) with the fitted model and the results from the 3GPP model [3GPP38.900] is given in Table A.21 (the 3GPP parameters are listed in Table A.20). In general, the models agree well with the measurements. However, the 3GPP model predicts a 12 ns shorter DS for the 10 GHz frequency band. Figure A.74: Empirical CDFs of RMS delay spread Top Left: Top Right: Bottom Left: Bottom Right: DS for m distance range and relative thresholds of 15, 20 and 25 db DS with and without bandwidth limitation (28.5 and 41.5 GHz), m distance range and relative threshold of 25 db. DS for m Tx-Rx distance and relative threshold of 15 db. DS for high-resolution processing, m Tx-Rx distance and relative threshold of 25 db Table A.19: Median and 95% Quantile of the RMS delay spread Freq. Γ 0.5 GHz, m HiRes, m HiRes, m (db) Median Q 0.95 Median Q 0.95 Median Q GHz GHz GHz mmmagic Public 161

182 Path loss and shadow fading The effective PL is defined as the instantaneous propagation loss, i.e., the received power in (negative) db without antenna gain normalized to 0 db. The transmit power is calibrated out of the measured channel coefficients. Since near omni-directional antennas were used during the measurements, the path powers can be used to calculate the PL by summing up the power of all MPCs. Due to the high-resolution processing, measurement noise is removed from the evaluation, which increases the accuracy in the parts of the measurement track with low received power. In order to remove the effects of fast fading from the statistics, neighbouring values within a distance of 1 m are averaged. These values are then used to parameterize a Hata model [Hat80] with parameters A, B, C and SF σ as PL = A log 10(d 3D ) + B + C log 10(f c ) + N(SF σ 2 ). (A-21) In this equation N(SF σ 2 ) is normal distributed variable with zero mean and variance SF σ 2. The 3GPP model [3GPP38.900] uses a dual-slope model that takes the influence of the ground reflection on the path loss into account. This is done with respect of a break-point d BP = 4 (h BS 1) (h MT 1) f c c. (A-22) However, with the given BS height of 5 m and MT height of 1.5 m, none of the three frequencies exceeded the break-point distance. Hence, only the first part of the 3GPP UMi path loss model can be validated. It is important to note that the channel is subject to strong fading. A regular fading pattern caused by the ground reflection cannot be clearly identified. Instead, there are many multipath components, which often contribute (in sum) more power than the direct path. The results are summarized in Table A.20. For the LOS channels, the PL model was parameterized to PL = 24.8 log 10(d 3D ) log 10(f c ) + N(3.7 2 ) (A-23) Compared to the 3GPP model, the measurements and the newly fitted model show good agreement for the three frequency bands. Ricean K-Factor The KF is defined as the ratio of the power of the direct path divided by the sum-power of all other paths. Some literature sources (e.g., [GME99], [Rap02]) define the KF with respect to the strongest path in the CIR which can originate from a dominant scattering cluster. In current channel models, however, the KF is defined as the power ratio between LOS and NLOS paths. Hence, to estimate the KF, the LOS path has to be detected. The empirical detection of the LOS path works in two steps: First, the power delay profiles (PDPs) of all MIMO sublinks are summed up. Then, a peak at the beginning of this sum-pdp, i.e., the first path that exceeds 1% of the total power, is detected. This ensures that small paths originating from noise at the beginning of the CIR are excluded. In the second step, the LOS delays of successive snapshots within a 1 m radius are compared and false detections are removed. From the remaining snapshots, the KF is calculated as K = P LOS P P LOS. (A-24) The KF is typically given in db and it is mapped to three parameters: KF μ and KF σ (both in [db]) to parameterize a normal distribution. The fitted distribution is KF = 5.3 log 10 (f c ) N(4 2 ). (A-25) The KF increases when the frequency increases. This is in line with the decreasing delay spreads. At higher frequencies, it seems that multipath components have less power and more power is received from the direct path. However, the impact of diffuse scattering and ground mmmagic Public 162

183 reflection fading needs to be further evaluated. At a carrier frequency of 60 GHz (and above), the multipath effect might even disappear completely [PWK+15]. This is in direct contrast to the 3GPP assumptions, which assume that the KF is independent of the frequency. Table A.20: Model parameters of HiRes evaluation Parameter Unit Combined 3GPP UMi Path Loss (PL) A db Path Loss (PL) 1 db m Path Loss (PL) C db Shadow Fading (SF) db Delay Spread (DS) μ log 10 s log10(fc) log10(fc) Delay Spread (DS) σ log 10 s log10(fc) K-Factor (KF) μ db log10(fc) K-Factor (KF) σ db Cross-Corr. DS vs. SF Cross-Corr. DS vs. KF Cross-Corr. SF vs. KF Delay scaling parameter DS corr. Dist. m SF corr. Dist. m KF corr. Dist. m Table A.21: Comparison measurement and model results Parameter Freq. Measurements Combined model 3GPP UMi GHz Q 0.05 Median Q 0.95 Q 0.05 Median Q 0.95 Q 0.05 Median Q 0.95 DS [ns] Γ = 25 db PL [db] m KF [db] Conclusions Statistical analysis results of the RMS delay spread (DS), the path loss, shadow fading and the Ricean K-Factor based on line-of-sight data of a multi-frequency channel measurement campaign in an UMi street canyon scenario has been presented. In an initial analysis, the data for the higher bands was reduced to the same bandwidth as used for GHz, namely 500 MHz. The same relative evaluation threshold was applied to the power delay profiles for all bands. Taking into account data for a distance range from 10 to 110 m, it reveals that the distribution of the DS for the lower bands (10, 28, 41 GHz) is very similar for evaluation thresholds of 15 and 20 db. For a threshold of 25 db, there is a trend to higher values for 10 GHz. A statistical analysis based on a reduced data set for the distance range m shows that the DS is tendentially lower at 82 GHz in comparison to the other frequencies. In a second analysis step, the multi-path components were extracted from the raw data. This reduced the influence of measurement noise on the evaluated parameters and allowed using the entire distance range for the data analysis. The trend towards longer delay spreads at lower mmmagic Public 163

184 frequencies could be confirmed. At the same time, there was a clear trend towards higher K- Factors at higher frequencies, i.e. more power gets allocated to the direct path and less multipath components can be found. When comparing the measurement results with the 3GPP model [3GPP38.900], path-loss and delay spread results seem to agree reasonably well. However, the strong and clear dependence of the K-Factor on the frequency is not included in the model. A.2.5 Measurement and simulation campaign in Daejeon An experiment including both ray tracing simulation and measurement was performed in Daejeon, Korea. Streets and buildings are the main components in the environment. The distance between the transmitter and receiver is up to 200 m. The experiment environment is categorized as urban micro (UMi) street canyon at 28 GHz. Several features of UMi street canyon scenarios have been identified in [mmmagic-d2.1]. UMi street canyon scenarios have high user density. In addition, the users are mainly pedestrians or slow vehicular users. Buildings on both sides of the street have normally four to seven stories. The length of a street is in the order of 100 m. In addition, street furniture such as lampposts, traffic signs, and trees is typically seen in the environment. A bird s view of Daejeon is illustrated in Figure A.75. It should be noticed that the ray tracing simulation only considers reflections, penetrations, and diffractions. Both line-of-sight (LOS) and non-los (NLOS) were simulated via raytracing. However, only NLOS scenario was measured. Figure A.75: Bird s views of environment for measurement and ray tracing simulation. The channel was measured via a wideband radio channel sounder, which transmits 250 Mega chip-per-second (Mcps) pseudonoise (PN) sequences. The channel sounders for measurement are shown in Figure A.76. The transmitter was located at a fixed position, which was 15 m above ground. Receivers were placed in thirty eight different locations (1.6 m above ground) in street canyons, which are illustrated in Figure A.75. Horn antennas with 24.5 dbi gain and 10-degree half-power bandwidth were used in both the transmitter and receiver. These antennas can be rotated pointing various directions in azimuth and elevation automatically via a synchronized triggering signal. Both the transmitter and receiver scanned in the azimuth and elevation directions. The transmitter scanned from -π to π in the azimuth and from π/3 to π/3 in the elevation. The receiver scanned from π/3 to π/3 in the azimuth and from 2π/9 to π/18 in the elevation. Other specifications of the channel sounders are listed in Table A.22. Measurement results were synthesized to generate 3-D omni-directional channel characteristics. In each measurement location, channel characteristics of rays such as power, delay, and angles were recorded. mmmagic Public 164

185 Figure A.76: Channel sounders for measurement campaign in Daejeon. Table A.22: Channel Measurement Settings Parameter Value Carrier frequency GHz Signal bandwidth 250 MHz Transmit power 29 dbm Antenna type Horn antenna Antenna Gain 24.5 dbi Antenna beamwidth 10 degrees (both azimuth and elevation) No. receiver locations 38 The ray tracing simulation was performed via the shoot-and-bounce method. Rays with 0.1 degree angular spacing were launched at the transmitter. Propagation effects on these rays such as reflections, diffractions, and penetrations were modelled using geometrical optics (GO) and uniform theory of diffraction (UTD). Then, the rays were traced up to a maximum number of reflections, diffractions, and penetrations until they reached the receiver. At the receiver side, a typical threshold 250 db of propagation loss was assumed such that rays attenuated by more than 250 db will not be considered in at the receiver. Delays, power, and angular information were the outputs of the ray tracing simulation. To mimic realistic environment as accurately as possible, 3-D geometrical information needs to be input to the ray tracing simulator. The 3-D layout of Daejeon is shown in Figure A.75. Moreover, frequency-dependent material properties were considered to model propagation characteristics at different frequencies. In this simulation, concrete was assumed for buildings and wet earth was assumed for the ground. In this case, relative permittivity values at 28 GHz for concrete and wet earth are ε= 6.5 and ε = 15, conductivity values at 28 GHz for concrete and wet earth are σ = and σ = [ITU-R P. 2040]. Other input parameters such as maximum numbers of reflections, penetrations, diffractions, and rays in total were configured as Table A.23. Larger values can be used for better approximation of the realistic environment, at the cost of longer simulation time. Table A.23: Ray Tracing Simulation Settings Parameter Value Ray spacing 0.1 degree Maximum number of reflections 12 Maximum number of penetrations 2 Maximum number of diffractions 1 Maximum number of rays in total 40 mmmagic Public 165

186 CDFs of the number of subpaths within clusters are depicted in Figure A.77. The mean value of ray tracing (excluding diffusive scattering and small objects) is 11.8, which is significantly less than the mean value 31.0 in measurements. One reason for this is that the maximum total number of subpaths considered in ray tracing simulations is limited by 40. If the maximum total number of subpaths is scaled to 104, same as the average total number of subpaths in measurements, the scaled mean number of subpaths within one cluster would be /40 = This is aligned with measurement results. In the current new 3GPP channel model [3GPP38.900], the number of subpaths within a cluster is constant 20, which may not be feasible for above 6 GHz channel models. In addition, it can be seen in Figure A.77 that the number of subpaths within one cluster follows a negative binomial distribution, which is different from the uniform distribution assumption used in [SR15]. Then, the probability mass function (PMF) f(m n ) of the number of subpaths within one cluster can be presented as f(m n ) = Γ(M n + r) M n! Γ(r) pm n(1 p) r, (A-26) where r is a positive real number and 0<p<1. For the ray tracing simulation, p = 0.18 and r = For the measurement, p = 0.06 and r = Figure A.77: CDFs of number of subpaths in measurement and ray tracing results.,ray tracing and measurement results match well. The mean and median values of the intracluster RMS delay spread are approximately 40 ns and 22 ns, respectively. These are larger than the current assumption in the new 3GPP channel model [3GPP38.900], which is 10 ns. Therefore, channel models for above 6 GHz channels should take intra-cluster delay spreads into account. Exponential distributions can be applied to fitting the distribution of intra-cluster delay spreads. mmmagic Public 166

187 Figure A.78: CDFs of intra-cluster RMS delay spreads of measurement and ray tracing. In Figure A.79 and Figure A.80, CDFs of intra-cluster angular spreads are presented. Since 3- D effects are considered, angular spreads should consist of azimuth angular spreads and elevation angular spreads. In Figure A.79, ray traced azimuth AoA spread and measured azimuth AoA spread have a similar trend, their median values are approximately degrees. Also, similar trend can be observed in intra-cluster zenith AoA spread in Figure A.80, which has approximately 5 degrees of spread. However, it can be observed that there is a small gap between the ray tracing result and the measurement result in Figure A.80. This can be caused by that elevation angles are not fully stochastic and they have certain dependency on the distance between the transmitter and receiver. Locations of ray tracing and measurement do not fully overlap, which is responsible for the gap. However, the gap is within 2 degrees, which is sufficiently small to be neglected. Also, lognormal distributions can be used to model angular spreads in both azimuth and elevation. Figure A.79: CDFs of intra-cluster azimuth AoA spreads of measurement and ray tracing. mmmagic Public 167

188 Figure A.80: CDFs of intra-cluster zenith AoA spreads of measurement and ray tracing. A.2.6 Ray-tracing simulation in Madrid-grid and Aalborg city layouts The ray-tracing simulations were performed in Micro Urban type of environment. Two different layouts were used: Madrid-grid, which is regular layout and Aalborg city layout which is realistic layout of typical European city. In Figure A.81 and Figure A.82, we show the 2D and 3D view of the maps. Figure A.81: 2D and 3D layout of Madrid-grid with locations of 5 transmitters. mmmagic Public 168

189 Figure A.82: 2D and 3D layouts of Aalborg city for 5 transmitters. mmmagic Public 169

190 The simulations were performed with WinProp ver.13 (AWE Communication) [AWE]. The 3D Standard Ray Tracing (SRT) model is used for simulation. The SRT model does not use discretization of map (i.e. breaking up buildings into tiles) for more accuracy. However, the tradeoff is a significant increase in simulation time [AWE]. The deterministic interaction model uses Fresnel Equations for reflection and transmission prediction and GTD (Geometrical Theory of Diffraction) or UTD (Uniform Theory of Diffraction) for diffraction prediction. Diffuse scattering were disabled to speed up simulation. The electrical parameters of materials (permittivity, permeability and conductivity) will be from Recommendation ITU-R P.2040 [ITUR ]. The model allows also prediction of reflection (and scattering) from the ground, which is very important to consider in radio propagation prediction. The 3D Standard Ray Tracing parameters of simulation were the following: Maximum number of reflections: 2 Maximum number of diffractions: 2 Maximum number of reflections and diffractions: 2 Diffuse scattering disabled Coherent superposition of rays (with consideration of phase) Maximum path loss of ray: 200dB Maximum number of rays per pixel: 40 Maximum dynamic range per pixel: 100dB Walls and ground: concrete material (electrical parameters based on ITU-R P.2040) Resolution: 5m The 5 Transmitters were used in both layouts with different locations. The simulation parameters were the following: TX antenna height: 5m RX antenna height: 1.5m TX and RX antenna characteristics: isotropic The frequencies: 2GHz, 5.6GHz, 10GHz, 28GHz, 39GHz, 73GHz The frequency dependence of different channel characteristics and parameters is discussed below. Interaction type The distribution of five strongest interaction types is shown on Figure A.83 for Madrid-layout and Figure A.84 for Aalborg-layout. The figures show the percentage number of 5 different interactions types with the strongest received power. There are the most significant interaction types versus frequency for different layout type. mmmagic Public 170

191 Figure A.83: The number of 5 strongest interaction types in Madrid-grid layout. Figure A.84: The number of 5 strongest interaction types in Aalborg-city layout. In the LOS case, we can observe almost no frequency dependence in the number of interactions in Madrid-grid. In the Aalborg-city in case of LOS, we can observe a small decrease of single diffraction and increase of double reflection with increasing of frequency. In NLOS case, the dependence is much stronger. We can observe in both layouts significant reduction of double diffraction and increase of single diffraction + single reflection with increasing of frequency. There are differences in numbers between both layouts due to differences in both environments. The results indicate that we can expect also frequency dependence in some of the Large Scale parameters especially in NLOS case. Pathloss exponent (PLE) Pathloss exponents for LOS and NLOS cases are shown in Figure A.85. Figure A.85: Pathloss exponents for LOS and NLOS for both layouts. The PLE dependence on frequency is very small for LOS and NLOS. We can conclude that PLE is generally not dependent on frequency. mmmagic Public 171

192 Shadow factor (SF) Shadow Factor is showed in the Figure A.86. Figure A.86: Shadow Factor for LOS and NLOS for both layouts. We observe not frequency dependence in LOS case because we have strong direct path and different building structures and placement does not cause impact on shadowing. In the NLOS case we see the frequency dependence on Shadow Factor in case of Madrid-grid only. Delay spread (DS) The RMS delay spread is shown on Figure A.87. Figure A.87: RMS delay spread for both layouts. We don t observe frequency dependence of DS for LOS case. In case of NLOS case we see clear decreasing of DS with frequency for both types of layouts. Azimuth angle spread of departure (ASD) The RMS azimuth angle spread of departure is shown on Figure A.88. Figure A.88: RMS azimuth angle spread of departure for both layouts. In case of LOS, we generally do not see the dependence in frequency for ASD. In NLOS case, the dependence is visible especially for Aalborg layout but small. mmmagic Public 172

193 Azimuth angle spread of arrival (ASA) The RMS azimuth angle spread of arrival is shown on Figure A.89. Figure A.89: RMS azimuth angle spread of arrival for both layouts. In this case we see very small dependence on frequency for ASA in LOS case and small dependence for NLOS in both layouts. Zenith angle spread of departure (ZSD) The RMS zenith angle spread of departure is shown on Figure A.90. Figure A.90: RMS zenith angle spread of departure for both layouts. We can observe here no frequency dependence in LOS case and small dependence in NLOS case for both types of environment. Zenith angle spread of arrival (ZSA) The RMS zenith angular spread of arrival is shown on Figure A.91. Figure A.91: RMS zenith angle spread of arrival for both layouts. The observations are similar like in the previous case. The frequency dependence is mainly visible in NLOS case for ZSA. mmmagic Public 173

194 Cross-polarization ratio (XPR) The mean cross-polarization ratio (XPR) is shown in Figure A.92. In the simulation maximum XPR was set for 25dB emulating the Cross Polar Discrimination (XPD) of antenna. Figure A.92: Mean cross-polarization ratio for both layouts. We do not observe clear frequency dependence in LOS and NLOS for XPR for both layout types. In the presented results, we do not observe clear environment type (regular, not regular) impacts on LSPs. The frequency dependence is observed for the LSPs in NLOS case. A.2.7 Open square measurements at 17 GHz Sounder setup For this measurement, Rohde & Schwarz Signal generator and signal analysers were used together with the Rohde & Schwarz Channel Sounding Software. For this campaign, channel sounding was conducted at 17 GHz, with no external up- or down conversion being required. For synchronized measurements, transmitter and receiver can be connected in order to use the same 10 MHz reference signal. Either the reference signal may be used from one of the devices (with direct connection) or it may be obtained from external sources for generator and analyser, depending on the spatial distance between transmitter and receiver. For validation and calibration, the sounder is set up with a reference cable and the signal generator (TX) acting as the reference source. The sounding sequence is played on the generator and transmitted over cable to the signal analyser, which feds the sampled IQ data to the channel sounding software. The retrieved phase is stable over the measurement period with a standard deviation of Additionally, the setup if validated in an anechoic chamber with the used antennas attached to transmitter and receiver, confirming the measured channel impulse responses that are observed over cable and are expected. As sounding waveform, a 16 bit Frank-Zadoff-Chu sequence with a length of samples and a bandwidth of 160 MHz is used. Measurement description The measurements were again conducted on the Rohde & Schwarz premises. Figure A.93 gives an aerial view of the measurement locations. The measurements for RX 4 and RX 5 are presented in Section A.3.2. mmmagic Public 174

195 Figure A.93: Aerial view for open square measurement at R&S. Google, GeoBasis DE/BKG. The square is about 50m x 50m in size. Including the adjacent street and buildings forecourt, it is roughly 65 m x 65 m and mostly surrounded by buildings. In addition, lamp posts, a small building, cars and trees cause additional obstacles. As the measurements were conducted in winter, the trees had no leaves, thus blocking less of the signal passing through. For all measurements, the Transmitter (TX) was placed outside the 5 th floor of the building across the street. The antenna was tilted down to point in direction of the centre of the square. The receiver was placed at five different locations on the square, three of which feature a direct LOS (RX1, 2, 6), one with obstructed LOS (RX 3) and one with no LOS (RX 7). An overview of the measurement campaign key information is given in Table A.24. Averaging is performed in post processing to lower the influence of noise. Table A.24: Measurement details for open square measurements in Munich Carrier frequency 17 GHz Bandwidth 160 MHz No. of Tx positions 1 No. of Rx positions 5 No of LOS links 3 No. of OLOS links 1 No. of NLOS links 1 Tx (BS) height ~ 14 m Rx (MS) height 2 m Tx antenna Vivaldi, vertically polarized Rx antennas Vivaldi, vertically polarized Dipole, vertically polarized mmmagic Public 175

196 Figure A.94: Averaged PDPs for OS LOS scenarios RX 1 and RX 2. First, receiver positions RX 1 and RX 2 are evaluated. The measured PDP for both scenarios is shown in Figure A.94. Both show a clear LOS peak with additional multipath components up to about 325 meter of overall travel distance. With the directive Vivaldi antenna, less path clusters can be observed, indicating that the multipath components observed using the dipole antenna arrive from different angles. At position Rx 3, the Vivaldi antenna receives even less multipath components, with only highly attenuated paths reaching the receiver. The PDP is shown in Figure A.95. Also, for this OLOS scenario, the dipole antenna receives a higher signal power in addition to numerous multipath components. While the higher received power is unintuitive at first look, it may be explained due to multiple paths being scattered at the tree that obstructed the direct LOS path. The antenna patterns are not de-embedded, which can lead to a lower power received by the Vivaldi antenna. Figure A.95: Averaged PDP for OS OLOS scenario RX3. For an additional measurement, the receiver antenna was placed at position RX 7 with two different orientations. The orientations B8 and B12 correspond to the antenna pointing towards Building 8 and Building 12 as shown in Figure A.93. Figure A.96 presents the results for the NLOS scenario at position RX 7. mmmagic Public 176

197 Figure A.96: Averaged PDP for OS NLOS scenario RX7. In this NLOS scenario, reflections from different angles arrive at the receiver with the directional measurements facing B8 (yellow) and B12 (red) each show only a subset of the paths received with the dipole antenna (blue). As expected, certain paths or clusters of paths arrive from different angles at different delays. This feature may be used to deploy beam tracking at the UE. Compared to the O2I NLOS scenario, the open square NLOS scenario generally features more distinct paths. Furthermore, the reception power is slightly higher for each path as well as the overall signal power. Figure A.97: Measured path loss without antenna gain for different receiver locations in OS scenario. Figure A.97 illustrates the measured path loss for Vivaldi and dipole antenna at the individual receiver positions for the OS scenario. Though there are not enough measurements to derive mmmagic Public 177

198 statistically relevant data, the received power between directional and omnidirectional receive antenna only differs substantially in the OLOS and NLOS scenarios, as expected. For all LOS links, the difference between expected FSPL and measured path loss was evaluated. Table A.25 compares this for the receiver positions for the outdoor-to-indoor and open square scenarios. All signal powers are related to only the first received peak (representing the LOS component). While the measurement in the anechoic measurement chamber coincides with the expected value perfectly, the outdoor measurements conducted in real scenarios differ by a few db in received signal power. For unobstructed OS measurements, the received signal power is higher than expected. This can be explained with multipath components that arrive only a few nanoseconds after the LOS component and are not detected due to the time resolution of the channel sounder not being fine enough with the given bandwidth of 160 MHz. These components may originate for example from the ground (c.f Section A.4.3) or the trees. Table A.25: Evaluation of measured and expected path loss for Vivaldi antenna Measurement Location TX Power RX Power Distance Exp. FSPL Measured Channel Loss Difference Measuring Chamber 18 dbm dbm 1.44 m 60.2 db 60.2 db 0.0 db RX 1 (OS, LOS) 18 dbm dbm 80 m 95.1 db 92.6 db -2.5 db RX 2 (OS, LOS) 18 dbm dbm 85 m 95.6 db 93.1 db -2.5 db RX 3 (OS, LOS/Tree) 18 dbm dbm 75 m 94.6 db 98.0 db 3.4 db RX 4 (O2I, LOS) 18 dbm dbm 50 m 91.0 db 95.0 db 4.0 db RX 6 (OS, LOS) 18 dbm dbm 71 m 94.1 db 94.9 db 0.8 db Table A.26: Evaluation of measured and expected path loss for Dipole antenna TX Power RX Power Measurement Location Distance Exp. FSPL Measured Channel Loss Difference Measuring Chamber 18 dbm dbm 1.44 m 60.2 db 60.3 db 0.0 db RX 1 (OS, LOS) 18 dbm dbm 80 m 95.1 db 91.4 db -3.7 db RX 2 (OS, LOS) 18 dbm dbm 85 m 95.6 db 93.0 db -2.7 db RX 3 (OS, LOS/Tree) 18 dbm dbm 75 m 94.6 db 89.0 db -5.5 db RX 4 (O2I, LOS) 18 dbm dbm 50 m 91.0 db 89.7 db -1.3 db RX 6 (OS, LOS) 18 dbm dbm 71 m 94.1 db 95.9 db 1.9 db A.2.8 Outdoor measurements at 3, 17 and 60 GHz Measurement scenario The objective of this outdoor multi-frequency measurement campaign carried out in the 3, 17 and 60 GHz frequency bands is to evaluate the frequency dependence of large scale parameters (LSPs) such as the channel delay spread (DS). The measurement campaign was conducted in the industrial area Techn hom in Belfort where Orange Labs premises are located. The aerial map view of the measurement environment is depicted in Figure A.98. It shows the three distinctive transmission points that were considered during the measurement: TxA, TxB and TxC. At each transmission point, the TX antenna was mounted on top of a van at 2.5 m above ground level and a set of LoS and NLoS measurements was performed with the RX antenna, kept at mobile user level at 1.5 m height, placed at different positions. A1 to A16 denote the RX antenna positions for TxA, B1 to B11 for TxB and C1 to C8 for TxC. The distance between the TX and the 35 RX positions ranged from 16 to 200 m. mmmagic Public 178

199 Figure A.98: Aerial map view of the measurement environment. Setup and procedure The measurements were performed using a wideband channel sounding system. An arbitrary waveform generator (AWG) produced successive 8192-length wideband sequences at 1 Gbpsrate with a 125 MHz channel bandwidth. Figure A.99 below displays the TX and RX measurement equipment. mmmagic Public 179

200 Figure A.99: Tx and Rx measurement equipment. Depending on the RX antenna radiation patterns, different measurement setups were considered according to Table A.27. Setup 1 (S1) refers to the use of omnidirectional RX antennas at 3, 17 and 60 GHz whereas Setup 2 (S2) corresponds with the use of directional RX antennas at 17 and 60 GHz. Five measurement configurations were then considered. At 3 GHz, an omnidirectional antenna with 1 dbi gain was used at each end. During the measurement, the TX antenna was fixed and the RX antenna was slightly moved over a few tens of λ λ denotes the signal wavelength while CIRs were being collected. We note h n(τ) the n th collected CIR where τ denotes the relative time delay and N the total number of collected CIRs. At 17 and 60 GHz, antennas with larger gains were used to compensate for the increased propagation losses. On the TX side, 13 and 15 dbi gain antennas, with 30 HPBW each, were used at 17 and 60 GHz respectively. On the RX side, an omnidirectional antenna and a 90 HPBW directional antenna were used for both frequency bands. At these two frequencies, the measurement procedure was the same. Both the omnidirectional and the 90 HPBW RX antennas were pointed towards 4 distinctive directions with a 90 angular step. In each RX direction, the 30 HPBW directional TX antenna was scanning the entire azimuth dimension while CIRs were being collected on each of the two RX antennas at the same time. With an angular step equal to the antenna HPBW, a total of 12 directions were considered at the TX side. We note Omni h p, q and Dir h p, q the CIRs collected in the p th TX and q th RX directions by the omnidirectional and the 90 HPBW RX antennas respectively. mmmagic Public 180

201 Table A.27: Measurement setup parameters Setup Parameters Frequency Bands [GHz] Bandwidth TX Power RX Antenna TX Antenna Polarization S1 3 GHz 17 GHz 60 GHz 17 GHz 60 GHz S MHz (at 60 GHz, bandwidth reduced from 250 to 125 MHz in processing) 1 dbi, Omni. 1 dbi, Omni. from 20 dbm (60 GHz) to 30 dbm (3 GHz) 0 dbi, Omni. 13 dbi, 30 HPBW 0 dbi, Omni. 15 dbi, 30 HPBW Vertical 7 dbi, 90 HPBW 13 dbi, 30 HPBW 6 dbi, 90 HPBW 15 dbi, 30 HPBW Data Processing Omnidirectional PDPs were computed from each measurement setup of each frequency band. For S1 at 3 GHz, the omnidirectional PDP was computed as the average power, associated with each delay, between the N CIRs. PDP 1 N N h n n 1 For S1 at 17 and 60 GHz, the omnidirectional PDP was obtained by summing the transmitted powers from all 12 TX directions and averaging the received profiles in the 4 different RX since we had an omnidirectional RX antenna. PDP q 1 p 1 Omni h p, q For S2 at 17 and 60 GHz, the omnidirectional PDP was computed as the sum of the transmitted powers from all 12 TX directions and the received powers in all 4 RX directions since we had a 90 HPBW RX antenna. Results PDP 4 12 q 1 p 1 The channel DS was computed at each of the 35 different RX positions and for each of the 5 configurations mentioned earlier with a 20 db threshold on the omnidirectional PDP. The obtained values are summarized in Table A.28. Dir h p, q mmmagic Public 181

202 Delay Spread DS [ns] Table A.28: Computed DS values S1 3 GHz 17 GHz 60 GHz 17 GHz mmmagic Public 182 S2 60 GHz TxA LoS A A A A A A A Average TxA NLoS A A A A A A A A A Average TxB LoS B B B B Average TxB NLoS B B B B B B B Average TxC LoS C C C Average TxC NLoS C C C C C Average Not performed, Low Dynamic Range

203 Figure A.100 and Figure A.101 show the cumulative distribution functions (CDFs) of the DS values presented in Table A.28 in both LoS and NLoS conditions respectively Probability LoS S1 : 3 GHz S2 : 17 GHz S1 : 17 GHz S2 : 60 GHz S1 : 60 GHz Delay Spread [ns] Figure A.100: LoS CDFs of DS values Probability NLoS S1 : 3 GHz S2 : 17 GHz S1 : 17 GHz S2 : 60 GHz S1 : 60 GHz Delay Spread [ns] Figure A.101: NLoS CDFs of DS values. Generally, it appears that the DS is statistically hardly frequency-dependent. We assume an open square environment when the TX is placed at TxA and TxB and a street canyon scenario when it is placed at TxC. In average, in the open square scenario, the values at 3 and 17 GHz are very similar and we observe a slight decrease of about 20% of this parameter at 60 GHz. Such a behavior is illustrated by the typical PDP example shown in Figure A.102. However, for the street canyon scenario, the channel DS displays an increasing behavior with the frequency. For instance, as illustrated in the typical example of PDPs shown in Figure A.103, the DS values at 17 and 60 GHz are higher than the values at 3 GHz. mmmagic Public 183

204 Relative Amplitude [db] S1 : 3 GHz S2 : 17 GHz S1 : 17 GHz S2 : 60 GHz S1 : 60 GHz Relative Time Delay [ns] Figure A.102: PDPs with TX at TxA and RX at A14. Relative Amplitude [db] S1 : 3 GHz S2 : 17 GHz S1 : 17 GHz S2 : 60 GHz S1 : 60 GHz Relative Time Delay [ns] Figure A.103: PDPs with TX at TxC and RX at C4. In addition, at 17 and 60 GHz, the CDFs also confirm that the computed DS values are statistically independent of the measurement setup. Finally, Table A.28 was reproduced for different threshold values including 15 and 25 db. In Figure A.104 and Figure A.105, the histograms corresponding to the average DS values for the different configurations are represented with different threshold values in both LoS and NLoS conditions respectively. The conclusions remain the same as above regardless of the chosen threshold value. mmmagic Public 184

205 Delay Spread [ns] LoS S1 : 3 GHz S2 : 17 GHz S1 : 17 GHz S2 : 60 GHz S1 : 60 GHz Threshold Values [db] Figure A.104: LoS DS average values with different thresholds. Delay Spread [ns] NLoS S1 : 3 GHz S2 : 17 GHz S1 : 17 GHz S2 : 60 GHz S1 : 60 GHz Threshold Values [db] Figure A.105: NLoS DS average values with different thresholds. These results were compared to the 3GPP TR channel DS model for frequency above 6 GHz. Figure A.106 shows that the DS values obtained from our measurements are quite low and less correlated with the frequency. This may be explained by the limited multi-path richness of the experimented channel in our measurement campaign due to the short separation distances between the TX and the RX units. mmmagic Public 185

206 Delay Spread [ns] GPP LoS 3GPP NLoS LoS Measurements NLoS Measurements 30 Figure A.106: 3GPP TR DS model above 6 GHz Vs measurement results. A.3 Outdoor-to-indoor Scenarios Frequency [GHz] A.3.1 O2I measurement campaigns at 3, 10, 17 and 60 GHz Measurement scenario The radio propagation channel was investigated at 3, 10, 17 and 60 GHz in an outdoor to indoor (O2I) scenario. Measurements were conducted using a wideband channel sounder to derive channel parameters such as building penetration losses and channel delay spread values. The measurement scenario, displayed in Figure A.107 on the left side, describes the O2I propagation scenario where the Tx antenna was located outside and mounted on the roof of the white van at 2.5 meters above ground level. The Rx antenna was placed at mobile user level inside the building, Orange Labs premises in Belfort, as indicated by the double-ended orange arrow. The distance between the Tx antenna and the building exterior walls was about 10 meters. The building layout is shown in Figure A.107 on the right side, where each number i (i = 1 38) corresponds with an Rx position. Different Rx locations were considered during the measurements. These included large offices (LO1 and LO2), corridors (Co), a break room (BR) and a residential flat (Fl). Fl consists of a living room and small offices separated by partition walls. BR, adjacent to Fl, is separated with the latter by a bearing wall. Finally, we have LO1, LO2 and Co, located on the same side of the bearing wall as BR, and separated with each other by partition walls. The exterior walls of the building are made of concrete and the windows of standard double-layered glass or coated double-layered windows. Measurement setup The channel sounding system operated in four different frequency bands with a bandwidth of 125 MHz. The distances between the Tx antenna and the different Rx positions were between 10 and 25 meters. The data were collected in three distinctive setups depending on the Rx antenna radiation patterns. As path loss increases with frequency, Rx antennas with narrower beams but larger gains were used at higher frequencies. Therefore, Setup 1 (S1) used omnidirectional (Omni.) antennas at 3, 10 and 17 GHz. Setup 2 (S2) used directional antennas with 90 half-power beamwidth (HPBW) at 17 and 60 GHz and finally, Setup 3 (S3) used highly directional antennas with 20 HPBW at 60 GHz. The Rx measurement equipment used in the different setups is presented in Figure A.108. mmmagic Public 186

207 Figure A.107: O2I scenario and building layout. a) Co: corridor; b) and c) Fl: residential flat; d) BR: break room; e) LO1 and LO2: large office. Table A.29: Measurement setup parameters Setup Parameters Frequency bands [GHz] Bandwidth Tx Power Parameter Values 3 GHz 10 GHz 17 GHz 60 GHz MHz (at 60 GHz, bandwidth reduced from 250 to 125 MHz in processing) 20 dbm at 60 GHz to 30 dbm at 3 GHz Rx Antenna S1: 1 dbi Omni. S1: 0 dbi Omni. S1: 0 dbi Omni. S2: 7 dbi 90 HPBW S2: 6 dbi 90 HPBW S3: 20 dbi 20 HPBW Tx Antenna 1 dbi Omni. 0 dbi Omni. 7 dbi 90 HPWB 10 dbi 50 HPBW mmmagic Public 187

208 Figure A.108: Rx equipment. a) 3, 10 and 17 GHz Omni. antennas; b) 90 HPBW antenna at 17 GHz; c) 20 and 90 HPBW antennas at 60 GHz. For S1, the omnidirectional RX antenna was mounted on a rotating arm describing a circular trajectory during the measurement. The radius of the trajectory was 2λ, 4λ and 6λ at 3, 10 and 17 GHz respectively, where λ is the wavelength. For S2, the 90 HPBW RX antenna was pointed towards 4 different directions with 90 (antenna HPBW) angular step, partitioning the space into 4 sectors. In each sector, the antenna remained static while the data were collected. For S3, the 20 HPBW RX antenna was mounted on a panoramic tripod head and manually moved by the operator to describe a 3-D trajectory during the measurement. The azimuth and elevation angles were recorded by an inertial unit. A 3-D cloud of channel impulse responses (CIRs) at irregular angular intervals in azimuth and elevation was collected. The interval between two consecutive azimuth/elevation points was about 2 to 3. Data Processing For S1, let h n(τ) be the n th CIR recorded during the measurement, N the total number of CIRs and τ the relative delay. N is approximately 400 to 500. The omnidirectional PDP is the average power associated with each delay: N 1 2 PDP h n. (A-27) N For S2 and S3, the omnidirectional PDPs need to be synthesized from the measurements performed using directional RX antennas. The following approach is adopted for the synthesis. In this approach, CIRs are first measured at different angular orientations with an angular step with equal to the RX antenna HPBW. The omnidirectional PDP is obtained by summing the received powers from all pointing angles since an angular step equal to the antenna HPBW ensures that the gain of the synthesized omnidirectional antenna is the directional antenna maximum gain. Therefore, for S2, let h k(τ) be the CIR measured in the k th sector and K the total number of sectors (K=4). The omnidirectional PDP is obtained by summing the received powers from all K sectors: n 1 mmmagic Public 188

209 PDP K h k k 1 2. (A-28) For S3, a regularly spaced P Q grid of CIRs with a 20 step width (antenna HPBW) in the azimuth and elevation domain is synthesized from the cloud. P and Q are the number of azimuth and elevation pointing angles respectively, where P=18 and Q=2 correspond to an azimuth range from -180 to 180 and an elevation range from -20 to 20, respectively. For a specific azimuth-elevation couple indexed by (p, q), h p,q(τ) is defined. The CIR in the cloud, whose angular orientation is the closest to the orientation indexed by (p, q), is assigned to h p,q(τ). The omnidirectional PDP is given as follows: Q P 2. PDP (A-29) q 1 p 1 h p, q Measurement results For each setup, the PDPs were computed assuming antennas with omnidirectional radiation patterns at the Rx side. This required the synthesis of equivalent omnidirectional characteristics from the measurements performed with directional Rx antennas i.e. S2 and S3 as described above. The building penetration losses, defined as an additional attenuation compared to the free-space path loss, as well as the channel delay spread values were then determined for the four frequency bands and at the 38 Rx positions when the measured CIR instantaneous dynamic range was at least 20 db. Table A.30 and Table A.31 summarize the average values for each location. Table A.30: Building penetration loss values Average Penetration Losses [db] Values 3 GHz 10 GHz 17 GHz 60 GHz Co Fl BR LO LO Low dynamic range Table A.31: Channel delay spread values Average Delay spread [ns] Values 3 GHz 10 GHz 17 GHz 60 GHz Co Fl BR LO LO Low dynamic range The results show a strong variation of the penetration loss depending on the window material composition. This is particularly the case for BR and Fl, which are separated by a bearing wall. The incidence angles from the Tx antenna were also similar in these two rooms. As shown in Figure A.109, where the attenuation values for the Rx positions located just behind the exterior wall are summarized, non-coated glass windows in BR favour the radio signal propagation (around 5 db attenuation), while the coated-glass windows in Fl cause strong signal attenuation (more than 30 db). Furthermore, at 60 GHz, there was not enough dynamic range for most of mmmagic Public 189

210 the measurements performed in Fl, so the attenuation was not quantifiable. In LO1, L O2I values behind the non-coated glass windows remain relatively low (10 db), although higher than in BR. This increase could be associated with the different incidence angles between these two rooms. L O2I values recorded in LO2, which is equipped with coated glass windows, and Co are relatively high (15 30 db) but overall lower than the values in Fl. The electromagnetic waves propagating into BR certainly pass through the partition walls that separate the latter from LO2 and Co. The transmission through these walls causes additional attenuation, which could also explain the scarcity of L O2I values at 60 GHz at these locations, as in Fl. An analysis based on the directions of arrival is ongoing in order to confirm these preliminary physical and geometrical interpretations. 40 Attenuation [db] GHz 10 GHz 17 GHz 60 GHz R14 R17 R20 R21 R22 R23 R24 R34 R38 RX Positions Figure A.109: Building entry losses. There is hardly any frequency dependence of the penetration values in BR or LO1 equipped with non-coated glass windows, whereas they increase with the frequency in Fl where the window glass is silver-coated. In Co and LO2, the trend is the same as in BR, where their main propagation paths most likely come from. Regarding the channel delay spread, the computed values remain very low (below 30 ns) and more or less uniformly distributed across the different frequency bands with a very slight decrease at 60 GHz. Figure A.110 shows a typical example of PDPs for the same Rx location. mmmagic Public 190

211 Relative Amplitude [db] GHz 10 GHz 17 GHz 60 GHz B Relative Delay Time [ns] Figure A.110: PDPs at 3, 10, 17 and 60 GHz at Rx position 25. A.3.2 O2I measurement campaign at 2.44, 5.8, and 58.7 GHz Introduction Extensive measurements have been conducted to determine the dependence of building penetration loss on propagation path elevation angle relative to the building exterior wall. This is useful for base station antennas, which are at substantially different height than terminals inside a building. Furthermore, to provide proper modelling for e.g. earth to space intersystem sharing scenarios in ITU-R it is essential to account also for the elevation angle dependence. Measurement campaign This measurement campaign, depicted in Figure A.111, provides results for the traditional type of building (having non-coated type of window glass) in the Kista area of Stockholm for different elevation angles and radio frequencies in the range 2-60GHz. The measurements were performed with the receiver antenna (Rx) is located about 1 m outside an open window at floors 4, 7 and 8 and the transmit antenna (Tx) is located about 1.8 m above the floor on floor 4. Both horizontal and vertical orientation of the Tx and Rx antennas were used. At floor 8 the elevation angle is around 70 degrees. Consequently, the Rx antenna has to be horizontally oriented for both the vertical and the horizontal polarizations (outside floor 4 at the exterior wall) in order to point the antenna lobe towards the exterior wall segment outside floor 4. For Rx locations at floors 4 and 7 the antenna orientations are the same as corresponding polarisations. For both vertical and horizontal polarisations, the Tx antenna is oriented to maximize the received power. For the vertically polarized Rx antenna at floor 8, the Tx antenna orientation for maximum power is horizontal close to the wall (Tx points 1-5) and vertical for the rest locations (Tx points 6-13). Carrier frequency Table A.32: Basic data of measurement setup 2.44 GHz 5.8 GHz GHz Bandwidth 80 MHz 150 MHz 2 GHz Transmit power 10 dbm 10 dbm 10 dbm Antenna Tx Electric dipole 2dBi Electric dipole 2dBi Electric dipole 2dBi mmmagic Public 191

212 Antenna Rx Electric dipole 2dBi Electric dipole 2dBi Electric dipole 2dBi Tx Antenna height Rx Antenna height 1.8 m (ref : indoor floor) 1.0 m (ref : indoor floor) Figure A.111: Outdoor to indoor measurement scenario (left). The indoor Rx locations at floors 4, 7 and 8 are marked with blue-filled circles and the Tx locations at floor 4 are marked with a red-filled circles. On the right hand side, a photograph of the building is shown. Analysis The elevation angle is determined as follows. First the point where the line connecting the Rx and Tx intersects the exterior wall is found. This point is projected vertically down to the wall segment at floor four. The elevation angle is now given by the line connecting this point to the Rx antenna. The elevation angle θ is the complement (θ = 90 φ) of the polar angle φ used in spherical coordinate systems. The measurements are calibrated by free space measurements accounting for any antenna pattern effects at Rx due to the elevation angle. The elevation angle variation of antenna gain was found to be within 1 db. In Figure A.112 the building entry loss is shown for the different elevation angles and frequencies (corresponding cumulative distributions are shown in Figure A.113). Points p1, p8 and p11 are omitted as the elevation angle is ambiguous for these points. Maximum, median and minimum losses are presented in three separate graphs. It was found that the elevation dependent loss L el may be modelled with a simple linear expression L el = k abs (θ)(db), (A-30) where k (db/ 90º) is a model parameter and θ is the elevation angle. This function has been fitted to the measurement data assuming that k is frequency independent. The model parameter mmmagic Public 192

213 k ranges from 24 db/90º to 31 db/90º. The standard deviation between model and measurements is in the range 1.35 to 2.22 db. Based on these results mmmagic proposes to use the linear model above with k greater than 24 db/90º for modelling the elevation angle dependence of building entry loss for traditional type of buildings as it is both simple and accurate in comparison with measurement data. Figure A.112: Measured building entry loss (loss in excess of free space loss) versus elevation angle for the different frequencies together with the fitted linear model. The top graph corresponds to minimum loss (2.5% level); the middle graph corresponds to median loss; and the bottom graph to maximum loss (97.5% level). mmmagic Public 193

214 Figure A.113: Cumulative distribution functions of measured building entry loss for the different elevation angles and radio frequencies. The left hand graphs show distributions where Tx locations p1, p8 and p11 are excluded. The right hand graphs show distributions for all Tx locations. A.3.3 O2I measurement campaign at 17 GHz Sounder setup The channel sounder setup is described in Section A.2.7. For the outdoor-to-indoor measurements, the transmission link crossed a street, making direct reference wiring between transmitter and receiver unfavourable. Therefore, external GPS receivers were used for reference and trigger source on the transmitter and receiver side. Additionally, the results were time-aligned in post-processing when using averaging. mmmagic Public 194

215 Measurement description The transmitter used a Vivaldi antenna specified up to 22 GHz. The antenna has a gain of 9 dbi and a directivity of 11 dbi at 17 GHz. At the receiver, an identical Vivaldi as well as an additional dipole antenna were used. This allowed to compare the resulting power delay spectrum for a directed and omnidirectional reception in order to make assumptions about the direction of arrival of some of the paths. The measurements were conducted at the Rohde & Schwarz premises in Munich, Germany. The transmission link was established between two buildings on each side of a street (Mühldorfstraße). The buildings on each side of the street are 20 meters apart, with a part of one building standing back another 22 meters. The outdoor-to-indoor channel sounding measurement was conducted at one LOS and one NLOS point in an empty room on the ground floor. For the LOS measurement, the link distance was about 50 meters. The transmitter was located just outside the 5th floor on the east side of the street. The receiver was located on the ground floor of the recessed building on the opposite side of the street. An aerial image of the location can be seen in Figure A.114. The first Rx position was located behind the window in a LOS scenario. The second position was located at the very end of the room without in a NLOS scenario. Table A.33 gives a summary of the measurement campaign. Table A.33: Summary of the 17 GHz measurement campaign in Munich outdoor to indoor Carrier Frequency 17 GHz Bandwidth 160 MHz No. of Tx positions 1 No. of Rx positions 2 No. of LOS links 1 No. of NLOS links 1 Tx (BS) height ~14 m Rx (MS) height 2 m Tx antenna Vivaldi, vertically polarized Rx antennas Vivaldi, vertically polarized Dipole, vertically polarized mmmagic Public 195

216 Figure A.114: Aerial view for outdoor to indoor measurement at R&S. Google, GeoBasis DE/BKG. Figure A.115 shows the PDPs for a single snapshot as well as the average over 1000 snapshots distributed over 21 seconds of the LOS link TX 1 RX 4. Taking the transmit power, cable losses and antenna gain into account, the expected path loss of 91 db over 50 meter is exceeded by 4 db in this O2I LOS scenario, which may be explained with additional attenuation caused by the office window. Figure A.115: Single snapshot (left) and averaged (right) PDPs for O2I LOS TX1-RX4. Averaging is used in post-processing to enhance the visibility of individual paths or make certain paths visible at all. Multiple paths below -110 db are received when using the omnidirectional dipole antenna. Some weak peaks at the end of the plot (path distance more than 275 meter) can be observed with the dipole antenna. These paths may be caused by multiple reflections, resulting in high attenuation and high delay. In general, the delay spread is higher when using omnidirectional antennas as more paths are received. For mobile communication with devices like cell phones, a higher delay spread has to be assumed and to be taken account when designing the waveform. For point to point communication links, antennas with higher directivity may be used which results in smaller delay spread for the channel. mmmagic Public 196

217 Figure A.116: Averaged PDPs for O2I NLOS RX5. For the NLOS scenario (receiver location RX 5), the power delay spectrum is below the noise level and can only be observed using averaging. Figure A.116 shows the averaged PDP. The PDP features roughly two clusters of paths. The first cluster has the higher receive power and consist of at least three distinct paths. The second cluster exhibits a travel distance of more than 100 meters and mostly consists of what seems like a single path plus some later signal components with lower power. Back-and-forth reflection in the street canyon may be an explanation for this cluster. A.4 Reflection, scattering and blockage A.4.1 Specular reflections Measurement setup A 2 GHz wide baseband signal is generated using Keysight M9099 Waveform Creator software and used to configure a Keysight M8190A arbitrary waveform generator (ARB). This is then used to drive I & Q ports on a Sivers IMA up-converter, thus producing a 60 GHz modulated carrier. At the receiver, a Sivers IMA device down-converts the 60 GHz signal to an IQ IF signal and this is then captured and processed using a high performance digital oscilloscope (DSO), MSOS804A. The Keysight VSA & Waveform Creator channel sounding function operates by repeatedly transmitting a single carrier signal bearing a modulated waveform. The waveform has excellent auto-correlation properties, and a low peak-to-average power ratio. The bandwidth and duration of the modulating waveform may be varied to suit the channel measurement required. Spectrum shaping can be applied to reduce out of band interference when transmitting the signal in a live environment. For the scattering measurements, the transmitting antenna was a vertically polarized rectangular horn antenna, while a circular horn was connected though an orthomode transducer at the receiver allowing simultaneous horizontal and vertical polarizations to be recorded. Both antennas had a gain of 20 dbi and a 3 db beamwidth of 16. Figure A.117 shows a picture of the hardware. mmmagic Public 197

218 Figure A.117: 60 GHz Keysight channel sounder. The transmitter and the receiver were mounted on a 1.00 m x 0.70 m x 0.72 m trolley using two poles as illustrated in Figure A.118 Both the transmitter and the receiver antennas were fixed and pointed at the same point on the wall creating an isosceles triangle between the antennas and the surface to be measured. In this way, the incident angle was equal to the reflected angle, and given the large 2 GHz bandwidth the specular scattered component could be isolated from the measured power delay profile and analysed. Both antennas were placed at a height of 1.5 m. Using this setup, it was possible to measure the specular scattering component for incident angles of 15 ο and 30 ο. For the 30 ο measurements, the separation distance of the trolley from the wall was approximately 1m, as depicted in Figure A.119 (a). In some chosen locations, it was physically impossible to reduce the distance to the wall in order to measure wider angles of 45 ο and 60 ο and, for that reason, it was necessary to increase the separation between the transmitter and the receiver antennas from 1 m to 3 m, as it is shown in Figure A.119 (b). The channel sounder for these measurements has been configured for external trigger, which is sourced from a shaft encoder mounted on a wheel of the measurement trolley. This creates a trigger every 4 mm, with the equipment capable of recording at a sample rate above walking pace. Figure A.118: Trolley used to conduct the scattering measurements. mmmagic Public 198

219 0.98 m 0.95 m Document: H2020-ICT mmMAGIC/D TROLLEY 1 m 1.13 m 3 m 3.3 m TROLLEY (a) (b) Figure A.119: Graphical illustration of the trolley measurements for (a) 30 0 and (b) The measurement results for three different building materials are presented in this work. These include a very rough outdoors surface (dressed stone block), an outdoors surface of medium roughness (Bath stone), and a smooth surface of non-metallized window. All the measured surfaces are described in Table A.34. Table A.34: Measured wall types for analysing diffused scattering effects at 60 GHz Building Type Walls Name Location (Indoor/Outdoor) Dressed Stone Block Outdoor Bath Stone Outdoor Glass Indoor and Outdoor A.4.2 Non-specular scattering Another important propagation mechanism in the mm-wave bands is diffused scattering at nonspecular directions. In order to measure the non-specular diffuse scattered multipath components, the wideband channel sounder was pulled along in the shape of an arc with a radius of 2 m around different indoor and outdoor building surfaces from an angle of approximately 10 ο up to about 80 ο from the normal to the wall. All measurements were performed at 60 GHz with a bandwidth of 2 GHz, and a selection of rough and smooth building surfaces have been characterized. These included three indoor wall materials (internal block wall with plaster finish; write-on-sheet wall; and a single glazed window) and three outdoor wall materials (dressed stone wall; Bath stone wall; and Penzance sand stone & red oxide wall). Details of these materials are presented in Figure A.120. mmmagic Public 199

220 Figure A.120: Wall materials used in the non-specular study. The transmitter was set at a static location at an angle between 10 and 80 from the normal (with steps of 5 ) while the trolley was pulled along the arc. The Keysight channel sounder has been used for these measurements and has been configured for external triggering, which is sourced from a shaft encoder mounted on a wheel of the measurement trolley. This creates a trigger every 4 mm, with the equipment capable of recording at a sample rate above waking pace. The antennas of the transmitter and of the receiver were constantly pointing towards the centre of the arc. The graphical representation of the measurement set-up depicted in Figure A.121, whereas Figure A.122 illustrates an example from the outdoor measurements that have been conducted for characterizing the non-specular reflections. TX 80 TX 70 TX 50 TX 60 RX TX 40 TX 30 Figure A.121: Graphical description of the arc diffuse scattering measurement. mmmagic Public 200

221 Figure A.122: Transmitter and Receiver Setup for an arc measurement following the chalk line. The main purpose of those measurements was a qualitative investigation of the effect of surface roughness on the wave scattering at non-specular reflections. Results presented in Chapter 3 demonstrate this effect and furthermore develops a method for modelling the relative permittivity of various building materials. In order to evaluate and assess the potential dependency of diffused scattering on the transmitter-receiver separation, the angle of incidence (transmission angle) and the surface type, two building materials have been tested: The red stone wall (which represents a rough wall surface) and a concrete pillar which corresponds a smooth wall surface. The measurements carried out at an incidence angle of 15, 30 and 45 degrees whereas the radius of the measured arc was set to 2, 4 and 6 meters respectively. To measure the power received at all directions, the Tx was placed to the fixed angle -with respect to the normal from the wall (~0 ο to 90 ο ) and the Rx was placed on a trolley and moved along the arc. Exemplary measurements from for the concrete pillar and the red stone wall measurements are illustrated in Figure A.123 and Figure A.124, respectively. Figure A.123: Diffused scattering measurement snapshot Concrete Pillar. mmmagic Public 201

222 Figure A.124: Diffused scattering measurement snapshot Redstone wall. A.4.3 Ground reflection Ground reflection measurement in a courtyard This measurement campaign evaluates the impact of ground reflection for mm-wave channels by investigating channels at 28 GHz and 39 GHz. The two-ray ground-reflection model (TWGRM) implies constructive and destructive interference between the line of sight (LOS) and ground reflected (GR) path, depending on the given geometry of the scenario. This effect may be observed at distances below the crossover distance d c = 4π h th r. At lower λ frequencies, for example below 6 GHz, the crossover distance is low and the cell size is big. At high frequencies, moving towards the millimeter band, the crossover distance gets large while the cell size shrinks. Therefore, the ground reflection effects have to be modelled carefully. The two ray interference path loss can be expressed as PL tri = 20 log 10 ( 4πd λ 1 + Γ ejφ 1 ), Γ = sin θ e r cos 2 θ sin θ + e r cos 2 θ, (A-31) where θ is the angle between the ground and the ground reflected path. The measurements conducted with the Rohde & Schwarz Channel Sounder aimed to evaluate the ground reflection contribution in a real environment. The effects were investigated on flagstone/concrete and grassland. Additionally, the effect of a watery surface was evaluated. Measurement setup A 2 GHz bandwidth signal was used. The high signal bandwidth should reflect the possibility to use broad bandwidth channels in the mm-wave range. The signal was transmitted using an R&S SMW200A with a maximum sampling rate of 2.4 GHz. As receiver, an R&S FSW43 was used in combination with an R&S RTO. The channel impulse response (CIR) was calculated by the R&S 5G Channel Sounding software. The chosen frequencies can both be directly addressed by the used devices, thus no external up- or downconverters were used in the measurements. For all measurements, transmitter and receiver were connected to a 10 MHz reference clock. Depending on the scenario, transmitter (Tx) and receiver (Rx) are either directly connected via cable or two Synchronomat devices by Fraunhofer HHI. The Synchronomats are able to provide a stable reference clock signal with minimal deviation between the two devices. As transmitter and receiver antennas, identical standard gain horns for 26.5 GHz to 40 GHz with mmmagic Public 202

223 a gain of 10 dbi were used. The 3 db beamwidth is given with 54.2 for the E-plane and 54.4 for the H-plane. The setup is depicted in Figure A.125. Figure A.125: Setup Block Diagram. Two different scenarios were considered. First, the effect of constructive and destructive interference was evaluated by measuring the CIR at different distances. Second, the individual paths are resolved in the CIR at fixed positions, identifying power and phase differences. Two-ray ground-reflection path loss over distance These measurements explored the constructive and destructive interference between the LOS and GR path. A measurement series consists of multiple individual measurements at different distances along a straight line. Figure A.126 shows the two test tracks at the Rohde & Schwarz premises in Munich, Germany. Figure A.126: Measurement Locations (grassland). mmmagic Public 203

224 The blue line indicates the path of movement of the Rx for flagstone/concrete, while the green line shows the path for grassland measurements. The Rx locations are at distances between 15 m to 35 m, with Tx and Rx mounted at about the same height of 86 cm and 83.5 cm above the ground facing directly towards each other. The transmitter remained fixed, while the receiver was moved on a trolley. With this setup, transmitter and receiver used a single reference clock, connected by cable. The maximum path difference between LOS and GR is less than 10 cm, which cannot be resolved to individual paths using 2 GHz of bandwidth. Thus, both paths contribute to the first peak in the CIR. The first peak is then time-gated and is a combination of LOS and GR power. Figure A.127 shows the received power levels of the first peak in the CIR for 28 GHz (top) and 39 GHz (bottom) on flagstone/concrete in blue. Each point is averaged over 50 CIR snapshots. The measurement points are connected using a cubic spline interpolation for presentation purposes. In addition to the measured values, also the free space path loss and two-ray groundreflection model are plotted in the figure. Both models are plotted for the centre frequencies 28 GHz and 39 GHz, respectively. Given the 2 GHz sounding bandwidth, the received signal power for the first peak (combining LOS and GR) in total adds up to about 5 db above expected free space path loss (FSPL) for constructive interference and about 9 db below expected FSPL for destructive interference at both frequencies with flagstone/concrete surface. Figure A.127: Received signal strength over distance on flagstone/concrete: measured power (blue), FSPL (red), TRGR model (black). Unfortunately, the ground is not fully flat, contributing to deviations of the periodicity of constructive and destructive interferences. Additionally, the high bandwidth of the sounding signal contributes to the fact, that especially the destructive interference is reduced compared to the model for a single CW signal. Since the locations for destructive interference are frequency dependent, different parts of the signal spectrum experience different levels of interference, leading to frequency-selective fading. Transmitting CW signals and investigating interference for single frequencies were not in scope of this measurement campaign. The behaviour for individual frequencies is assumed to follow the two-ray ground-reflection model. Figure A.128 shows tworay ground-refection model plots for the upper and lower frequency in addition to the centre frequency, illustrating the frequency dependency. mmmagic Public 204

225 Figure A.128: Two-ray ground-reflection model for different frequencies with measured values on flagstone/concrete: measured power (blue), FSPL (red), TRGR model upper, lower and centre frequency (black, green, light blue. As previously mentioned, the same measurements were conducted on grassland in a range of 15 to 30 meters with 25 cm steps. Given the structure of grassland, more scattering of the signal was expected. In the plot, the relative permittivity was substituted by dry earth in order to act as reference. Figure A.129 shows the measured values alongside the reference simulation. The measured data show no real correlation to the simulated values, based on a flat dry earth structure. At least for the given geometry in the measurement scenario, the two-ray model does not seem to apply to grassland as good as for a flagstone or concrete surface. It can be deducted that multiple reflection points with different path lengths each sum up at the receiver, instead of only a single reflection point in the ideal two-ray ground-reflection model. Figure A.129: Received signal strength over distance on grassland: measured power (blue), FSPL (red), TRGR model (black). The measurements show that in a real environment the two ray ground reflected path has a huge impact on the received signal power over small deviations in distance. The effect may be mmmagic Public 205

226 lowered by using higher bandwidths. However, due to the frequency-selective behaviour, frequency-selective fading will be introduced. Ground reflection on grass and concrete / flagstone In an additional measurement, the ground reflection path is resolved as an individual path in the CIR using a different geometry leading to a bigger path difference between LOS and GR. Again using 2 GHz of sounding bandwidth, the transmitter was placed at a higher location. By placing receiver and/or transmitter higher, the path difference between LOS and GR can easily be increased. Again, two different surface substances were chosen for the measurement: stone/concrete and grassland. Additionally, the concrete surface was covered with water to visualize the effects of changing the permittivity and flatness of the ground surface. Figure A.130 shows different locations for transmitter and receiver for the concrete surface measurements. Positions for the grassland measurements are shown in Figure A.126. Detailed information on all measurement positions are provided in Table A.35. Table A.35: Positions for Measurement Campaign Pos Tx-Rx Freq TX RX height LOS distanctance GR Path dis- [GHz] height Tx1-Rx Tx2-Rx Tx2-Rx Tx3-Rx Figure A.130: Measurement Locations (concrete surface). For the measurements using a grass surface, the transmitter was positioned at a height of 5.95 m, the receiver at 1.96 m. The horizontal distance was m, leading to a path difference of 1 meter. The positions were chosen carefully, in order to prevent additional reflections on buildings or other objects that might impair the analysis. mmmagic Public 206

227 Due to a greater distance of the transmitter and receiver in the concrete/flagstone scenario, the transmitting antenna was positioned slightly higher. With a height of 2.10 m of the receiving antennas, the path difference is again about 1 m at 28 GHz measurement. The path difference is about 90 cm at 39 GHz, because the antenna was positioned at a height of 1.96 meters in this case. Figure A.131: PDP for grassland (blue) and flagstone (red). Figure A.131 shows the PDP of the grassland (blue) compared to the concrete surface (red) for 28 GHz and 39 GHz. In post-processing, all graphs were averaged over 100 PDPs. All measurements show a clearly visible LOS peak with the ground reflection peak/path after 1 m and 90 cm, respectively. Comparing both measurements, the reflected power is greater from the concrete surface compared to grassland. Additionally, the reflected power on grassland at 39 GHz is considerably smaller than at 28 GHz. The power of the LOS paths fits well with the expected values, taking into account the antenna pattern and the free space loss. Table A.36 illustrates the individual points of power in dbm. Pos. Tx-Rx LOS power GR power Tx1-Rx1 28 GHz dbm dbm Tx1-Rx1 39 GHz dbm dbm Tx2-Rx2 28 GHz dbm dbm Tx2-Rx2 39 GHz dbm dbm Table A.36: Path power for different measurements. For the comparison between dry and wet ground, the transmitter was positioned on the 5th floor (Tx3). The receiver was placed in the centre of the square (Rx3). An area of 5 m x 8 m in front of the receiving antenna was wetted for the measurement. As can be seen from these measurements, the ground reflection can be identified as the strongest multipath component below the LOS power given enough resolution and potentially be mitigated using time gating. Ground reflection measurements in a street canyon (industrial area) Description and scenarios The ground reflection measurements were conducted in a street canyon scenario in an industrial area. The transmission point is denoted by BS. The antenna was placed on top of a van at 2.5 m above ground level. The RX antenna was kept at mobile user level at 1.5 m height. is the Rx was mounted on a trolley pushed by an operator along a linear trajectory. The distance between Rx and Tx ranged from 30 m to 160 m. The operator was either behind or in front of the Rx antenna, leading to a LOS or obstructed measurement scenario as shown in Figure A.132. mmmagic Public 207

228 Figure A.132: Measurement environment and scenario. The measurements were performed using a wideband channel sounding system. An arbitrary waveform generator (AWG) produced successive 8192-length wideband sequences at 1 Gbpsrate with a 125 MHz channel bandwidth. An impulse response was measured every 200 ms corresponding to a distance around 20 cm between two successive impulse responses. The measurement setup is summarized in Table A.37. Table A.37: Measurement Setup Parameters. Measurement Setup Parameters S1 3 GHz 17 GHz 60 GHz 17 GHz 60 GHz S2 Frequency Bands [GHz] Bandwidth 125 MHz TX Power from 20 dbm (60 GHz) to 30 dbm (3 GHz) RX Antenna 1 dbi, Omni. 0 dbi, Omni. 0 dbi, Omni. 7 dbi, 90 HPBW 6 dbi, 90 HPBW TX Antenna 1 dbi, Omni. 13 dbi, 30 HPBW 15 dbi, 30 HPBW 13 dbi, 30 HPBW 15 dbi, 30 HPBW Polarization Vertical Omnidirectional antennas or sectorial antennas with vertical polarization were used at each end. At 3 GHz, an omnidirectional antenna with 1 dbi gain was used at each end. At 17 and 60 GHz, antennas with larger gains were used to compensate for the increased propagation losses. On the TX side, 13 and 15 dbi gain antennas with 30 HPBW were used at 17 and 60 GHz, respectively. On the RX side, an omnidirectional antenna and a 90 HPBW directional antenna were used for both frequency bands. The distance between the Rx omni and Rx directive antenna was around 10 cm and both antennas were located along a line orthogonal to the measurement run. Figure A.133 and Figure A.134 display the TX and RX equipment and antennas. mmmagic Public 208

229 Figure A.133: Tx antenna at 17 GHz (left) and Tx measurement van (right). Figure A.134: Rx antennas at 17 GHz (left) and trolley with Rx antenna at 60 GHz (right). Results The path gain, inverse of path loss, is calculated from the impulse response. We did not apply any sliding averaging generally used to suppress the small-scale fading as we wanted to highlight the interferences between the direct path and other reflected paths such as the ground reflection. The gain is calculated according the following equation: Gain 10 log 10 h d (A-32) As illustrated in Figure A.135, a typical impulse response is composed by a first strong component and by secondary weak components attenuated by at least 15 db compared to the first one. Therefore, the channel gain is strongly correlated with the amplitude of the first component which results from positive or negative interference between non-resolved paths. Taking into account the specific environment depicted in Figure A.132, where the propagation is guided by the ground and parallel vertical walls, we can assume that the direct path interferes with the ground and wall reflections. mmmagic Public 209

230 data1 data2 data3-115 Amplitude (db) Relative delay (ns) Figure A.135: Three typical impulse responses separated by 5 cm at 17 GHz with interferences in the main component. Figure A.136, Figure A.137 and Figure A.138 show the channel gain as function of the distance at 3, 17 and 60 GHz respectively. Following conclusions can be drawn from the inspection of the three figures: - The channel gain is impacted by the small-scale fading with fading node inter-distances increasing with the Tx-RX distance. This phenomenon is typical when a direct path interferes with the ground reflection. It appears at all frequencies. The fading node inter-distance decreases with the wavelength. As the Tx-Rx distance is the same for the three frequencies, the repetitive fading pattern is clearly visible at 60 GHz, a little less at 17 GHz and hardly detectable at 3 GHz. - The measurement environment is a pedestrian area with a flat concrete ground that should favour a strong and stable ground reflection. However, the repetitive fading pattern is not regular and the small-scale fading experienced by the omnidirectional and directional antennas is not always correlated. The small-scale fading may arise from interferences between the direct path, the ground reflection and reflection on vertical walls. Vertical walls are not regular with large windows and there were some urban objects or cars along the measurement trajectory. Vertical walls or urban objects can create strong but non-stable reflections. - Due to the limited bandwidth and large antenna aperture, it is impossible to resolve the direct and reflected paths and therefore to estimate how each path contributes to the channel gain. Anyway, a LOS street guided environment with a non-obstructed Tx-RX line cannot be modelled as a free space environment regardless of the frequency. The channel gain difference with the free space loss ranged roughly between -10 db and +5 db. - When the user is placed between the Tx and Rx antennas (blocked scenario), all the abovementioned paths are attenuated and the repetitive fading pattern disappears. In that case, the channel gain can be simply modelled by the free space loss plus a 10 db extra loss. Therefore, the ground/wall reflection effect may change in the presence of blockers (people, car, etc) or non-regular ground. Future measurements are planned in a non-pedestrian LOS urban street-guided environment mmmagic Public 210

231 Los Blocked Free Space -75 Amplitude(dB) distance (m) Figure A.136: Channel gain vs distance at 3 GHz, Rx omni antenna Omni Dir Free Space Los Blocked Free Space Amplitude(dB) Amplitude(dB) distance (m) distance (m) Figure A.137: Channel gain vs distance at 17 GHz Omni Dir Free Space Los Blocked Free Space Amplitude(dB) Amplitude(dB) distance (m) distance (m) Figure A.138: Channel gain vs distance at 60 GHz. mmmagic Public 211

232 A.4.4 Blockage Measurements Description and scenarios Additional measurements to what is described in Section A.1.1 were performed in order to evaluate model blockage. Measurements were carried out at a central frequency of 62 GHz and 83.5 GHz with a bandwidth of 6 GHz. A phantom reproducing a human (torso and head) is blocking the line of sight between transceivers. Figure A.139: Blockage measurement in the office room. In the left hand side picture, red (resp. black) rectangles indicate desks (resp. storage cabinets and bookshelves). The measurement scenario is depicted in Figure A.139 with the receiver, the phantom, and the transmitter located behind the back of the phantom (right) and a floor plan of the measurement (left) is shown. The phantom is comprised of the head and trunk, which are made of polyvinyl chloride (PVC). A grid of six measurement points is considered in GHz (black dots in the left hand side of Figure A.139) whereas 15 measurement points are considered in GHz (black and red dots). At each frequency, the measurements are carried out with and without the phantom. Results Figure A.140 shows an example of power delay profile with and without the phantom between the transceivers. It can be seen from the figure that the presence of the phantom attenuates considerably the power level of the LOS component with a loss higher than 15 db. For the sake of comparison, we evaluate the shadowing loss of the LOS path at the same positions (i.e., black points indicated in the floor plan of Figure A.139) for the two investigated frequency bands. Table A.38 shows the shadowing loss values. ΔP 0 is the difference in the power level of the first arriving components without and with the phantom between transceivers. P1,, P6 are the positions of the Rx antenna and are indicated in the left hand side of Figure A.139. mmmagic Public 212

233 Figure A.140: Example of power delay profile with(out) the phantom. Table A.38: Shadowing loss values. All values are expressed in decibels (db). P1 P2 P3 P4 P5 P6 ΔP 0 (62 GHz) ΔP 0 (83.5 GHz) In the following analysis, P6 is omitted as the LOS path was not actually blocked by the phantom in that position. The shadowing loss of the LOS path varies from db to db at 62 GHz, whereas it ranges from 6.30 db to db at 83.5 GHz. Regarding the position P3, we can notice a difference of the LOS path shadowing (ΔP 0 ) when comparing the two bands. This is probably due to a discrepancy on phantom positioning, since the measurement campaigns were carried out at two different times. Hence while the positions P1, P2, P4, and P5 are always in obstructed condition and position P6 in LOS for both campaigns, the obstruction of the position P3 could differ. This explains the difference on shadowing loss values for this specific measurement point. This could also be the reason why the shadowing loss is lower at the higher frequency at some positions (i.e., P1, P3, and P4). A.5 LSP frequency correlations for multi-frequency simulations The simulations presented in Sections 2.1.5, 2.2.2, A.1.4, and A.2.2 are used to examine the large scale parameter (LSP) correlation across frequencies. In 3GPP [3GPP38.900], as well as in the mmmagic proposal, see Table 4.1, the LSPs are same across the frequency. In this section, we test this approximation by calculating the correlations of the LSP at different frequencies. If the LSPs are same, or scaled according to frequency dependent mean values, the correlations are all exactly one. Examples of some LSPs at different frequencies are shown in Figure A.141 where each point corresponds to one simulated channel at different frequencies. If the model approximation is accurate these points should lie on a line. Figure A.142 presents mmmagic Public 213

234 the correlation between the simulated LSPs in airport LOS and in street canyon LOS environments. In both of these cases, the simulations include four centre frequencies from 15 to 83.5 GHz leading to six non-zero frequency offsets. Most of the LSPs are highly correlated across frequencies with only a few exceptions. In the street canyon, all of the correlations are fairly high even with the largest frequency offset between 15 and 83 GHz. In the airport, the elevation spreads, ESD and ESA, have relatively low correlations across frequencies indicating that the model approximation might not be valid for all LSPs in all environments. Nevertheless, since most of the LSPs have relatively high correlations, it can be concluded that the model approximation is fairly good and the LSPs can be assumed to be same (or scaled according to frequency dependent mean values, if applicable). It should be noted that correlation between frequencies is not same as frequency dependency of LSPs, as studied in section 3.1. The correlation between frequencies only affects how the LSPs change as a function of the frequency relative to the mean values. Figure A.141: Correlations between 15 GHz ASD and 83.5 GHz ASD in airport (lhs.) and 61 GHz K-factor and 83.5 GHz K-factor in airport (rhs.). Figure A.142: Correlations between LSPs in airport (lhs.) and in street canyon (rhs.). A.6 Cross-polarization ratio in indoor and outdoor environments A.6.1 Measurement campaigns The multipath components (MPCs) cross-polarization ratio (XPR) is analysed based on measurements in sections 2.1.3, 2.2.1, 2.2.5, A.1.3, A.2.1, and A.2.3. This analysis is based the MPC XPR model presented in [KGM+16] and parametrization presented in [KJN+17]. The work in [KJN+17] includes measurement results from many projects done by Aalto University in several mmmagic Public 214

235 indoor and outdoor environments at frequency bands from 15 to 83 GHz. The indoor environments include airport and shopping mall as well a cafeteria and a coffee rooms. The outdoor environments are open square and four different street canyon environments. The mapping between environments and measured frequency bands is presented in Table A.39. Table A.39: Measurement environments and frequency bands [KJN+17] PLACE 15 GHZ 28 GHZ 60 GHZ 83 GHZ Shopping mall (SHOP) X X X Airport (AIR) X X X X Coffee room (C1) X X X Cafeteria (C2) X Open square (SQR) X X X X Street canyon 1 (SC1) X X Street canyon 2 (SC2) X Street canyon 3 (SC3) X X X X Street canyon 4 (SC4) X A.6.2 Multipath detection The measurements are done with a VNA-based sounder with one omni-directional antenna and a horn antenna that is rotated in azimuth. The main polarisation measurements are done with vertical-to-vertical polarization and the cross-polarisation with horizontal-to-vertical (horn-toomni) polarization. Different horns, with equal azimuth and elevation beamwidth, are used for the main and cross-polarisation measurements. Also, every frequency band uses different antennas but the antenna gains and azimuth and elevation beam-widths are equal. The multipaths are determined from the power angular delay profiles (PADPs). A multipath is defined as a local maximum in each PADP. Each path is assigned with a main (M i) and a cross-polarisation (C i) amplitude well as delay and azimuth angle. [KJN+17] The measurement noise level limits the available information on the MPC properties. A noise threshold level P th is defined above the noise and everything below the noise threshold is omitted. Sometimes an MPC is detectable for the co-polarized PADP while it is below the noise floor for the cross polarized PADP and, therefore, unavailable. Three different kinds of MPC XPR data samples can be identified: o o o MPCs with both co- and cross-polarised components above the noise threshold. XPR is the difference between main and-cross-polarisation amplitudes, i.e., XPR = M i-c i. MPCs with the co-polarized component above the noise threshold but with the cross-polarised component below the threshold. XPR is known to be larger than the difference between the main polarisation amplitude and the noise threshold, i.e., XPR> M i-p th. MPCs with cross-polarised component above the noise threshold but with the co-polarized component below the threshold. XPR is known to be less than the difference between the noise threshold and the cross-polarisation amplitude, i.e., XPR=< P th -C i. One example measurement result is shown in Figure A.143. mmmagic Public 215

236 Figure A.143: Main polarization PADP and its power delay profile in the cafeteria at centre frequency of 63 GHz. Detected MPCs are shown with markers and the noise threshold level is shown with black dash line [KJN+17]. A.6.3 MPC XPR Model Typically, XPR is modelled with a log-normal distribution. The mean and standard deviation values constants with different values for different environments. In [KGM+16] [KJN+17], it is shown that this simple model does not fit the measurement data very well and the parameter values are strongly dependent on the measurement dynamic range. Since the dynamic range is different with every sounder and frequency, the mean and standard deviation parameters are incomparable between different frequencies and different measurement campaigns. It is also shown that this simple model, with constant mean value, underestimates the multipath XPR for the most significant strong paths and, therefore, overestimates the total power in the crosspolarisation components. In order to take into account the fact that strong and weak paths have different XPR statistics, the MPC XPR is modelled as log-normal distribution with excess loss dependent mean value and a constant standard deviation [KGM+16] [KJN+17] XPR [db] = N(μ, σ) (A-33) μ = αe + β, E β/α (A-34) μ = 0, E > β/α, (A-35) where α, β, and σ are the model parameters and E is the multipath excess loss, i.e., the difference between free space path loss (FSPL) at the multipath delay and the path main polarisation amplitude. Eq. (A-35) ensures that the mean XPR is always zero or positive. By making the mean value a function of the multipath excess loss ties the XPR to the amount of depolarization that the path experiences as it is reflected, scattered, or diffracted. A.6.4 MPC XPR model parameters In order to take into account all MPCs, including those that do not have both main and crosspolarisation components above the noise threshold, a Tobit maximum likelihood estimation is used to calculate the parameter estimates α, β, and σ [KJN+17]. Figure A.144 shows one example of fitted MPC XPR model and the measured XPR values. mmmagic Public 216