Multi-User MIMO Channel Reference Data for Channel Modelling and System Evaluation from Measurements Christian Schneider, Gerd Sommerkorn, Milan Narandžić, Martin Käske, Aihua Hong, Vadim Algeier, W.A.Th. Kotterman, Reiner S. Thomä Institute of Information Technology, Ilmenau University of Technology, Germany {christian.schneider}@tu-ilmenau.de Carsten Jandura Vodafone Chair Mobile Communications Systems Technical University of Dresden, Germany carsten.jandura@ifn.et.tu-dresden.de Abstract Novel channel models conceived in IST-WINNER, COST273 or standardisation bodies introduce reference scenarios to evaluate the link- and system-level performance of multi-antenna concepts. Although, parameter of those listed models are publicly available underlying measurement data are only rarely accessible. This paper presents an extensive multiuser and multi-base station MIMO measurement campaign in an urban macro cell scenario. The measurement setup allows high-resolution path parameter estimation through the use of dedicated measurement antennas. Furthermore, a framework for the extended use of channel sounding data is presented. Parts of the measurement data and estimated multipath parameters will be made publicly available. Index Terms channel sounding, measurement data, high resolution multipath parameter estimation, spatial channel modelling, reference scenario I. INTRODUCTION n order to allow comparison of performance evaluations, by I simulation, of different multi-antenna communication systems, the definition of reference scenarios is required. Channel models developed within COST [1][2], WINNER [4], or standardisation bodies[5] rely on scenario classifications like macro and micro cell, indoor and outdoor. Furthermore, based on system-deployment schemes, such scenarios are grouped into e.g. wide area, metropolitan area and local area concepts [4]. The parameter tables for these channel models for simulations on system or link level are based on the analysis of, often dedicated, radio channel sounding data. While the parameter tables are usually freely accessible, the measured channels are not. Recently, within the COST project various reference scenario sets have been presented and are made accessible via the web site [1]. However, what is lacking is a consistent view on the chain from channel sounding to processing for various applications, e.g. parameters for channel models or simulations including access to measurement data. Under this premise the presented work flow and measurement data in this paper can be understood as reference data for various application fields. In particular the used measurement configuration targets many system aspects that are of interest for the current standardization process, e.g. configurations as multiple base station and multiple users, relaying as well as frequency and bandwidth. Furthermore the presented data offers huge potential for scientific research, because of the considered system setup and quality of the data. In this paper, we present an extensive multi-user and multibase station MIMO measurement campaign in Section II. Furthermore in this section we discuss the outcome of such a campaign, including the data validation and notes how to share parts of the data. In Section III various fields of usage of realistic channel sounding data are discussed. First results based on the proposed data and their analysis is shown in Section IV. The paper ends with the summary in Section V. II. MULTI-USER MULTI-BASE STATION MIMO CHANNEL MEASUREMENT CAMPAIGN A. Scenario and Measurement System Our measurement campaign focused on gathering realistic channel data in an urban macro cell environment in the 3GPP Long Term Evolution band (LTE [5]). In Table 1 and Table 2, the measurement and antenna setups are highlighted. Channel sounding is performed at 2.53 GHz in a band of 2x4 MHz. Figure 1 Radio environment; BS2@25m, and reference points: 21 (upper right) and 13a (lower right).
Measurement Campaign Scenario Location Urban macro cell City centre, Ilmenau, Germany 3 Base stations, 1 relay Measurement setup station, 22 tracks, full MIMO Channel Sounder Type Transmit power @ PA Centre frequency Bandwidth CIR length CIR sampling Snapshot rate MIMO sub-channels AGC switching Positioning [dbm] [GHz] [MHz] [ μs] [samples] [Hz] [#] RUSK TUI-FAU Medav, GmbH 46 2.53 2x4 MHz 6.4 641 >75 944 (16 58 eff.) in MIMO sub-channels Odometer and GPS Table 1 Scenario and measurement setup To allow high resolution path parameter estimations, dedicated antenna arrays at Tx and Rx side are used. On the Tx side (base station), an uniform linear array is used with 8 dual- of polarised (H/V) elements, each of which consists of a stack 4 patches in orderr to form a narrow transmit beam in elevation. At the mobile (passenger car), a circular array with 2 rings of 12 patches with H/V polarisations is used. Additionally, a MIMO cube is placed on top, see Figure 3. The mobile acts as Rx. In Figure 2, a map of the city centre of Ilmenau is shown including the position of the base stations, the relay station, and the routes. For each of the tracks and for each measured snapshot, geo-dataa information based on GPS, odometer and separated distance measurements via laser are available. The accuracy around the start and end points for each track is approximately. 1 m and along the route 1m. During the measurement the position in height is only traced via GPS with low accuracy, therefore an interpolation is necessary and decreases the accuracy in the cases where the height along the track is changing. The information about the building/street structure and the landscape profile (height) has been acquired from the land registry office and 3D laser scans, respectively. In total the measurement campaign covers 3 base station Tx Array Rx Array SPUCPA 2 12 Name (type) PULPA8 + MIMO-Cube Height [m] 25, 15 and 3.5 1.9 Beamwidth (3dB) : azimuth [ ] 1 Omni elevation [ ] 24 8 Tilt (down) [deg.] 5 Mobility [m/s] 3-5 Table 2 Antenna setup positions with 25m and 15m height and additionally a relay point in the middle of the scenario. The intersite distance between the base stations is found to be for BS1-BS2 = 68m, BS2-BS3 = 58m and BS3-BS1=64m. More than 2 individual tracks with more than 12 measurement runs were performed. B. Campaign Outcome The result of the measurement campaign is composed of not only the raw data acquired by the channel sounder itself rather additional information about the measurement system settings including the antenna arrays and accompanying data are needed for a full description, see Figure 4. The latter one N Base station Relay station Meas. track 25m 2 BS2 17 2 41 4a 4 39 3 Figure 2 Sketch of the scenario, routes, and base stations 15 16 BS3 14 22 23 4 13 13b 13a 21 12c 12b 12a 25 11 7 5a 5b 5 24 1b 28 1a 6 8 1 BS1 Rx: SPUCPA 2x12 with MIMO- the Tx: PULPA8, h/v-polarized Cube on top, h/v- polarized Figure 3 Pictures of Tx and Rx antenna arrays we introduce as meta data holding the knowledge about geographical positions, heights and orientations of the used antenna arrays as well as descriptions of the surrounding environment represented by media data. The reliability of measurement raw data is always a crucial point, since a large number of errors might be occurred during the measurements s. At one hand the measurement technique itself can be affected by hardware defects and at the other hand the operator might have forgot to adjust all settings in the
Figure 4 Schematic outcome of the reference campaign correct way both situations will lead to more or less useless measurement data. Additionally external influences like disturbing wireless systems in the same frequency range can be responsible for useless measurement data. That's why a so called quality check validating the raw data stored by the channel sounder should always be performed. Dealing with a huge amount of data and for better accessing them, there's a need for a rough characterization of the raw data for further processing. For that simple non-directional analysis items have been introduced called basic data. Since the measurements have been performed with customized antenna arrays an alternative representation of the mobile radio channel by RIMAX based parametric data sets [8][9][1] is given with a default analysis step. These parametric data contains an antenna independent set of parameters for the -125-13 -135 Ilmenau City Environment Channel Sounder measurement settings measurement arrays Raw Data channel matrices out of the box Raw Data channel matrices approved by quality check Basic Data analysis items Default Analysis Meta Data geographical data media data Quality Check Parameter Estimation application of RIMAX Parametric Data specular components DMC components measurement remainder specular and dense multipath components. With that additional feature the flexibility for the usage of the reference campaign is significant increased due to the chance of reconstructing the wave field (and therefore the channel matrices) for arbitrary application antenna arrays at both link ends. C. Data Validation An important and unfortunately often neglected part of a measurement campaign is the validation of the obtained data. The measurement equipment may be affected by errors that did not occur during the calibration of the system. This means although the system functions correctly at the beginning of the measurement one should not assume that this is the case throughout the whole campaign. In general the validation of measurement data is a challenging task. This is due to the complexity of the measurement system, errors can occur either by operating errors or malfunctioning of the hardware. Therefore the data should be validated on measurement side by on-line processing and later more intensive by post-processing procedures. Since a MIMO channel sounder examines the structure of the channel both in the time and the spatial domain a proper data validation scheme needs to incorporate those domains as well. Therefore, the method presented in this section uses the high resolution parameter estimation framework RIMAX. RIMAX uses a data model that describes the observation of a MIMO channel as the superposition of specular propagation paths (SC) and dense multipath components (DMC). The presence of DMC can be explained by diffuse scattering or otherwise unresolvable specular paths. (1) For the validation of measurements the physical meaning of DMC is of lower importance, it is in fact used as a method to improve the estimation of the SC. The parameter vector contains the parameters direction of arrival (DoA) and departure (DoD) as well as time-delay of each specular path. The contribution of all paths can be calculated using the following equation: T φ T,, T, R φ R,, R, (2) Power [db] -14-145 ca. 8dB with being the complex path weight and the vector of complex exponentials in the frequency domain that are induced by the time-delay of the l th path. -15 T (3) -155.2.4.6.8 1 normalized delay Figure 5 PDP of measurement and remainder after subtraction of estimated paths,,, and,,, denote the steering vectors of the receive and transmit arrays respectively for the DoA and DoD of the l th path.
maximum signal-to-remainder-ratio [db] 9 8 7 6 5 4 3 2 1 correct channel order scrambled channel order 5 1 15 2 25 snapshots Figure 6 Developing of the maximum SRR of measurements with correct and scrambled channel order,,,,,,,,,,, T (4),,,,,,,,,,, T (5) For simplicity this equation applies only for the nonpolarimetric case, however all of the following conclusions are also valid in the case of full-polarimetric array beampattern and path weights. It is intuitive that the data model of RIMAX heavily depends on the order of the array elements of the Rx and Tx respectively. For the measurement this order is defined by the multiplexer switching sequence at the Rx and Tx. Therefore, it is mandatory that the same order is used for parameter estimation that was used during the measurement. In case of multiplexer malfunctioning it cannot be guaranteed that both channel orderings are the same. A typical failure of the multiplexers is that they are not switching at all because of hardware error or that they are switching in a different order because the wrong multiplexing table was loaded into the channel sounder software. A further cause of mismatching channel ordering is the loss of the MIMO-synchronisation. This synchronisation ensures that both Rx and Tx know which array element is currently enabled on either side. This is necessary in order to make sure that the transmitter is only switching to the next element after the receiver switched all of its elements. If for any reason the channel order is scrambled, the parameters estimator will not be able to determine the path parameters correctly. Since RIMAX is a maximum-likelihood estimator, it will still estimate some specular paths, but when they are subtracted from the measured data they will not significantly reduce the power which is to be expected when the multiplexer switching and parameters are correct. In order to decide whether the path parameters are valid or not a suitable metric must be introduced. Based upon experience with visual inspection of estimator results, the maximum ratio of the power-delay-profile (PDP) of the measured channel impulse response and the remainder after subtraction of the specular paths is used (maximum signal-to-remainder-ratio SRR). The relevant PDPs are thereby calculated as the mean PDPs for all channels. High resolution parameter estimation is in general a computational complex task that requires both a lot of memory and computation time. However, for the purpose of data validation it is not necessary to estimate all propagation paths but only a few to demonstrate that the estimation procedure will work. Therefore, RIMAX is configured to estimate a reduced number of paths and to use limited bandwidth. Depending on the propagation environment the estimation of 5 to 1 specular paths seems to be sufficient, as long as individual paths exist that are distinct in all parameter dimensions. To further speed up the estimation RIMAX is only used to estimate the paths for the first 1 snapshot of a measurement file, after that an extended Kalman filter is used to track the parameters of the paths [15]. Figure 5 depicts an example of the PDP of the measured CIR of a snapshot and the PDP of the remainder after the subtraction of the estimated paths. In this case a total number of 5 paths were estimated. Even with this low number of paths it is possible to subtract approximately 8dB of power at a normalized delay of ca..2. This is a strong indication that the multiplexing sequence is correct at both the Rx and Tx. Figure 6 illustrates the development of the maximum SRR for a part of a measurement route. Here, two cases are distinguished. The first is the SRR using the correct channel order (blue line). It can be seen that it is always in the range of 6dB to 8dB, indicating correct multiplexing. The second case uses a scrambled channel order (red line), more precisely the channel order does not start with the first element of the transmitter and the transmitter switched in the middle of a receiver switching sequence. This results in the SRR not exceeding 2dB. Therefore, it can be concluded that a certain value of the SRR can be used to decide whether the measured data is valid or not. During the validation of all measurement files of the underlying measurement campaign in this paper a value of 3dB turns out to be a proper threshold. Finally, it can be said that the use of high resolution parameter estimation with proper limitation of the number of estimated paths - is a promising approach to validate measurement data before it is used for further processing. D. Sharing the Campaign Parts of the above described measurement campaign will be accessible for any researcher and institution. This will include the above described outcomes, including raw and meta data, as well as basic and parametric data from RIMAX. All data sets are collected and processed with high quality by considering current standards and available tools. Nevertheless no warranty can be given for improbable but possible mistakes or ambiguities, because of the experimental character of a channel sounding measurement campaign.
The aim in sharing the campaign to many researchers is to offer the possibility to apply realistic data with high quality and high resolution multipath estimation results to their research work in channel modelling and in particular in system and algorithm design evaluation. Based on this already in an early stage of the research the evaluation can be conducted under realistic conditions. This is important especially for advanced mobile communication concepts considering multiple access of multiple users to multiple base stations equipped with multiple antenna elements. Under these premises the spatial-temporal channel characteristic plays a significantly role, hence realistic data has to be considered for reliable system design and evaluation. Furthermore within the COST framework this measurement data will be placed as a reference scenario [2]. Based on the extensive measurement campaign 9 files describing the links from 3 base stations at the same height of 25m to 3 tracks in the city of Ilmenau are selected. The complete data set for the 9 files as described above is publicly available by accepting the Terms of Use at [16], whereby the raw data import functions from the RUSK channel sounder [17] is free accessible by a license agreement. 1b 1b BS1 BS2 BS3 1b 55dB 45dB MT - 2 4 MT 1b- 2 4 6 35dB MT - 2 4 distance [m] Figure 7 Dynamic along 9 measurement tracks 6 5 4 3 6 5 4 3 6 5 4 3 dynamic [db] dynamic [db] dynamic [db] III. DATA FOR CHANNEL MODELLING AND SYSTEM EVALUATION Within a consistent channel sounding work-flow various application fields for the measurement data can be identified. Those applications are basically grouped into analysis and extraction of individual channel parameters, into channel modelling and simulations for performance studies of various algorithms or complete systems. Furthermore the applied methods in these groups depend on the underlying data source: on the one hand the measurement data Figure 4 and on the other hand the results of the high resolution parameter estimation [8][9][1]. The later offers, compared to the pure measurement data, an antenna independent usage of the data, in particular access to the spatial domain of the multipath channel. However this method is only available if the specific antennas are used during the measurement. Studies for channel and propagation analysis will guide to fundamental understanding of the mobile multipath channel. Well-known characterization as described in [13] can be applied for any type of antenna structures on both sides. Whereby receive power, delay spread and geometry factor are only few important examples. In the case of MIMO and/or polarimetric antenna elements the characterization should be extended to MIMO capacity and eigenvalue analysis. The aforementioned studies consider the direct use of the measurement data. By performing measurements with specific antenna arrays, high resolution path parameter estimation can be applied. This allows further studies of directional parameters as DoA and/or DOD and clustering of multipath parameters in various dimensions. The latter step offers a high potential for a closed description and characterisation in the spatio-temporal domain. Reliable Channel Models are essential for network planning and link design. For the design of Next Generation Wireless Networks specified in WINNER or 3GPP LTE the SCM [5], SCME [6] and more advanced WINNER channel models were defined [4][7]. Using the measured and processed impulse responses a statistical analysis of parameters corresponding to the requirements, e.g. model structure, small and large scale parameters can be carried out. This leads to a more generalized approach to the measured channel data without neglecting the real scenario. Furthermore an extension of these models towards cooperative MIMO schemes and the influence of the dense multipath components (DMC) is possible. Another application for the use of channel sounding based measurement data is the calibration and tuning of existing propagation simulation tools. On the one hand one can tune path loss and fading characteristics and on the other hand it is possible to calibrate the results of deterministic ray tracing or ~launching simulators for the actual scenario.
Link and system level simulations are a basic part in the development and evaluation of new algorithms and transmission strategies. Those simulations can be performed based on direct use of measurement data, on channel models and channel reconstruction. In all cases they will benefit from the use of real world data [11]. This is in particular important if the system under test aims to exploit the spatial domain (MIMO configurations). The implementation of the raw data as part of the channel module in the software or hardware channel emulator is mapping the properties of the realistic measurement drive on the simulation machine. It acts like a drive or device test in the box, where even complete hardware demonstrators can be tested. One can compare different algorithms or scheme for the same channel realisations under the constraints of measurement antennas and side specific tracks. Furthermore it is possible to reconstruct the pattern of the electromagnetic waves impinging the measurement antenna. Based on the estimation results of the RIMAX algorithm [9][1] the behaviour of other antennas can be investigated under realistic conditions. With this antenna embedding new antenna designs can be evaluated e.g. in terms of spatial multiplexing capacity. IV. RESULTS ON DATA ANALYSIS The following results are derived from the 9 raw data files, which will be publicly available. For these files the dynamic range along the measurement track is shown in Figure 7. On the left side the coverage of each base station to the 3 tracks is highlighted separately (orientation of base stations is depicted by colored arrows). Each base station can cover all 3 tracks, whereby the dynamic (maximum peak to off-peak ration in CIR) is typically larger than 3dB resulting in a good raw data quality for further processing. On the right side in Figure 7 the dynamic ranges are shown from the perspective of the mobile P(Delay Spread < Abscissa) 1.9.8.7.6.5.4.3.2 1%: 13.1 5%: 31.9 9%: 92.3 1%: 24.2 5%: 44.5 9%: 7.7 1%: 7.2 5%:.7 9%:1.1 1%: 23.8 5%: 49.1 9%:114.4 1%: 38.6 5%: 61.7 9%: 88.9 1%: 3.1 5%: 71.2 9%:144. BS1-1-MT1b-, full DR BS1-1-MT1b-, 2 db BS2-1-MT1b-, full DR BS2-1-MT1b-, 2 db.1 BS3-1-MT1b-, full DR BS3-1-MT1b-, 2 db 5 1 15 2 25 3 Delay Spread, σ [ns] τ Figure 9 CDFs of per-snapshot delay spread (DS), calculated for same route of the mobile terminal toward different base stations. terminal to the 3 base stations. Different colors identify the base stations in the same manner as in the left column of the figure. In order to acquire better observation of the channel, the RUSK TUI-FAU sounder uses dynamic AGC, meaning that each individual measurement channel, defined as a combination of the one Tx and one Rx antenna element in antenna-switched concept, has independent gain control. This feature requires proper interpretation during the channel characterization: in Figure 8 it is showed that cross-polar losses are partially compensated and therefore the observed dynamic ranges are similar for co-polarized (VV and HH) and cross-polarized combinations of Tx and Rx antenna elements. The measurement data are taken sequentially over the same routes for the mobile terminal toward different base stations. P(Dynamic Range < Abscissa) 1.9.8.7.6.5.4.3.2.1 1%: 8.8 5%: 16.1 9%: 29. 1%: 1.5 5%: 18. 9%: 28.8 1%: 15.9 5%: 25.9 9%: 37.7 1%: 12.1 5%: 22.4 9%: 36.4 VV VH HV HH P(Delay Spread, Max. Excess Delay < Abscissa) 1.9.8.7.6.5.4.3.2.1 1%: 15.9 5%: 5.3 9%: 94.7 1%:11. 5%:29. 9%:49. 1%: 32.1 5%: 66.5 9%:11. 1%:22. 5%:. 9%:97. Delay Spread, full DR Max. Excess Delay, 2 db Delay Spread, full DR Max. Excess Delay, 2 db 1 2 3 4 5 6 Dynamic Range, DR [db] Figure 8 CDFs of per-snapshot dynamic ranges, for different combinations of Tx and Rx antenna polarizations. 2 4 6 8 1 12 14 16 18 2 Delay, τ [ns] Figure 1 CDFs of Delay Spread and Max. Excess Delay for 9 selected measurement routes.
Figure 9 shows Delay-Spread (DS) distribution over the route 1b- for 3 used base stations. As noted in [14] parameters determined from Power-Delay-Profile (PDP) show significant dependence on available Dynamic-Range (DR). This is also illustrated on Figure 9: lower DS values are obtained if DR of PDP is limited to 2 db. The extraction procedure and the underlying processing assumptions have significant influence on the derived results. Hence careful processing of the data should be performed to acquire results matching the goals for modeling, analysis or simulation. When all values of DS and Maximum-Excess-Delay (MED, from every measured route of the mobile terminal toward different base stations) are used to generate probability distribution (Figure 1), we get the stochastic equivalent of measured radio-environment (scenario). When DR was set to 2 db, DS values corresponding to different space-time samples become comparable. This complies with WINNERmodeling approach, and therefore obtained results should be comparable with previously reported parameters for C2 scenario (typical urban macro cell) [4]. Since WINNER model provides median DS parameters separately for LoS (41 ns) and NLoS (234 ns), we can confirm the matching with the computed median of 5.3 ns under the assumption that selected routes mainly exhibit the LoS condition. Calculated values show that in 99% DS is lower than 166 (152) ns, and MED is lower than 195 (87) ns, depending weather full or limited (to 2 db) DR is used. Furthermore high resolution multipath parameter estimation based on RIMAX [9] [1] has been applied. Preliminary results are shown for the 9 measurement files along the tracks. Because of the computational complexity unfortunately not all snapshots in each of the files could be performed in time. Figure 11 shows the DoA spreads at the mobile. On the left hand side the results are shown in the map along the tracks. Each map indicates the results form base stations view jointly for 3 selected tracks. It is interestingly to note, that one particular base station position (BS2) offers for this 3 measurement tracks always very high azimuth spreads. Not shown here, but it can be expected that the MIMO capabilities for these links will be very good. On the right side of Figure 11 the results are shown jointly for the 3 base stations seen from the mobile along the individual tracks. Here the inter-site channel situation for the mobile routes can be studied. In general the characteristics of the azimuth spreads at the mobile are changing quite frequently, which indicates changing situation within the propagation channel in terms of the multipath scattering and richness. Furthermore in the lower map for the track MT - a considerable correlation between the developing of the azimuth spreads for the links from BS1 and B2 is found. Further studies are required here. Besides the azimuth characteristics at the mobile also the elevation is of interest. These results are highlighted in Figure 12. For the track MT - lower elevation spreads with not to many changes in the developing compared to the other two 1b MT - 2 4 1 5 azimuth spread @Rx [deg] 1b MT - 2 4 4 2 elevation spread @Rx [deg] 1b BS1 BS2 BS3 8 6 4 2 MT 1b- 1 2 3 1 5 azimuth spread @Rx [deg] 1b BS1 BS2 BS3 45 3 15 MT 1b- 1 2 3 4 2 elevation spread @Rx [deg] 1b Figure 11 DoA azimuth spreads at the mobile along the measurement tracks MT - 2 4 distance [m] 1 5 azimuth spread @Rx [deg] 1b Figure 12 DoA elevation spread at mobile along the measurement tracks MT - 4 2 1 2 3 distance [m] elevation spread @Rx [deg]
tracks are found. In general the elevation spreads seems to oscillate between 2-3, but also easily reaches more than 45. This underlines the importance especially in MIMO applications of the elevation in characterising and modelling of mobile channels. V. SUMMARY In the baseline of the paper an extensive channel sounding campaign has been presented. The MIMO measurements were focused on configurations for multi user and multi base station applications in an urban macro cell scenario. The frequency range covers the deployment for LTE. Furthermore a sketch of the channel sounding work-flow is discussed including an approach for data validation. The data can be used as base for various applications, e.g. detailed channel analysis and channel modelling as well as on soft- and hardware simulators for linkand system-level evaluations of advanced multi-antenna concepts. First measurement campaign outcomes are shown. The results indicated a very good measurement data quality in terms of dynamic range coverage. A delay spread and maximum excess delay analysis support the findings from the WINNER project for the C2 scenario. High resolution path parameter estimation are applied and offered interesting insight to the spatial characteristics of the gathered measurement data. The measurement campaign will be introduced as reference scenario within the COST 21 frame work. Parts of the measurement data including raw data, meta data and estimated multipath parameters will be publicly available. [9] M. Landmann, W. Kotterman, R.S. Thomä, Estimated Angular Distributions in Channel Characterisation, in EUCAP 27, Edinburgh, UK. [1] Richter A., On the Estimation of Radio Channel Parameters: Models and Algorithms (RIMAX), Ph.D. dissertation, TU-Ilmenau, Ilmenau, Germany, May 25. [11] U. Trautwein, C. Schneider, R.S. Thomä, Measurement Based Performance Evaluation of Advanced MIMO Transceiver Designs, EURASIP Journal on Applied Signal Processing 25, No.11, pp.1712-1724. [12] C. Schneider, U. Trautwein, W. Wirnitzer, R.S. Thomä; Performance Verification of MIMO Concepts using Multi-Dimensional Channel Sounding, European Signal Processing Conference, EUSIPCO 26, Florence, Italy, September 26. [13] T. S. Rappaport, Wireless Communications: Principles and Practice, 2nd Edition, Prentice Hall, 22. [14] M. Narandžić, M. Landmann, C. Schneider, and R. S. Thomä: "Influence of Extraction Procedures on Estimated Wideband MIMO Channel Parameters", Proc. of IST Mobile & Wireless Communication Summit, Budapest, Hungary, July 1-5, 27. [15] Salmi, J.; Richter, A.; Enescu, M.; Vainikainen, P. & Koivunen, V. Propagation Parameter Tracking using Variable State Dimension Kalman Filter Proc. VTC 26-Spring Vehicular Technology Conference IEEE 63rd, 26, 6, 2757-2761 [16] www-emt.tu-ilmenau.de [17] www.channelsounder.de VI. ACKNOWLEDGEMENT This work was supported in part by the research excellence cluster UMIC at RWTH Aachen. The authors would like to thank Volker Jungnickel from FhG-HHI, Berlin, for the support with the antenna arrays. Furthermore Carsten Jandura from Vodafone AG for the support with the measurement car, and Medav GmbH for providing parts of the sounder equipment. REFERENCES [1] L. Correia, Ed., Mobile Broadband Multimedia Networks, Academic Press, 26. [2] http://grow.lx.it.pt/web/cost21/ [3] P. Almers, E. Bonek, A. Burr, N. Czink, M. Debbah, V. Degli-Esposti,H. Hofstetter, P. Kyosti, D.Laurenson, G. Matz, A. Molisch, C. Oestges,and H. Ozcelik, Survey of channel and radio propagation models for wireless MIMO systems, in EURASIP Journal on Wireless Communications and Networking, 27. [4] https://www.ist-winner.org [5] http://www.3gpp.org/ [6] D. S. Baum, J. Salo, G. Del Galdo, M. Milojevic, P. Kyösti, and J. Hansen, An interim channel model for beyond-3g systems, in Proc. IEEE VTC 5, Stockholm, Sweden, May 25. [7] Narandzic,Milan, Schneider,Christian, Thomä,Reiner S, Jämsä,T., Kyösti,P., Zhao,X., Comparison of SCM, SCME, and WINNER Channel Models, IEEE VTC27-Spring, Dublin, Ireland, April 27. [8] R.S. Thoma, M. Landmann, A. Richter, et. al., Multidimensional High- Resolution Channel Sounding, in T. Kaiser et. al. (Ed.), Smart Antennas in Europe - State-of-the-Art, EURASIP Book Series on SP&C, Hindawi Publishing Corporation, Vol. 3.