Channel Measurements for Evaluating Massive MIMO Precoding

Transcription

1 Channel Measurements for Evaluating Massive MIMO Precoding Stephan ten Brink 1 1 Institute of Telecommunications University of Stuttgart Workshop on Smart Antennas Technical University of Berlin Institute of Telecommunications Prof. Dr. Ing. Stephan ten Brink

2 Contributors MIMO research at the Institute of Telecommunications Xiaojie Wang (MIMO and 5G techniques) Felix Fellhauer (WLAN IEEE ay: channel model, beam training) Maximilian Arnold, Marc Gauger (massive MIMO, measurements) plus students as acknowledged in respective webdemos Stephan ten Brink MIMO Channels /59

3 Webdemos... presentation uses webdemos that were created over the past 3 years at the Institute of Telecommunications demos can be accessed worldwide for teaching, textbook-style examples of, e.g., MIMO capacity for presentation of student theses, Sphere Decoder for illustrating ongoing research, e.g., Massive MIMO Stephan ten Brink MIMO Channels /59

4 Outline 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

5 Agenda 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

6 Why are measurements important? Birth of a new system, from a PHY layer perspective... spectrum becomes available channel measurements channel model is agreed upon modulation and coding is agreed upon standard ratified Stephan ten Brink MIMO Channels /59

7 Main objective of channel models channel models should guide communication engineers to build good systems how detailed do we have to describe the real world to arrive at engineering solutions for practical systems? to arrive at performance estimates (capacity) to guide algorithm design (implementation) Stephan ten Brink MIMO Channels /59

8 MIMO Timeline, 1995-today v99r U;;; today Pprefspatial)multiplexP:) beamformingg)phased)arraysg)radar;) vector)detection)in)mufcdma:crossftalk MIMOasphereadetection massivedmimo bmarzettaw massive6mimo6again pmarzettaw Theory algorithms:detection spatial)multiplex)bblastkg) PUP)MIMO)capacity)bFoschiniG)Telatark Vf:Df:HfBLAST) spatial)multiplex MIMO)based)on)succ") with)app)soft)detection cancel")detection optical)mimo) spaceftime)block)codes) bmultimode)fibersk be"g")alamoutik softasphereadetection iterativeamimoadetectiona P2PaMIMOacapacityaachievedausinga andadecoding APPadetectionaandatailoredaLDPCa pushaforaclosedaloopatechniquesa bchannelafeedbackq P2PaEigenmodeaa diversityavsamultiplexinga beamforming tradeaoff dirtydpaperdcoding bbroadcastw P2MPdbMUw-MIMO indcellulardcontext networkdmimo,dcompd bcellularw interferencedalignment bcellularw mmwave6mimo optical6mimo6 pmulticore6fibersw HhybridH6MIMO massive6mimo6with6 HW6impairments massive6mimo6 Cloud/Fog6radio6access with6low/res6dac/adc Channel) Models LOSG)phased)arrays bbeamformingk i"i"d")complex)gaussian) bergodic:blockffadingk TX:RX)antenna) correlations keyhole)effects WLANa802)11na 4x4aMIMOamodel detaileddaoa,daod LTE measurements MIMOdmodel LTE MIMOdmodel massive6mimo6measurements pratesc6pairwise6orthogonalityw 60GHzC6wideband 802F11ad66model6 p60ghzc6feedbackw 802F11ay66model6 pray6tracing/basedw a (very incomplete and) subjective view: some milestones in MIMO theory/algorithms channel modeling Stephan ten Brink MIMO Channels /59

9 What did I do in 1996?... MIMO CDMA (no MIMO yet...) it took me 5 more years to start working on MIMO... Stephan ten Brink MIMO Channels /59...

10 MIMO Timeline, v99r U;;; Pprefspatial)multiplexP:) beamformingg)phased)arraysg)radar;) vector)detection)in)mufcdma:crossftalk Theory algorithms:detection spatial)multiplex)bblastkg) PUP)MIMO)capacity)bFoschiniG)Telatark Vf:Df:HfBLAST) MIMO)based)on)succ") cancel")detection spaceftime)block)codes) be"g")alamoutik optical)mimo) bmultimode)fibersk spatial)multiplex with)app)soft)detection Channel) Models LOSG)phased)arrays bbeamformingk i"i"d")complex)gaussian) bergodic:blockffadingk TX:RX)antenna) correlations keyhole)effects Stephan ten Brink MIMO Channels /59

11 MIMO Timeline, MIMOasphereadetection softasphereadetection iterativeamimoadetectiona andadecoding pushaforaclosedaloopatechniquesa bchannelafeedbackq diversityavsamultiplexinga tradeaoff P2PaMIMOacapacityaachievedausinga APPadetectionaandatailoredaLDPCa P2PaEigenmodeaa beamforming WLANa802)11na 4x4aMIMOamodel Stephan ten Brink MIMO Channels /59

12 MIMO Timeline, massivedmimo bmarzettaw P2MPdbMUw-MIMO indcellulardcontext networkdmimo,dcompd bcellularw dirtydpaperdcoding bbroadcastw interferencedalignment bcellularw detaileddaoa,daod LTE measurements MIMOdmodel LTE MIMOdmodel Stephan ten Brink MIMO Channels /59

13 Intro - MIMO Timeline, 2010-today 2010 today massive6mimo6again pmarzettaw HhybridH6MIMO mmwave6mimo massive6mimo6with6 HW6impairments optical6mimo6 pmulticore6fibersw Cloud/Fog6radio6access massive6mimo6 with6low/res6dac/adc massive6mimo6measurements pratesc6pairwise6orthogonalityw 60GHzC6wideband 802F11ad66model6 p60ghzc6feedbackw 802F11ay66model6 pray6tracing/basedw Stephan ten Brink MIMO Channels /59

14 Agenda 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

15 Webdemo webdemo theses/mimo_capacity/index.php?id=0 Stephan ten Brink MIMO Channels /59

16 MIMO Model Stephan ten Brink MIMO Channels /59

17 Spatial Correlations (more detailed: Weichselberger-model, considers correlations on channel) Stephan ten Brink MIMO Channels /59

18 Example: Kronecker Corr. Model in n WLAN IEEE802.11n; antenna spacing λ/4, typ. office webdemo theses/80211n_channel_model/index.php?id=0 Stephan ten Brink MIMO Channels /59

19 Example: Kronecker Corr. Model in n WLAN IEEE802.11n; antenna spacing λ, typ. office webdemo theses/80211n_channel_model/index.php?id=0 Stephan ten Brink MIMO Channels /59

20 Capacity i.i.d. complex Gaussian matrices H... Stephan ten Brink MIMO Channels /59

21 Numerical Example, Ergodic linear growth with nb of antennas; correlations reduce capacity Stephan ten Brink MIMO Channels /59

22 Numerical Example, Ergodic 4x4, 8x8 channel with/without correlations Stephan ten Brink MIMO Channels /59

23 MIMO Detectors, BER performance of linear (ZF, MMSE...), non-lin. (ML, APP) can be studied with this simple channel model webdemo theses/mimo_sphere/index.php?id=10 Stephan ten Brink MIMO Channels /59

24 MIMO Detectors, Mutual Information can get very close to capacity with soft output detection webdemo theses/mimo_sphere/index.php?id=10 Stephan ten Brink MIMO Channels /59

25 Example: Sphere Detector 4x4 QPSK generating a candidate list to derive soft information webdemo theses/mimo_sphere/index.php?id=4 Stephan ten Brink MIMO Channels /59

26 Outcome for Open Loop simple i.i.d. complex Gaussian/Kronecker correlation models fine for studying MIMO detection performance for designing tracking algorithms etc. combined with time-/frequency-fading models but still some shortcomings how to account for feedback (beyond waterfilling)? does not help with, e.g., studying beam training algorithms in 60GHz mmwave WLAN systems... Stephan ten Brink MIMO Channels /59

27 Agenda 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

28 WLAN ad, 1x1 WLAN ad, 60GHz, 1x1 up to ca. 5Gb/s one spatial stream (beamforming) based on phased array channel model extends n model to 11ad includes geometric scenarios, antenna model spatial dependence of channel impulse response according to a geometric set-up distributions of AoA, AoD, ToA generated based on specific scenario closed loop: feedback based on received signal strength/codebook indices to steer beams Stephan ten Brink MIMO Channels /59

29 WLAN ay, 4x4 Hybrid MIMO WLAN ay, 60GHz, 4x4 up to 28Gb/s, hybrid MIMO: 4 spatial streams, phased arrays spatial distributions generated based on integrated ray-tracer for each channel realization (LOS, 1st and 2nd order reflections) scenario defined by room geometry and user drop area analog front-end (with phased arrays) becomes part of channel Stephan ten Brink MIMO Channels /59

30 Antenna Models, IEEE ad, ay specified in channel model ad: simplified Gaussian beams (red) ay: actual phased array beams (green) Stephan ten Brink MIMO Channels /59

31 Different Geometric Scenarios geometric 3D-models: conference room, cubicle, living room, plus in.11ay: open area, street canyon, hotel lobby STA: station on table; AP: access point at ceiling Stephan ten Brink MIMO Channels /59

32 IEEE ad, 1x1 5-dim. tapped delay line model h(τ,ϕ T,ϑ T,ϕ R,ϑ R ) dependent on steering angles at TX, RX distributions of AoD, AoA, and ToA modeled by ray-tracing performed once, off-line, to get distributions for each scenario thus, geometry accounted for by specific PDFs Stephan ten Brink MIMO Channels /59

33 IEEE ay, 4x4 Beam Training ray-tracer included in channel model; runs for each channel realization and user drop position objective: beam training optimization to maximize rate 2x2 MIMO; discrete codebooks (7 entries per antenna; 7 different beam directions per sub-array; x-axis TX combination, y-axis RX combination; sparse... Stephan ten Brink MIMO Channels /59

34 Outcome channel models get more and more involved to allow algorithm design (beam training) not only distributions, but actual physical geometry parts of TX/RX (analog front-end, phased array antenna model) are included in channel models massive MIMO, precoding channel models may need to be even more detailed for performance analysis and algorithm design Stephan ten Brink MIMO Channels /59

35 Agenda 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

36 Index-based vs full CSI Feedback index-based/coarse feedback (e.g. MU-MIMO, LTE) well defined array geometry needed codebook-based (only a table index is fed back) pre-defined patterns, mostly based on "LOS"-like beamforming low feedback-rate, ok in FDD full CSI feedback (prerequisite for true massive MIMO) no well-defined array geometry needed phase, or phase/amplitude feedback possible in TDD exploiting reciprocity Stephan ten Brink MIMO Channels /59

37 Principle of Massive MIMO base station with N antenna elements channel matrix user 1 K user positions user k large-scale antenna array systems (LSAS/massive MIMO) massive number of N antennas at base station (50s s) few antennas (1-4) at K terminals Stephan ten Brink MIMO Channels /59

38 Principle of Massive MIMO Simplified DL model can be written as y K 1 = SNR (H H) K N s N 1 + n K 1 with H = [h 1,N,...,h K,N ] C N K closed-loop scheme base station learns matrix channel H in UL (TDD assumed) applies (e.g. MMSE-)precoding matrix in DL widely used simulation model for channel matrix H i.i.d. complex Gaussian Stephan ten Brink MIMO Channels /59

39 Precoding webdemo research/mmimo/index.php?id=2 Stephan ten Brink MIMO Channels /59

40 Precoding (MRC/MF) different pre-coding schemes Stephan ten Brink MIMO Channels /59

41 Precoding (ZF) different pre-coding schemes Stephan ten Brink MIMO Channels /59

42 Precoding (MRC/MF), LOS non-scattering (i.e. LOS) channel, 64-ant lin. array effect of precoding reduces to classic beamforming Stephan ten Brink MIMO Channels /59

43 Precoding (ZF), LOS influence of ZF precoding illustrated TX power normalized Stephan ten Brink MIMO Channels /59

44 Precoding (MRC/MF), NLOS same precoding, but now rich-scattering environment (NLOS) hot spots (challenge for channel modeling... clusters) Stephan ten Brink MIMO Channels /59

45 Agenda 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

46 Evaluating Two-User Orthogonality spatial correlation coefficient δ i, j at any two user positions i, j h δ i, j = i h j 2 h i 2 h j 2 vector h i contains N = 64 complex channel weights from N base station antennas to single receive antenna at terminal position i a cross-correlation of δ i, j = 0 means positions i, j are orthogonal to each other base station could address terminals i, j with no cross-talk at the same time, on the same frequency resource Good two-user orthogonality is prerequisite for multi-user orthogonality as desired for efficient massive MIMO-operation Stephan ten Brink MIMO Channels /59

47 Intro Open Loop 60GHz WLAN MIMO Closed Loop Measurements Spider Antenna Summary Antenna Array at Base Station Scalable N = 64-element antenna array (LTE-signal format, 2.59GHz, 20MHz BW) driven by a proprietary FPGA-baseband prototype (Nokia Bell Labs) connected to remote radio head, 20dBm per antenna element 8 parallel RF-lines switched through entire array in less than 50ms Stephan ten Brink MIMO Channels /59

48 Outdoor Measurement Area unwrapped baseband phase of different positions in [rad] in m antenna array position on rooftop facing South S in m Phase measurements for various positions in the field (horizontal array) to verify set-up Stephan ten Brink MIMO Channels /59

49 Two-User Orthogonality Measurements, Outdoor 10 0 Outdoorpminp6dBpSNR i.i.d. horizontalpoutdoor planarpoutdoor verticalpoutdoor Averagepcorrelationpcoefficientp NumberpofpantennaspN cross-correlation of randomly picked channel vector pairs, planar versus horizontal and vertical horizontal array stays close to idealized i.i.d. Gaussian model Stephan ten Brink MIMO Channels /59

50 Outcome for now, only focused on pairwise orthogonality already performed some sort of area measurements to study, e.g., orthogonality clusters, need more detailed measurements Stephan ten Brink MIMO Channels /59

51 Agenda 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

52 More detailed channel measurements most MIMO measurements thus far detailed study at positions of input/output ports AoA, AoD, etc... estimation energy distribution over antenna angles different approach: over area; why? find orthogonality clusters for scheduling feedback formats (quantization of channel information in FDD) improve channel models etc. Stephan ten Brink MIMO Channels /59

53 Spider Antenna Set-up Spider antenna set-up for 3D channel measurements indoor 2 2 2m 3 ; can be extended to m 3 (outdoor) Stephan ten Brink MIMO Channels /59

54 Indoor, Measurement Set-Up cupboards7(closed) cupboards metal7plate7for NLOS7experiments y TX7antenna m transmission 7direction spider antenna electric7 motor 7to7drive7 antenna RX lab desk z measurement area7for7 spatial sampling x 9.5m area/volume spatial sampling NLOS scenario through metal plate in front of array Stephan ten Brink MIMO Channels /59

55 Verification: x,y-wavefront Images 1x1; (ideal) LOS baseband phase h sim = λ 4πr e j2π r λ sub simulated (left) vs actually measurement baseband phase, horizontal x, y-plane TX/RX on same height; ripples at distance λ Stephan ten Brink MIMO Channels /59

56 Verification: x,z-wavefront Images simulated (left) vs actually measurement baseband phase, vertical x, z-plane Stephan ten Brink MIMO Channels /59

57 16-Antenna Precoding, LOS simulation vs measurement (MRC precoding, target user at (x, y) = (0.1m, 0.2m) received signal at each coordinate y = HPs, MRC-precoding vector P = 1 M H H 16-antenna linear array, LOS scenario Stephan ten Brink MIMO Channels /59

58 16-Antenna Precoding, LOS vs NLOS Comparison of a LOS and a NLOS scenario with MRC precoding on (x, y) = (0.6m, 0.2m), 16-antenna linear array other spatial measures can be derived; just the beginning... Stephan ten Brink MIMO Channels /59

59 Agenda 1 Introduction 2 Open Loop MIMO: Channel Model and Detection 3 Example: 60GHz WLAN Channel Model 4 Closed Loop MIMO: Channel Model and Precoding 5 MIMO Measurements 6 Measurements with a Spider Antenna 7 Summary Stephan ten Brink MIMO Channels /59

60 Summary mmwave, beamforming, massive MIMO... goes quickly beyond classic stochastic channel modeling ray-tracing found its way into current channel models potential for 3D area/volume channel measurements to improve channel models still some surprises to be discovered... Stephan ten Brink MIMO Channels /59

61 Thanks for your attention Stephan ten Brink MIMO Channels /59

62 XXX xxx Stephan ten Brink MIMO Channels /59

63 Backupslide 2 Backupslide 2 Stephan ten Brink MIMO Channels /59