Channel Estimation and Multiple Access in Massive MIMO Systems. Junjie Ma, Chongbin Xu and Li Ping City University of Hong Kong, Hong Kong

Transcription

1 Channel Estimation and Multiple Access in Massive MIMO Systems Junjie Ma, Chongbin Xu and Li Ping City University of Hong Kong, Hong Kong 1

2 Main references Li Ping, Lihai Liu, Keying Wu, and W. K. Leung, "Interleave Division Multiple-Access," IEEE Trans. Wireless Commun., vol. 5, no. 4, pp , Apr Peng Wang, Jun Xiao, and Li Ping, "Comparison of Orthogonal and Non- Orthogonal Approaches to Future Wireless Cellular Systems," IEEE Vehicular Technology Magazine, vol. 1, no. 3, pp. 4-11, Sept Junjie Ma and Li Ping, Data-aided channel estimation in large antenna systems, IEEE Trans Signal Processing, June Chongbin Xu, Peng Wang, Zhonghao Zhang, and Li Ping, "Transmitter design for uplink MIMO systems with antenna correlation," IEEE Trans. Wireless Commun., vol. 14, no. 4, pp , Apr

3 Contents Introduction Channel estimation at the BS Channel estimation at MTs Multiple access: OFDMA, SDMA, IDMA and NOMA Conclusions 3

4 Contents Introduction Channel estimation at the BS Channel estimation at MTs Multiple access: OFDMA, SDMA, IDMA and NOMA Conclusions 4

5 Inspiration from Professor Adachi H. Andoh, M. Sawahashi and F. Adachi, Channel estimation filter using timemultiplexed pilot channel for coherent RAKE combining in DS-CDMA mobile radio, IEICE Transactions on Communications 81 (7), , D. Ishihara, J. Takeda, and F. Adachi, Iterative channel estimation for frequency-domain equalization of DSSS signals, IEICE Trans. Commun., vol. E90-B, no. 5, pp , May G. Gui, W. Peng and F. Adachi, Improved adaptive sparse channel estimation based on the least mean square, IEEE, Wireless Communications and Networking Conference (WCNC),

6 CDMA systems interference among users Professor Adachi is a pioneer in CDMA systems. He made tremendous contributions in the development of 3G CDMA systems in Japan. In principle, there is interference among different users in CDMA. Therefore, CDMA can be also seen as a non-orthogonal multiple access system. F Adachi, M Sawahashi and H Suda Wideband DS-CDMA for next-generation mobile communications systems, IEEE Communications Magazine,

7 Iterative channel estimation Professor Adachi is also a pioneer in iterative channel estimation. He has made inflation contributions using the frequency domain equalization approach. The complexity of his approach is surprisingly low, which provides an attractive option for practice. D. Ishihara, J. Takeda, and F. Adachi, Iterative channel estimation for frequency-domain equalization of DSSS signals, IEICE Trans. Commun., vol. E90-B, no. 5, pp , May

8 SDMA and multi-user gain The current OFDMA system is orthogonal. How about the future evolution path? Orthogonal or non-orthogonal? How to optimize multiple access techniques in massive MIMO environments? The following is from information theory the capacity for a SDMA system: sum-rate ~ min(n BS, K N MT ) log(1+snr), where K is the number of users. This gain can be achieved using multi-user transmission. Peng Wang, and Li Ping, "On maximum eigenmode beamforming and multi-user gain," IEEE Trans. Inform. Theory, vol. 57, no. 7, pp , Jul

9 Impact of CSIT However, the accuracy of CSIT is crucial here. This will be the focus of my talk. 9

10 Assumptions: CSIT for the downlink joint beamforming individual detection MT 1 correlation BS correlation channel MT k correlation MT K correlation For the downlink, decoding is done individually. Accurate CSIT is crucial so as to avoid interference. The system should be orthogonal. 10

11 Assumptions: CSIT for the uplink joint detection individual beamforming MT 1 correlation BS correlation channel MT k correlation MT K correlation For the uplink, accurate CSIT is not crucial. Interference can be suppressed at the BS via joint detection. The system can be non-orthogonal. 11

12 Difference between down-link and up-link Assume TDD. Accurate CSIT is possible for the down-link. Therefore down-link can be orthogonal. Accurate CSIT is difficult for the up-link. Therefore up-link should be nonorthogonal. down-link up-link interference 12

13 Contents Introduction Channel estimation at the BS Channel estimation at MTs Multiple access: OFDMA, SDMA, IDMA and NOMA Conclusions 13

14 Accurate channel estimation is crucial at the BS Accurate CSIT is required for the downlink. With TDD, downlink can be estimated at the BS. Pilot contamination is a serious problem for channel estimation in massive MIMO. Data aided channel estimation provides an efficient solution to the pilot contamination problem. 14

15 Pilot contamination pilot data user 1 user 2 user 3 user 4 interference In a multi-cell system, the pilot symbols from neighboring cells may interference each other, which reduces the accuracy of channel estimation. This effect is refereed to as pilot contamination. 15

16 Pilot contamination and capacity received SNR with accurate CSIT received power with antenna contamination number of BS antennas number of BS antennas The received power is seriously affected by pilot contamination. 16

17 Impact of pilot length Pilot contamination is caused by the correlation among pilots. Its effect reduces with pilot length as: E p p H 2 E p1 p2 1 H In principle, pilot contamination can be mitigated by increasing the pilot sequence length J p. In practice, this is not desirable, because increasing J p will reduce the effective data rate. frame length < channel coherent time J p J p pilot symbols J d data symbols Can we reduce pilot contamination without affecting data rate? 17

18 Data aided channel estimation pilot data user 1 user 2 user 3 user 4 The key of the pilot contamination problem is the correlation among pilots. The data signals have much lower interference (since their length is longer). More accurate results can be obtained by using data for channel estimation. 18

19 Iterative data aided channel estimation decoder ˆd 1 data detector ĥ 1 d y1 d 1 channel estimator y, p p 1 1 pilot symbol data symbols However, data symbols are not known initially. We can use partially detected data for channel estimation based an iterative procedure. Junjie Ma and Li Ping, Data-aided channel estimation in large antenna systems, IEEE Trans Signal Processing, June

20 Data-aided channel estimation Channel estimation phase: H H h ˆ d d h + d d h noise Data detection phase: hˆ y H H H 1 1 j hh H H H 2 hˆ 1 hˆ ˆ ˆ ˆ ˆ 1 h1 h1 h1 h1 hˆ h hˆ ˆ d j d j d j noise self contamination cross contamination 20

21 Cross-contamination Cross contamination is now caused by the correlation among data rather than pilot. hh ˆ hh ˆ ˆ H 1 2 H 1 1 H H d1 d2 h2 h2 H H d1 d1 h1 h1 other terms 0 when N. other terms The cross-contamination effect is inversely proportional to data length J d, and also to the a priori reliability (1-v d. ). J 1v H 2 E d1 d2 1 H 2 E d1 d1 d d. 21

22 Self-contamination When d 1 is not perfectly known, ĥ and ˆ 1 h1 h1 are NOT independent. Such dependency can be modelled as h hˆ 1 hˆ z where z is independent of ĥ 1 H H H H H hˆ 1 hˆ ˆ ˆ 1 h1 h1 and is a random variable. Then hˆ h hˆ hˆ hˆ hˆ z 1,when N Hence self-interference does not vanish when N. We call this effect self-contamination For the pilot based scheme, =1 and no such effect exists. 22

23 Impact of correlation among pilots Cross contamination: ˆ 2 H E hh H 2 E hh ˆ ˆ J d vd 1 Self-contamination: (unique to a data-aided scheme) Both are inversely proportional to J d 2 H E hˆ 1 h1 hˆ 1 vd 2 ˆH E hhˆ J d v β 1 : large scale factor of h 1 ; β 2 : large scale factor of h 2 When v d = 0, self-contamination vanishes while cross contamination converges to a positive constant. Junjie Ma and Li Ping, Data-aided channel estimation in large antenna systems, IEEE Trans Signal Processing, June d

24 SINR performance Consider an L-cell System. Each cell contains one user SINR = 1 2 L H H 2 2 ˆ 1 1 ˆ 1 ˆ ˆ 1 i 1 N0 i1 self-interference cross-interfernece noise contamination E E h h h E h h E h hˆ 1 L 2 2 L vd i / 1 1 i / 1 N0 / 1 1 vd i1 1 v x J d i1 4 conventional interference 1 M Contamination induced distortion 1/J d Conventional cross-cell interference 1/M 24

25 Simulation results: 1 user per cell BER r = 1 r = 100 r = 1 r = 1 perfect CSI simulation prediction conventional pilot-based SVD blind estimation data-aided 4 iterations strong interference even for an extremely high pilot power r = r power of pilot symbol power of data symbol SNR (db) Settings: 1 =1, i =0.2 for i1, N=128, J p =1, 64-QAM. { i } are large scale fading factors. The SVD method is from the following reference. R. R. Muller, et al, Blind Pilot Decontamination, IEEE Journal of Selected Topics in Signal Processing on Signal Processing for Large-Scale MIMO Communications,

26 Contents Introduction Channel estimation at the BS Channel estimation at MTs Multiple access: OFDMA, SDMA, IDMA and NOMA Conclusions 26

27 Accurate channel estimation is not crucial at MTs Coarse statistical channel information, such as a covariance matrix, is sufficient in the up-link massive MIMO. Without accurate CSIT, interference is inevitable. IDMA is an efficient interference cancelation technique, and hence a natural way to realize NOMA. Chongbin Xu, Peng Wang, Zhonghao Zhang, and Li Ping, "Transmitter design for uplink MIMO systems with antenna correlation," IEEE Trans. Wireless Commun., vol. 14, no. 4, pp , Apr

28 Statistical channel information Statistical channel information refers to partial knowledge of the channel. A typical case of statistical channel information is a correlation matrix containing the power distribution of on different eigen-directions. It does not contain phase information. A correlation matrix can be obtained by taking cross-correlation of the received signals on different antennas. It changes slowly and can be estimated with much lower cost (compared with full CSIT). The overheads related to CSIT (such computational cost and pilot usage) can be greatly reduced in this way. 28

29 Non-orthogonal mode transmission Statistical channel information, only partial CSIT is available. The system is characterized by the following properties. Partial CSIT is very useful. It still can provide close to optimal performance. The channel cannot be fully orthogonalized. There is interference among different users. This leads to the non-orthogonal mode transmission. Interference cancelation techniques are required to suppress in this case. initial interference 29

30 Generalized NOMA Non-orthogonal scheme are necessary to achieve the ultimate multi-user capacity. The advantage of NOMA becomes noticeable in the high rate regime. sum rate K= K=8 K=4 K=2 K= sum power (db) 4 4 MIMO single cell equal rate for all users K is the number of users S Tomida and K Higuchi Non-orthogonal access with SIC in cellular downlink for user fairness enhancement ISPACS Peng Wang, Jun Xiao, and Li Ping, "Comparison of orthogonal and non-orthogonal approaches to future wireless cellular systems," IEEE Vehicular Technology Magazine, Sept

31 Mutual information analysis N BS /N MT = 4 FCSIT = full CSIT SWF = statistical water filing NP = no precoding (no-csit) Chongbin Xu, Peng Wang, Zhonghao Zhang, and Li Ping, "Transmitter design for uplink MIMO systems with antenna correlation," IEEE Trans. Wireless Commun., vol. 14, no. 4, pp , Apr

32 CSIT for the uplink joint detection individual beamforming MT 1 correlation BS correlation channel MT k correlation MT K correlation For the uplink, beamforming is individually done, so approximate CSIT is sufficient to ensure good performance. Decoding can be jointly and interference can be suppressed. 32

33 Joint detection via IDMA IDMA is a low-cost technique that facilitates joint detection. With IDMA, the signals are separated by user-specific interleavers. Channel estimation, MUD and decoding can be performed jointly and iteratively in an IDMA receiver. decoder ˆd 1 MUD ĥ 1 d y1 d 1 channel estimator y, p p 1 1 Li Ping, Lihai Liu, Keying Wu, and W. K. Leung, "Interleave Division Multiple-Access," IEEE Trans. Wireless Commun., vol. 5, no. 4, pp , Apr

34 IDMA performance 10 0 BER BS antennas 16 users total rate = 16 MRC 10-4 IDMA Eb/N0(dB) With a large number of users, conventional MRC detection performs poorly. Iterative IDMA detection is a low-cost, high performance option. 34

35 Contents Introduction Channel estimation at the BS Channel estimation at MTs Multiple access: OFDMA, SDMA, IDMA and NOMA Conclusions 35

36 Overview OFDMA is suboptimal for MIMO. Orthogonal SDMA can do better but is still sub-optimal. In general, any orthogonal scheme is suboptimal for MIMO. Theoretically, NOMA can achieve MIMO capacity but, practically, interference can still be a problem. IDMA is a low-cost interference cancelation technique, and hence a natural way to realize SDMA and NOMA. 36

37 Balanced and unbalanced MIMO Ideally, we want a true MIMO system, with large numbers of antennas at both ends. Such a setting can provide a huge rate gain. rate N In practice, however, we can only mount a limited number of antennas on a handset. Such a setting can achieve large power gain but not rate gain. power N 37

38 OFDMA In conventional OFDMA, only one user is allowed to transmit on each subcarrier at a given time in a cell. OFDMA with massive MIMO achieves good power gain. However, the related rate gain is less impressive. rate Can we do better? N 38

39 Space-division multiple access (SDMA) In massive MIMO, more users can be supported using the SDMA via ZF. sum-rate ~ min(n BS, K N MT ) log(1+snr). However, ZF requires accurate CSIT. rate K=3 K=2 K=1 N 39

40 IDMA and NOMA IDMA does not require accurate CSIT in the upper link. It can acquire and refine channel information iteratively. SDMA-IDMA thus provides a robust implement technique NOMA. initial interference S Tomida and K Higuchi Non-orthogonal access with SIC in cellular downlink for user fairness enhancement ISPACS Peng Wang, Jun Xiao, and Li Ping, "Comparison of orthogonal and non-orthogonal approaches to future wireless cellular systems," IEEE Vehicular Technology Magazine, Sept Y Chen, J Schaepperle and T Wild, Comparing IDMA and NOMA with superimposed pilots based channel estimation in uplink PIMRC

41 Multi-user (non-orthogonal) gain We can view IDMA-SDMA as a NOMA or a non-ideal SDMA. With iterative processing, it can potentially provide the similar gain as ideal SDMA. sum-rate ~ min(n BS, K N MT ) log(1+snr). initial interference among users rate K= K=2 K=1. N Peng Wang, and Li Ping, "On maximum eigenmode beamforming and multi-user gain," IEEE Trans. Inform. Theory, vol. 57, no. 7, pp , Jul

42 Simulation Results: 4 Users per Cell conventional pilot-based BER data-aided: 4th iteration SVD blind estimation SNR (db) Settings: 1 =1, i =0.2 for i1. J p =1. 64-QAM. Other parameters are the same as the previous figure. { i } are large scale fading factors. 42

43 Contents Introduction Channel estimation at the BS Channel estimation at MTs Multiple access: OFDMA, SDMA, IDMA and NOMA Conclusions 43

44 Conclusions Down-link requires accurate CSIT. Pilot contamination is a problem in this case, but it can be mitigated by a data aided channel estimation technique. Up-link requires only coarse statistical CSIT, provided that iterative detection is used at the BS. OFDMA is suboptimal for massive MIMO. Orthogonal SDMA can do better but is still sub-optimal. Theoretically, NOMA can achieve MIMO capacity but, practically, interference can be a problem. IDMA is a low-cost interference cancelation technique, and hence a natural way to realize SDMA and NOMA. 44

45 Pilot overhead user 1 user 2 user 3 user 4 user 5 user 6 user 7 user 8 pilot data An challenge how to design pilots for NOMA? In a massive MIMO system with many users, it can be very costly to allocate orthogonal positions to all users. 45

46 Superimposed pilots user 1 user 2 user 3 user 4 user 5 user 6 user 7 user 8 pilot and data This problem can be resolved in SDMA-IDMA using superimposed pilots. This technique is discussed in the paper below and we are current working on this issue. Chulong Liang, Junjie Ma and Li Ping, Rate maximization for data-aided channel estimation in multi-user large antenna systems, under preparation. 46