Interference Mitigation by MIMO Cooperation and Coordination - Theory and Implementation Challenges

Transcription

1 Interference Mitigation by MIMO Cooperation and Coordination - Theory and Implementation Challenges Vincent Lau Dept of ECE, Hong Kong University of Science and Technology

2 Background 2

3 Traditional Interference Management in Cellular Systems Frequency Reuse: Static spectrum sharing; TDM within each cell (GSM) Adjacent cells use different frequency Widely used for voice communication Coverage radius reduced and low spectral efficiency CDMA/OFDMA & Scheduling: Universal frequency reuse is possible Interference Averaging: Spread spectrum in time-domain (CDMA) or frequency domain (OFDMA) Statistical multiplexing of timefrequency More signal processing and complicated resource allocation The system is highly interferencelimited F3 F1 F2 Frequency Reuse (reuse 3) F1 F2 F3 F1 F2 F3 F1 F2 F3 F1 F2 F3 F1 F3 F1 F2 CDMA OFDM 3

4 Spatial Interference Mitigation using MIMO Technology Multiple antennas provides spatial degrees of freedom for interference mitigation Beamforming has been widely used for spatial interference mitigation SINR is a function of T, R and q Transmi(ers p1, t1 p2, t2 H 1,2 H 1,1 Receivers r 1 r 2 p3, t3 H 2,3 H 3,3 r 3 t: trade-off between controlling interference to others and maximizing signal power r: trade-off between suppressing interference and maximizing signal power T p = [ p,..., p ] (power vector) lk, 1 T= { t,..., t } (Tx beamforming vectors) 1 R = { r,..., r } (Rx beamforming vectors) H 1 L L L (channel matrix from Tx k to Rx l) 4

5 Successive Interference Cancellation (SIC) and Dirty Paper Coding (DPC) y SIC for uplink receiving Decoder User Encoder User 1 ˆb 1 ˆb 2 Decoder User Encoder User 2 Decoder User L ˆbL DPC for downlink transmitting [M. Costa, TIT83] DPC Encoder 1 DPC Encoder 2 DPC Encoder L Concatenation It is convenient to simply view DPC as successive interference pre-cancellation at the Tx. Pre-coder 5

6 Achievements and Limitations of Existing Interference Mitigation Technologies Achievements Beamforming + DPC achieves the capacity of single cell MIMO downlink Beamforming + SIC achieves the capacity of single cell MIMO uplink Simple beamforming scheme (e.g., zero-forcing) + user selection achieves a large portion of the capacity Limitations The inter-cell interference is not handled and is treated as noise The performance of the system is limited by inter-cell interference The cell edge users suffers from bad performance This motivates the idea of inter-cell interference (ICI) mitigation by cooperation and coordination between BSs 6

7 Theory of Co-MIMO 7

8 Basic Concepts of Cooperative and Coordinative MIMO Multi-cell environment with frequency reuse factor 1 Joint processing Optical fiber Optical fiber interference I I Optical fiber I I I I Ø Multiple BSs collaborate to mitigate ICI or align interference for cancellation. Ø Improves cell-edge performance and overall throughput. 8

9 (1) Cooperative Zero-Forcing Beamforming (ZFBF) h 2 h 1 ZFBF on a Larger dimension Signal Space à achieve full DoF e 2 e 1 g t The direction of the projection gives transmitter vector, while the norm 1 1 square of the projection gives the equivalent channel The mutual interfering multi-user channel is decomposed into multiple parallel 2 independent SISO channels: T 1 The optimal power allocation is easy, e.g., Water-filling is optimal for sum rate maximization with sum power constraint T 2 g = ht g 2 = ht R 1 R 2 9

10 (2) Coordinated Zero-Forcing (IA) Consider a MIMO cellular network consisting of three BSs and three MSs, where each node has 2 antennas and each BS delivers 1 data stream. 10

11 (2) Coordinated Zero-Forcing (IA) Breakthrough performance Interference alignment (IA) achieves the optimal throughput scaling law w.r.t. SNR in MIMO interference networks and MIMO-X networks [V. R. Cadambe, S. A. Jafar, and S. Shamai, TIT 08], [V. R. Cadambe and S. A. Jafar, TIT08]. Famous for the saying No matter how many people come to share the cake, everyone can get a half. Limitations (K/2)log SNR The feasibility issue arises without infinite dimension symbol extension [C. M. Yetis, S. A. Jafar, and A. H. Kayran, TSP 10]. Requires perfect CSI: Small channel estimation errors can be tolerated, while larger errors reduce the diversity gain significantly. [A. Sezgin, S. A. Jafar, and H. Jafarkhani, GlobalCOM 09 ] 11

12 Different Levels of Cooperation Co-MIMO with full cooperation All CSI and data symbols are collected at a central processor (CP) à heavy backhaul consumption A virtual MIMO BC/MAC is created for the transmission of all users A user is served by all BSs a big MIMO! Coordinated MIMO without Cooperation The CP only collects CSI à limited backhaul consumption The CP jointly optimize the user scheduling, power allocation and beamforming vectors of all BSs A user is only severed by one BS, the signals from other BSs are treated as interference Co-MIMO with limited cooperation In practice, the capacity of backhaul is limited Co-MIMO is more expensive compared with Coordinated MIMO We have to quantize the data symbols and/or only exchange part of the data symbols 12

13 Standardization in LTE 13

14 Related Standard Activities in LTE Advanced Coordinated multipoint transmission/reception (CoMP) Improve coverage, cell-edge throughput, and/or system throughput CoMP schemes (especially DL) identified in LTE-A (3GPP TR ) Joint processing/transmission (multipoint transmission to single UE) (a) Coherent transmission (joint precoding) (b) Non-coherent transmission (independent precoding) Coordinated scheduling/beamforming (data is transmitted only from one cell site, and scheduling/beamforming is coordinated among cells) (a) Coordinated scheduling (ICIC with fast cell selection (FCS)) (b) Coordinated beamforming (space-domain fast ICIC) 14

15 Practical Challenges and Issues 15

16 Practical Challenges and Issues Backhaul constraint Backhaul latency Capacity constraints CSI Acquisition Synchronization Timing Sync Frequency offsets Sync 16

17 Backhaul Constraint Backhaul capacity requirements Co-MIMO requires the exchange of a large amount of information Depending on the backhaul technology, the backhaul capacity may or may not support Co-MIMO operation Methods to reduce the Backhaul burden for Co-MIMO: Quantizing baseband signals Cell Clustering (Q: What s the optimal number of cooperative cells) Optimized data sharing via a rate splitting Backhaul Latency requirements Backhaul Latency must be small compared to channel coherent time Typical Backhaul Latency in LTE system 17

18 CSI Estimation Quality Objective: Accurate channel estimation (CE) with moderate overhead is basis of any advanced Co-MIMO scheme. Challenges: High # of channel components per cooperation area Multi cell environment with strong inter cell interference Strong variation of path loss over different cells Fast outdating of CSI Larger CSI delay due to cooperation Pilot Pollution between Cells Possible Research Activities: Exploiting channel reciprocity Overhead reduction schemes (compression techniques) Accurate CSI estimation under strong inter cell interference Codebook design under strong variation of path loss Robust CSI prediction schemes 1/1/

19 CSI Feedback design for Co-MIMO Feedback of channel information Allows transmitter adaptation and enables interference avoidance Consumes reverse link capacity: tradeoff performance gain vs. reverse link penalty Some approaches: Hierarchical feedback: provide more information on stronger (more relevant) transmitters Feedback compression: Lossless vs. lossy Channel tracking: only provide feedback info for channel evolution Feedback combined with channel prediction Two-step scheduling: Coarse CSI (e.g., SINR) feedback for scheduling and refined CSI feedback for power allocation and pre-coder calculation 1/1/

20 Timing Synchronization For Co-MIMO, the transmission from multiple BSs must be synchronized. OFDM with CP allows some margins in timing sync. However, large distances between the base stations result in large timing offsets which may exceed CP length and lead to inter-symbol interference Sync using primary clock reference Primary clock boards of cooperative BSs is synchronized to a common external reference clock Commercial Rubidium- and crystal-based reference clocks can be phaselocked to the GPS. Network synchronization Suitable for indoor BSs where GPS signal may be blocked Precise timing protocol (PTP) specified in IEEE 1588 standard is a good candidate Packet delay spread must be kept small to achieve high accuracy of timing sync: give higher priority for PTP packets 20

21 Carrier Frequency offsets (CFO) Synchronization CFO causes inter-carrier interference in OFDM systems Challenges of CFO Sync for Co-MIMO: Each receiver need to estimate multiple CFOs of multiple BSs CFO compensation at the Rx side requires high complexity equalizer CFO compensation at the Tx side needs CFO feedback from the receiver 21

22 Performance under Practical Constraints 22

23 Flexible Partial Cooperative MIMO Consider a MIMO downlink cellular network with B Basestations (BSs) and N mobile users (MSs) [H. Huang and V. Lau]. 23

24 Flexible Partial Cooperative MIMO Performance and insight: Suppose the BSs are symmetrically distributed along a line and the pathloss grows exponentially w.r.t. distance. 24

25 (1) Simulation Results: CDF of downlink user rates for different backhaul capacities per BS. No Coop Full Coop Tradeoff between user rates and backhaul capacity 25

26 USRP Platform Overview The universal software radio platform (USRP) demonstrates a wireless radio system using a general purpose PC Key components of the USRP platform are: General purpose PC Universal software radio platform (USRP) board Transceiver board Indoor antennas E100 Embedded platform GNU Radio (software development toolkit) Hydra (wireless testbed which supports most features specified in the IEEE n standard) Tx Transceiver Board Rx Host PC USB USRP Board Tx Transceiver Board Rx Connection Diagram of the USRP Board 26

27 Hardware Configuration of Co-MIMO Uplink BS #1 MS #1 Host PC BS Setup USB USRP Board XCVR2450 Board XCVR2450 Board Tx/Rx Tx/Rx y 1 h 11 Tx/Rx x 1 XCVR2450 Board USRP Board USB Host PC MS Setup LAN cable BS #2 h 21 h 12 MS #2 Host PC USB USRP Board XCVR2450 Board XCVR2450 Board Tx/Rx Tx/Rx y 2 h 22 Tx/Rx x 2 XCVR2450 Board USRP Board USB Host PC 27

28 Upper-level Protocol and Physical Layer scheme MS1: MS2: TS from MS1 TS from MS2 Data from MS1 Data from MS2 Time diagram of Data transmission from MSs BS1 BS2 Send CSI to BS1 via Socket communication BS1 BS2 BS1 calculates ZF receive vectors r1,r2 and Send r2 to BS2 BS1 BS2 BS2 send r2^t*y2 to BS1, BS1 decode x1 using r1^t*y1+ r2^t*y2 Cooperative receiving at the BSs (only for decoding of x1) Physical layer scheme: OFDM with 56 data subcarriers Cooperative receiving based on Zero-Forcing (ZF) detection at BSs Constant transmit power and uniform power allocation over subcarriers at MSs 28

29 (1) System Parameters for the Co-MIMO Experiment Specification Value MCS index 3 Modulation 16QAM Coding Rate R 1/2 Spatial stream 1 Data rate 26.0 Mb/s Distance 2m No. of packet 1000 packets Data bits in a packet 1000 bits / packet Signal transmit gain 5000 to 9000 Interference transmit gain 6000, 9000 Maximum transmit power 50 mw Carrier frequency 2.4 GHz Transmit antenna type 5 dbi omni-directional Receive antenna type 5 dbi omni-directional 29

30 (1) Co-MIMO Experiment Results Verify the cooperative gain under some practical constraints: Ø CSI estimation error Ø Timing/CFO sync error Ø Practical modulation and coding scheme 30

31 (1) Implementation Issues Large CSI feedback delay Each CSI training costs about 10ms CSI Feedback costs about 10ms If exhaustive search is used to find the best BF vector in the codebook, it takes about 1ms for each subcarrier even when the codebook size is reduced to 256 According to channel correlation tested over time, the feedback delay must be kept within 70ms to achieve 20db suppression gain. average time correlation over 56 subcarriers X: 7 Y: per 10ms Solutions: 1. Use structured codebook to reduce the searching time 2. Only feedback CSI for two subcarriers by exploiting the channel correlation over subcarriers Conclusion: CSI feedback delay is a serious problem in practical implementation! 31

32 (2) System Parameters for the Coordinated MIMO Experiment Specification Value MCS index 1 Modulation QPSK Coding Rate R 1/2 Spatial stream 1 Data rate 13.0 Mb/s Distance 2m No. of packet 5000 packets Data bits in a packet 1000 bits / packet Signal transmit gain 6000 to 9000 Interference transmit gain 6000 to 9000 Maximum transmit power 50 mw Carrier frequency 2.4 GHz Transmit antenna type 5 dbi omni-directional Receive antenna type 5 dbi omni-directional 32

33 (2) Coordinated MIMO Experiment Results 10 0 Rx1 PER - Tx2 Interference Curve (MCS1) (Tx1 Gain = 9000) Rx1 Packet Error Rate (PER) 10-1 Verify the coordination gain under some practical constraints: Ø CSI estimation error and feedback delay Ø Timing/CFO sync error, Ø Practical modulation and coding scheme 10-2 Coordinated MIMO Random Beamforming Tx2 Interference 33

34 (3) Energy Saving Performance [Cui 04] Tx 1 Rx 1 Tx 2 Tx 1 Rx 1 Tx 1 to Rx 1 without any help (SISO link) Cooperative MIMO without CSIT (MISO) Using Alamouti scheme with BPSK modulation in both typology Due to NO P cct Due to P cct effect Transmission energy per bit (exclude P cct ) Total energy per bit (include P cct ) Not always the more Tx antennas, the better! Not always the more cooperative BSs, the better! Effect of P cct is significant when P cct is non-negligible from the total power consumption 34

35 Conclusion 35

36 (1) Summary of Co-MIMO and Coordinated MIMO Coordinated MIMO: Moderate performance gain Light backhaul burden Insensitive to Sync error Easier for distributed Implementation Optimal transmission scheme is unknown Co-MIMO: Largest performance gain Heavy backhaul burden Sensitive to Sync error Centralized Implementation Optimal transmission scheme is known What s the best tradeoff between coordination and cooperation? 36

37 (2) Summary of Major Implementation Challenges Advanced backhaul technologies to meet the requirement of Co-MIMO CSI Estimation and Feedback Overhead of training Feedback delay makes CSI outdated Scalable Implementation How to adapt the cooperative level according to the network state and users requirement? Adapting the number of cooperative cells for each user Adapting the amount of information exchanged : From full cooperative to coordinated MIMO How to make users operating at different cooperative levels co-exit? How to balance between centralized and distributed control? Co-MIMO and Coordinated MIMO in fast fading channel A thorough study on the performance gain of open loop Co-MIMO is still lacking 37

38 (3) Summary of the Green Aspects of Co-MIMO and Coordinated MIMO Energy per bit metric Circuit power consumption P cct is constant as long as transmission rate 0 Further complicates the Co-MIMO and Coordinated MIMO design The choice of # of Tx antennas is a combinatorial problem Not always use all the Tx antennas Not always involve in more cooperative BSs Effect of P cct is significant when P cct is non-negligible from the total power consumption 38

39 (4) Techniques Beyond Physical Layer: Interference-Aware Networking Physical Layer Technologies Coordinated MIMO Spatial interference mitigation via signal processing techniques Infrastructure based MIMO/OFDM Networks, Multi-channel Mesh, Multi-hop Cooperative Systems Cooperative MIMO resolve interference issues in Cellular Systems via data cooperation System Level Technologies Interference-Aware Networking Exploit knowledge of interference profile in MAC and networking protocol designs 39