Resilient Multi-ser Beamforming WLANs: Mobility, Interference, and Imperfect CSI Presenter: Roger Hoefel Oscar Bejarano Cisco Systems SA Edward W. Knightly Rice niversity SA Roger Hoefel Federal niversity of Rio Grande do Sul Brazil
From SISO to MIMO 2 IEEE 802.11n: 2009 600 Mbps 4x4 MIMO 40 MHz 64QAM C 10-15 April 2016 // San Francisco - SA
From SISO to MIMO 3 In MxN MIMO, capacity increases with min(m, N) M: # of TX antennas N: # of RX antennas C2 C1 C Real World: Low number of antennas due to form factor of mobile devices and cost C4 C3 10-15 April 2016 // San Francisco - SA
Multi-user MIMO 4 Simultaneous spatial sharing of medium by multiple users. Capacity Gain: The system can serve as many users as antennas at the. C2 C1 Objective: Remove the constraint on the # of antennas at client devices. M-MIMO: Beamforming at TX side. C4 C3 10-15 April 2016 // San Francisco - SA
5 Multi-user MIMO C2 The IEEE 802.11ac amendment (2013) specifies optional M-MIMO operation: Maximum of four users and two spatial streams (SS) per user. C1 C4 C3 Throughput Gains Shown in prior works: [Tse05,Yoo06,Aryafar10,Balan12,Shepard12] However, in this work we identify one key challenge 10-15 April 2016 // San Francisco - SA
Inter-Stream Interference 6 In practice, the accuracy of beam-steering weights used to precode the TX signal depends on ser mobility C2 Environmental mobility Quantization C1 Out-of-cell interference Challenge: High susceptibility of the M-MIMO performance with inter-stream (multi-user) interference. 10-15 April 2016 // San Francisco - SA C4 C3
Paper Contributions 7 Design and evaluate one protocol to enable resilient M-MIMO by removing the adverse effects of inter-stream interference due to outdated and inaccurate Channel State Information at Transmit Side (CSIT) using M-MIMO bit rate selection and loss recovery. CHRoME: Channel Resilient Multi-user beamforming 10-15 April 2016 // San Francisco - SA
Roadmap 8 Background and Motivation Protocol Description CHRoME Channel Resilient Multi-user beamforming Objective: reduce the effects of Inter-Stream Interference Protocol Evaluation: in the throughput. Trace Driven Emulation using Over the Air Channel Measurements Conclusions 10-15 April 2016 // San Francisco - SA
Inter-Stream Interference 9 C2 C1 Ideal M-MIMO scenario: C3 Fully or partially suppressing of the interference to maximize the SINR at the users. Accurate CSIT for beam-steering weight calculation; Beamformed transmission within the channel coherence time. 10-15 April 2016 // San Francisco - SA C4
Quantization Out-of-cell interference Imperfect Bemforming: Inter-Stream Interference ser mobility Environmental mobility 10 C2 C2 C1 C1 C3 C3 C4 Single stream decodable Multiple streams C4 non-decodable noise at user2 Inaccurate CSIT at user2 Throughput Penalty 10-15 April 2016 // San Francisco - SA Inaccurate CSIT
11 CHRoME CHannel Resilient Multi-user beamforming (CHRoME) Avoids or resolves the problem of co-channel interference using: Inter-stream/Out-of-cell interference estimation and MCS adaptation: Beamformed probing for just-in-time MCS selection prior to data transmission. Fast soundless recovery: One-time fast retransmission with liberated antenna resources 10-15 April 2016 // San Francisco - SA MCS -> Modulation and Coding Scheme
12 MCS Selection in M-MIMO MCS must be reduced due to imperfect beamforming!! Throughput 64-QAM 64-QAM BPSK Inaccurate CSIT decodable non-decodable MCS selection is increasingly more difficult for M-MIMO compared to S-MIMO: SINR depends on channels to other concurrent users; Inter-stream/out-of-cell interference need to be taken into consideration`. 10-15 April 2016 // San Francisco - SA
13 MCS Selection in M-MIMO 64-QAM 64-QAM BPSK Inaccurate CSIT decodable non-decodable Basiline Protocol: MCS selection based on CSIT [Halperin11,Shen12] Transmitter learns channel matrix (vectors to all receivers) and infers the post processing SÌNR (e.g., projection onto null space of the vectors to the other users) to select the MCS. Drawbacks: SINR depends on channels to other concurrent users Inter-stream/out-of-cell interference need to be taken into account. 10-15 April 2016 // San Francisco - SA
C2 Limitations 14 C1 of CSIT-Based MCS Selection for data transmission C4 C3 Illustrative example explicit feedback Gradual CSI degradation + other factors DL Sound L Feedback DL - Data Channel observed at the user, which incorporates measured interference Best MCS, and real channel unknown time 10-15 April 2016 // San Francisco - SA
C2 Limitations 15 C1 of CSIT-Based MCS Selection for data transmission C4 C3 Interference C3 Illustrative example implicit feedback plink training sequences (No explicit information) DL Trigger L Training The is not aware of any source of interference at users since CSI is obtained from previous uplink transmissions 10-15 April 2016 // San Francisco - SA DL - Data time
16 Limitations of CSIT-Based MCS Selection for data transmission Gradual Gradual CSI degradation CSI degradation + other factors Illustrative example explicit feedback CHRoME - Introduces a mechanism for just-in-time multi-user MCS selection which implements a beamformed probe that captures the real channel observed at the users DL Sound L Feedback DL - Data time Best (highest possible) MCS, and real channel unknown 10-15 April 2016 // San Francisco - SA
(1) Probing-Based MCS Selection 17 C2 Short beamformed probing frame just prior data transmissiom using previously collected CSI C1 C3 C4 DL Sound L Feedback DL - Data sers measure the precoded probe s SINR and select the highest possible MCS (considering all current sources of interference) 10-15 April 2016 // San Francisco - SA time
(3) Probing-Based MCS Selection Feed back MCS selection: 18 C2 PN Sequences C1 C3 C4 DL Sound L Feedback DL - Data time PN bit sequence: transport the MCS index 10-15 April 2016 // San Francisco - SA
(4) Probing-Based MCS Selection 19 C2 Correlatable symbol sequences C1 C3 C4 Advantages of PN Sequences: No decoding required (no preamble or data processing) 6.35 μsec Highly reliable (detected at low SINR, i.e., -6 db) Low feedback overhead introduced 10-15 April 2016 // San Francisco - SA
CHRoME: Evaluation 20 CHannel Resilient Multi-user beamforming 2 Methodology 1 3 Trace-Driven Emulation Emulation based on over-the-air channel measurements we collect Enables repeatability for fair comparison 4 10-15 April 2016 // San Francisco - SA
Evaluation 1 2 2 Methodology Indoor channel traces - conference 4 3 3 Indoor channel traces - conference rooms/labs/offices environment 15,000+ frame transmissions per scheme 15,000 frame transmissions 1 1 A P 4 4 2 2 3 3 1 1 A P 4 4 2 2 3 3 21 15,000 frame transmissions over/under/accurate selection 1 1 A P 4 4 2 2 3 3 1 1 A P 4 4 2 2 3 3
Evaluation 22 Methodology Indoor channel traces - conference rooms/lab/office environment. 4 15,000+ frame transmissions per scheme Very-High Throughput (VHT) 802.11ac frame 802.11ac timings 2 1 3
23 Probing-Based MCS Selection 2 MCS selection accuracy in real indoor channels 1 3 4 MCS selection solely based on CSIT Conservatively decrease MCS Over-selection nder-selection CHRoME? Ground truth found by measuring per subcarrier SINR during actual data transmission and mapping to MCS 10-15 April 2016 // San Francisco - SA
24 Probing-Based MCS Selection BL = Baseline CSIT-Based BLc = Baseline CSIT-Based Conservative (BL-1) CH = CHRoME 2 1 3 MCS selection accuracy 4 Out-of-cell interference -70 to -90 dbm Explicit Feedback Implicit Feedback Conclusions: 1. Higher accuracy of CHRoME compared to the baselines. 2. Much higher gain in implicit since the does not consider the interference at the STAs in CSI estimation. (Bottom) No out-of-cell interference 10-15 April 2016 // San Francisco - SA
Probing-Based MCS Selection Basic Conclusion: 1. Much higher accuracy of CHRoME compared to the baselines since there is no interference and there is more room to make mistakes due to outdated CSI. (Top) Out-of-cell interference 4 2 25 MCS selection accuracy BL = Baseline CSIT-Based BLc = Baseline CSIT-Based Conservative (BL-1) CH = CHRoME 1 3 No out-of-cell interference 10-15 April 2016 // San Francisco - SA
Probing-Based MCS Selection 26 2 1 Percent gain of CHRoME From 7% to 280%. MCS selection accuracy (Throughput) # succ. received data bits 4 3 (Top) Out-of-cell interference 7.3% 42.2% total time (incl. overhead) 498% 280% (Bottom) No out-of-cell interference 52.6% 88.9% 32.2% 64% BL = Baseline CSIT-Based BLc = Baseline CSIT-Based Conservative (BL-1) CH = CHRoME 10-15 April 2016 // San Francisco - SA
27 Probing-Based MCS Selection MCS selection accuracy (Adaptation response) 2 1 The baseline mostly overselects whereas CHRoME follows closely the ideal MCS selection 4 3 Best MCS (Ground truth) CHRomE Baseline MCS Inde ex Sample Index 10-15 April 2016 // San Francisco - SA
Retransmissions in 802.11-based networks 28 # TX antennas: 4 4 users with 1 Rx antenna each 4x4 ACK DIFS CW = 31 slots Sounding Re-Tx 4x4 Corrupted packets require re-contention after doubled backoff window and sounding the channel again time Packet loss + Additional time until retransmission Stale CSI Resound! High overhead introduced 10-15 April 2016 // San Francisco - SA
M-MIMO Fast Packet Recovery 29 4x4 ACK Re-Tx 4x2 ACK DIFS CWmin = 15 slots Exploit liberated antenna resources to obtain diversity and power gain, increasing robustness Select configuration that minimizes the time to retransmit the corrupted packets 4x1 TDMA vs. 4x2 M-MIMO time 10-15 April 2016 // San Francisco - SA
Fast Recovery: 30 MCS Selection The performs MCS selection based on the report of individual inter-stream interference components piggybacked in the ACK control frame. 4x4 ACK Re-Tx 4x2 ACK DIFS CWmin = 15 slots 10-15 April 2016 // San Francisco - SA time ACK control frame piggybacks SINR for each individual stream
31 Fast Recovery Advantages: reduce the overhead Obviate the need to re-sound the channel Avoid doubling backoff window of CSMA mechanism. Disadvantages Neglect higher multiplexing gain during retransmission (e.g. 4x4 vs 4x2) Beamforming with Increasingly outdated CSIT 10-15 April 2016 // San Francisco - SA
Fast Recovery 32 802.11ac: always uses all DoF in re-tx and re-tx.is always successful TDMA 4X1: diversity gain with overhead penalty. M-MIMO 4XR: retransmission to R users with outdated CSI Similar performance of M-MIMO and TDMA. Performance depends on two major factors: (i) Retransmission success rate; (ii) Incurred overhead / overhead reduction. 10-15 April 2016 // San Francisco - SA
Fast Recovery 33 CHRoME s lowest throughput is at least that of the best performing scheme 10-15 April 2016 // San Francisco - SA
Conclusions: CHRoME CHannel Resilient Multi-user beamforming 34 Probing-based MCS selection MCS selection mechanism that assesses the channel and inter- stream interference affecting each user, and adapts rate accordingly Fast, soundless M-MIMO recovery Immediate retransmission mechanism that precludes the need to re-sound the channel by leveraging liberated DoF at the transmitter Take away message: Incorporating knowledge with respect to co-channel interference into protocol decisions leads to substantial mitigation of its effects 10-15 April 2016 // San Francisco - SA
Thank you for your kind attention! Oscar Bejarano obejarano.rice@gmail.com Roger Hoefel Edward Knightly 10-15 April 2016 // San Francisco - SA