DL MU MIMO Error Handling and Simulation Results Month Year doc.: IEEE 802.11-yy/0847r0 DL MU MIMO Error Handling and Simulation Results Authors: Date: 2010-03-15 John Doe, Some Company
Outline Motivation Overview of simulated schemes DL MU MIMO with scheduled ack DL MU MIMO with polled ack Overview of MAC protection Overview of error handling at the AP Medium access behavior Simulation scenarios and parameters Simulation results Summary
Motivation Downlink Multi-user MIMO is identified as a key technology to improve the overall network performance Two DL MU MIMO Ack mechanisms were proposed in 11-09-1172/r0 The goals of the simulation are To study the behavior, including the error recovery behavior, of the two DL MU MIMO ack mechanisms To study the medium access behavior of the AP when transmitting a DL MU MIMO burst To evaluate the performance difference of the two DL MU MIMO response mechanisms in different scenarios
DL MU MIMO with scheduled Ack The AP contends for the medium using EDCA Once a contention is won, the AP transmits a downlink MU MIMO burst to multiple STAs Each data packet defines an offset value such that each STA knows when to transmit back a BA Scheduled Ack introduces less overhead but a STA needs to schedule its BA transmission Data (STA1) Data (STA3) BA Data (STA2) SIFS SIFS or RIFS pad
DL SDMA scheduled ack schemes do not have an error recovery mechanism If a STA does not correctly receive an A-MPDU, it does not transmit a BA There will be a gap in the scheduled response sequence When a BA is transmitted at certain data rates, there is not enough time for the AP to transmit even a NULL data frame to fill the gap Other STAs that have not correctly set the NAV may try to contend for the medium during the gap Data (STA1) Data (STA2) Data (STA3) BA SIFS SIFS SIFS BA
DL MU MIMO with Polled ACKs The AP contends for the medium using EDCA Once a contention is won, the AP transmits a downlink MU MIMO burst to multiple STAs One STA will transmit a BA immediately after receiving the data packet The AP sends BAR frames to poll the remaining STAs for BAs Polled Ack introduces more overhead yet STAs don’t need to schedule BA transmissions Data (STA1) Data (STA3) BA Data (STA2) BAR SIFS pad
Error recovery for the polled ack scheme If the AP senses the medium as idle PIFS after transmitting a BAR frame, it transmits a BAR frame to poll the next STA to which it has transmitted a packet in the DL SDMA burst PIFS Data (STA1) BAR BAR BAR Data (STA2) Data (STA3) BA SIFS BA
Overview of MAC protection With MAC protection: One RTS/CTS exchange at the beginning of a TXOP One RTS sent to a random STA, the STA replies with a CTS Without MAC protection RTS Data (STA1) BAR Data (STA2) Data (STA3) BA CTS BA SIFS SIFS or RIFS BA SIFS SIFS or RIFS
Medium access behavior at the AP As long as one BA is received by the AP, the AP treats the DL SDMA transmission as a success CW = CWmin If no BA is received from any STA, the AP treats the DL SDMA transmission as a failure CW = (CW+1)*2-1 Backoff (CWmin) Data (STA1) Data (STA1) Data (STA2) Data (STA2) Data (STA3) BA BA BA BA SIFS RIFS RIFS BA
Simulation scenario and Traffic pattern One AP and multiple STAs in one BSS Fully loaded network: Bi-directional video conferencing traffic (UDP) Study the network capacity in terms of the network saturation throughput
Simulation parameters One AP (4 antennas), three STAs (each with 2 antennas) TXOP limit: 3 ms 20MHz: 52 data subcarriers, 4 pilot tones SIFS=16 us, RIFS=2us, aSlotTime=9 us Data packet size: 1500 bytes CWmin=7, CWmax=63 for AC_VI; CWmin=15, CWmax=1023 for AC_BE Data rates: With BF only: 802.11a, 16QAM, r=1/2 for control rate (BAR/BA), 802.11n MCS15 (64QAM, r=5/6, nSS=2) for data rate With DL SDMA: 802.11a, 16QAM, r=1/2 for control rate (BAR/BA), 802.11n MCS7 (64QAM, r=5/6, nSS=1) for downlink data rate 802.11a, 16QAM, r=1/2 for control rate (BAR/BA), 802.11n MCS15 (64QAM, r=5/6, nSS=2) for uplink data rate Assumption: each STA needs 2 antennas to receive one spatial stream in DL SDMA (MMSE precoding and MMSE receiver for resolvable LTFs) Training with implicit feedback Comparison: DL SDMA with scheduled ACK (SIFS) DL SDMA with scheduled ACK (RIFS) DL SDMA with polled ACK
Without MAC protection (AC_VI) Throughput achieved by polled ack scheme is 7-9% higher than scheduled ack schemes
Without MAC protection (AC_BE)
With MAC protection (AC_VI and AC_BE)
Simulation scenario (a hidden-node case) Fully loaded network Bi-directional UDP traffic The AP can transmit simultaneously to STA1, 2, and 3 STA4 cannot decode DL SDMA packets that are intended for STA1, 2, and 3 STA2 has high PER This can be viewed as STA2 experiences frequent collisions from hidden nodes in another BSS STA2 is always the second one to respond a BA STA4 AP1 STA1 STA2 STA3
In this scenario, a gap may occur in the response sequence for the scheduled ack scheme STA4 cannot decode the DL SDMA packets and thus does not set the NAV STA4 and STA1 are hidden nodes (STA4 cannot receive BA from STA1) STA4 waits for EIFS after receiving the DL SDMA burst If scheduled ack is used, because STA4 may transmit during the gap in response sequence, the remaining BA will collide with STA4’s transmission EIFS = 94us Data (STA1) BAR Data (STA2) Data (STA3) BA SIFS BA
Polled ack scheme performs better than scheduled ack schemes in this scenario
Discussion of the hidden-node scenario Polled ack scheme performs better than scheduled ack schemes STA4 cannot set NAV based on DL MU MIMO packets After receiving a BAR frame transmitted from the AP, STA4 can set the NAV correctly Because STA4 won’t transmit during the gap in the response sequence, the AP can successfully receive the remaining BA (no unnecessary retransmission needed) PIFS EIFS = 94us Data (STA1) Data (STA1) BAR BAR BAR Data (STA2) Data (STA2) Data (STA3) Data (STA3) BA BA SIFS BA SIFS BA
Summary Three DL MU MIMO response schemes Error recovery Polled response scheme implements an error recovery mechanism Scheduled response schemes do not have an error recovery mechanism MAC protection One RTS sent from the AP to a randomly selected STA AP’s medium access behavior The AP initiates exponential backoff only when no BA response is received (i.e. treated as a traditional transmission failure) The AP initiates success backoff when at least one BA response is received (i.e. treated as a traditional transmission success)
Conclusion With MAC protection, the three response mechanisms have similar performance Transmitting RTS/CTS for every TXOP lowers MAC efficiency The overhead of RTS/CTS becomes more significant if the data rate is higher or the TXOP size is smaller Polled ack mechanism is more robust In general, polled ack performs better than scheduled ack schemes when there is no MAC protection Even for some STAs that are not able to decode DL MU MIMO packets, the BAR frame can still set the NAV at those STAs The polled ack mechanism is preferred over scheduled ack mechanisms Lower complexity More robust (Consistent performance is required to support QoS traffic)
Straw Poll #1 Do you prefer the polled ack scheme over the scheduled ack scheme? Yes No
Straw Poll #2 Do you support the AP medium access behavior as described below? The AP shall consider the DL MU MIMO transmission as a failure when no BA response to the DL MU-MIMO TX is received 2) The AP may consider the DL MU MIMO transmission as a success when at least one BA response to the DL MU-MIMO TX is received Yes No