Performance Evaluation of MU-MIMO WLANs Under the Impact of Traffic Dynamics
Knightly, Edward W
Master of Science
Downlink multiuser MIMO (DL MU-MIMO) has shown a great promise to enhance the performance of a wireless network by serving multiple users simultaneously. While researchers have focused on tackling the challenges involved in deploying these networks in the real world, little is known about their performance under the impact of traffic dynamics. In this paper, we evaluate the performance of DL MU-MIMO under the impact of traffic dynamics. Specifically we study how the system behaves under closed loop (FTP/TCP flows) and open loop traffic (UDP flows). We extend the network simulator ns 3 to include support functionalities for 802.11ac compliant DL MU-MIMO simulations. By performing extensive simulations, we find that 802.11ac MU-MIMO achieves extremely small queue lengths ($\approx$ 2 packets) under TCP flows. Moreover, the contention based channel access reduces the user set size for downlink transmission thereby affecting the multiplexing gain as well. Consequently, the downlink performance is extremely poor. Unfortunately we find that even under an ideal network and channel conditions, the system performs worse than a SISO system. We show that while the downlink performance can be improved by enabling frame aggregation for uplink transmissions, the multiplexing gain remains constrained due to the contention based channel access mechanism. This shows that 802.11ac MU-MIMO has severely limited throughput under closed loop traffic. Furthermore, we find that with the AP's co-ordination the system performance shifts closer to the maximum achievable throughput which is not possible under a contention based access mechanism. Next we show that with the help of a multi-user uplink, 802.11ac compliant DL MU-MIMO achieves the maximum achievable throughput under closed loop traffic. Finally, we explore how the system performs under open loop traffic burstiness. We find that the aggregate downlink throughput shows a exponential decrease with an increasing amount of traffic burstiness. While the system does achieve a high throughput at lower levels of traffic burstiness, at higher levels there is no gain from simultaneous data transmission.