Multi-User MIMO Transmission
MU-MIMO (Multi-User Multiple Input Multiple Output) achieves substantial capacity gains by using precoding to support multiple, concurrent data streams to a group of clients. Precoding comprises of computing the transmitter’s antenna gains and phases from the channel state information (CSI), i.e., the channel matrix in which each element represents the magnitude and phase offset for each transmitter-receiver antenna path. In this way, each receiver can simultaneously decode its streams
To realize the capacity gains, in addition to precoding, the AP must also select the (i) mode: the number of transmit antennas and collective number of receiving antennas, and (ii) users: the set of receiving antenna(s), i.e., clients. For each transmission, the ideal mode and user set is channel dependent and therefore their selection would require CSI for all receivers. However, due to large overhead, it is practically infeasible to collect CSI for all potential receivers prior to each transmission. The additional overhead could substantially mitigate the benefits of MU-MIMO transmissions.
User Selection and Rate Estimation without Sounding
The key to efficient MU-MIMO transmissions is the amortization of sounding overhead over the transmitted data. In this project we present PUMA (Pre-sounding User and Mode selection Algorithm): a method of mode and user selection prior to channel sounding. The key techniques of PUMA are threefold: (i) estimation of expected per-user datarate based on theoretical MU-MIMO system scaling, (ii) characterization of relative cost from overhead for each potential mode, and (iii) calculation of expected aggregate throughput for a potential group of users given a particular mode. Through this pre-sounding estimation process, PUMA selects the best mode and group of users by maximizing throughput with respect to overhead. While estimating the expected throughput using the theoretical MU-MIMO system scaling is not perfect, it is sufficient for mode and user group selection because the indoor Wireless LAN (WLAN) environment usually results in well conditioned channel matrices due to the prevalence of multi-path effects.
PUMA with IEEE 802.11ac
The following figure depicts the timeline of two separate transmissions from a 3-antenna transmitter to either three (sub-figure (a)) or two single antenna receivers (sub-figure (b)). Each stream consists of 10 aggregated full size (1,500 byte) packets. The overhead (channel sounding, channel feedback, and acknowledgment) for both transmissions is sent at the base rate although the resulting data rates are different. For this example, we assumes that each of the three potential users have 18 dB omnidirectional SNRs.
PUMA computes the aggregate throughput of the potential modes given the expected datarate and node backlog. Through this computation, PUMA identifies that the aggregate throughput of sub-figure(a) is 145 Mbps while the aggregate throughput of sub-figure(b) is 161 Mbps. This example highlights a common yet counterintuitive result that PUMA identifies. A MU-MIMO transmission does not always benefit from using the most antennas. Not only does the protocol overhead increase with additional antennas, but also the per-user MU-MIMO SINR decreases resulting in lower per-user datarates.
Publication
N. Anand, J.K. Lee, S.J. Lee, and E. Knightly, “Mode and User Selection for Multi-User MIMO WLANs without CSI,” in Proceedings of IEEE INFOCOM 2015, Hong Kong, China, April 2015.