Our key activity was to study and gain insight and better understanding of the correlation between the attainable sum-rate gain and the user selection mechanisms, by analytically exploring the potential throughput gains from different approaches of users’ selection strategies for MU-SIMO, downstream traffic (from AP to users) systems. We also explore the overhead required for CSI gathering for each of these user selection approaches and its trade-off with the attainable sum-rate gain.
Specifically, we analyze three approaches: the first approach relies on random selection of users for the next transmission opportunity (e.g., in a round robin fashion). The number of these randomly selected users matches the number of transmitting antennas (to attain the full degrees of freedom). CSIs are collected and data are being transmitted to these selected users according to their channel states. Note that this approach is similar to the common approach adopted by many standards such as IEEE 802.11, in which the users are chosen in a round-robin fashion. The second approach opportunistically selects a small subset of users with high channel gains, collects CSIs only from this small subset, and transmits to them, again based on their channel states. Note that this approach still collects CSIs from the scheduled users, similar to the first approach, yet the users are selected opportunistically based on their channel quality, and not randomly as in the first approach. Analyzing this approach affords us better understanding on the performance attained by selecting users opportunistically, solely based on their channel gain, without considering their channel correlation. The third approach also selects users opportunistically, yet instead of adopting the transmitted beam directions based on the selected users, the AP determines the transmission beams’ directions a priori and the users are selected distributively, based on the beam directions predefined by the AP. Note that since the beams are selected a priori, there is no need to collect CSIs from the selected users, only to estimate their expected Signal to Noise plus Interference Ratio (SINR)to adapt their respective rate, i.e., this approach compromises sum-rate as it chooses the transmission beams in advance, yet requires only minimal information exchange between the AP and the users (only SNR acquisition by the AP and not complete CSIs).
In order to analyze, get some insight and compare the performance of the three approaches on the same ground rules, we devise a distributed protocol for each suggested scheme. Specifically, in addition to the typical round-robin user selection procedure, we devise two additional prototype protocols. In the first, termed MINOS (MU-MIMO Interference-less Null-steering Opportunistic Scheduler), prior to each downstream transmission, each user estimates its channel gain based on the transmitted training sequences (pilots) sent by the AP. Users with preferable channel state (channel gain above a predefined threshold) signal the AP. Subsequently, CSIs are collected from these users in the same manner determined by the IEEE 802.11ac standard. The transmission to this set of users is performed utilizing Zero-forcing Beam-Forming (ZFBF). Note that MINOS’s user selection procedure solely relies on the channel gain between the AP and the selected receivers, overlooking the correlation (mutual orthogonality) between the selected users’ channels, which is expected to affect the attainable sum-rate. However, even though as far as overhead is concerned, MINOS is similar to the round-robin scheme as all the attending-for-transmission users are polled for CSI, in contrast to the randomly selected users, MINOS selects users opportunistically, and in particular the ones with high channel gains, hence is expected to attain higher throughput with similar overhead. In the second algorithm, termed DEMOS (Distributed Economic MU-MIMO Opportunistic Scheduling), before each transmission, the AP announces the beam directions in which it is going to transmit next (in fact it announces only a single direction and the users construct the rest of the directions accordingly). Based on these expected transmission directions, each user can estimate its expected SINR in each direction. Users with expected high SINR in one of the directions (SINR above a predefined threshold), signal the AP. In turn, the AP will transmit to these signaled users which are backlogged in their preferred direction. Note that DEMOS compromises on the attainable performance of each transmission opportunity as the transmission directions are chosen a priori without prior knowledge of the channel states of the selected users. Nonetheless, DEMOS requires negligible overhead, as the attending users are not polled for their CSI, which can potentially compensate for the reduction in transmission rates.
We thoroughly analyze the three scheduling algorithms, providing explicit expressions for the attained SINR. It is important to note that while current literature includes several scheduling and user selection algorithms for the above problem, a key contribution of this study is a detailed analysis of the performance under the suggested schemes when there is a moderate number of users. Specifically, using novel tools in algebra of random variables, we give exact expressions for the SINR under all three protocols, and discuss the trends and key insights one can gain from this analysis. We believe that the tools utilized to analyze MINOS and DEMOS are also applicable to other algorithms, hence will enable rigorous analysis of various other practical MU-MIMO procedures.