Learning based link adaptation in multiuser MIMO-OFDM

Link adaptation in multiple user multiple-input multiple-output orthogonal frequency division multiplexing communication systems is challenging because of the coupling between user selection, mode selection, precoding, and equalization. In this paper, we present a methodology to perform link adaptation under this multiuser setting, focusing on the capabilities of IEEE 802.11ac. We propose to use a machine learning classifier to solve the problem of selecting a proper modulation and coding scheme, combined with a greedy algorithm that performs user and spatial mode selection. We observe that our solution offers good performance in the case of perfect channel state information or high feedback rate, while those scenarios with less feedback suffer some degradation due to inter-user interference.