Link adaptation in MIMO-OFDM with non-uniform constellation selection over spatial streams through supervised learning

Supervised learning has been used to develop practical link adaptation algorithms for MIMO-OFDM under an equal rate per stream assumption. In this paper we develop supervised learning algorithms that select from non-uniform rates per stream. We show that the straightforward application of existing supervised learning link adaptation algorithms exhibits complexity that scales with the number of spatial streams. Therefore, we propose a decoupled stream link adaptation algorithm which reduces the complexity below the original supervised learning algorithm with uniform spatial streams. We further show that the performance loss of decoupled link adaptation is reduced in systems with non-uniform constellations per spatial stream. IEEE 802.11n and uncoded MIMO-OFDM simulations are used to validate the proposed algorithms.