A Supervised Learning Approach to Adaptation in Practical MIMO-OFDM Wireless Systems

Abstract

MIMO-OFDM wireless systems require adaptive modulation and coding based on channel state information (CSI) to maximize throughput in changing wireless channels. Traditional adaptive modulation and coding attempts to predict the best rate available by estimating the packet error rate for each modulation and coding scheme (MCS) by using CSI, which has shown to be challenging. This paper considers supervised learning with the k-nearest neighbor (k-NN) algorithm as a new framework for adaptive modulation and coding. Practical k-NN operation is enabled through feature space dimensionality reduction using subcarrier ordering techniques based on postprocessing SNR. Simulation results of an IEEE 802.11n draft-compatible physical layer in flat and frequency selective wireless channels shows the k-NN with an ordered subcarrier feature space performs near ideal adaptation under packet error rate constraints.