Supervised-Learning-Aided Communication Framework for Massive MIMO Systems With Low-Resolution ADCs

This paper considers a massive multiple-input-multiple-output (MIMO) system with low-resolution analog-to-digital converters (ADCs). In this system, inspired by supervised learning, we propose a novel communication framework that consists of channel training and data detection. The underlying idea of the proposed framework is to use the input-output relations of a nonlinear system, formed by a channel and a quantization at the ADCs, for data detection. Specifically, for the channel training, we develop implicit and explicit training methods that empirically learn the conditional probability mass functions (PMFs) of the nonlinear system. For the data detection, we propose three detection methods that map a received signal vector to one of the indexes of possible symbol vectors, according to the empirical conditional PMFs learned from the channel training. We also present a low-complexity version of the proposed framework that reduces a detection complexity by using a successive-interference-cancellation (SIC) approach. In this low-complexity version, a symbol vector is divided into two subvectors and then these two subvectors are successively detected using SIC. When employing the proposed framework with one-bit ADCs, we derive an analytical expression for the symbol-vector-error probability. One major observation is that the symbol-vector-error probability decreases exponentially with the inverse of the number of transmit antennas, the operating signal-to-noise ratio, and the minimum distance that can increase with the number of receive antennas. Simulations demonstrate the detection error reduction of the proposed framework compared to existing detection techniques.