Interference Suppression for Uplink Massive MIMO: A Trainable Integrated Receiver Design

Abstract

Interference suppression for uplink massive multiple-input multiple-output (MIMO) systems has been a vital topic in practice. We revisit this problem by advocating a trainable integrated interference suppression receiver design framework using both pilot and data symbols. Specifically, we develop semi-factorized and fully-factorized variational expectation maximization (VEM) solutions which efficiently employ the posterior statistics of the interference-plus-noise (IPN) covariance matrix and data symbols for performance enhancement. Then, a novel information entropy-based data filtering strategy is introduced to further improve the estimation accuracy of both channel and IPN covariance matrix. Finally, two model-driven networks are constructed by unfolding the proposed VEM algorithms to learn the optimal values of trainable parameters adaptively. Simulations verify the superior interference suppression capabilities of the proposed networks which can outperform the ideal interference rejection combining (IRC) receiver with known channel and IPN covariance matrix, and can also approach the ideal expectation propagation (EP) and variational Bayesian (VB) based receivers.