Frequency Selective Hybrid Precoding Based on Adaptive Gradient Algorithm in mmWave Systems

Abstract

Hybrid precoding can combat severe attenuation of the millimeter wave (mmWave) link by leveraging large-scale antenna array, while permitting practicable circuits with low power consumption and hardware cost. Although existing near-optimal algorithms have approached the performance of fully-digital precoding, their complexities are still very high. In this paper, we reconsider the problem of frequency selective hybrid precoding and propose an equivalent neural network architecture of point-to-point hybrid precoding for orthogonal frequency division multiplexing (OFDM) multi-input multi-output (MIMO) systems. Under this new architecture, the elements of the digital-and analog- precoders can be regarded as the connecting weights of a single hidden layer neural network. Inspired by the backpropagation (BP) algorithm in feedforward neural networks, we propose an adaptive gradient (AG)-based BP algorithm for hybrid precoding in this new architecture. The numerical simulation results demonstrate that the proposed algorithm can achieve the performance of the unconstrained fully-digital precoding with lower complexity compared with the the existing near-optimal alternating minimization algorithms.