Training-Based Hybrid Precoding Scheme for Multiuser Massive MIMO-OFDM

Abstract

In this letter, we propose a training based hybrid precoding scheme for multiuser massive MIMO-OFDM time-division duplexing (TDD) systems, where we combine hybrid precoding design with channel estimation in a mixed-timescale structure. Firstly, we design a training based fully-digital precoder/combiner using the statistics of the transmitted and received signal under minimum mean square error (MMSE) criterion through an iterative process between downlink (DL) and uplink (UL), from which the radio frequency (RF) precoder/combiner is extracted using principal component analysis (PCA). Secondly, a low dimensional baseband (BB) channel is estimated and the BB precoder/combiner is designed by block diagonalization (BD) with the estimated BB channel. Thirdly, we discuss the transmission strategy for our proposed scheme in block-fading channels and limited feedback systems. Finally, we evaluate the sum spectral efficiency (SSE) and bit error rate (BER) performance of our proposed scheme and prove that our scheme has better performance compared to the existing schemes.