Robust Frequency Selective Precoding for Downlink Massive MIMO in 5G Broadband System

In this article, we study the downlink precoding problem under imperfect channel state information (CSI) in 5G broadband system in which all subcarriers share one precoder. We propose a robust frequency selective precoding (RFSP) algorithm by maximizing the expected weighted sum-rate with the per-Antenna power constraints at the transmitter. To derive our algorithm, we first employ some approximation techniques to obtain a lower bound of the expected sum-rate, then transform it into the problem of minimizing the expected mean square error (MSE). The RFSP is obtained by applying the block coordinate descent (BCD) method over the precoding matrix, weighted matrix, and the hypothetical decoder matrix alternately. Furthermore, we propose unfolding the RFSP into a deep neural network (dubbed U-RFSP) to reduce its computational complexity. Experimental results show that the sum-rate achieved by the proposed RFSP is higher than other non-robust precoders and the computational complexity of the U-RFSP is substantially reduced compared with the RFSP while maintaining nearly the same performance.