Exploiting Channel Spatial Feature Extraction for Low Complexity R-WMMSE Precoder Design

The increasing scale of antenna arrays in massive multiple-input multiple-output (MIMO) systems poses a significant challenge to computation-affordable downlink precoding design. In this paper, we focus on designing low complexity precoding algorithm for the classic non-convex weighted sum rate (WSR) maximization problem. By performing spatial feature extraction on the MIMO channel and selecting a proper number of data streams for multiplexing, we develop two novel WSR-oriented precoders, namely the channel feature reduced weighted minimum mean squared error (CF-RWMMSE) and the channel feature enhanced R-WMMSE (CFE-RWMMSE). The two precoding methods are obtained in the form of iterative algorithms with closed-form updates, and their convergence is readily ensured. Importantly, as the dimension of the precoder is significantly reduced, the proposed precoder designs achieve an $\mathcal{O}\left(D^{2} / N^{2}\right)$ complexity reduction over the classic R-WMMSE precoder, where D and N denote the total number of data streams and receive antennas, respectively. Numerical results show that the proposed precoders can achieve a WSR performance close to that of R-WMMSE in perfect channel acquisition case, and even outperform the R-WMMSE in aging channel case.