Randomized Two-Timescale Hybrid Precoding for Downlink Multicell Massive MIMO Systems.
Xihan Chen,An Liu,Yunlong Cai,Vincent K. N. Lau,Min-Jian Zhao
DOI: https://doi.org/10.1109/tsp.2019.2926017
IF: 4.875
2019-01-01
IEEE Transactions on Signal Processing
Abstract:Although massive multiple-input multiple-output (MIMO) promises high spectral efficiency, there are several issues that significantly limit the potential gain of massive MIMO, such as severe inter-cell interference, huge channel state information (CSI) overhead/delay, high cost and power consumption of RF chains, and user fairness. Several precoding schemes have been proposed for massive MIMO but there still lacks a systematic solution to address all of the aforementioned issues especially for the multicell case. In this paper, we propose a randomized two-timescale hybrid precoding (RTHP) scheme for the downlink transmission of multicell massive MIMO systems. RTHP allows time sharing among multiple sets of analog precoders to achieve better tradeoff between sum throughput and fairness. Moreover, the time-sharing factors and analog precoders are adapted to the long-term channel statistics to mitigate inter-cell interference with reduced CSI overhead and RF chains. The digital precoder is used to support multiuser MIMO at each base station based on local low-dimensional effective CSI only. As such, RTHP resolves all of the aforementioned issues. We formulate the optimization of RTHP as a general network utility maximization problem (GNUMP), which is a two-timescale stochastic non-convex problem, and consequently challenging to solve. By exploiting some approximation techniques, we convert the problem into a more tractable form and subsequently develop an online algorithm, which is shown to converge to stationary solutions of the approximate GNUMP. Finally, simulation results verify the advantages of the proposed scheme over baseline schemes.