Learning-Based Joint Channel Prediction and Antenna Selection for Massive MIMO with Partial CSI

This paper investigates the massive multi-input multi-output (MIMO) system in practical deployment scenarios, in which, to balance the economic and energy efficiency with the system performance, the number of radio frequency (RF) chains is smaller than the number of antennas. The base station employs antenna selection (AS) to fully harness the spatial multiplexing gain. Conventional AS techniques require full channel state information (CSI), which is time-consuming as the antennas cannot be simultaneously connected to the RF chains during the channel estimation process. To tackle this issue, we propose a novel joint channel prediction and AS (JCPAS) framework to reduce the CSI acquisition time and improve the system performance under temporally correlated channels. Our proposed JCPAS framework is a fully probabilistic model driven by deep unsupervised learning. The proposed framework is able to predict the current full CSI, while requiring only a historical window of partial observations. Extensive simulation results show that the proposed JCPAS can significantly improve the system performance under temporally correlated channels, especially for very large-scale systems with highly correlated channels.