Deep Learning Based Channel Estimation for Asymmetrical Full-Digital System

In asymmetrical full-digital systems where the number of uplink receive antennas is less than that of the downlink transmit antennas, the uplink-downlink channel inconsistency poses a challenge to channel estimation. To address this challenge, we employ the topology of nested array in the uplink receive array. Specially, a two step-angle estimation net (Ts-AEnet) method based on deep learning is developed to estimate the direction of arrival and a least square problem is solved to obtain path gain. Then the downlink channel can be reconstructed using those estimated parameters. Finally, simulation results confirm the effectiveness of the proposed method.