A CVNN-Aided Anti-Interference Channel Estimation for Massive MIMO Systems

This paper presents a complex-valued neural-network (CVNN) aided channel estimation method for massive MIMO systems, improving the channel state information (CSI) estimation under high power interference scenarios. Different from traditional interference mitigation techniques, which require the knowledge of channel statistics, the proposed CVNN method takes received pilots in the frequency-domain as the input, and outputs predicted channel coefficients. The CVNN is trained with synthetic signals generated by a Python-based massive MIMO simulator. The proposed method is evaluated on the channel estimation performance, the model generalization, and the impact of the frontend impairment. Besides evaluation, this paper also includes a detailed discussion of the advantage of using complex-valued network. The result shows that with the aid of CVNN, the channel estimation performance can be improved by 3.16× with shorter pilots compared to traditional channel estimation methods, and the proposed method is resilient to frontend impairments.