A Deep Learning-Based Approach for Channel State Information Estimation in Massive Multiple-Input Multiple-Output Networks

Abstract

Massive Multiple Input Multiple Output (M-MIMO) networks require accurate channel state information (CSI) estimation to enhance spectral efficiency and optimize wireless communication. However, traditional CSI estimation techniques suffer from high estimation errors, interference, and inefficient resource utilization, limiting network performance. The current challenge lies in improving CSI estimation accuracy while reducing feedback overhead and channel interference in M-MIMO networks. Existing methods struggle with high bit error rates (BERs) and suboptimal spectral efficiency, leading to degraded network performance. Hence, this paper proposes a Deep Learning-based Channel State Estimation and Feedback Optimisation (DL-CEFO) approach that leverages Deep Q-Network Reinforcement Learning (DQRL) for improved CSI accuracy, interference mitigation, and channel feedback overhead reduction. The findings show that DL-CEFO achieved 26.31% improvement in normalized mean squared error (NMSE), 26.31% increase in sum rate, and 77.29% BER reduction, significantly enhancing M-MIMO network reliability, when compared with the Compressive-Sensing and Learning-to- Optimise (CSI-L2O) approach. The DL-CEFO outperforms CSI-L2O by dynamically optimizing CSI feedback and interference mitigation.