Machine Learning-Based Channel Estimation for Multi-RIS-Assisted mmWave Massive-MIMO OFDM System in a Dynamic Environment

Abstract

Wireless communication in high obscurity environments is not effective especially for millimeter-wave (mmWave) frequencies due to propagation challenges. It is therefore, necessary to deploy multiple reconfigurable intelligent surfaces (multi-RISs) and massive multiple-input multiple-output (massive-MIMO) systems to circumvent the propagation challenges in such environments for effective communication. However, deploying massive-MIMO multi-RIS system increases the dimensionality of the channel, and using conventional channel estimation (CE) methods for estimating this multi-RIS-assisted channel is infeasible. In literature, machine learning (ML)-based CE methods have been proven to yield more accurate CE results for massive-MIMO RIS-assisted systems. Thus, in this paper, the ML-based CE method called the denoising convolutional neural network-gated recurrent unit (DnCNN-GRU) scheme is proposed for estimating the uplink cascaded time-varying massive-MIMO multi-RIS-assisted channel. The proposed scheme was then benchmarked with other CE methods to prove its superiority. It achieved a performance closer to the lower bound oracle least square (LS) estimate with about 1 dB gap in terms of the normalised mean squared error (NMSE).