Performance Analysis and Deep Learning Design of Short-Packet Communication in Multi-RIS-Aided Multiantenna Wireless Systems

Abstract

This paper studies short-packet communication (SPC) in multi-reconfigurable intelligent surface (RIS)-assisted multi-antenna wireless systems. In this system, a sensor node communicates with another sensor node through the help of an access point (AP) and two sets of distributed RISs. Aiming to enhance system performance, we combine the best RIS selection strategies with maximum-ratio transmission (MRT) beamforming designs to improve the transmitted signal and selection combining (SC) or maximum-ratio combining (MRC) to increase the received signals. Closed-form expressions for the block error rate (BLER) BLER, throughput, latency, and reliability of the receivers over Rayleigh fading channels are derived to evaluate the system performance. Numerical results show that, in the first transmission phase, employing MRC with optimal phase shift (OPS) occurring at RIS can help AP achieve the best BLER performance, while SC with OPS provides better BLER performance than MRC and uncertain phase shift (UPS). In the second transmission phase, the reflective-path beamforming design shows better performance than the direct-path beamforming design, where OPS also attains outstanding performance when compared to UPS. Aiming for real-time system configurations with high reliability along with minimizing costs and resources, we propose a deep learning (DL) approach to optimize the number of reflective elements at each RIS or the number of RIS in a set of distributed RISs. Our work shows that the prediction results of the DL framework match the analytical derivations and the deep neural network (DNN) can help the systems save overhead and resources while satisfying the requirement of real-time communication.