Development of optimization techniques for wireless communication systems

Abstract

In this study, we investigate advanced optimization strategies for fully connected reconfigurable intelligent surface (FC-RIS) assisted downlink communication systems operating under ultra-reliable low-latency communication (URLLC) constraints. Our focus is on maximizing the finite blocklength (FBL) achievable rate in a multi-user multiple-input single-output (MU-MISO) setting, where joint optimization of beamforming vectors, RIS phase shifts, and blocklength allocation is critical. We address the complexity of the problem arising from high-dimensional search spaces and the presence of Rician fading. Our investigation begins with a deep learning-based framework that employs gradient-based optimization to handle the non-convexity of the system. We then extend this approach by exploring reinforcement learning methods and employ the twindelayed deep deterministic policy gradient (TD3) algorithm to jointly optimize active beamforming and blocklength under FC-RIS-aided URLLC scenarios. Unlike earlier works that primarily focused on single connected RIS or standard DDPG methods, our TD3-based approach adapts effectively to the dynamic environment and achieves superior rate and reliability outcomes. In the final part of our work, we propose a hybrid model that integrates convolutional neural networks (CNN) and long short-term memory (LSTM) networks for initial configuration prediction, followed by refinement using the Successive Convex Approximation (SCA) method. These models demonstrate significant performance gains in terms of throughput and resource utilization compared to traditional techniques. Together, these results affirm the promise of intelligent optimization techniques in next-generation RIS-assisted wireless systems.