Joint Beamforming Design and Sensing in Satellite and RIS-Enhanced Terrestrial Networks: A Federated Learning Approach
Sonia Pala, Keshav Singh, Chih–Peng Li, Octavia A. Dobre, Trung Q. Duong
Abstract
This paper presents a novel analytical framework for minimizing transmit power in satellite and terrestrial integrated networks using reconfigurable intelligent surface (RIS) technology within integrated sensing and communication systems. We employ a cutting-edge federated deep reinforcement learning approach, utilizing a federated deep deterministic policy gradient (F-DDPG) algorithm, to tackle the complex non-convex power minimization problem effectively. The approach leverages federated learning to dynamically adapt to network changes, ensuring compliance with beamforming designs, multiple target signal-to-interference-plus-noise ratio thresholds, and RIS phase-shift requirements through an effective feedback loop. In particular, we propose an F-DDPG algorithm that outperforms existing benchmarks such as the federated deep Q-network (DQN), centralized DDPG, and conventional DDPG and DQN methods. Through simulations, we demonstrate that integrating RIS significantly lowers base station (BS) power requirements against both random configurations and non-RIS setups. The optimal RIS configuration with 60 elements achieves a 6.3% reduction in BS transmit power compared to the random RIS scenario and a 34.2% reduction compared to the no-RIS setup. Additionally, our results demonstrate that increasing the number of RIS elements markedly improves sensing capabilities while maintaining the same level of transmit power.