Deep Reinforcement Learning-Based Joint Resource Allocation and Trajectory Design for UAV-Assisted Multi-Cell ISAC Systems

Abstract

The rapid development of multi-cell integrated sensing and communication (ISAC) systems requires new architectural and algorithmic solutions to tackle two key challenges: severe inter-cell interference that degrades communication quality for edge users, and the difficulty of maintaining reliable sensing performance in dynamic environments. This paper proposes an uncrewed aerial vehicle (UAV)-assisted multi-cell ISAC framework. Therein, the UAV acts as a flying base station to support terrestrial base stations in jointly satisfying users' communication and sensing requirements. To minimize the total transmit power while ensuring communication quality and sensing accuracy, a bi-level deep deterministic policy gradient-based joint resource optimization (DDPG-JRO) algorithm is developed. In the outer level, UAV trajectory optimization is formulated as a Markov decision process, where the trajectory decision for each time slot is updated through rewards from the inner level. In the inner level, user scheduling is determined based on the serving channel gain principle, and optimal power allocation is obtained through semidefinite relaxation and Gaussian randomization. The obtained inner-level solution is incorporated into the outer-level reward, thereby coupling power control, scheduling, and trajectory design within a unified learning framework. Numerical results validate the superiority and efficiency of the proposed DDPG-JRO algorithm.