Multi-Objective ISAC for Underwater Acoustic Networks: A Meta Deep Reinforcement Learning Approach

This paper studies an underwater acoustic (UWA)-based integrated sensing and communication (ISAC) system, in which an autonomous underwater vehicle (AUV) transmits collected environmental data to a buoy and simultaneously senses the unauthorized mobile target. The objective is to maximize both the expected communication rate and the expected sensing signal-to-noise ratio, and to minimize the expected power consumption over a given navigation period, through joint optimization of the AUV’s beamforming and trajectory. Constraints including the navigation mission, collision avoidance, the AUV’s maximum transmit power, and the effects of long propagation delays on signal transmission are explicitly incorporated. To address this problem, we propose an intelligent joint optimization scheme called meta multi-objective policy gradient (Meta-MOPG). Specifically, Meta-MOPG first introduces a double-critic-based deep deterministic policy gradient algorithm to support continuous-domain decision-making. An objective-preference vector is then incorporated to capture the trade-off among communication, sensing, and power consumption performance. To enhance training efficiency, we further put forth a multi-step experience replay mechanism that evaluates each joint decision by considering both immediate and future rewards. In addition, a meta-learning approach is developed to improve the generalization ability of Meta-MOPG across different objective-preference vectors. Simulation results show that Meta-MOPG yields more Pareto-efficient solutions than other schemes without acquiring prior knowledge of the target’s mobility pattern, while maintaining stronger robustness across various simulation setups. We also demonstrate the generalization ability of Meta-MOPG for unseen tasks.