Integrated Sensing and Communication for Energy-Efficient Low-Altitude Economy Based on Meta Deep Reinforcement Learning

This paper investigates a new class of terrestrial integrated sensing and communication (ISAC) schemes that support the energy-efficient low-altitude economy (LAE) applications based on meta deep reinforcement learning (DRL). Specifically, we consider an LAE-oriented ISAC system, in which a ground base station (GBS) provides communication and navigation services for authorized unmanned aerial vehicles (UAVs), as well as senses the low-altitude airspace to monitor the unauthorized mobile target. By jointly optimizing the GBS's beamforming and authorized UAVs' trajectories, we aim to maximize the expected energy efficiency over a given flight period, subject to constraints including the average sensing signal-to-noise ratio requirement, UAV mission completion and collision avoidance, and the maximum transmit power at the GBS. To circumvent this sequential decision-making problem, we first put forth a new ISAC scheme for LAE, termed Deep Energy-efficient LAE-ISAC (DeepESC), leveraging DRL techniques. In DeepESC, a new reward function and a constrained action selection policy are introduced to satisfy various constraints. Also, an episodic experience replay mechanism is developed to derive efficient joint optimization strategies across the whole flight period. The gist is to employ experience sets each containing all experiences from a complete episode to jointly train the neural network, assign priorities to these sets, and more actively sample newer experience sets. Besides, to enable rapid adaptation to unseen flight periods with fewer experience samples, we propose the Meta-DeepESC scheme by integrating a new meta-learning approach into DeepESC. Simulation results demonstrate that compared with other schemes, Meta-DeepESC (i) achieves higher energy efficiency while satisfying all constraints, (ii) converges faster, and (iii) exhibits better generalization across different flight periods.