Abstract
This paper studies an integrated sensing and communication framework, in which an unmanned aerial vehicle (UAV) concurrently serves mobile users and sensing targets with the assistance of a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS). To analyze the performance of this system, an admission control problem is formulated that aims to maximize the number of served sensing targets. Due to tight coupling and its non-convex nature, the problem is transformed to a Markov decision process (MDP) form, based on which a recurrent deep deterministic policy gradient (RDPG) agent is trained to jointly optimize the UAV flight trajectory, STAR-RIS coefficients, as well as the transmit and receive beamforming at the transceivers. Concerning the frequent displacement of the UAV and thus the considerable dynamism of the system, we further enrich the trained RDPG model for better adapting to the system variations by integrating a meta-learning technique. Numerical results exhibit at least 30% enhancement in average admission rate of sensing targets with the assistance of STAR-RIS. Additionally, the proposed adaptive resource allocation scheme brings about 25% superiority in average, over the existing soft actor-critic (SAC) counterpart available in the literature.
Funding Information
  • European Union through the Horizon Europe Project COVER (101086228)
  • Horizon Europe Project UNITE (101129618)
  • Horizon Europe Project INSTINCT (101139161)
  • Horizon Europe project TWIN6G (101182794)
  • Agence Nationale de la Recherche (NF-YACARI 22-PEFT-0005)
  • Agence Nationale de la Recherche (CHIST-ERA-22-WAI-04)
  • Agence Nationale de la Recherche (ANR-23-CHR4-0003-01)

This publication has 25 references indexed in Scilit: