Enhancing fairness in energy efficiency and peak age of information in low earth orbit-aided unmanned aerial vehicle-integrated sensing and communication networks: Attention-aided multi-agent proximal policy optimization approach

초록

The integration of unmanned aerial vehicles (UAVs) into space-air-ground integrated networks (SAGIN) has emerged as a promising paradigm for enabling flexible and efficient integrated sensing and communication (ISAC) systems. However, simultaneously optimizing energy efficiency (EE) and peak Age of Information (PAoI) fairness in such systems remains a critical challenge, particularly under the heterogeneous demands of non-orthogonal multiple access (NOMA) networks with partial observability and dynamic user environments. In this paper, we investigate the joint maximization of time-average EE and PAoI fairness in a UAV-aided ISAC system within SAGIN, formulating a multi-variable optimization problem over UAV trajectory, beamforming vectors, user equipment (UE) association, and mode selection, subject to energy budgets, minimum sensing requirements, quality of service guarantees, and UAV mobility constraints. The problem is reformulated as a decentralized partially observable Markov decision process (Dec-POMDP) and addressed by a proposed attention-aided multi-agent proximal policy optimization (A-MAPPO) algorithm. The attention mechanism is specifically designed to appropriately weigh different UEs’ contributions and cope with a varying set of observable UEs under partial observation. Performance evaluations demonstrate that A-MAPPO outperforms baseline methods including MADDPG, traditional MAPPO, Multi-Agent Dynamic QMIX (MADQMIX), and gradient search (GS) in joint EE and PAoI fairness optimization. Furthermore, analysis of the trade-off between time-average EE and PAoI fairness reveals that achieving PAoI fairness is generally more feasible than achieving EE fairness.

키워드