Power Efficient Resource Allocation for ISAC: Combing Lyapunov Optimization and DRL

Radar and wireless communication systems are two important applications of modern electromagnetic theory. However, the serious shortage of spectrum resources in recent years poses a significant challenge for the development of both systems. Integrated sensing and communication (ISAC) has emerged as...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2023 IEEE Globecom Workshops (GC Wkshps) S. 1135 - 1140
Hauptverfasser: Wang, Haodong, Wang, Zifan, Chen, Yawen, Lu, Zhaoming, Wen, Xiangming
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 04.12.2023
Schlagworte:
Computational modeling
Deep Reinforcement Learning
Integrated Sensing And Communication
Lyapunov Optimization
Network Stability
Power demand
Radar detection
Resource Allocation
Simulation
Stability analysis
Wireless communication
Wireless sensor networks
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Radar and wireless communication systems are two important applications of modern electromagnetic theory. However, the serious shortage of spectrum resources in recent years poses a significant challenge for the development of both systems. Integrated sensing and communication (ISAC) has emerged as a promising solution to address this issue. Due to shared infrastructure and spectrum, optimal resource allocation is essential to meet the quality requirements of both sensing and communication, while maximizing the system efficiency. This paper presents a combinatorial online optimization framework based on Lyapunov theory and deep reinforcement learning (DRL) for a typical ISAC system with time-varying wireless channels and random user arrivals. The framework aims to minimize power consumption while ensuring the performance of both communication and radar detection. It first applies Lyapunov optimization to decouple a long-run stochastic mixed-integer nonlinear programming problem into deterministic subproblems in each time frame. Subsequently, it combines model-based optimization with model-free DRL to solve these subproblems. Simulation results demonstrate that the proposed framework can stabilize the data backlog queue within a short computation time and significantly reduce power consumption.
DOI:10.1109/GCWkshps58843.2023.10464593