Energy-Efficient Semantic Communication for Aerial-Aided Edge Networks

Zheng, Guhan and Ni, Qiang and Navaie, Keivan and Pervaiz, Haris and Kaushik, Aryan and Zarakovitis, Charilaos (2024) Energy-Efficient Semantic Communication for Aerial-Aided Edge Networks. IEEE Transactions on Green Communications and Networking. ISSN 2473-2400

Text (Author accepted final version)
Author_accepted_final_version.pdf - Accepted Version
Available under License Creative Commons Attribution-NonCommercial.
Download (892kB)

Abstract

Semantic communication holds promise for integration into future wireless networks, offering a potential enhancement in network spectrum efficiency. However, implementing semantic communication in aerial-aided edge networks (AENs) introduces unique challenges. Within AENs, semantic communication strategically substitutes part of the communication load with the computation load, aiming to boost spectrum efficiency. This departure from traditional communication paradigms introduces novel challenges, particularly in terms of energy efficiency. Furthermore, by adding complexity, the use of a semantic coder based on machine learning (ML) in AENs encounters real-time updating challenges, further amplifying energy costs in these complex and energy-limited environments. To address these challenges, we propose an energy-efficient semantic communication system tailored for AENs. Our approach includes a mathematical analysis of semantic communication energy consumption within AENs. To enhance energy efficiency, we introduce an energy-efficient game-theoretic incentive mechanism (EGTIM) designed to optimize semantic transmission within AENs. Moreover, considering the accurate and energy-efficient updating of semantic coders in AENs, we present a game-theoretic efficient distributed learning (GEDL) framework, building upon the foundations of the renewed EGTIM. Simulation results validate the effectiveness of our proposed EGTIM in improving energy efficiency. Additionally, the presented GEDL framework exhibits remarkable performance by increasing model training accuracy and concurrently decreasing training energy consumption.