Cognitive UAV-IRS planning for semantic-aware mobile edge computing networks

Abstract

Unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) networks offer a powerful solution for enhancing communication efficiency in resource-constrained environments and managing compute-intensive tasks. However, the inherent limitations of UAVs, such as restricted data storage, computation capability and battery capacity, hinder the maximum communication efficiency. For the first time, this paper investigates a semantic-aware mobile edge computing (SMEC) network, where task data is semantically compressed at the users and processed at edge computing servers. This approach aims to significantly reduce the transmission and storage overhead in UAV, and improve task performance in low signal-to-noise ratio (SNR). To further enhance transmission robustness and task performance, we incorporate a UAV-carried mobile intelligent reflecting surface (IRS). The objective is to minimize system costs while maintaining task performance, which requires the joint optimization of UAV trajectories, server pairings, user assignments, and IRS reflecting elements. This problem is NP-hard, posing significant computational challenges. To address the complexity of the formulated problem, we propose a novel cognitive UAV-IRS planning strategy based on deep reinforcement learning (DRL), where the UAV can infer the task intentions of the users. Simulation results demonstrate the effectiveness of our intelligent scheme, showing rapid convergence in solving the complex optimization problem. Comparative analysis with benchmark schemes reveals a substantial reduction in system costs and more robust task performance achieved by our proposed approach.