Semantic-AI-Based Trajectory Design of Multiple UAV Base Stations in Sparse and Mobile User Environments

Designing an efficient and equitable communication service policy for sparsely distributed mobile users across extensive areas poses a considerable challenge in the field of trajectory planning for multiple Uncrewed Aerial Vehicles (UAV) Base Stations (BS). The challenge arises due to the dispersed nature of User Terminals (UTs) and the restricted sensor range of the UAVs, which frequently results in overlooking the communication requirements of certain edge users. In response to this challenge, a fairness model has been proposed to prioritize edge users and ensure a balanced user experience. Furthermore, an innovative UAV-BS cooperation algorithm has been introduced to effectively manage sparse observation features and enhance the UAV-BSs’ understanding of the environment through a node-level attention mechanism and a semantic-level aggregating mechanism. Additionally, the proposed enhances coordination among UAV-BSs through a CTDE (Centralized Training with Decentralized Execution) method. The simulation results demonstrate that the proposed algorithm outperforms the state-of-the-art methods up to 36% in communication rate and 33% in fairness.

I. Introduction

In recent years, the substantial benefits of low cost, extensive coverage, and adaptable deployment [1] have positioned UAV base stations (UAV-BSs) as pivotal components in emergency rescue scenarios. As such, research has increasingly concentrated on multiple UAV-BS deployments to facilitate dynamic coverage of the target area and augment the data rate of the communication system. For instance, cooperative trajectory design for multiple UAVs is conducted to maximize throughput while upholding service fairness. In [2], [3], and [4], UAV trajectory and resource association are jointly optimized. To investigate the optimal cooperative policy for multiple UAVs, certain studies have employed Multi-Agent Deep Reinforcement Learning (MADRL) algorithms [5], [6].