Semantic-Aware Dynamic and Distributed Power Allocation: a Multi-UAV Area Coverage Use Case

TL;DR

This paper proposes a semantic-aware power allocation framework for multi-UAV networks using a novel deep reinforcement learning algorithm, improving observation quality and scalability in dynamic aerial-terrestrial communication scenarios.

Contribution

It introduces the SAMA-D3QL algorithm that incorporates semantic data quality into power allocation, advancing adaptive and scalable solutions for UAV-based network coverage.

Findings

SAMA-D3QL outperforms traditional methods in simulation.

The approach enhances observation quality in UAV networks.

The framework demonstrates high scalability and adaptability.

Abstract

The advancement towards 6G technology leverages improvements in aerial-terrestrial networking, where one of the critical challenges is the efficient allocation of transmit power. Although existing studies have shown commendable performance in addressing this challenge, a revolutionary breakthrough is anticipated to meet the demands and dynamism of 6G. Potential solutions include: 1) semantic communication and orchestration, which transitions the focus from mere transmission of bits to the communication of intended meanings of data and their integration into the network orchestration process; and 2) distributed machine learning techniques to develop adaptable and scalable solutions. In this context, this paper introduces a power allocation framework specifically designed for semantic-aware networks. The framework addresses a scenario involving multiple Unmanned Aerial Vehicles (UAVs) that collaboratively transmit observations over a multi-channel uplink medium to a central server, aiming to maximise observation quality. To tackle this problem, we present the Semantic-Aware Multi-Agent Double and Dueling Deep Q-Learning (SAMA-D3QL) algorithm, which utilizes the data quality of observing areas as reward feedback during the training phase, thereby constituting a semantic-aware learning mechanism. Simulation results substantiate the efficacy and scalability of our approach, demonstrating its superior performance compared to traditional bit-oriented learning and heuristic algorithms.