Semantic RF Waveform Adaptation in Flying Ad Hoc Networks via Hybrid Knowledge Bases and Deep Reinforcement Learning

By directly encoding the semantic meaning into each RF waveform symbol, we can significantly reduce the communication overhead. Particularly, semantic communication in Flying Ad Hoc Networks (FANETs)—formed by unmanned aerial vehicles (UAVs)—is a promising application for enabling efficient airborne networks. This paper introduces a semantics-aware approach to optimize OFDM (Orthogonal Frequency Division Multiplexing) waveforms, leveraging Deep Q-Network (DQN)-based reinforcement learning (RL) with nonlinear optimization, enhanced by an innovative hybrid knowledge base shared by all nodes. Our method incorporates <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">semantic importance</i> through adaptive parameter tuning and constrained optimization. The hybrid knowledge base allows for efficient representation and utilization of semantic information. We propose a framework that jointly optimizes modulation order and cyclic prefix length while considering the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">semantic importance</i> of transmitted data. The system is formulated as a Markov Decision Process (MDP) with scalable state/action spaces. This algorithm enhances the system’s ability to handle diverse and dynamic communication scenarios, adapting to varying semantic importance levels.