DRL-Based Power Optimization for Hybrid FSO/THz-Enabled UAV Systems Using IR-HARQ

Abstract

This letter addresses the transmit power minimization challenge in unmanned aerial vehicle (UAV)-assisted hybrid free-space optical (FSO)/terahertz (THz) systems integrated with incremental redundancy hybrid automatic repeat request (IR-HARQ), a cornerstone for 6G's ultra-reliable low-latency communications (URLLC). We propose a deep reinforcement learning (DRL)-driven framework, leveraging proximal policy optimization (PPO), to adapt power allocation across retransmissions via an agent-learned policy dynamically. This ensures reliable packet delivery under stringent delay bounds while accounting for channel impairments, including atmospheric attenuation, scintillation fading, and beam pointing errors. The system model incorporates SNR-based FSO/THz switching, with FSO as the primary link and THz as backup, evaluated through outage probabilities tailored to IR-HARQ, chase combining HARQ (CC-HARQ), and automatic repeat request (ARQ). Simulations across diverse environmental conditions reveal the proposed DRL-IR-HARQ hybrid achieves up to 0.7 dBm savings over THz-only baselines and conventional HARQ protocols, underscoring its robustness for energy-efficient 6G aerial backhauls and disaster-resilient networks.