A Learning Automaton MAC Protocol for Directional FANETs With Throughput Enhancement and Fairness Control

Flying Ad Hoc Networks (FANETs) leveraging millimeter-wave (mmWave) communications and directional antennas offer high-capacity and low-latency connectivity for uncrewed aerial vehicles (UAVs). However, dynamic topologies, deafness, and hidden terminal problems make medium access control (MAC) protocol design challenging. Existing approaches, ranging from traditional contention-based schemes to learning-enabled methods, often suffer from inefficiency, high overhead, or excessive computational complexity. This paper proposes the Fairness-Aware Learning Automaton-based Directional MAC (F-LA-DMAC), a distributed protocol that maximizes spatial reuse while ensuring equitable channel access. We first establish a theoretical throughput upper bound using a conflict-graph model. We then design F-LA-DMAC, where each node updates its transmit/receive probabilities through local environmental feedback. A fairness mechanism further balances throughput and fairness by dynamically adjusting a tunable parameter. Simulations show that F-LA-DMAC achieves up to 86% higher throughput and 67% lower delay compared with deep reinforcement learning baselines (DDMAC and DDDMAC) in dense networks. Pareto frontier analysis reveals that F-LA-DMAC sustains a Jain’s fairness index of 0.96–0.98 while operating near peak throughput. We also analyze computational complexity and demonstrate that F-LA-DMAC incurs only linear runtime and memory overhead, making it suitable for embedded UAV platforms. In contrast, DDMAC and DDDMAC require costly neural inference and specialized hardware due to their polynomial complexity. Finally, FPGA-based experiments confirm the protocol’s feasibility in practice and validate its effectiveness in achieving the throughput–fairness trade-off.