Adaptive Integration of Distributed Deep Q-Networks for Enhancing OLSR Routing in Dynamic Mobile Ad-Hoc Networks

Adaptive routing in Mobile Ad-Hoc Networks (MANETs) poses considerable difficulty owing to the network's dynamic characteristics, lack of stable infrastructure, and swift topology alterations. The Optimized Link State Routing (OLSR) protocol provides a proactive routing mechanism via topology dissemination and MultiPoint Relay (MPR) selection. Nevertheless, it exhibits diminished responsiveness to real-time topology alterations, as it depends on periodic updates and does not explicitly account for link quality. This paper suggests the incorporation of the Deep Q-Network (DQN) methodology into OLSR as a reinforcement learning strategy to improve routing adaptability and efficiency. The DQN model employs network metrics like latency, ETX, buffer occupancy, and neighbor count as state inputs, with actions determined by Q-values obtained via environmental interactions. Simulations conducted with NS-3 and PyTorch demonstrate that OLSR-DQN enhances Packet Delivery Ratio (PDR) by as much as 20%, decreases delay by 15–25%, and markedly boosts throughput in dynamic MANET situations. Keywords: MANET, OLSR, Deep Q-Network, adaptive routing, reinforcement learning