Deep Reinforcement Learning Based Medium Access Control Protocol for LoRa Networks

LoRa is a key enabling technology for large-scale Internet of Things (IoT) deployments due to its long-range and low-power capabilities. However, its pure ALOHA Medium Access Control (MAC) protocol suffers from high collision rates and poor scalability under dense network conditions. This paper proposes an intelligent MAC protocol based on Deep Q-learning (DQN) to enhance reliability and learning efficiency in multichannel LoRa networks. Where each node develops its own policy to select the optimal combination of channel, spread factor (SF), and transmission power (TxP), based on previous transmission outcomes, enabling distributed optimization without centralized coordination. The proposed DQN-based model is implemented using LoRaSim and compared against conventional Q-learning and ALOHA under identical network conditions. Simulation results demonstrate that intelligence-based MAC protocols maintained higher PDR than traditional ALOHA as the number of nodes increased. Moreover, DQN-based MAC achieved faster convergence, up to 40% fewer collisions compared to ALOHA, and about 25% compared to Q-learning across various network densities. These results confirm that Deep Reinforcement Learning(DRL) offers a scalable solution for intelligent MAC design in dynamic LoRa networks.