A Lightweight Decentralized Reinforcement Learning Based Channel Selection Approach for High-Density LoRaWAN

It is predicted that the number of IoT devices will reach 125 billion by 2030. The collisions among IoT devices will increase as the number of devices grows, which will reduce the communication performance of the IoT systems. To avoid collisions among high-density IoT devices, some researchers studied the solutions to adapt communication parameters of devices including channels to the environment. However, most of them are difficult to be applied to IoT devices with limited memory and computational ability. To solve the disadvantages of the existing work, we propose a lightweight decentralized reinforcement learning-based channel selection approach for high-density IoT systems in this paper. In our proposed approach, IoT devices can select appropriate channels only based on Acknowledge (ACK) information to avoid collisions among IoT devices with low computational complexity and memory requirement. Performance in terms of frame success rate (FSR) is evaluated by experiments using LoRa devices in the real world. Experimental results show that our proposed approach can efficiently avoid collisions in high-density LoRaWAN both with and without varying available channels compared to other lightweight reinforcement learning-based channel selection approaches.