Ensuring Reliability for LoRa Networks Using a Reinforcement-Learning-Assisted Time Division Duplex Protocol

Long range (LoRa) is one of the most widespread members of the low-power wide area network technologies family. Its ability to cover long distances combined with the robustness to radio frequency propagation effects make it an ideal solution in the deployment of low-cost sensor networks. However, two of the most critical issues that LoRa networks have to deal with are their scalability and reliability. The default medium access mechanism is the Pure-Aloha, which while ensuring low-energy consumption in end devices (EDs), due to its distributed and simple operation, does not offer any protection to packet collisions. As a result, the contention-based Pure-Aloha offers satisfactory operation in situations of low- and medium-offered load, while when the interconnected devices and the offered load increase, the number of collisions increases as well, resulting in a decrease of transmissions reliability. In this article we present a novel reinforcement learning (RL) assisted time division duplex protocol for LoRa networks. The operation of the proposed protocol is based on a combination of mechanisms and techniques that ensure the reliability required based on a predetermined value, combined with low-energy consumption for the EDs. The more important employed operating elements of the proposed protocol are the time division separation between uplink and downlink transmissions, the use of the channel activity detection mechanism, as well as the implementation of an RL mechanism to adjust basic parameters of the network operation. The effectiveness of the proposed protocol was evaluated through a number of simulation scenarios, where the results showed a significant improvement both in the reliability and fairness, as well as in the energy consumption of the EDs, compared to the existing and other relevant protocols from the literature.