Deep Learning-Based Channel Adaptive Resource Allocation in LoRaWAN
Arshad Farhad, Daeho Kim, Jeong-Sun Yoon, Jae-Young Pyun
Abstract
Resource management of LoRa-enabled devices deployed on a large scale is challenging due to the underlying propagation conditions. LoRaWAN recommends an adaptive data rate (ADR) mechanism to manage the device resources such as spreading factor (SF) and transmission power. However, due to the sudden changes in the propagation environment, ADR cannot take appropriate measures to predict and take evasive measures to alleviate the massive packet loss caused by the unsuitable SF. Deep learning can be applied for resource management in LoRaWAN, allowing intelligence in the devices to learn the underlying propagation conditions and act intelligently by transmitting uplink packets with proper SF. This paper proposes a novel deep learning-based resource allocation method using a gated recurrent unit (GRU) to alleviate the packet loss caused by improper SF. The proposed LoRaWAN resource allocation has offline and online modes. By leveraging the pre-trained GRU model in the online mode, the proposed method allocates an appropriate SF for a device before uplink transmission, improving the packet success ratio by 11%.