Deep Reinforcement Learning based Resource Allocation in Dense Sliced LoRaWAN Networks

Long-Range Wide Area Network (LoRaWAN) is a rapidly expanding communication system for Low Power Wide Area Network (LPWAN) in the Internet of Things (IoTs) deployments. It employs an Adaptive Data Rate (ADR) scheme that optimizes data rate, airtime, and energy consumption. Recently, the use of Network Slicing (NS) in LoRa Wannetworks is being widely studied and a hot topic for the latest research in the literature. Network resources must be efficiently assigned to IoT devices in an isolated manner in order to handle and support specific Quality of Service (QoS) requirements for each slice. However, in dense LoRaWAN networks, the ADR scheme is insufficient for efficient resource allocation to meet the QoS requirements of each slice. In this article, we propose a DRL-based approach for intra-slicing resource allocation in dense LoRa Wannetworks. In each slice, we implemented multi-agent DRL that allocates Spreading Factor (SF) and Transmission Power (TP) to IoT devices to meet QoS requirements, i.e. we replaced the conventional ADR scheme with multi-agent DQN with different reward function design for each slice according to QoS requirements. Experimental results realized in real conditions show that our approach outperforms the existing ADR scheme for all the slices.