Services Provisioning By Using Intelligent Learning For Long Range Wide Area Network (LoRaWAN)

Abstract

The exponential growth of Internet of Things (IoT) services and ecosystems recently emerged with a novel type of communication network known as Low Power Wide Area Network (LPWAN). This standard enables low-power long-range communication at a low data rate. Besides, Long Range Wide Area Network (LoRaWAN), is a recent standard of LPWAN that incorporates LoRa Wireless into a networked infrastructure. Consequently, Quality of Service (QoS) efficient service provisioning is a major challenge due to the highly dense network environment, the limited battery lifetime of LoRa-based End Devices (EDs), spectrum coverage, and data collisions. Intelligent and efficient service provisioning is a dire need of a network to streamline and address these problems. This study proposes a novel and Intelligent Learning (IL) based framework for efficient service provisioning without placing any extra burden on the network and its resource constraint LoRaWAN EDs. The proposed framework intelligently learns from varied underlying potential parameters such as real-time Packet Error Rate, data throughput, data delay, data collisions, and energy consumption to improve the overall network performance. The proposed framework is extensively simulated and evaluated with current state-of-the-art benchmark algorithms using standard and extended evaluation metrics. Slotted Aloha with Markov chain model mitigates collision and enhances the performance of LoRaWAN by 38% in terms of data throughput. Results of Slotted Aloha with Markov chain model are compared with Pure Aloha used by conventional LoRaWAN. Adaptive Scheduling Algorithm (ASA) with Gaussian Mixture Model (GMM) is extensively compared with conventional LoRaWAN and Dynamic PST (Priority Scheduling Technique). ASA with GMM enhanced performance in terms of delay by 5% in the LoRaWAN environment. Dynamic Reinforcement Learning Resource Allocation significantly reduced the energy consumption of EDs by 20% measured in Joules. Results of Dynamic Reinforcement Learning Resource Allocation is compared with conventional LoRaWAN and Adaptive Priority-aware Resource Allocation (APRA). The proposed work is properly cross-validated to utterly show unbiased results.