ADRL: A reconfigurable energy-efficient transmission policy for mobile LoRa devices based on reinforcement learning

Abstract

LoRa is a wireless communication technology for low-power and long-range Internet of Things (IoT) networks. Although it was not initially designed for mobile IoT devices, its applicability has quickly expanded to support mobility. LoRaWAN, its network protocol, includes an Adaptive Data Rate (ADR) mechanism to minimize energy consumption while optimizing data rates and Time-on-Air values. However, ADR struggles to adapt effectively in highly dynamic and unpredictable environments, leading to slow convergence and performance degradation. These challenges are particularly evident in scenarios with frequent changes in signal conditions, such as urban areas with non-line-of-sight (NLoS) or environments with rapid channel variations. To address these limitations, we introduce ADRL, a novel transmission mechanism that proactively selects the most suitable transmission configuration of LoRa devices under dynamic and variable conditions. To this end, the proposed method (i) forecasts the received signal strength of end devices using a lightweight k-nearest neighbors (KNN)-based approximation, and (ii) employs deep reinforcement learning to create a reconfigurable controller on the network server side for selecting the transmission parameters that minimize energy consumption and ensure a given probability of successful packet decoding. Moreover, we leverage duty-cycle information to maximize the delivery ratio and comply with the associated regulations. We compare our method to state-of-the-art works using a realistic simulator in various urban environments. The results show that ADRL effectively balances energy consumption and performance requirements, reducing average energy expenditure by up to 15% and increasing the packet delivery ratio by up to 57%.