A deep learning-based Adaptive Data Rate algorithm for LoRaWAN networks

Abstract

LoRaWAN is emerging as a key protocol for several Internet of Things (IoT) applications, as it enables long-range communication with low power consumption between a large number of end-devices. LoRaWAN end-devices are characterized by a number of configurable transmission parameters, whose values need to be carefully selected, as they significantly influence the network performance. The Adaptive Data Rate (ADR) algorithm recommended by Semtech dynamically adjusts the transmission parameters of LoRaWAN end-devices to improve the network reliability while keeping energy consumption low. However, ADR is a rule-based algorithm not suitable for dynamic IoT scenarios in which the network conditions can be highly variable and the end-devices move around in the sensing area. In contrast, deep learning techniques appear a promising solution to set the transmission parameters of LoRaWAN end-devices in such dynamic IoT environments, thanks to their ability to learn from data a non-linear, state-dependent model. For this reason, this paper proposes a deep learning-based mechanism, called Rel-ADR, that dynamically tunes the transmission parameters of LoRaWAN end-devices to improve the transmission reliability, while maintaining low power consumption in dynamic and dense networks. The paper presents the design of Rel-ADR and the results of an extensive comparative performance evaluation between Rel-ADR and existing approaches in the literature, obtained through OMNeT++ simulations in realistic scenarios.