Multi-Agent Q-Learning Algorithm for Dynamic Power and Rate Allocation in LoRa Networks

Abstract : In this paper, we consider a Low Power Wide Area Network (LPWAN) operating in a licensed-exempt band. The LoRa network provides long-range, wide-area communications for a large amount of objects with limited power consumption. In terms of link budget, nodes that are far from the collector suffer collisions caused by nodes that are close to the collector during the data transmissions. Chirp Spread Spectrum (CSS) modulation is adopted by assigning different Spreading Factors (SF) to active sensors to help reduce destructive collisions in LoRa network. In order to improve the energy efficiency and communication reliability, we propose an application of multi-agent Q-learning algorithm in the dynamic allocation of power and SF to the active nodes for uplink communications in LoRa. The main objective of this paper is to reduce power consumption of the uplink transmissions and to improve the network reliability.