Are Existing Analytical Models for LoRa Networks Accurate?

Long range (LoRa) is one of the most flourishing low-power wide-area (LPWA) technologies, designed to satisfy the requirements of IoT (Internet of Things) devices. Several researches have recently proposed complex analytical models encompassing both single-cell and multi-cells LoRa networks, and that can be used for scalability evaluation or network dimensioning. These models rely on either stochastic geometry tools or enhanced modelling of Aloha channel access, to derive closed-from expressions of LoRa performance metrics such as packet delivery ratio (PDR) or data extraction rate (DER). However, the obtained expressions need to be validated by realistic simulations as pointed out by some of the authors themselves [1]. We also noted that different models seem to yield different results. In this paper, we briefly introduce the two main analytical modelling approaches and summarize their mathematical results. We then proceed to fill the validation gap by customizing the packet-level simulator LoRaSim and running extensive simulations in order to assess the accuracy of these models. Our study shows that the stochastic geometry-based models provide accurate DER values that match the simulation results.