Search for: All records

LoRaWAN is a popular Long-Range Low-Power wireless communications protocol that is enabling many IoT applications worldwide, with more networks growing both in size and number around the world. To effectively plan and operate these networks, it is necessary to have tools that reliably quantify, measure, and predict the connection quality provided by LoRaWAN receivers. Being able to reliably quantify connection quality would allow LoRaWAN adopters to answer questions such as, “What does ‘good coverage’ mean?”. Reliably measuring coverage would allow for questions like “What is the quality of network coverage in a given area?”, to be answered, while predicting connection quality would allow adopters to answer questions such as “What would the coverage quality be if we deployed an additional wireless receiver in this location?” This paper proposes a novel data-driven approach to connection quality modeling that is tailored for LoRaWAN with the following features. First, connection quality is quantified by the packet reception rate (PRR), as opposed to the traditional received signal strength typical of generic radio planning tools. The PRR more closely captures what network operators and users ultimately care about. Next, we leverage a large set of original data to fit a model for PRR. This dataset is unique in two ways. First, it includes transmissions that were transmitted but not received by any gateway, eliminating an otherwise persistent source of bias in empirical estimates of wireless connectivity. Second, it includes features derived from high-fidelity terrain topology extracted from LiDAR point clouds. Our model includes both feature extraction and estimation. We evaluate our model out-of-sample, including in regions entirely disjoint from the training data, and show that it is considerably more accurate than common benchmark wireless propagation models. Finally, we demonstrate how our model can be used to provide coverage maps in a real-world network.