Pearl - A More Reliable LoRaWAN

Abstract

LoRaWAN is the dominant unlicensed low-power wide-area networking (LP-WAN) technology for large-scale IoT deployments, celebrated for its potential to offer decade-long battery life. However, in real-world urban environments, highly dynamic wireless channels degrade this promise as their high variance lead to significant packet loses. These losses are often caused by just a few incorrectly decoded bits, despite the presence of error-correcting codes, resulting in disproportionate retransmission overhead and wasted energy. This thesis introduces Probabilistic Error-Aware Repair of LoRaWAN (PEARL), a new decoding architecture that reduces packet failure and improves communication efficiency in LoRaWAN. Pearl estimates the posterior probability that each received symbol is correct by modeling the channel as AWGN and using the Rician distribution, then converts these symbol-level probabilities into bit-level likelihoods. Using these, Pearl probabilistically selects the most likely transmitted codeword, enabling correction of multiple bit errors beyond what traditional forward error correction alone allows. Evaluated on a campus-scale deployment covering 0.18 km², Pearl improves client battery life by 1.3à , achieves 2.045à higher throughput, and reduces latency by 52.7% compared to standard LoRaWAN implementations.