LoRaWAN Localization Using Neural RSS and SNR Modeling

The patent introduces a neural network model as an intermediary between the received signal measurements (RSS, SNR) and location determination. This neural network processes the signal measurements and produces location estimates with uncertainty quantification, thereby improving location accuracy while maintaining relatively simple device implementation. The neural network acts as a mediator that transforms raw signal data into reliable location information even in non-line of sight conditions.

The patent implements feedback mechanisms where the neural network model is trained using ground truth location data and then deployed to make predictions. The system continuously refines its predictions by comparing estimated locations with actual locations (when available) and adjusting the model parameters accordingly. This feedback loop enables the system to improve location accuracy over time while maintaining computational efficiency.

The patent performs preliminary actions by pre-training the neural network model offline using extensive ground truth data. Once trained, the model is deployed to edge devices where it can make location predictions with high accuracy using only simple signal measurements. This preliminary training phase separates the computationally intensive work from the actual deployment, allowing low-power devices to achieve high accuracy without consuming excessive energy during operation.

The patent replaces traditional signal-based location methods (which rely on complex signal processing and multiple reference points) with a neural network-based approach. This substitution allows the system to achieve high location accuracy using simple RSS and SNR measurements, thereby reducing the computational power and energy consumption required for location determination while maintaining or improving accuracy.