Toward a Digital Twin of a LoRa Transmitter: a GPU-Based Parallel Simulation Approach

LoRa is a relevant communication technology that can be analyzed and improved by tools like Digital Twins and simulators. For these tools to be used effectively, it is desirable that they can operate on a large scale. This work aimed to implement a LoRa physical-layer simulation using TensorFlow and Sionna, exploring the parallel GPU simulation and the compatibility with tools such as Sionna Ray Tracing, AI and ML models. To validate the implementation, the obtained outputs were compared with a reference GNU Radio simulator, confirming functional correctness. Computational efficiency analyses evaluated processing time, resource usage (CPU, RAM, and GPU), and per-stage execution time. These analyses showed that the proposed simulator achieved the goal of exploring GPU parallelization, considerably reducing the processing time required in larger simulations.