Deep Learning Enhanced Spectrum Sensing for LoRa Spreading Factor Detection

Over the last decade, there has been a significant increase in the demand for wireless sensor applications, in line with the increased connectivity of objects in the Internet of Things concept. As the number of wireless sensors continues to grow, there is a corresponding increase in the demand for spectrum resources which will eventually lead to overcrowding, packet collisions and interference ultimately resulting in a reduction of the wireless sensor network performance level. Obtaining a detailed understanding of the radio spectrum is critical in preventing collisions in wireless sensor networks. This requires knowledge of the level of spectrum occupancy and the specific radio modulation schemes employed, enabling the implementation of effective collision mitigation strategies. This paper proposes a LoRa spreading factor detection scheme based on Convolutional Neural Network (CNN) classification of spectrograms. Our results show that the developed algorithm shows a high degree of performance in detecting radio modulations, both in high SNR and low SNR conditions, resulting in a detection accuracy of over 98%.