A Hybrid Deep Learning Spectrum Sensing Architecture for IoT Technologies Classification

In recent years, there has been a significant expansion of the Internet of Things concept, because of its versatile and extensive range of applications across various industries such as transportation, logistics, smart metering, or agriculture. The rapid expansion of Internet of Things networks has resulted in a limitation of radio frequency spectral resources, giving in turn a substantial rise in packet collision rate, interference level, bit error rate parameters and reducing the communication channel capacity altogether. Employing spectrum sensing techniques can aid in overcoming these challenges by providing key information on the radio spectrum occupancy and the modulation schemes used. This data can be leveraged to develop collision avoidance schemes, which can enhance the performance level of wireless networks increasing the network capacity. This paper proposes and evaluates a hybrid deep learning spectrum sensing technique for simultaneous detection and classification of LoRa and Sigfox Internet of Things communication technologies in the 868 MHz frequency band. The developed hybrid deep learning spectrum sensing architecture is capable of effectively detect and classify multiple IoT technologies represented by LoRa communications using different spreading factors and Sigfox within the 868 MHz frequency band, achieving an average precision of 85.51 % and an average intersection over union metric of 0.77.