Wireless Technology Identification Using Continuous Wavelet Transform and Deep Learning

The identification of wireless technology is vital due to the rising use of wireless devices and the coexistence of multiple technologies. It enables dynamic spectrum access, facilitating efficient spectrum sharing among multiple users and applications within the same frequency bands, optimizing utilization, and minimizing interference. This study proposes a novel wireless technology identification approach that utilizes continuous wavelet transform (CWT) and a lightweight convolutional neural network (CNN). First, we use the CWT to generate the spectrogram images from the in-phase and quadrature (IQ) samples collected from various communication technologies. These spectrogram images are then fed to CNN to identify the coexistence of different communication technologies. The proposed model achieves an accuracy of 90.6%, surpassing the benchmark models. In addition, our model has a compact size of only 1.04 MB, making it notably lighter than other models. We further evaluate the time it takes for our model to make inferences on a Raspberry Pi 5 and other hardware, demonstrating its efficiency and practicality for real-world deployment.