Identification of Interfering Signals in Radios using Machine Learning Techniques
FM radios, detection of interfering signals, audio signal processing, spectral sensing, MFCC (Mel Frequency Cepstral Coefficients
This work offers a comprehensive view of the role and importance of radio in the Brazilian historical and commercial context. The problem of illegal radio stations stands out, which harm legal stations, reducing the quality of communication and interfering with the ability to reach more listeners. To deal with this challenge, it is proposed to use machine learning methods in conjunction with feature extraction techniques from audio signals to identify interference generated by other FM radios. In this work, interference signals were not treated simply as noise, with a clear differentiation between AWGN noise and interference from other radios. To this end, unique feature extraction techniques were explored, such as methods based on properly adapted spectral sensing, the MFCC method, first-order and extended-order statistical methods. Furthermore, strategies that use Autoencoder networks and Convolutional Neural Networks to classify radio signals that reach receivers were explored. For this study, under these conditions, solutions with baseband and passband signals were explored, as well as situations with multiple sources of interfering signals, so that the proposed models can deal with challenging scenarios. Finally, tests were carried out to validate the capacity of the proposed methods in computer simulation environments and in real environments, using Universal Software Radio Peripheral to generate signals that propagate through the communication channel.