An Anomaly Detector Using Filtering Stockwell Transform and Siamese Convolutional Neural Network in Radio Monitoring (AnoFSTSCNN)

Abstract

Detecting anomalous behavior in the radio spectrum is a demanding task due to high levels of interference behaviors caused by massive wireless devices. Stockwell transform (ST) has grown in popularity for the analysis of nonstationary and nonlinear signals, but it is yet to be adequately explored in the radio monitoring domain. An approach that consists of FST (filtering Stockwell transform) and SCNN (siamese convolutional neural network) for radio spectrum anomaly detection (AnoFSTSCNN) is proposed in this paper. Four types of anomaly behaviors, including tone, chirp, pulse, and noise frequency modulation (FM) are simulated using MATLAB. Simulation results show that the AnoFSTSCNN has a 13.18% and 29.52% improvement in detection performance over short-time Fourier transform (STFT) and ST, respectively. The proposed approach provides a possible reference solution for radio monitoring and anomalous signal detection.