Multi-Frequency Mutation Detection in Broadband Electronic Signals Using SVD Denoising, Improved Mask EMD, and PSO-Optimized SVM

Abstract

When wideband electronic signals undergo multi-frequency mutations, the mutation times, amplitudes and phase changes of each frequency component are different, making it extremely difficult to accurately distinguish the characteristics of the mutated signals from the complex signal background. To this end, a detection method integrating the improved mask EMD and PSO-SVM is proposed. The singular value decomposition method is adopted to denoise wideband electronic signals, and the steady-state components and abrupt components of the signals are analyzed. The improved mask EMD method is adopted to extract the intrinsic modal function components of each order, extract the multi-frequency instantaneous frequencies and instantaneous energies of the corresponding components of the wideband electronic signal, construct the multi-frequency mutation feature set, and input it into the PSO-SVM detection model to capture the steady-state components and mutation components in the signal, and realize the multi-frequency mutation detection of wideband electronic signals. The experimental results show that the research method adopts the combined processing of improved mask and EMD, which reduces the signal reconstruction error compared with the traditional EMD method and eliminates the modal aliasing phenomenon. Combined with the SVM classifier optimized by PSO, the F1-score reached 0.92 on the same test set, which was significantly better than the machine learning non-orthogonal signal detection method. The indicators such as the instantaneous bandwidth (12.58Hz), frequency resolution (0.18Hz) and dynamic range (100dB) of this method are all superior to those of the baseline method, providing an effective solution for the detection of sudden changes in broadband signals in complex communication environments.