Denoising Algorithm Research Of Railway Signals Based On Sparse Decomposition

With the rapid development of the railway, understanding in real time the damage state of turnout by turnout vibration signal and detecting correctly parameter of track circuit frequency shift signal have become an important guarantee on the railway traffic safety.In field acquisition, the turnout vibration signal and track circuit frequency shift signal may be mixed with noise, which seriously interfere injury recognition of turnout and detection of track circuit frequency shift signal. Therefore, denoising of the above signals is a problem to be solved. The traditional denoising algorithms can not obtain satisfactory denoising effect. Hence, based on the strong anti-noise of sparse decomposition, which is applied to denoise of the turnout vibration signal and track circuit frequency shift signal and detect of track circuit frequency shift signal in low SNR. The main research contents of the thesis are as follows:First of all, for the non-obvious structure characteristics of High speed turnout vibration signal, this thesis designs a gabor overcomplete dictionary and proposes a denoising algorithm of the turnout vibration signal based on improved particle swarm optimization (IPSO) optimized spare decomposition. Simulation results demonstrate that the algorithm not only can filter out noise of turnout vibration signal effectively, but also has better denoising quality compared with the wavelet threshold based on EMD algorithm, wavelet threshold algorithm and Fastica algorithm.Afterwards, for the characteristics of track circuit frequency shift signal, this paper designs a cosine overcomplete dictionary, combines the two-phase with spare decomposition, and proposes denoising algorithm and detection algorithm of the track circuit frequency shift signal based on spare decomposition. Simulation results show that the former can effectively remove noise from low SNR frequency-shift signal to improve the detecting SNR by lOdB, which is far better than the wavelet threshold denoising based on EMD and wavelet threshold denoising algorithms; the later can detect the carrier frequency and low frequency under strong noise background and has a lower detection error rate, which is far superior to the most popular ZFFT algorithm.Finally, the content and existing problems of the paper is separately summarized and researched, and the further research is given an expectation.