Processing Of The Non-stationary Signal Based On The Wavelet Analysis Technology

In the last two decades, the development of information science and signal processinghad been tremendous. This thesis is about a research using wavelet analysis technology toanalyze a non-stationary signal with an emphasis on de-noising. The aiming of the research isto study the correlation between the selection of wavelets selection and the type of nan-stationary signal, and the selection of control parameters used in the wavelet analysis and thecharacteristic of signal being analyzed. Commonly used classical signal de-noising methods(such as the single time domain method, the single frequency domain method, the Fouriertransform) can only be used for stationary signal analysis. For signals containing nan-stationary factor, one has to use some time-frequency methods such as the wavelet transform.The main work discussed in the thesis include:Through sampling a signal with different non-stationary factors, separately using theclassical Fourier transform method, some traditional filters (Butterworth, Chebyshev, Elliptic),and the wavelet transform to analyze the signal. The analyzing performance is discussedthrough several criteria (the Signal to noise ratio SNR, the root mean square error RMSE, andthe smoothness of the signal). Both the modulus maximum and the threshold methods areused in the wavelet de-noising and their results are compared in the thesis. Simulation showsthat it is important to choose the form of wavelets and the controlling parameters according tothe characteristic of the signal to be analyzed.