| Summary: | 碩士 === 國立雲林科技大學 === 電機工程系 === 102 === In this thesis, the methods for mixed signal separation based on fractional Fourier and short-time Fourier transform analysis are proposed. In the first part, we determine the best order of fractional Fourier transform for separating the mixed signals in a theoretical approach. Then, the results based on the Gaussian fitting approach reported in our previous work are validated. In the second part, consider that the multi-component Gaussian linear chirp signals may overlap in a single fractional order. The separated signals may have low correlation coefficients with their original versions. Because the spectrogram of a Gaussian linear chirp signal is elliptically distributed, we use a simplex downhill search method to obtain the spectral distribution of each signal and reconstructed the signal using the short time inverse Fourier transform. In the third part, we analyze the best order of fractional Fourier transform of the circular fringe pattern in a theoretical approach. We propose an iterative approach to search for the best fractional order by recording the maximum projection of fractional Fourier transform of signals. Then we search for the best order with a local maximum through a differential operation, and find center point of the circular fringe pattern by using the spatial filtering process. The results show that the parameters of reconstructed circular fringe pattern are close to that of the original one.
|
|---|
