| Summary: | 碩士 === 國立海洋大學 === 電機工程學系 === 88 === The time-frequency distribution (TFD ) can accurately transform signals into two-dimensional time-frequency domain. It has extensive application in the nonstationary fields of sonar, seismology and underwater acoustic.
Most of the current high resolution time-frequency methods are based on the structure of the Cohen time-frequency distribution function. However, since the time-frequency distribution has the bilinear property which will produce the cross terms factor treated it as the noise. This thesis presents that when analyzing the signals in terms of time-frequency distribution, we vary the time-frequency distribution by changing the parameter values in its kernel, and based on the derived relationship between time-frequency signal resolution and parameter, a fast adaptive processing algorithm is implemented to obtain the optimum resolution time-frequency signals in a short time. The analytical ability of the time-frequency distribution will be influencing at low signal-to-noise ratios(SNR''s). We propound a method based on the subtractive-type algorithms and combine it with the adaptive time-frequency distribution, the masking adaptive time-frequency distribution (MATFD), to enhance the analytical ability of the time-frequency distribution at noisy environment. Finally, by substituting the kernel of optimum parameter for signal recognition structure, we acquire accuracy in nonstationary signal recognition in high interference channel. And we achieve the speech recognition structure using the time-frequency distribution.
|
|---|
