Method Study Of Adaptive Time-frequency Analysis And Time-frequency Attributes Extraction

Time-frequency analysis is a essential study hotspot in nowadays signal processing field, which is applied widely in many fields such as economy,national defence,science and technology and life. It has also important application in seismic data analysis,for example,seismic spectral decomposition ,instantaneous attributes extraction,etc. As one of the best tools to analyze the non-stationary signals, time-frequency analysis method shows a good picture in the joint time-frequency domain and makes the change of frequency along with the time clearer. At present , a great many time-frequency analysis methods are studied, which are up to dozen kinds from short-time fourier transform(STFT) to wavelet transform(WT), S transform(ST), Wigner–Ville distribution (WVD), etc. But it is found that the above methods based on traditional theories take on many insurmountable faults ,and it is very difficult to improve temporal resolution and frequency resolution at the same time.Therefore there are also many problems in the time-frequency attributes extraction . The aim of the thesis is to find a best method according to time-frequency localization precision and cross-terms repressing, which is applied in time-frequency attributes extraction.On the base of studying time-frequency analysis fundamental theories and traditional methods, the thesis develops adaptive time-frequency analysis method. It is a method that uses adaptive signal processing measure to complete signal time-frequency analysis to acquire better time-frequency information and tracking effect. Based on several shapes of adaptive time-frequency analysis, adaptive optimum kernel time-frequency representation (AOK) and time-frequency analysis method based on local wave decomposition(LWD) are proposed. Moreover they are used to extract instantaneous attributes and seismic spectral decomposition attributes ,and are compared with traditional methods. The fundamental principle of different methods is studied, and testing computations are implemented with a modeling seismic signal and real seismic data. In the end the results of the above methods are compared.Linear time-frequency representation including STFT, WT, ST and bilinear time-frequency distribution including WVD, smoothed pseudo-Wigner distribution (SPWVD),cone-kernel time-frequency distribution(CKD) are studied in the thesis. On the base of the results, it can be concluded that time-frequency localization precision of linear time-frequency representation is lower and cross-terms of bilinear time-frequency distribution is more severe. Wherefore adaptive time-frequency analysis methods are studied thoroughly.AOK time-frequency representation uses short-time ambiguity function (STAF) and adaptive kernel along with time change, has adaptability to arbitrary length and kind singal without whatever prior information. It is a nearly perfect method which can repress cross-terms of multi-component signal time-frequency analysis, and compare beauty with WVD in time-frequency resolution. AOK representation acquires optimal effect between time-frequency localization precision and repressing cross-terms.Time-frequency analysis method based on LWD which is another adaptive time-frequency analysis method studied in the thesis decomposes signal into finite intrinsic mode functions (IMFs) by LWD, and then time-frequency analysis is implemented to every IMF, and finally results of all IMFs time-frequency analysis are plused. This method can gain ideal effect in repressing cross-terms, keeping resolution and acquiring significative frequency.After the above time-frequency analysis methods studied, they are used to extract several time-frequency attributes including instantaneous attributes and seismic spectral decomposition attributes.On the base of the results, it can be concluded time-frequency attributes that are extracted with linear time-frequency representation are poor in precision and rule. Time-frequency attributes extracted with bilinear time-frequency distribution improve precision enormously ,but are severe in cross-terms disturbance .Time-frequency attributes extracted with AOK are optimal according to time-frequency localization precision, cross-terms repressing and sequences detection validity with respect to the modeling signal. Wigner distribution based on local wave decomposition (LWVD) represses cross-terms generated by WVD and offers higher precision in calculating separate spectrum and extracting instantaneous frequency. AOK is applied into seismic spectral decomposition to improve precision and repress cross-terms in the thesis; it hews LWD's new way in seismic signal processing that LWD is first used to calculate separate spectrum.