Abstract:The common predicted deconvolution can improve the resolution of seismic data in seismic data processing under the precondition of two basic hypotheses: (1)seismic wavelet is minimum phase; (2) reflectivity sequence is random process. There is large difference between above hypotheses and real seismic wavelets. For that reason, the authors presented a new mixed phase deconvolution on the basis of the predecessors studied. The implemented process is as follows:(1)the amplitude spectrum of seismic wavelet is fitted by least square method according to the relationship between the autocorrelation of seismic records and autocorrelation of seismic wavelet, and then the autocorrelation of seismic wavelet is further computed; (2)designing a minimum phase filter by using computed autocorrelation of seismic wavelet, then adding a delay to dominant diagonal of autocorrelation function, decomposing a minimum phase filter into short filter and convoluting with a longer filter, which resulted in series of mixed phase wavelets and mixed phase filter; (3)convoluting the resulted mixed phase filter with raw seismic traces and selecting a certain judging rule to choose a optimal mixed phase filter and associated mixed phase wavelet. It is known by model tests and real data processing that deconvolution in separate widows is better than in single window and mixed phase deconvolution is better than minimum phase deconvolution. It should pay attention to the relationship between the broadness of the window and length of operator when computing the seismic wavelet in separate windows in order to obtain better deconvolution results.