Abstract:Conventional least-squares deconvolution technique is valid only when the conditions of both minimum-phase wavelet and a white reflectivity series are satisfied; but it is invalid in the case of unknown mixed-phase pulse. A new least-squares deconvolution method suitable for unknown finite duration mixed-phase pulse is proposed here.The predicted duration of a mixed-phase wavelet must be offered before the method is used, The tests of synthetic data show that this method brings good result and is insensitive to the predicted wavelet duration. It is also proved that the least-squares deconvolution method for unknown mixed-phase pulse is equivalent to gapped smoothing error filtering, which can be decomposed into a forward multi-step prediction error filtering and a backward multi-step prediction error filtering. It can be foreseen that this new deconvolution technique will finds its practical applications in seismic data processing.