For around twenty years, the signal processing community has developed, for digital communication, different methods of blind deconvolution making it possible to retrieve the unknown convolutive components of one known signal. In this work, we study the way to apply these blind deconvolution technics to seismology in order to estimate the source function or the site effects from only a small numbers of surface seismic records. Based on the theory of coda wave and on a minimum phase deconvolution method, we develop an original algorithm for coda waves. We first apply it to the records of the Kursk's underwater explosion (12/08/2000). After showing that such an explosion is roughly a minimum phase source, we estimate, from the obtained minimum phase source function of the Kursk, the depth and the power of the explosion. We then apply this method to the Rambervillers's earthquake (22/02/2003, Ml=5.4). From the estimated minimum phase source time function, we retrieve not only the duration and the seismic moment but also the site effect at recording stations located several kilometers away from one another. In order to release the minimum phase assumption, we use a higher order statistic blind deconvolution on the stationnarized coda signal. From the multi-station stacked tricorrelation, we obtain a source time function consistent with the source function estimated after deconvolution by empirical Green function, which is in agreement with the minimum phase seismic source function. These first results show that these methods seem to be very powerful, and they should be applied on a set of records of different events from different stations through global inversion scheme.
Keywords: Seismology, blind deconvolution, higher order statistics, tricorrelation, minimum phase, stationnarization, coda waves, source time function, site effects.