IEEE SIGNAL PROCESSING MAGAZINE [27] JULY 2012 Digital Object Identifier 10.1109/MSP.2012.2189999

nsupervised sig-nal processing has been an excit-ing theme of research for at least three decades. It finds the potential application in practi-cally all fields where well-estab-lished techniques of digital signal processing have been employed, including telecom-munications; speech and audio processing; image, radar, and sonar; and biomedical signals. Among these classical prob-lems, geophysical signal pro-cessing has played a prominent role in the development of unsupervised methods. In fact, the field of unsupervised process-ing can be said to have started with the early application of Wiener’s theories to seismology. Wiener filters involve second-order statistics (SOS), specifical-ly correlation and power spectrum density, and are built over the theoretical framework of linear systems and Gaussian signals. In contrast, unsupervised filtering uses higher-order statistics (HOS), enabling the weakening of classical assumptions about the systems and signals under study. Thus, in the last few years, there has been an increasing interest in moving from corre-lation-based methods to unsu-pervised techniques in seismic signal processing. This article reviews some key aspects of two important branches in unsupervised sig-nal processing: blind deconvo-lution and blind source separation (BSS). It also gives an overview of their potential application in seismic process-ing, with an emphasis on seis-mic deconvolution. Finally, it presents illustrative results of the application, on both synthetic and real data, of a method for seismic deconvolution that combines techniques of blind decon-volution and blind source separation. Our implementation of this method contains some improvements over the original method in the literature described in [1]