Application Of BSS In Seismic Signal Denoising

In the seismic exploration of gas and oil resources, the random noise ofenvironmental vibration and industrial alternating current is inevitably received bygeophones which increase the difficulty in understanding of layer structures.Considering the spherical diffusion attenuation and absorption of inelastic media, theenergy of seismic wave finally received by the geophones often become extremelyweak as the random noise is comparatively strong that bring much difficulty inseismic data processing.As a new signal processing technique, blind source separate can recover theoriginal seismic signal and extract the noise under low S/N situation. This methodonly based on the assumption that the original signal and noise are statisticindependence which meets most of the situations.The recovered signal of blind source separation is inconsistent in scale and thesequence appears uncertain. Whiting method could normalize the signal to solve thefirst question and the second problem of sequence can be solved by comparing thesimilarity of noised-signal and recovered-signal. However, the noise-signal may nothave any similarity with the original waveform of seismic data if the noise is toostrong thus the method failed. Regarding that the waveform of original seismic signalhas comparatively strong consistency, the difference of contiguous seismic recordswon’t be large and this ensures the effectiveness of the separation of signal and noise.The whiting mixture matrix is an orthogonal matrix and could be resolve into aproduct of a series of Givens matrix. Then the blind source separation could be seenas a process of rotating mixed signals which continues till every signal becomeindependent and each turn of the rotate angle is determined by the minimum mutualinformation method.In order to test the effectiveness of the method, it is used for the separation of twomixed speech signal to restore the original independent signals. Then the method usedto deal with a seismic profile which is seriously disturbed by noise. When processingthe profile, put two adjoining traces of seismic data, and get a seismic signal and anoise signal, so that the trace-number of resulting seismic signals is less than seismicdata that before separation. Considering that separation method is based on theinstantaneous linear mixed model, wavelet transform is used to convert noisy seismicwaves from time domain to frequency domain before de-noising processing seismicdata, then transformed into the de-noising the time domain. From the seismic profilewe can see that the strata structure de-noised is clearer. Some of the micro stratastructure originally drowned by noise is appeared, and signal to noise ratio is significantly increased.