Page 353 - Acquisition and Processing of Marine Seismic Data

344 6. DECONVOLUTION

FIG. 6.29 Determination of suitable deconvolution operator length from the autocorrelation of an input seismogram.
(A) Schematic display of a trace’s autocorrelogram indicating the length of the first isolated amplitude package (suitable
deconvolution operator length). (B) Generally, autocorrelations of a number of successive shot gathers are analyzed together
to determine the operator length more precisely. The first transient zone on the autocorrelation traces is expected to have
almost similar characteristics for all seismic traces of the shots, representing the autocorrelation of the source wavelet.

second zero crossing point of the autocorrelo- the remaining n lags are zeroed out. Further-
gram of the input seismogram is a general more, as the α increases, the amplitude spectra
convention. Fig. 6.31 shows statistical deconvo- of the input and deconvolution outputs become
lution results with an operator length of similar. In the example in Fig. 6.31,the de-
n ¼ 100 ms applied to a minimum phase wave- convolution is ineffective on the wavelet after
let for different prediction lags. Specifically, α ¼ 86 ms because all lags of the autocorrela-
α ¼ 6 ms and α ¼ 18 ms correspond to the first tion are passed to output. Fig. 6.32 shows
and the second zero crossing times of the auto- statistical deconvolution results with an opera-
correlation in Fig. 6.31, respectively. tor length of n ¼ 100 ms applied to a minimum
If the prediction lag is not unity, the predic- phase seismogram for different prediction lag
tive deconvolution produces a wavelet of finite values. Resolution of the deconvolution output
length instead of a zero-lag spike. As a general is inversely proportional to the prediction
implication, for a given input wavelet with a lag value.
length of (α + n), a deconvolution operator with Resolution power of a predictive deconvolu-
alengthof n and a prediction lag of α truncates tion output in a noise-free environment can be
the wavelet in the deconvolved output into a controlled by prediction lag. A unit prediction
compressed wavelet of α ms length. The first lag (α ¼ sampling rate) corresponds to spiking
α lags of the autocorrelation are passed, while deconvolution and represents the maximum