Program Ich(Identification and Characterization)
by Don Wagner January 21, 1993Identification & Characterization of Signal and Noise
Seismic data suffers more from the presence of noise than the absence of signal. To remove noise, it is essential to identify and characterize what is signal and what is noise. The method of identification and characterization is due to Dennis Bjerstedt of Amoco Canada who successfully used it for over a decade as an interpreter, and who wrote USP analysis programs to quantify his procedure. Recently, Kurt Marfurt generalized Bjerstedt's call to arms by writing program ICH (Identification & Characterization). The basis of the program lies in the interpreter's ability to differentiate between signal and noise, or at least see one or the other on seismic data.
Figure 1 shows a shot record with moveout applied.
This shot record shows weak signal coherency but is dominated by coherent noise and a salt-and-pepper look that appears to be random noise. However, what appears as random noise is actually aliased coherent noise. This statement is true for most all seismic data.
In Figure 2, we picked a sample of the coherent noise events visible on Figure 1. These linear noise events were picked with the XSD plotting program, saved in a pick.file, and input to program ICH. We ran ICH as follows:
- ich -N r1 -P r1.npicks -M -O r3.ich.n.
where r1 is the record in Figure 1, r1.npicks the noise events picked from this record, -M a flag requesting a strip of data centered on our picks, and r3.ich.n the output file plotted in Figure 2. Alternatively, we can pick the signal events from Figure 1 and process those thru ICH to get Figure 3 on the following page. Now we have identified and characterized separately both the signal and noise.
The task then is to view these characterizations in a domain that permits separation of signal from noise. We opt for the f-k domain and compute the noise and signal spectrum, respectively in Figures 4 and 5. Note how the signal, as expected, parallels the frequency axis at k = 0, or horizontal position 128, whereas the noise is greatest for negative dips in quadrant 2. We have the option of drawing a polygon around the negative dip noise in Figure 4 and muting all energy inside the polygon; or we can draw a polygon around the signal in Figure 5 and mute all energy outside the polygon.
We choose to construct a polygon that encases signal around the k = 0 axis (horizontal position 128), avoids the horizontal banding of low frequencies around vertical position 451, and avoids the large negative dip noise event that begins at horizontal position 128 and dips to the left into quadrant 2.
Figure 6 shows the f-k spectrum of Figure 1 after applying this polygon mute to save signal and reject noise. We used the processing flow :
- fft2da -N r1 | polymute -P r1.fks -M out -O fk.signal.
where fft2da transforms to the f-k domain, and polymute masks all data outside the polygon.
Finally, we inverse transform to the time-space domain in Figure 7, and compare this f-k filtered result with the original Figure 1. The processing step for entire process:
- fft2da -N r1 | polymute -M out -P
Now, in Figure 7, we are able to see the signal that had been masked by coherent noise.
Thus Program ICH provides a way to characterize signal and noise. We can transform characterized data sets into different domains to search for a basis to separate signal from noise. Often the f-k domain is suitable for separating characterized signal from characterized noise. One can then construct a mask to mute unwanted events in the new domain, transform back to the space-time domain, and continue processing.
Return to the USP Home Page.Frame Doc to html conversion by Tony Garossino