Simultaneous sources: blending & deblending

Classical acquisition waits for each shot’s response to decay before firing the next. That forces a minimum listening time (typically 6-10 s for deep targets) and caps crew productivity. Simultaneous (or "blended") acquisition fires two or more sources close enough in time that their responses overlap in the recording, then deblends in processing to separate them.

The enabling trick: a random dither

A source fires once per shot, so every trace of one shot record shares that shot’s firing time. Inside a common-shot gather the second source is a complete, coherent shot record shifted by one delay, and no coherence filter can reject it there. The trick is to change the delay from shot to shot: source B fires $\\tau\_k = \\bar\\tau + J u\_k$ after source A on shot $k$, with $u\_k$ random in \[-1, 1\]. Take one trace from each of many shots (a common-receiver, common-offset or CMP gather), align it on A’s firing times, and A’s reflections line up while B’s land at a different time on every trace. Align it on B’s firing times and the roles reverse.

In Figure 1.6 one ocean-bottom node records 60 shots from each source. Start with no dither and raise $J$, then step the iterations and watch the separation climb in (e).

Basic deblending algorithm (iterative)

With no dither, three iterations separate source A to only 2.8 dB: B lines up in A’s frame nearly as well as A does. With $\\pm 200$ ms of dither the same three iterations reach 26.1 dB, the floor set by the filter’s own damage to the curved shallow reflections. The algorithm the figure runs (Mahdad et al., 2011) is:

1. Cut the continuous record at each source’s firing times, giving $d\_A$ and $d\_B$.
2. Estimate A with a coherence filter that treats B as noise, $\\hat a = F\_A(d\_A)$. The figure’s $F$ is a median across neighbouring shots read along the source’s moveout (Huo et al., 2012); production codes more often use sparsity in a curvelet or Fourier domain.
3. Re-blend: shift $\\hat a$ by each shot’s delay $\\tau\_k$ into B’s frame and subtract it from $d\_B$. What remains is mostly B.
4. Estimate B from that residual, $\\hat b = F\_B\\bigl(d\_B - \\hat a(t + \\tau\_k)\\bigr)$, re-blend it into A’s frame, subtract it from $d\_A$ and filter again.
5. Iterate until both estimates stop changing.

A small dither is not fatal, only slow: at $\\pm 20$ ms the first pass leaves A at 6.0 dB and eight iterations bring it to 22.1 dB. A large dither separates the sources in one or two.

Productivity gain

Two sources fired within one listening time can double the shot rate, and $N$ sources can raise it up to $N$-fold. The realised gain is smaller: the cross-talk left after separation grows with the number of sources and separation costs processing time, so returns diminish beyond two to four sources. Wide-azimuth marine surveys use several source vessels around one streamer spread, and blending lets those sources fire without waiting for one another. Ocean-bottom surveys, whose receivers record continuously, blend most aggressively.

References

• Beasley, C. J., Chambers, R. E., Jiang, Z. (1998). A new look at simultaneous sources. SEG Annual Meeting Expanded Abstracts, 133-135.
• Berkhout, A. J. (2008). Changing the mindset in seismic data acquisition. The Leading Edge, 27(7), 924-938.
• Huo, S., Luo, Y., Kelamis, P. G. (2012). Simultaneous sources separation via multidirectional vector-median filtering. Geophysics, 77(4), V123-V131.
• Mahdad, A., Doulgeris, P., Blacquière, G. (2011). Separation of blended data by iterative estimation and subtraction of blending interference noise. Geophysics, 76(3), Q9-Q17.