 Solution

At each source position, couple the vibrator to the ground, transmit a known frequency sweep $s(t)$ and record the ground response on synchronized geophones. Record a pilot or measured source-force reference and sufficient listening time for delayed arrivals. Repeat sweeps where beneficial and retain geometry and timing information.

If the Earth’s impulse response is $h(t)$, a linear model is $d=s\ast h+n$. Vibroseis correlation cross-correlates the record with the known sweep:The sweep’s autocorrelation $R_{ss}$ is a compact, approximately zero-phase wavelet, so the long chirp is compressed into identifiable delayed events. In the frequency domain this multiplies $D(\omega )$ by $\overline{S(\omega )}$, yielding $H(\omega )|S(\omega ){|}^2$ plus noise. It is pulse compression, not an exact recovery of $h$ at frequencies never transmitted. Spectral shaping or stabilized deconvolution can further compensate the finite source bandwidth and sidelobes.^[ref]