We have constructed a channel complex model at a scale of 1:10,000 by stacking 3D-printed polylactide layers with negative relief meandering channels. This model was subjected to an ultrasonic common-offset acquisition in a water tank (with the water filling the channels), and the result was treated as a zero-offset 3D acoustic reflection seismogram, receiving a deterministic deconvolution and a poststack migration as data treatment. We then developed an algorithm to yield volumes of estimated two-way time layer thickness from multiple-frequency volumes obtained through the short-time Fourier transform. The estimated thicknesses were compared with the measurements of the physical model obtained through X-ray computed tomography. Despite the strong signal attenuation and imaging issues, the results were rather satisfactory, increasing the confidence in using spectral decomposition for quantitative seismic analysis.