Real acquisitions / independent references
The raw data,
and what it cost to trust it.
The two radar acquisitions behind every measurement on this site, at full extent, with the surveyed chambers beneath them and the registration and timing checks that had to pass before any of it counted.
Loading the saved acquisition reports…
Put the survey
in three dimensions.
These models describe excavated spaces using published measurements. They give a future radar prediction something specific to get right—or miss.
Exactly what this model simplifies
The mesh uses local metres and simplified connected volumes. It is not a georeferenced or watertight engineering model.
Measure before
interpreting.
These results use the supplied complex data. They test input integrity, geometry and image measurements. None is a field depth detection.
Two clocks. Different meanings.
Three 24-column strips retain every native azimuth row. Curves average Fourier power across columns, normalized separately for display. The horizontal axis is image frequency scaled by the processing PRF; it is not ground-vibration frequency.
Numbers, conventions and limits
Can the tracker recover a known shift?
Known translations are applied to a 512 × 512 crop of the real complex image. The same patch tracker used in the mechanism studies measures 25 fixed, interior patches for each input. Both signs, both axes and a zero input are retained.
| Injected (row, col) px | Median measured px | 95th percentile error px | ≤ 0.02 px? |
|---|
These translations are controlled software inputs, not measured ground vibration. Patches from one crop are dependent. Passing a 0.02 px calibration tolerance does not demonstrate sensitivity to the 0.005 px motion gate or resolve subsurface depth.
Every patch measurement (.csv) ↓Follow every figure back to its input.
Small derived images, sample hashes, settings and every calibration trial travel with this page. Full private rasters stay outside the repository.