Explore the measurements from four anchors.
Understand range error, stability and what changed
when the tag returned to the same location.
Operator-confirmed coordinates; equal UWB antenna heights.
Illustrative layout, not a live position map.
Total received ranges
Across 16 actual capturesEligible for statistics
From 33,854 selected raw rangesSelected captures
P1–P9 plus a return to P1Detailed chart plates
With a complete 24-page PDF01 / THE FINDINGS
The frozen correction reduces overall error,
with different results by anchor and location.
Based on eight equally weighted point–anchor means from P9 and the return to P1.
Average sample MSE within each point–anchor cell first, then weight cells equally. This differs from the point-mean metric above.
Evaluated on data excluded from fitting
had lower absolute mean error
A3 at P9 also worsened after correction. The model remains an engineering candidate for further evaluation.
Not deployed as permanent calibration02 / ANCHOR COMPARISON
RMSE of two held-out point means per anchor.
Centimetres; lower values mean smaller errors.
| Anchor | Before correction | After correction | Point-level results |
|---|---|---|---|
| A1(0, 0) m | 16.25 | 12.36 ↘ | 2/2 cases improved |
| A2(2, 0) m | 27.34 | 28.00 ↗ | 0/2 cases improved |
| A3(0, 2) m | 11.68 | 9.84 ↘ | 1/2 cases improved |
| A4(2, 2) m | 10.36 | 2.99 ↘ | 2/2 cases improved |
Use the frozen y = a·u + b fitted to P1–P8 and correct with (y − b) / a. Neither P9 nor the return to P1 was used to adjust parameters. Raw P9 statistics were viewed before fitting, so validation was not blind.
03 / WHAT THE DATA TELLS US
On the return to P1, A2 had an SD of about 2.67 cm but a bias of +34.35 cm. A stable reading still needs its offset checked.
WITHIN-CAPTURE DISPERSIONAt the return to P1, the A1–A4 means changed by +24.16 / +32.81 / −11.14 / +5.68 cm. Boot changes, resets, power interruptions and a board fall occurred between captures, so a cause cannot be isolated.
BETWEEN-CAPTURE CHANGESThe operator confirmed the coordinates and equal antenna heights. No independent survey or reference-position uncertainty is available. These results characterize the archived test conditions.
REFERENCE GEOMETRY04 / A CLOSER LOOK
Raw and corrected bias, within-capture spread and absolute-error P95 in one view.
Compare raw and corrected readings within selected error bands.
See how the means shifted under changed conditions.
An appendix of 3,492 sampled snapshots, separated by solver state.
05 / READING THE RESULTS
All analysis uses archived observations.
No new capture or model refit.
Bias is the mean error against the reference range. A positive value means over-reading. SD describes spread within a capture. RMSE combines offset and dispersion; lower values mean smaller errors.
All results use the operator-confirmed layout, not independently surveyed reference distances.
The ten selected captures contain 33,854 raw ranges, of which 33,669 are used for statistics. Thirty basic-invalid and 155 before-status records are excluded from metrics, with all originals retained.
Six problematic captures are archived separately with their exclusion reasons. No additional positive-outlier trimming or gap filling was applied.
Survey antenna phase-center geometry, record antenna orientation, power supply and temperature, and repeat measurements under controlled conditions. Set acceptance criteria before measuring.
P95/P99 are empirical percentiles of received data. Cycle coverage is not RF packet loss. X,Y are sampled snapshots, not every fix. These stationary captures do not identify a dynamic transfer function.
