UpToWhereChecked against work other engineers signed
We validate the engine by replicating studies that were published by somebody else: their coordinates, their antenna heights, their filed report. That is a harder test than agreeing with another online calculator, because a field survey has a named surveyor and a date on it. Both replications below found bugs in our engine. Both are listed.
Case 1. A filed microwave path survey, Portland, Oregon
Verdict: match"Microwave Path Survey Report", job NA210924-57622PH1, Aviat Networks. Lookout Point to Water Bureau Filtration Plant, 11.2 GHz, antenna centrelines at 150 ft and 145 ft on field-verified coordinates. Field survey 2021-12-01 by Larry Song, report by Jun Li. Filed as Multnomah County Exhibit A.150. PDF
Diagram, not to scale
Elevations and angles land inside the measurement noise of a professional survey, and the verdict and the Fresnel conclusion agree.
| Quantity | Their published value | Ours | Delta |
|---|---|---|---|
| Ground elevation, Lookout Point | 2650.89 ft (807.99 m) | 806.00 m | -1.99 m |
| Ground elevation, Filtration Plant | 722.88 ft (220.33 m) | 220.00 m | -0.33 m |
| Path length | 8.16-8.17 mi | 8.159 mi | inside their own spread |
| True azimuth A to B | 260.91° | 260.88° | -0.03° |
| Vertical angle at A | -2.61° | -2.61° | 0.00° |
| Verdict | clear at the centerlines | CLEAR | match |
| Fresnel criterion (100% F1 at k=4/3) | passes | 1262% of F1 | match |
| Free-space loss | 135.83 dB | 135.81 dB | -0.02 dB |
What the replication found in our engine
Bearings were not compass azimuths
We returned the raw atan2 value, so every westward path reported a negative bearing: this link came out as -99.1° where the surveyor writes 260.91°. That number is printed in point-to-point results and in the link matrix, so it was user-facing and wrong. Normalised at the source and pinned by a test.
Distances ran about 0.3% short
We computed distance on a fixed 6371 km sphere. The WGS84 geodesic for this path is 13131.3 m and we returned 13095.4 m. Every mode now uses the local WGS84 radius of curvature along the path, which brought this pair to 13131 m: metre-level agreement with Vincenty. The correction runs from -0.07% at the Canaries to +0.28% at 60°N.
A caveat about their document, not ours
The source PDF prints Lookout Point’s field-verified elevation as 2560.89 ft on its site pages and 2650.89 ft in the engineering table, a transposed digit, and it prints the path length three ways (8.14, 8.16, 8.17 mi) in three sections. We compared against the engineering value, because that is the one their own RF calculations used. Saying which figure was compared is the difference between a comparison and a claim.
Case 2. A sightline somebody photographed
Verdict: visible, +13.6 mMarc Bret’s Beyond Horizons catalogue documents the Pico del Teide to Pico de las Nieves sightline, about 110 km over open sea between two Canary islands, photographed 2014-12-04.
Diagram, not to scale
Here the ground truth is a photograph, so the only acceptable answer is visible. Ours: visible, clearance +13.6 m over 110.7 km. A marginal-but-real geometry, which is exactly what a photographed long-range sightline should look like: if it came back with a hundred metres to spare, the engine would be flattering itself.
| Quantity | Their published value | Ours | Delta |
|---|---|---|---|
| Verdict | photographed visible, 2014-12-04 | VISIBLE | match |
| Distance | about 110 km | 110.7 km | consistent |
| Clearance over the bulge | not published | +13.6 m | marginal and real |
Reproducing it
Summit coordinates have to be snapped to the local maximum of the elevation model. The first attempt used coordinates about 30 m off Teide’s crest; the observer then sat below the ridge immediately in front of it and the path read blocked 30 m out. Sharp peaks need the snap, and a tool that hides this from you is hiding the thing most likely to make your own answer wrong.
What these two cases do and do not establish
- They establish that the geometry is right: elevations, geodesic distances, compass azimuths, vertical angles, curvature at the refraction constant each mode uses, and the Fresnel clearance test.
- They do not establish a propagation model. We do not compute received signal level, fade margin or availability, at any tier, on purpose.
- They do not make the terrain data something it is not. Copernicus GLO-30 is a surface model at 30 m: canopy and buildings are part of the terrain, and a 30 m cell does not contain your mast.
- Two cases is two cases. More replications go here as they are done, including the ones that disagree with us.
The engine that produced these numbers is the one you would be using
Same code path, same datasets, same printed provenance on every export. Studio is 49€/month.