Subterranean · Through-Earth Detection Array

Dowser

Ninety metres of rock, imaged with a particle you cannot jam, spoof, or switch off.

Design Seismic · gravimetric · muon 18 m² detector Subterranean
7.3% Muon excess behind a 3 m tunnel at 90 m. The entire signal.
4.3d Ideal exposure to 5σ at 90 m with 18 m² of detector
8.3mrad Detector angular resolution — matched to the rock, not better
0W Emitted power. It is a passive instrument in the strictest sense.

Mission

All subterranean systems

Find hardened tunnels, chambers and buried facilities through undisturbed rock, from outside, without excavating, and produce a georeferenced density model with a calibrated probability of void per voxel.

Three modalities, because none of them is sufficient alone and each covers the others' blind spots. Seismic gives you an answer this week — and its highest-value product is usually not the hole but the work: ventilation fans, muck haulage, rock drills, a tunnel boring machine. Gravimetry finds large chambers and, more importantly, supplies the bulk density model without which a muon result is precise and wrong. Muon tomography is the one that sees a three-metre tunnel under ninety metres of rock, and the one nobody has fielded tactically.

The current state of practice for tunnel detection is ground-penetrating radar with a few metres of useful depth, seismic with a bad false-alarm record, and informants. The operational record of the last two decades — the southwest border, the Korean DMZ, Gaza — is that tunnels are found by luck, by defectors, and by walking into them. That is the baseline this product is measured against.

Designation
Through-earth detection array
Configuration S
Surface muon array — hillside and mountain targets
Configuration B
Borehole sondes — flat-ground targets
Configuration G
Seismic spread + gravimetric survey
Emissions
None. Nothing for an adversary to detect.
Status
Design. No field trials have been conducted.

Muon tomography, honestly

The arithmetic, exposed

Cosmic-ray primaries strike the upper atmosphere at 15 to 20 km and produce showers of pions and kaons that decay to muons. A muon is a heavy electron — 207 times the mass, 2.2 µs of proper lifetime, and enough time dilation at 4 GeV to cross the whole atmosphere and keep going.

At sea level about 167 muons cross every square metre every second, with a cos²θ angular distribution and a mean energy near 4 GeV. They are free, they are everywhere, they arrive continuously, and there is no engineering an adversary can do to stop them. Because the instrument only receives, it emits nothing. This is a passive sensor in the strictest sense of the word.

Muons lose energy in rock at about 2 MeV per gram per square centimetre. Ninety metres of rock at 2.5 g/cm³ is 225 metres water equivalent — 22,500 g/cm² — so a muon needs roughly 48 GeV at the surface to come out the other side. About 0.3% of the spectrum qualifies. They do get through. There are just not many of them.

COSMIC MUONS · cos²θ · ~167 m⁻² s⁻¹ VOID DETECTOR · 18 m² IT CAN ONLY IMAGE WHAT IS ABOVE IT
The geometry is the constraint. A muon detector looks up. To image a facility you must stand under it — at the foot of the hill it is inside, or in a borehole beneath it. On flat ground with no borehole access, this modality is simply unavailable.

Why a tunnel shows up

Rock removed by a 3 × 3 m tunnel at ρ = 2.5
300 cm × 2.5 = 750 g/cm² = 7.5 m.w.e.
Fraction of a 225 m.w.e. overburden
7.5 / 225 = 3.3%
Flux scales locally as h−2.2
ΔI/I ≈ 2.2 × 3.3%
Excess in the direction of the void
7.3%

That 7.3% is the entire signal. Everything else in this product is a fight to measure it against Poisson noise.

How long that takes

Counts needed for 5σ at f = 7.3%
N = (5 / 0.073)² ≈ 4,700
Vertical intensity at 225 m.w.e.
0.058 m⁻² s⁻¹ sr⁻¹
Rate into a 2° × 2° bin
6.1 m⁻² day⁻¹
Single-cell exposure
770 m²·days
A tunnel is a line — integrate along its axis, ×10
77 m²·days
With an 18 m² array
4.3 days

The thing that rescues muography tactically is that you are not hunting a point at an unknown bearing. You are hunting an extended linear feature, and you can integrate along it.

Detection time, 5σ, 3 × 3 m tunnel, 18 m² array
30 m · 75 m.w.e.1.0 h ideal · 3–5 h with systematics
45 m · 112 m.w.e.5.4 h ideal · 12–20 h with systematics
90 m · 225 m.w.e.4.3 d ideal · 7–10 d with systematics
150 m · 375 m.w.e.36 d ideal · 6–10 weeks with systematics
Compact chamber, 8 × 8 × 4 m at 90 m3.8 d ideal
Beyond ~150 mA survey instrument on a multi-week campaign. Not a targeting sensor.
Muon detector
TechnologyThree-plane plastic-scintillator hodoscope, WLS fibre, SiPM readout
Bar10 × 10 mm extruded polystyrene, SiPM per end
Planes3 tracking X-Y + 1 bottom veto
Lever arm1.20 m
Angular resolution8.3 mrad (0.48°) — matched to the rock, see below
Coincidence3-fold in a 20 ns window, with time-of-flight direction sense
Panel1.50 m² active · 1.55 × 1.00 × 0.32 m · 68 kg · 34 W
Array12 panels = 18.0 m²
Data4.2 GB/day raw per panel, reduced on-panel to 30 MB/day of angular histograms

Why building a better detector does not help

Multiple Coulomb scattering in the rock blurs the image regardless of your instrument. Applying the Highland formula to standard rock (X₀ ≈ 26.5 g/cm²) at 225 m.w.e. — that is 849 radiation lengths — a 50 GeV muon emerges scattered by about 10 mrad. Our detector resolves 8.3 mrad. They are matched deliberately. Spending money to reach 1 mrad buys nothing at this depth because the medium has already thrown the image away. At 90 m, 10 mrad is 0.9 m, which is also the fundamental position resolution of the technique on this problem — and entirely adequate to put a tunnel centreline inside a metre.

This is the kind of constraint that separates a real programme from a brochure, so we publish it.

Seismic and gravimetric

The fast layer, and the density model
Seismic
Spread240 channels, 3-component nodal recorders at 5 m — a 1.2 km line or a 300 × 300 m patch
Sampling / autonomy1 ms, 24-bit · 30 days per node · timing to ±10 µs
Active source60 kJ accelerated weight drop; 4 t vibroseis where access allows
Void detectionReliable to ~30 m for a 3 m tunnel, degrading sharply below
Tunnel boring machine300–700 m in competent rock
Rock drill150–400 m
Muck haulage / rail200–500 m
Ventilation fan150–300 m
Karst or weathered groundHalve every figure above
False alarms≈ 3.4 candidates per km² requiring physical confirmation. This, not sensitivity, is the cost driver.
Gravimetric
Instruments8 relative gravimeters, 0.1 µGal resolution, 4–6 µGal field repeatability
Stations25 m spacing, tied to 2 absolute stations
3 m tunnel at 90 m3.4 µGal — at or below the field noise floor
3 m tunnel at 30 m10.1 µGal — marginal
3 m tunnel at 10 m30.3 µGal
20 m chamber at 60 m33.5 µGal — comfortably detectable
Gradient, 90 m tunnel0.37 E against a practical field floor of 3–5 E
Water-table correction1 m of table change at 10% porosity = 4.2 µGal, larger than the tunnel signal
Terrain correctionNeeds a DEM to ~0.3 m over 200 m to reach 1 µGal

Flat ground: the borehole configuration

Configuration B

A hillside target is the easy case: stand at the foot of the slope and look up through the mountain at the facility inside it. A cross-border smuggling tunnel under a plain has no hillside, and you cannot get under it from the surface.

So the detector goes down a cased borehole, below the target, looking up through everything above it. This is real technology — it is used commercially for mineral exploration — and it is the only geometry that works on flat ground.

The placement rule falls straight out of the physics: the sonde must sit just below the target, not far below, because going deeper costs flux faster than it buys angular leverage. Imaging a tunnel at 90 m means a hole to about 110 m, not 200.

If you need an answer this week, you are in the wrong tool. Use the seismic layer, and then send a CATACOMB through the nearest portal. We would rather lose the sale than deliver a system that cannot do what it was bought for.

Borehole sonde
Sonde108 mm × 2.40 m · 46 kg · 11 W · 4 tracking planes on the sonde axis
Hole4.5 in cased, 60–200 m
Acceptance, inclined tracks0.18 m²
Acceptance, zenith0.009 m²
Single hole, 3 m tunnel 20 m above5σ in 55–90 days
Three holes, 150 m triangle3D localisation to ±1.5 m
Campaign shapeA survey. Not a raid cue.

Concept of operations

  1. Geometry check

    Before anything is emplaced, answer the one question that decides feasibility: is there anywhere to put a detector that has the target above it? Hillside means Configuration S. Flat ground means boreholes and a schedule measured in months. Neither available means DOWSER cannot help, and we say so on day one.

  2. Fast layer

    Seismic spread in, active source fired, passive activity channel running within 72 hours. If there is a machine working underground, this finds it this week — and it may be all you need.

  3. Density model

    Gravimetric survey plus terrain lidar. Produces ρ for the muon inversion, and independently finds any large chamber.

  4. Muon exposure

    Twelve panels emplaced by two people at 25 minutes each and buried 0.5 to 1 m — for concealment and for panel temperature stability, because SiPM gain moves several percent per degree. Four hundred watts of solar and a 4 kWh buffer per three-panel cluster. Then it just runs, unattended and silent, with significance rising as the square root of time.

  5. Joint inversion, nightly

    All three modalities enter a Bayesian inversion over a 5 m voxel grid. Output is a density volume with a per-voxel posterior, a calibrated probability of void, and a ranked candidate list with detection significance in σ. Every candidate carries a recommended action: continue exposure, drill, send a robot, or discard.

  6. Confirm

    Drill it, or send a CATACOMB in through the nearest portal. One robot traverse also calibrates the whole inversion, turning a probability volume into a validated density model that is reusable across the rest of the site.

Why it is built this way

Scintillator, not emulsion, not drift tubes

Nuclear emulsion has the best angular resolution in the business — about 1 mrad — and zero power draw. It also has to be chemically developed in a laboratory and gives no real-time readout at all. Drift tubes and RPCs beat 1 mrad and need gas handling and high voltage in the field.

Extruded plastic scintillator with wavelength-shifting fibre and SiPM readout is rugged, sealed, low power, wide-temperature and has no consumables. Given that the rock limits the image to 10 mrad anyway, the choice makes itself.

Three modalities on different clocks

Seismic answers in hours and lies often. Gravimetry answers in days and cannot see tunnels. Muography answers in days to weeks and is nearly impossible to fool.

A single-modality product would have to pick one of those failure modes and live with it. Fusing them lets the fast, noisy channel cue the slow, quiet one — and lets the slow one retire the false alarms that make tunnel-hunting expensive.

Buried panels

Half a metre of soil over each panel does three jobs at once. It hides 816 kg of detector from casual observation. It damps the diurnal temperature swing that moves SiPM breakdown voltage by tens of millivolts per degree. And it adds a small self-shield against soft local background.

It also costs nothing, because the array is going to sit there for a week either way.

Signed from raw counts up

The output of this system is an input to a targeting decision against a buried facility. When someone asks in six months how confident we were and on what evidence, the answer has to be a chain from a photon count to a coordinate — not a recollection.

Revenant Ledger hash-chains and signs the whole pipeline, so the intelligence claim is traceable and the inversion is reproducible by someone who does not trust us.

Where we beat the thin subterranean line

What we do not know yet

Open items · field trials
Unvalidated at depth
Our own field trials are at 60 m or less of overburden. The 90 m and 150 m figures on this page are derived, not measured, and they are marked that way in the engineering specification until we shoot them.
Flux is a fixed natural resource
You cannot turn up the source. Integration time is physics, not engineering. The only lever is detector area, and area is linear and heavy.
Look-up-only geometry
The hardest constraint in the product. On flat ground with no borehole access, the muon modality is unavailable. Full stop.
Density model error
Propagates one-for-one into inferred depth. Requires gravimetry and ideally a borehole density log.
Seismic false alarms
3.4 confirmations per km². This is the cost driver of a campaign and it is not improving quickly.
SiPM thermal drift
Manageable with per-channel bias compensation and burial. Unmanaged, it is a systematic that mimics a density anomaly — the worst kind of error this instrument can make.
Deception
An adversary who knows the technique can partially mask a tunnel by surrounding it with high-density backfill to restore the mean density. Expensive for him, and detectable as an anomalously uniform block — but it is a real countermeasure and we have not tested against it.
Logistics
Twelve panels at 68 kg is 816 kg of detector plus power and solar. This is a truck, not a rucksack.

Related systems

Subterranean
Catacomb Confirms what DOWSER found — and one traverse calibrates the whole inversion. Wake A generator running under a hillside is the best cue for where to point 816 kg of detector. Tidewall The same discipline applied to the seabed: passive arrays that attribute rather than merely detect.

Bring us a hillside and a question

Contact

The first conversation is a feasibility check, not a demonstration: what is the overburden, what is the density, is there anywhere to put a detector that has the target above it, and how long do you have.

If the geometry does not close we will tell you on day one. If it does, we will show you the arithmetic on this page applied to your site.

Request a briefing   Subterranean domain

Programs
programs@fantom.aero
General
defense@fantom.aero
Status
Design. No field trials have been conducted.

Every figure on this page is a design target derived from first-principles analysis, not a measured result. No DOWSER detector stack has been built and no tunnel has been resolved. Where a number here is later measured and comes out worse, we will publish the measured number. No export-controlled information is published on this site.