FANTOM DEFENSE Fantom Defense — home

Sub-T · Self-Emplacing Surveillance Pod

Wake

Four hundred days in the ground. Forty-six seconds a day of speech.

In design WAKE · WAKE-A 1.1 W Sub-T
400d Buried endurance at a 1.1 W mean draw
0.3mW Cost of transmitting. Listening is the whole budget.
5.3×10⁻⁴ Emission duty cycle — 46 seconds per day
0.06K Residual surface thermal anomaly, against 1–3 K of clutter

Mission

All subterranean systems

Watch a tunnel portal, a suspected buried command post, or a road network over a hardened facility — continuously, for more than a year, with no logistics, no revisit, and no emissions until there is something worth the risk of transmitting.

The historical precedent is worth naming, because it is exact. ADSID — the air-delivered seismic intrusion detector of Operation Igloo White, 1967 — was dropped from an aircraft, embedded itself in the soil, left an antenna disguised as a plant, and reported truck movements on the Ho Chi Minh Trail. The concept was right in 1967.

What has changed is four hundred days of endurance instead of weeks, a classifier on the pod instead of a relay to a mainframe in Thailand, and a signed record instead of an operator's judgement. What has not changed is the physics of burying a radio, and we have not pretended otherwise anywhere on this page.

Designation
Self-emplacing surveillance pod
Sensing
Seismic · acoustic · magnetic · RF energy
Decision
On the pod. 12-class CNN, 220 kB int8.
Exfil
240 MHz DSSS burst, 0.5 W, 7.7 s, on a corroborated match only
Design point
Talking is free. Listening is the entire cost. So listen always and speak almost never.
Status
Design. No prototype has been built.

400 days at 1.1 W, done honestly

10,560 Wh. Everything follows from that.
Where the 1.1 W goes
Seismic — 3-axis, 24-bit ΣΔ, 500 Hz, continuous95 mW
Acoustic — 4 channels, 24-bit, 4 kHz140 mW
Magnetic — 3-axis fluxgate, 0.1 nT/√Hz, 100 Hz210 mW
RF energy detector, 5% duty35 mW
Edge classifier — gated front end, full model at 4% duty185 mW
Timing and housekeeping — TCXO, RTC, supervisor28 mW
NAND ring-buffer writes65 mW
Scheduled receive windows — 1 s every 6 h3 mW
Burst transmit — 6 reports/day at 0.5 W × 7.7 s0.3 mW
Regulator and leakage, 8%78 mW
Sub-total839 mW
Design margin, 30%261 mW
Allocation1,100 mW
What that costs in battery
ChemistryLi-SOCl₂ primary
Cell — D size, 19 Ah68 Wh / 100 g
Cells183
Nameplate12,440 Wh
Self-discharge over 400 d−1.1%
Cold derate at 5 °C soil−6%
End-of-life cutoff−8%
Usable10,570 Wh
Endurance at 1.1 W400 days
Cell mass and volume18.3 kg / 12.7 L

The single most useful line in the first table is the transmit line. Burst exfiltration costs 0.3 mW — three hundredths of one percent of the budget. Talking is free; listening is the entire cost. Which means the emissions discipline that makes this pod survivable is also, conveniently, the cheapest thing about it.

And this is where the requirement drives the airframe

12.7 litres of lithium cells do not fit in an A-size sonobuoy tube, which has 11.0 litres of internal volume in total. There is no packaging trick that fixes that. So the baseline pod grows to 166 mm × 1.05 m and 26.4 kg — dropped from a SPECTER bay, an MQ-9 pylon, a CH-47 ramp roller, or emplaced by hand — and WAKE-A exists at the A-size envelope for the missions where launcher compatibility is worth more than endurance. Both numbers are on this page because the alternative is a customer discovering the mass in a shipping crate.

Variants
WAKE — envelope166 mm × 1.05 m · 26.4 kg
WAKE — energy12,440 Wh nameplate / 10,570 usable
WAKE — endurance400 days at 1.1 W
WAKE-A — envelope123.8 × 914 mm A-size sonobuoy · 9.1 kg
WAKE-A — energy3,100 Wh nameplate / 2,640 usable
WAKE-A — endurance100 days at 1.1 W · 340 days at a 25% armed duty
WAKE-A — deliveryAny A-size sonobuoy launcher: P-8, MH-60, MQ-9 dispenser
Operator power trades
Baseline1.10 W → 400 days
Magnetometer disabled0.82 W → 538 days
Chip-scale atomic clock added1.22 W → 361 days
What the clock buys1 µs/day, enabling inter-pod TDOA on a seismic event
What the clock costs39 days of mission life. That is the trade, in the unit the operator cares about.

Self-emplacement

And the failure that matters most
SURFACE 0.5 m 1.0 m 1.5 m BURIED ANTENNA · −25 dB THROUGH 1 m PENETRATOR BODY 166 mm × 1.05 m 30 W AUGER STAGE < 12 Wh TO FINAL DEPTH SOIL AT 1 m: ±5 °C ANNUAL SWING · 30-DAY LAG
Burial helps the battery. At one metre, soil temperature swings only about ±5 °C a year and lags the surface by a month. The pod lives in a benign, low-cycling thermal environment — worth several percent of capacity over 400 days against a surface asset baking and freezing daily.
Delivery and emplacement
Release300–18,000 ft AGL
DescentDrogue-retarded to 28 ± 4 m/s
Accuracy12 m CEP from 3,000 ft with a guided decelerator; 45 m unguided from 800 ft
ImpactNose plants 0.3–0.6 m in medium soil
Auger stage30 W helical tip to a final 0.8–1.4 m in 12–30 min, under 12 Wh
BackfillVibratory settling of soil into the annulus
Soil temperature at 1 m±5 °C annual swing, lagging the surface by ~30 days

The biggest single risk in the product

On rock, hardpan, pavement or frozen ground, it does not bury

It lies on the surface, where its thermal, visual and radio signatures are all far worse and its detection probability by a casual patrol goes from negligible to high. There are three mitigations and only the first is any good: select the emplacement point from imagery for soil, not only for field of view. Second, the pod computes its achieved depth from the integrated accelerometer penetration profile and reports it in the first status burst — so the operator knows within minutes whether he has a buried asset or a surface one, rather than finding out on day 200. Third, accept the loss; they are cheap enough to over-drop.

Classification at the edge

Revenant Forge

The discriminators that actually work are physics, not pattern matching.

Track-pad passing frequency — speed divided by pad pitch — is a hard, unambiguous separator between tracked and wheeled vehicles with no training-data dependence at all. A diesel firing-order harmonic comb has its fundamental at RPM over sixty times half the cylinder count, so the spectrum tells you the cylinder count and the RPM, which tells you the vehicle class. Rotorcraft announce themselves with a blade-pass frequency.

And a rock-solid 50 or 60 Hz mechanical line with harmonics is a generator set. A generator running continuously under a hillside with no building on top of it is the single best signature of a buried command post that exists. That is the highest-value classification in this entire domain, and it is why a WAKE hit is a legitimate cue for pointing 816 kg of DOWSER detector at a slope.

The magnetic channel does the disambiguation. A ferrous vehicle produces a dipole transient falling as one over range cubed; a three-tonne truck at fifteen metres gives ten to forty nanotesla against a 0.1 nT/√Hz floor, and the zero crossings give direction and a coarse speed. It is what separates a truck went past from the wind moved a tree.

Classifier — honest performance
Model220 kB int8 CNN, log-mel + spectral flux, 1.0 s frames, 0.5 s hop, 12 classes
Held-out data, sites seen in training91% top-1
Site never seen before74% top-1
Unseen site, after 72 h on-pod adaptation86% top-1
WhySoil coupling changes the source-to-sensor transfer function by more than the difference between vehicle classes
Detection ranges · competent soil, quiet site
Heavy tracked vehicle250–600 m
Wheeled truck120–300 m
Running generator80–180 m
Rock drill150–400 m
Footfall, single walker20–40 m
Operating point
False alarm budget≤ 1 false report per pod per 30 days
Corroboration rule≥ 2 modalities. Single-modality hits are logged, never sent.
Pd — heavy tracked within 60 m0.93
Pd — running generator within 120 m0.88
Pd — single walker within 25 m0.61

We publish the 0.61. A pod that reports two thirds of the people who walk past it is what the physics and the false-alarm budget allow. A customer who is told 0.95 plans a mission that fails.

Burst exfiltration, and staying unfound

The link is easy. Not being heard is the hard part.
Exfil link budget · to a receiver at 4 km slant
Transmit power, 0.5 W+27.0 dBm
Buried antenna gain — detuned, lossy−6.0 dBi
Soil loss, 1 m of temperate loam at 240 MHz−25.0 dB
Free-space path loss, 240 MHz, 4 km−92.1 dB
Receive antenna gain+6.0 dBi
Received power−90.1 dBm
Noise floor, 2.4 kHz, NF 3 dB−137.2 dBm
Signal-to-noise ratio+47 dB

Forty-seven decibels of margin is not an invitation to shout further. It is margin we spend on survivability: 0.5 W, a 1023-chip DSSS waveform with 30 dB of processing gain, frequency hopping across 12 MHz, and transmission only inside a scheduled window when a receiver is actually overhead — from the BEACON ephemeris or a planned SPECTER orbit. The pod never radiates into an empty sky.

Signature, and what we can do about it
Thermal, raw1.1 W at 1 m → ΔT ≈ Q/(2πkr) = 0.18 K over ~1 m²
Thermal, mitigated0.06 K — aerogel jacket above, conductive path below, so heat goes down. Against 1–3 K of diurnal and vegetation clutter.
MagneticTitanium and composite housing, non-magnetic fasteners, steel only in the auger tip. Residual dipole < 0.02 A·m² → 4 nT at 1 m, under a typical 5–20 nT geological noise.
Magnetometer surveyA deliberate 1 m-grid survey will still find it.
Ground-penetrating radarFinds it. There is no mitigation and we will not pretend there is one.
VisualA 166 mm ground disturbance. Drop before precipitation; one rain cycle to blend.
RF46 s of emission per day, spread and hopped. A determined, co-located, correctly-tuned collector will eventually see a burst.
End of lifeCryptographic erasure, controlled cell venting, self-reported grid for recovery — or an inert 26 kg mass left in the ground.

Concept of operations

  1. Plan for soil

    Select the emplacement point from imagery for soil, not only for field of view. Soft ground is worth more than a good look angle, because a surface pod is a found pod.

  2. Deliver

    WAKE-A from any A-size sonobuoy launcher. WAKE from an aircraft bay, a pylon, a ramp roller, or by hand. Drogue-retarded to 28 m/s, 12 m CEP with a guided decelerator.

  3. Emplace

    Impact plants the nose. The 30 W auger drives it to final depth over twelve to thirty minutes for under twelve watt-hours — one tenth of one percent of the energy budget. Vibratory backfill closes the annulus.

  4. Report in

    One status burst: position, achieved depth, soil stiffness, battery, self-test. The operator now knows exactly what asset he has, on day one.

  5. Baseline

    Seventy-two hours of ambient learning. A site-specific background model and adapted thresholds recover most of the 91-to-74 percent domain-shift gap.

  6. Listen

    Four hundred days. Everything is logged locally to a hash-chained store. Almost nothing is transmitted.

  7. Report

    On a corroborated two-modality match, inside a scheduled receive window, a 384-byte burst in 7.7 seconds at 400 bps. Six a day is 46 seconds of emission.

  8. Re-task, then end

    On any overflight the pod accepts a new classifier, a new watch list, or a new operating point. At end of mission: erase, vent, report grid — or be abandoned deliberately, with that decision recorded.

Why it is built this way

240 MHz, and no mast

Soil loss is a steep function of frequency. Through one metre of soil at 0.01 S/m: about 52 dB at 900 MHz, 30 dB at 300 MHz, 21 dB at 150 MHz.

We accept 25 dB rather than expose a mast. ADSID's disguised-plant antenna was the tell that got ADSIDs found, and a 380 mm whip standing proud of the ground is a 380 mm whip standing proud of the ground. Wet clay is worse — 45 to 60 dB — and in wet clay the pod is emplaced shallower by design, trading signature for link. That is an explicit, planned trade rather than a surprise.

Corroboration before speech

Every transmission is a risk, so the operating point is set by the false alarm rate rather than by detection probability. A report requires agreement across at least two modalities — a seismic hit, a magnetic transient, an acoustic harmonic match. Single-modality hits are logged and never sent.

The whole product depends on not exfiltrating garbage, because a pod that cries wolf six times a day has a duty cycle a collector can find.

Heat goes down, not up

A 1.1 W source at one metre produces a steady-state surface anomaly of about 0.18 K over a square metre. A cooled long-wave imager with a 25 mK noise floor can resolve that in principle.

An aerogel jacket on the upper hemisphere and a deliberately conductive path to the base pushes the heat downward and drops the surface anomaly to about 0.06 K — comfortably inside the one-to-three-kelvin natural clutter of diurnal heating and vegetation. It is a cheap trick and it works.

Nothing is hidden about what finds it

Ground-penetrating radar will find this pod. A deliberate magnetometer survey on a one-metre grid will find it. A patient collector staring at the right band will eventually catch a burst.

The design assumes a distracted adversary, not an omniscient one, and that assumption belongs on the product page rather than in a footnote, because it is the assumption the mission is actually betting on.

Why the medium decides

What we do not know yet

Open items · prototype
Emplacement into the wrong soil
The largest single risk. Rock, hardpan, pavement or frozen ground turns a buried asset into a surface one. Mitigated by planning and by self-reported depth. Not eliminated.
Unvalidated endurance
No pod has yet come out of the ground after 400 days. Our longest continuous run is well short of it. The 400-day figure is a derived budget with a 30% margin, not a measurement, and it stays marked that way in the engineering specification.
Classifier domain shift
91% to 74% on an unseen site, recovering to 86% with on-pod adaptation. The residual is unsolved and it is a data problem, not an architecture problem.
One-shot mechanisms
The auger motor must work exactly once, twenty minutes after being dropped out of an aeroplane. Cell passivation after a year of microamp draw and NAND retention over 400 days are the other long poles.
Detectability
GPR finds it. A dedicated magnetometer survey finds it. A patient collector eventually sees a burst.
No recall
Once it is 1.2 m down, it is 1.2 m down. There is no abort and no retrieval short of a shovel.
Environmental and legal footprint
A 26 kg inert object left in a foreign country, potentially permanently. This is a real policy question that the engineering does not answer.

Related systems

Sub-T
Dowser Where a WAKE generator hit tells you to point 816 kg of muon detector. Barrow The same persistence trick, inside the tunnel: 14 months of listening on 1.8 mW. Specter The overhead receiver a WAKE burst is scheduled against — and one of its delivery aircraft.

Request a briefing

Contact

Bring us a site, a soil type and a watch list. We will bring the power budget, the ROC curve including the number we are not proud of, and the list of things that will find this pod.

Request a briefing   Sub-T domain

Programs
programs@fantom.aero
General
defense@fantom.aero
Status
Design. No prototype has been built; endurance is analysis, not measurement.

Every figure on this page is a design target derived from first-principles analysis, not a measured result. No WAKE pod has been built or emplaced. 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.