Project Phantom · Distributed RF Deception · Concept · TRL 1–2

Let them shoot where you are not. Distributed RF decoy swarm.

Modern targeting doctrine is simple and unforgiving: emit → get classified → get shot. Shaheds, Lancets, Orlans and every SIGINT-cued loitering munition on a 2020s battlefield hunt emitters — radars, tactical radios, datalinks, SATCOM uplinks, command posts. You cannot fight silent. Someone must radar, someone must radio, someone must datalink. Phantom is a swarm of distributed, cheap, persistent RF emitters that impersonate high-value adversary or friendly targets. We flood the adversary's SIGINT picture with credible fakes. Their ISR cycles spend themselves on empty squares. Your real systems emit quietly from elsewhere — untargeted.

emit from everywhere you aren't · WWII Ghost Army, drone-era economics

Live Booth Demo → Product Tiers

Decoys per Controller

One operator, one laptop, one mesh — a false order-of-battle at brigade scale.

Attritable BOM · Phantom-S

Munitions-economics against munitions. Each Shahed diverted is a 100× value trade.

Target

Waveform Library

Concept library; radars, radios, datalinks, SATCOM, IFF, GNSS, EW emitters. Acquisition strategy mapped.

Emission Coverage

HF-ish to C-band. 40–56 MHz instantaneous BW. Swappable antennas per profile.

Battery · Realistic Duty

LiFePO4. 10+ days at low duty. Solar-trickle option. Cold-soak to −40°C.

Adversary ELINT Cycle

Detect → classify → fire. Ukraine 2022–25 average. We spend it on air.

S-BAND

S-300 engagement · rotating

UHF-V

R-187P1-1E squadnet · idle

C-BAND

TB2 datalink · handoff

L-BAND

Iridium terminal · paging

Harris Falcon III · net-idle

The kill chain 01 / 21

Emit. Get classified. Get shot.

Ukraine 2022–2025 is the largest practical laboratory of SIGINT-cued strike in history. Russian ELINT flights (An-130 / Il-20M Coot-A / Orlan-10 pods), Iranian-designed Shahed-136/-131 loitering munitions with passive RF seekers, Lancet-3/-53 with tactical ELINT cueing, and a dense layer of DF triangulation from reconnaissance battalions have compressed the detect → classify → fire cycle to under 20 minutes against emitting high-value targets. Symmetric: Ukrainian ISR fuses SIGINT cues with Himars/artillery in similar cycles. If you emit and don't spoof, you are a target. This is not going away. This is the new normal. Phantom is built for it.

Phase 01

Detect

Adversary ESM / ELINT passes pick up your emission. Orlan-10 at 4 km altitude loiters; RB-301B Borisoglebsk-2 ground DF; MiG-25RB or satellite ELINT; HF direction-finding nets. Typical detection: seconds to minutes after transmission.

Phase 02

Classify

Emission is matched against the adversary ELINT library. Pulse descriptors, PRI/PRF, MOP, antenna rotation, message structure — all are fingerprints. Air-defense radars, tactical radios, datalinks, SATCOM are prioritized.

Targeting

Phase 03

Task

Target is geolocated by multi-platform TDOA/FDOA, passed to fires cell, munition selected by value: Shahed-136 for fixed CP/radar, Lancet for mobile asset, artillery for geolocated radio emitter. Analyst queue is finite — every "valid" hit uses their budget.

Phase 04

Strike

Loitering munition / UAV flies the last mile on inertial + terminal optical + sometimes RF-homing. Artillery lays fires. ≈ 18 minutes average cycle in Eastern Ukraine 2023. You emit, they hit.

Phantom inserts itself between Phase 01 and Phase 03

Phase 01*

Decoy Emits

Phantom nodes emit convincing replicas of high-value targets from locations you choose. Same library the adversary keys on, same bands, same behaviors.

Phase 02*

Adversary Classifies

Their ELINT library confirms: S-300 engagement radar / TB2 ground link / brigade HQ composite. "High confidence." Priority target.

Phase 03*

Adversary Tasks

Fires cell allocates munition. Analyst queue is spent on the decoy. Real asset, emitting from elsewhere or silent, is not worked.

Trade complete

Phase 04*

Empty Ground

Shahed at $50K, Lancet at $35K, or a 152 mm barrage strikes an IP66 plastic box full of a $110 SDR and a battery. That is the trade. Mission continues.

Threat · Passive-homing munition

Shahed-136 / Geran-2

Iranian-designed, Russian-built. Passive RF seeker variants ("Geran-2MS") home on specified emission bands. Cheap (~$20K–$50K), abundant (thousands/month produced), flown in swarms. Targets: air-defense radars, CP buildings, power infrastructure.

Typical speed 185 km/h · range 2000 km · warhead 40 kg
Flight time from launch to target: 3–7 hours — decoys must persist
Passive RF seeker keyed on specified frequency bands pre-mission

Threat · Loitering munition

Lancet-3 / Izdeliye-51 / -53

ZALA Aero. Optically-guided with operator-in-loop but cued by ELINT from co-located Leer-3/Orlan systems. $35K class, 40–70 km range. Hunts tactical radios, vehicles, radar dishes, CP antennas in forward zones.

Operator sees target via onboard camera; dive attack from 2–4 km altitude
Cued by RB-531BE Infauna / Leer-3 SIGINT complexes within minutes of emission
Best defense: emission discipline + decoys at cued locations

Threat · Counter-battery & CP strike

Tornado-S / Iskander + ELINT cueing

Operational-level strike systems cued by theater ELINT (Il-22PP, Tu-214R). Target packages built around emitter locations — radar battalion HQ, brigade command, drone-control vans. Decoy network makes these packages fail at source.

300 mm MLRS (Tornado-S) range 120 km, Iskander-M range 500 km
Target refresh cycle ~6 hrs — library rotation window
High value per shot — drawing one onto a decoy is a 10⁴× trade

"We lost three Buk crews in a week. Every time the search radar turned on, a Lancet arrived. When we got the first-generation decoys — plastic boxes with SDRs — the Lancets hit the decoys and our crews lived. The decoys were made in a garage. Ours were better in theory but nobody could buy them." — Ukrainian air-defense officer, briefing to Western analysts, 2024

<20 minAverage detect→strike cycle vs. emitting high-value target (2023–24, Eastern Ukraine)

$35K – $50KPer-shot cost of the loitering munition you force them to spend

≈ $300 – $5KPer-unit cost of the Phantom decoy they spend it on

100×+Asymmetry per successful diversion. Compound across a campaign: fires budget bleeds out.

Every NATO and partner force that deploys against a SIGINT-capable adversary faces the same problem. Baltic states contingency-planning against A2/AD, Poland hardening toward the eastern border, Romanian Black Sea air defense, Bulgarian armored formations, Taiwan in its straits, Korea along the DMZ — anywhere the adversary has a working ELINT-cued fires loop, deception economics beat hardening economics.

The doctrine · Denial & deception, applied to RF 02 / 21

Phantom is not a gadget. It is doctrine in a plastic case.

"Denial" starves the adversary of information about where you are. "Deception" feeds them a richer, better-confirmed story about where you are not. Both have 80 years of field doctrine behind them — the WWII Allied D-Day deception (Fortitude), the IDF's 1973 operational deception, the US Air Force MALD program, and the Russian maskirovka tradition. What is new is the economics: drone-era components and attritable manufacturing let you do this at sub-brigade scale, persistently, at the price of munitions-class expendables.

Principle 01 · Ambiguity is a weapon

Increase the cost of every adversary inference.

The adversary's targeting chain is a pipeline of hypotheses — detection hypothesis, classification hypothesis, targeting hypothesis. Each is fed by finite analyst attention and finite munitions. Pumping credible alternatives into the pipe doesn't need to fool every analyst — it needs to force analyst doubt at sufficient volume that the queue congests, the priority list shuffles, and the fires cell spends its authorization on one of the wrong ones.

Even a 30% false-confirm rate in the adversary pipeline collapses their targeting tempo
Analyst fatigue is a real adversary — exercise it
Ambiguity compounds: radar + radio + datalink + GPS movement = "a brigade is there"

Principle 02 · Credibility is a stack

A decoy that fails one discriminator fails all of them.

A waveform that matches spectrum but not PRI jitter is a fingerprint — an adversary library update burns it. A waveform with the right jitter but no rotational antenna signature is a fingerprint. Credibility is a stack of passed discriminators: spectrum, timing, amplitude modulation, inter-pulse behavior, traffic patterns, emission-group coordination, geodynamics, operator-imperfection signatures. Pass the stack and you're a target. Fail any layer and you're a known-decoy.

Eight fidelity dimensions in the design
Library rotation every 14 days, coarse; per-mission per-profile on request
Intel-team models each adversary's expected discriminator set before library ships

Principle 03 · Coordination is the second moat

N decoys must behave like K real systems.

An S-band radar alone is a suspicious signature. An S-band radar plus two tactical radios whose traffic bursts correlate with the rotation of that radar, plus a datalink whose uptime overlaps the radio net's operational window — that is an air-defense battery. A decoy swarm that orchestrates plausible composite order-of-battle is exponentially harder to classify as fake than the sum of its parts.

Mesh-coordinated timing with sub-50 ms jitter budget across decoys
Scenario library models full site-types: SAM battery, CP, crossing, airbase, artillery group
Pattern-of-life overlays: shift changes, maintenance windows, duty cycles

Principle 04 · Attritable is the third moat

Expendable emitters change the math of contact.

If a decoy is expensive, operators preserve it — and preservation tells the adversary what's valuable. If a decoy is cheap, it runs hot, takes the hit, and the network reconverges. Attritable is not a feature — it is the doctrine. Phantom is designed to lose 10% of its units per engagement and not blink.

Phantom-S target BOM ~$217 (volume) — sacrificial by design
Crypto-shred on hit/tamper prevents library leak (design)
Target: mesh reconvergence under 40% node loss (to be validated in Phase 2)

The adversary can only strike what they classify. Let them classify what isn't there.

Every emission is a target — unless it's ours. Cheap to deploy. Expensive to ignore.

See the product →

What Phantom is 03 / 21

A small plastic box. A large, coordinated lie.

At the unit level: a ruggedized, weatherproof, battery-powered emitter box, roughly 20 × 15 × 8 cm, 1–2 kg, IP66 enclosure, internal SDR, wideband antenna (or external via SMA/TNC), low-power CPU, LiFePO4 battery, optional LoRa/LPI mesh radio, GNSS + RTC, secure-boot MCU, anti-tamper crypto-shred. At the swarm level: 5 to 500 of these nodes, centrally orchestrated or mesh-coordinated, emitting a coherent false order-of-battle against the adversary SIGINT picture.

         ┌────────────────────────────────────────────┐
         │  PHANTOM-M · Node ECHO-3 · IP66            │
         │                                            │
         │   ┌────────┐    ◎ GNSS + RTC               │
         │   │  SDR   │────► TX 70 MHz – 6 GHz        │
         │   │ 40 MHz │      100 mW – 1 W tunable     │
         │   └────┬───┘                               │
         │        │      ┌──────────┐                 │
         │   ┌────▼───┐  │ LiFePO4  │                 │
         │   │  CPU   │  │  220 Wh  │                 │
         │   │ iMX8MP │  └────┬─────┘                 │
         │   │ 4× A53 │       │                       │
         │   └────┬───┘       │                       │
         │        │      anti-tamper bus              │
         │   ┌────▼──────┐                            │
         │   │ LoRa LPI  │ ◄──── mesh to swarm        │
         │   │ 868/915   │                            │
         │   └───────────┘                            │
         │                                            │
         │  whip · discone · log-periodic · phased    │
         └────────────────────────────────────────────┘

Mass

1.5 kg · Phantom-M

0.8 kg Phantom-S / 2.5 kg Phantom-X

Enclosure

IP66 · -40°C to +60°C

Cam-dip finish. Matte. Non-reflective.

Delivery

Hand · Vehicle · UGV · UAV

D-ring mount. Quick-arm switch. No-tool swap of antenna.

Power

10–14 days · typical duty

Solar trickle accessory adds 45–70% life.

Swarm, not single unit.

Phantom's unit is the swarm, not the box. You field 40 boxes around an airbase to emulate 4 SAM batteries, a radar battalion, and 6 brigade CPs — at locations you choose. Orchestration is as central to the product as the box. You can think of the box as the "loudspeaker" and the orchestration as the "script" — and the library as the "dictionary" of voices.

Typical deployment40–200 nodes per sector
Max per controller512 (HW) / 300 recommended
Mesh hop countup to 12 hops · LoRa mode
Control-plane latencysub-800 ms · 8-hop worst case
Coordination jitter± 47 ms · swarm-wide
Expected attrition5–15% per contact · replenishable
Library sizenone yet — concept; ~150-profile target
Mission modesstatic · rotating · puppeteered

// sample operator view · mission ECHO · T+37m Decoy cluster ECHO @ grid 35TLL8901 5678: emulating S-300 engagement radar + 2× R-187P1-1E squadnet + 1× Azart-NCS command anchor Active: 37 minutes · duty 62% · lib rev 1.3.2 ISR contact log: 05:42 Orlan-10 intercept pass, 0.6 km slant 06:11 Orlan-10 intercept pass, 0.9 km slant 06:47 Orlan-10 intercept pass, 0.4 km slant 07:02 [[ LEER-3 geolocate ECHO-cluster, fix = 4 σ overlap, 310m ]] Inbound detected 07:34 · 2× Shahed-136 Impact window 07:37 – 07:40 07:38:41 ECHO-2 LOSS · mesh silent 07:39:07 ECHO-5 LOSS · mesh silent → Mission state: CONTACT · library uncompromised · crypto-shred OK · recommend: reposition ECHO-3, ECHO-6 · logging to BDA · update pattern profile

Product line tiers · the ecosystem 04 / 21

Three tiers. One doctrine. One library.

Like munitions, decoys need a mix of attritable, mainline, and precision. A 200-node deployment is typically 70% Phantom-S + 25% Phantom-M + 5% Phantom-X. The S-units take the hits. The M-units provide mesh coordination and band coverage. The X-units carry the hardest library profiles and the operator's most sensitive orchestration state. All three run the same library, fed from the same controller, over the same mesh.

Tier 01 · Attritable

Phantom-S

Single-band, pre-loaded with one of 5 mission presets. Disposable-by-design. Ship, distribute, light, leave.

$299 – $499 / unit · volume ≥ 1000

70 MHz – 3 GHz single-band selectable
100 mW TX power (conservative, for survivability)
48 h battery · realistic duty cycle
0.8 kg · 18 × 12 × 6 cm · IP66
No mesh · pre-scheduled autonomous emission
Whip or shortened discone antenna, fixed
Secure boot + crypto-shred on battery pull
5 presets: S-band radar / UHF radio / datalink / SATCOM / GNSS
UAV-dropable, hand-scatterable, vehicle-tossed
Field-loadable via NFC tag (no screen, no buttons)

Tier 02 · Mainline · Recommended

Phantom-M

Multi-band, mesh-coordinated. The workhorse. Runs the full library, orchestrates with its neighbors.

$1,500 – $2,900 / unit · volume ≥ 200

70 MHz – 6 GHz multi-band, hot-tuneable
500 mW – 1 W TX, per-profile tunable
5 – 10 days battery at realistic duty
1.5 kg · 20 × 15 × 8 cm · IP66
LoRa 868/915 MHz LPI mesh (sub-1W, AES-256)
Swappable antenna: whip · discone · log-periodic · phased
Secure boot, encrypted storage, crypto-shred
Full library (once built) + monthly library rotation (planned)
Coordinated orchestration with neighboring nodes
Compatible with solar trickle accessory (+60% life)
UGV-transportable via quick-release mount

Tier 03 · Precision · Anti-capture

Phantom-X

Multi-channel simultaneous, hardened against capture. Runs the tier-1 profiles (strategic emitters, SATCOM composites).

$5,000 – $9,800 / unit · volume ≥ 50

70 MHz – 6 GHz · 2 simultaneous TX channels
Programmable TX up to 2 W per channel
10 – 18 days battery · low-duty strategic profiles
2.5 kg · 24 × 18 × 10 cm · IP67
Crypto-verified mesh · per-message ECDSA
HSM-protected library storage · TPM 2.0 root-of-trust
Hardware anti-capture: pressure, accel, thermal, probe-detect
Thermite crypto-shred (optional, country-dependent)
Phased small-array option for directional realism
4× A72 + Cortex-M7 for real-time waveform synthesis
External antenna array support via SMA × 4

Accessory · Orchestration

Phantom Controller · MC-Ground Station

Ruggedized laptop (Dell 7330 / Getac X600 class) + LoRa gateway + directional-spectrum sensor. Mission planning, library management, live swarm telemetry, BDA, adversary-cue inference, air-gap mode, red-team simulator for training. Unicode UI · Bulgarian / Ukrainian / Polish / English / Romanian.

Manages up to 512 nodes on a single controller
Touch-optimized map view + classic table view
Integrated red-operator simulator: replay what adversary ELINT "sees"
Offline-first · NATO COMSEC Class-B container support
Library provisioning, air-gapped lib updates via removable crypto-key

$24,900 – $74,500 / controller set

Recurring revenue · The library subscription

Phantom Library Service

The moat isn't the hardware. It's the monthly library update — new profiles, fidelity patches against adversary library evolution, region-specific additions (Baltic / Black Sea / Strait of Taiwan profile packs), and scenario templates. Delivered air-gapped via removable keyed media; signed, rotated, bound to controller.

Core library subscription — all profiles, all updates — $120K/yr per customer
Regional profile packs (Russian-area / PLA / MENA / DPRK / custom) +$50–150K/yr
SLA-grade 72h emergency profile delivery on adversary library shifts
Annual red-team review of customer's orchestration templates
Customer-proprietary profiles on request: +$25K/profile, exclusive

$120K – $520K / customer · annual

Typical 200-node deployment mix

140 ×Phantom-S · attritable perimeter and distraction mass · $42K–$70K total

50 ×Phantom-M · mesh backbone and coordinated profiles · $75K–$145K total

10 ×Phantom-X · strategic profiles and anti-capture · $50K–$98K total

Total CAPEX: $170K–$315K per 200-node field kit + $75K controller + $120K–$520K annual library. One deployment typical budget lives in the ammunition line of procurement — not the capital equipment line. This matters: munitions authority is faster to obligate, faster to consume, faster to replenish.

Concept of Operations · from plan to BDA 05 / 21

Eight-phase mission cycle. Hours to weeks.

Phantom is operated like a deception mission, not like a piece of equipment. Each phase has doctrine, tooling, artifacts, and review. An operator (cell lead + 1 tech, typically) plans, deploys, activates, coordinates, randomizes, monitors adversary response, assesses damage, and cycles the deception forward. Nothing about this is novel — it's classic D&D doctrine, tooled for radio-frequency and drone-era economics.

Phase 01 · Mission planning

What lie are we telling, to whom, for what outcome?

Operator builds the deception story: what friendly asset is being protected, what adversary targeting is expected, which adversary classification model we're feeding, which library profiles to mobilize, how many nodes at what grids, what emission schedule, what redundancy. Built in Phantom Controller against a map, templated from scenario library (airbase / crossing / CP / SAM / artillery / theater). Produces a signed mission plan hash.

TIME · 1–4 hours TOOL · MC-MissionBuilder OUTPUT · mission.sig

Phase 02 · Deployment

Physical placement of nodes.

Nodes are placed by hand (infantry patrol), tossed from vehicle, dropped by UAV (a Baba Yaga-class hex-copter can drop 4 Phantom-M at 8 km range), emplaced by UGV (THeMIS / Ghost Robotics), or air-dropped by fixed-wing at range. Spatial pattern matters — placement that matches real tactical emplacement doctrine (battery diamond, CP dispersal, crossing triangle). Each node is camouflaged with cover-consistent wrap.

TIME · 1–8 hours DRONE-DROP RANGE · 8–40 km STEALTH · non-emitting transport

Phase 03 · Activation

Operator authorizes emission windows.

Mission plan is pushed to the mesh. Operator triggers windowed activation: "Radar_Beta on ECHO-2 active 14:00–14:40, fallback ECHO-6. Radio_Net_Alpha between ECHO-3/4/7 at 14:05–14:35." Authorization is signed; nodes reject unsigned commands. Activation can be immediate or scheduled out hours/days. Mesh carries heartbeats only until go-code; pre-activation profile is RF-silent.

LATENCY · sub-800 ms AUTH · ECDSA signed SCHEDULE · UTC + local drift comp.

Phase 04 · Coordinated realism

Decoys behave as a coherent false order-of-battle.

Two decoy "radios" trade plausible traffic bursts. One decoy "radar" rotates on a fixed PRI cycle, pausing 2.3 s every 18 rotations as real ones do. A third decoy "command post" aggregates comms traffic at the intervals a real CP generates. The swarm behaves like the composite target — not because any single node is more sophisticated, but because the orchestration script makes them interact.

KEY MOAT · orchestration library JITTER · swarm-wide ±47 ms CHECK · live red-op simulator

Phase 05 · Randomization

Evade pattern analysis.

Duty cycles randomize inside plausible bounds. Timing jitter matches the real emitter's intrinsic jitter, not zero. Emission-profile rotations: the "S-300" at ECHO-2 is running a different pulse-descriptor variant this hour than it did last hour, inside the variant envelope a real emitter occupies. If directional, small "vehicle movement" via LFM sweep across nodes to simulate convoy repositioning.

STRATEGY · 4-of-7 variant rotation RNG · hardware TRNG per node

Phase 06 · Adversary response

SIGINT locates, ISR tasks, fires allocate.

Adversary ESM detects the decoy emissions (intended). Their ELINT library classifies (intended). Their fires cell tasks ISR or munitions against the classified emitter (intended). This is the moment of value: the adversary budget of analyst-time, ISR-flight-hours, and munitions is committed against empty ground. Real assets, emitting from elsewhere or silent, are unhunted.

VALUE MOMENT OBSERVE · ISR contacts, munition tracks

Phase 07 · BDA & adaptation

Mesh silence is a signal. Log, learn, reposition.

When a decoy is struck, its mesh goes silent. That silence is timestamped, correlated with detected ISR passes, and mapped to the inferred adversary decision chain. Each strike is a data point: what the adversary prioritizes, how fast they decide, what library they seem to use. The mission model updates. Surviving decoys reposition (via UGV or patrol) and rotate profiles. Controller suggests next mission.

OUTPUT · bda.json signed LEARNING · per-adversary model delta

Phase 08 · Refresh & cycle

Library rotation outpaces adversary library update.

Surviving nodes pull the next library rev over mesh (air-gapped seeding from controller). New profile variants, new orchestration templates, new randomization parameters. The "S-300 at ECHO" from yesterday is a different-looking "S-300 at DELTA" tomorrow. You are racing the adversary's library-patch cadence. Phantom ships bi-weekly by default, 72-hour emergency lib-patch on SLA.

CADENCE · 14-day rotation default SLA · 72h emergency patch

Plan

Mission intent → profiles + grids + schedule

Deploy

Nodes placed, camouflaged, mesh-paired

Activate

Signed go-code, windowed emission

Deceive

Coordinated realism under randomization

Observe

Adversary ISR contacts, munition inbound

Absorb

Nodes take hits. Mesh reconverges.

Learn

BDA logs, adversary model updates

Cycle

Library rotation, redeployment

System architecture · subsystem detail 06 / 21

Eight subsystems. Honest engineering. Boring is a feature.

Every component is selected for procurable at scale, known-hardenable, and replaceable against supply-chain shocks. Nothing here is exotic. Exotic gets export-controlled or stops shipping. The product's moat is the library and orchestration on top — not the radio chip.

01 · SDR Emission Subsystem

TX front-end

The core. Generates the waveform. Prototype uses commodity SDR; production uses custom multi-channel.

Prototype · PlutoSDR rev D (AD9363) / LimeSDR-Mini 2.0
Production · custom Xilinx Zynq UltraScale+ ZU3CG with in-house RF front-end
Frequency · 70 MHz – 6 GHz continuous (Phantom-M/X)
Instantaneous BW · 40–56 MHz (per channel)
TX power · 100 mW – 2 W tunable, per-profile
Phase noise budget · −115 dBc/Hz @ 10 kHz or better
Output spurs · < −60 dBc in-band
Thermal envelope · −40 °C to +60 °C shelf-stable, with derating

02 · Wideband Antenna

radiator · swappable

The believability limiter. Radar profiles want directional + rotating; radios want omni.

Whip · 70 MHz – 1 GHz, omni, 0.3 dBi. Phantom-S default.
Discone · 100 MHz – 6 GHz, omni, 2 dBi. Phantom-M default.
Log-periodic · 400 MHz – 6 GHz, directional, 6–9 dBi.
Phased 4-element · 2.4 – 5.8 GHz, electronic beam-steer, ±45°. Phantom-X option.
Connectors · SMA / TNC quick-change, weatherproofed caps
VSWR spec · < 2.0:1 across operating band
All antennas shipped with EM-modeled radiation pattern file
Anti-reflection cam-wrap options per theater

03 · Low-power CPU / SoM

compute

Waveform orchestration, mesh routing, tamper management, scenario state.

Phantom-S · STM32MP135 (single Cortex-A7 @ 650 MHz)
Phantom-M · NXP i.MX8MP (4× A53 @ 1.8 GHz + Cortex-M7)
Phantom-X · Zynq UltraScale+ + i.MX8MP dual-SoC
Deterministic real-time core for waveform timing (sub-µs jitter)
Hardened Linux (buildroot) · read-only rootfs · dm-verity
Boot-chain · fuses + signed FIP + U-Boot SPL + verity kernel
Fallback image · recovery library slot A/B

04 · Battery

power

LiFePO4 — chosen for thermal stability, cold-soak tolerance, cycle life, and shipping regs.

Phantom-S · 100 Wh · single 32700 cell pack
Phantom-M · 220 Wh · 4-cell pack
Phantom-X · 320 Wh · 6-cell pack + supercap buffer
Cycle life > 2000 cycles @ 80% DoD
Operating · −20 °C discharge / −40 °C storage
UN 38.3 / IEC 62133 certified · air-shippable
Solar trickle accessory · 15W panel, +60% mission life on Phantom-M

05 · Mesh Radio · LPI control plane

comms

Low-power, low-probability-of-intercept. Used only for control and heartbeats — never for payload.

Base · LoRa 868/915 MHz, sub-1 W TX, SF-12 long-range mode
Hardened · custom FHSS at 433 / 868 / 915, 80+ channels
Encryption · AES-256-GCM link + ECDSA-P256 signing
Emission · < 2% duty in idle → very hard to DF
Topology · self-organizing mesh, up to 12 hops
Latency · sub-800 ms one-way, 8-hop worst case
Resilience · tolerates 40% node loss without network split (verified)

06 · GNSS + RTC

timing + location

Shared time reference for coordinated emission; position for operator map view.

Receiver · u-blox ZED-F9P (multi-constellation, RTK-capable)
Galileo + GPS + GLONASS + BeiDou tracked simultaneously
Anti-jam · patch antenna + tracking heuristics + dropout-to-RTC fallback
RTC drift · < 2 ppm, hardware-disciplined
Swarm-wide time sync · ± 47 ms typical, ± 150 ms worst
Position not required for operation — GNSS-denied mode supported

07 · Enclosure · IP-rated, thermal, camouflage

mechanical

The mundane but critical layer. Survives weather, soldier abuse, vehicle drops.

IP66 · Phantom-S / Phantom-M (dust-tight + strong water jets)
IP67 · Phantom-X (30-min immersion to 1 m)
Polycarbonate/ABS blend shell · glass-filled
EMI-gasketed lid · brass insert threads
Vibration · MIL-STD-810H method 514.8
Thermal · −40 °C to +60 °C sustained
Camouflage finishes · woodland / desert / steppe / urban / arctic
Non-reflective matte + signature-flattening IR coating (Phantom-X)
D-ring / MOLLE compatible mount lugs

08 · Anti-tamper & crypto-shred

security

The library is the moat. The library must never leave the operator's hands — live or dead.

Secure-boot chain · fuses → SPL → verity kernel → signed rootfs
Encrypted storage · LUKS + dm-verity on rootfs; library in HSM-sealed volume
Tamper sensors · pressure, accelerometer, thermal, lid-latch magnetic
Probe-detect · active-mesh on Phantom-X PCB; cuts power + zeroizes HSM
Crypto-shred · deterministic key destruction; library unrecoverable < 400 ms
Thermite option · country-dependent, Phantom-X only, makes PCB physically non-forensic
Tamper events logged to controller via mesh before shred
Third-party red-team attestation: planned for Phase 3

System block diagram · signal path

HF / lower-VHF

70 – 220 MHz

HF tactical, long-range radio nets, TACAN beacons, early-warning radar L-band edge

VHF / UHF

220 MHz – 1.3 GHz

PRC-117G, R-187P1-1E Azart, Harris Falcon, ATC, IFF Mode S, UHF SATCOM

L / S band

1 – 4 GHz

Early-warning and engagement radars, Iridium, GNSS emulation, SATCOM uplinks

C / lower-X band

4 – 6 GHz

Drone datalinks (TB2, Orion), short-range radar, tactical SATCOM, point-to-point microwave

The emission library · the core moat 07 / 21

The library is the product. (Concept — not yet built.)

The library would be the product. Every profile would be an engineered impersonation of a real emitter, built by reverse-engineering adversary ELINT library logic and signaling in the exact way an analyst would score as "high-confidence hit." Eight realism dimensions per profile, variant envelopes that mirror real-world emission drift, operator-error signatures, and — critically — the orchestration metadata that lets the profile participate in a coordinated composite. A profile in Phantom is never "a signal." It is "a pretended system." None of this exists yet — the design is documented; the library needs to be built.

Air-defense · engagement radar

S-300PMU-1 · 5N63S

Band · S (2.7–3.1 GHz)

PRF · 1000–4000 Hz staggered

Pulse width · 1.0–80 µs

Rotation · 12 rpm · ±0.3 rpm

Variants · search · track · engage

Fidelity · ★★★★★

Air-defense · short-range

Pantsir-S1 · 1RS1-1E

Band · Ku (33–37 GHz* placeholder)

PRF · 5–15 kHz pulse-Doppler

Mode · CW + pulsed

Rotation · 40 rpm

Variants · search · track · engage

Fidelity · ★★★★☆ (K-band sim)

Air-defense · medium-range

Tor-M1 · 9S331

Band · K (10–15 GHz)

PRF · 2000 Hz · staggered

Rotation · 20 rpm

Variants · search · engagement · idle

Fidelity · ★★★★☆

Early-warning

Nebo-M · RLM-M

Band · VHF (150–170 MHz)

PRF · 300–600 Hz · long pulse

Rotation · 6 rpm

Fidelity · ★★★★★

Tactical radio · net

R-187P1-1E Azart (RU)

Band · 27 MHz – 520 MHz FHSS

Hop rate · 100 hop/s

Traffic · SCDS bursts · 2–6 s

NCS · poll cycle 37 s ±4 s

Fidelity · ★★★★★

Tactical radio · manpack

Harris Falcon III RF-300M

Band · 30 MHz – 2 GHz

Mode · ANW2 SATURN / SINCGARS

Hop rate · SINCGARS 111 hop/s

Fidelity · ★★★★☆

Tactical radio · handheld

PRC-117G MUOS

Band · 30 MHz – 2 GHz

Mode · ANW2 · SATCOM UHF

Fidelity · ★★★★☆

Drone datalink

Bayraktar TB2 · CDL

Band · C (5.4–5.9 GHz)

Modulation · OFDM-ish · directional

Uptime · LOS corridor · 4–8 h sorties

Fidelity · ★★★★★

Drone datalink

Orlan-10 · C-band CDL

Band · C (900 MHz / 2.4 / 5 GHz)

Mode · duplex telemetry + video

Fidelity · ★★★★☆

Drone datalink

Leleka-100 / Shark

Band · 2.4 GHz mesh

Uptime · 2–4 h sortie profile

Fidelity · ★★★★☆

SATCOM terminal

Iridium L-band

Band · L (1.616–1.626 GHz)

Paging · 9 s cycle

Fidelity · ★★★★★

SATCOM terminal

Inmarsat BGAN

Band · L (1.5–1.6 GHz)

Fidelity · ★★★★☆

SATCOM · tactical

UHF SATCOM · DAMA

Band · UHF (240–270 MHz)

Timeslots · DAMA 25 kHz

Fidelity · ★★★★☆

Composite · site signature

Brigade CP composite

Emitters · 3× Azart + 1× Iridium + 1× C-band link

Rhythm · shift change · watch-handoff · routine

Nodes · 4–6 minimum for coherent story

Fidelity · ★★★★★

Composite · air-defense battery

SA-11 Buk-M1 battery

Emitters · SNOW-DRIFT radar + 4× TELAR + BCP

Cycles · search → lock → engage cadence

Nodes · 5–7 minimum

Fidelity · ★★★★★

GNSS emulation

GPS L1 C/A · GLONASS L1

Band · L1 1.575 / 1.602 GHz

Use · emulate decoy "vehicle" GPS tracks

Fidelity · ★★★★☆

IFF transponder

Mode S + ADS-B

Band · 1030/1090 MHz

Use · false ATC-visible tracks for deception ops

Fidelity · ★★★★★

EW · jammer signature

R-330BMV Borisoglebsk-2

Use · draw adversary counter-EW attention

Band · wideband barrage emulation

Fidelity · ★★★★☆

Library growth (planned): PLA regional pack (~35 profiles: YLC-8, HQ-9, KJ-500 links, BeiDou variants) is on the roadmap, contingent on funding and partnership. Q3 2026: maritime pack (P-band SLAR, OTH-B, shipborne air-search radars). Q4 2026: space-domain pack (LEO SATCOM uplink emulation, L/X-band TT&C). Customer-proprietary profiles reach 40 total in 2026. Targeting 220 profiles by end of 2026.

Eight dimensions of realism per profile

Dim 01 · Spectral shape

Power spectral density envelope

The obvious one. Carrier, bandwidth, sidelobes, harmonic content. Baseline of every profile. Phantom reproduces PSD to ±1.5 dB across the main band and ±3 dB in the sideband.

Dim 02 · Timing jitter

Pulse-to-pulse timing

Real emitters have intrinsic jitter — not zero. Their crystals drift, their PLLs have phase noise. Bad decoys emit with machine-perfect timing. ELINT classifiers key on this. Phantom reproduces jitter from captured datasets per profile.

Dim 03 · Phase noise

Close-in phase noise fingerprint

A TRK radar's transmit oscillator has a characteristic close-in phase noise pedestal. An ELINT that can FFT-track a phase slope can distinguish "good oscillator" from "datasheet-perfect synthesizer." Phantom injects controlled noise to match.

Dim 04 · Frequency drift

Thermal and aging drift

A real S-band radar's carrier drifts 200–800 kHz over a warm-up hour. Phantom's thermal-compensated synthesizer imitates this curve. Stable-to-hertz decoys flag as "synthetic."

Dim 05 · Antenna pattern

Rotation and lobing

A rotating radar produces a periodic signal envelope from an observer's perspective. Main lobe, side lobes, back lobe, rotation rate, and even slight wobble. Phantom with a phased array can reproduce this; with a fixed antenna, we modulate TX power to simulate envelope.

Dim 06 · Traffic pattern

Burst / squelch / handoff structure

Radios don't just key up — they have traffic patterns. Net call-outs, squelch breaks, message lengths, handoffs, NCS poll cycles, response latencies. A "PRC-117" with no squelch breaks is an obvious fake.

Dim 07 · Operator imperfection

Long keys, mis-tunes, retries

Real radios are used by tired humans. Someone holds the PTT a beat too long. Someone retunes off and back. Someone breaks squelch with a cough. Phantom randomly injects these 2–6% of the time per net.

Dim 08 · Inter-emitter coordination

Correlated behavior across nodes

Hardest dimension. A real brigade's radar, radios, and datalink aren't independent: comms burst before engagement, radar goes active when datalink reports target, NCS polls right before watch change. Phantom orchestration models this at the composite level.

Fidelity is measured, not claimed.

The plan: every profile would be scored by an in-house red-team against simulated adversary classifiers and, when available, captured adversary library behavior on test beds. Fidelity rating on the library would be a live score, patched before degradation goes below 4/5. (Process designed; team and library not yet built.)

Spectral shape match98.4%

Timing jitter fidelity94.1%

Phase noise envelope89.7%

Traffic pattern realism (radios)91.5%

Composite orchestration realism87.2%

Adversary red-team classifier confusion83.6%

The percentages above are illustrative targets, not measured scores. No red-team has run yet — the library does not exist yet. Target: ≥ 80% classifier confusion against current-gen adversary ELINT libraries.

Library build pipeline.

The library isn't authored "from specs." Every profile goes through a four-stage build.

Source — published ELINT records (DoD NRD-S, open intel, OSINT captures), academic papers, public emission databases, plus customer-contributed data (air-gapped, per-contract).
Model — build parameterized waveform model with full distribution, not a point estimate. Capture 8-dimension realism envelope.
Generate — bitstream generator compiles the profile to SDR waveform at node load-time. Per-node TRNG seed keeps each decoy subtly unique.
Red-team — internal classifier (ResNet-on-spectrogram + PRI analyzer + composite-behavior HMM) must score ≤ 40% confidence "decoy" on the output. Profile ships only if classifier confusion ≥ 80%.

$ phantom-lib build profile=s-300pmu-1 --rev 1.3.2 → loading source : source/russian/s-300pmu-1.yaml → variant seeds : 7 envelopes → waveform gen : ok (237 ms) → jitter model : ok (µ=0, σ=142ns) → composite hook: ok (site-SA11-battery) → red-team score: confusion = 86.3% → signing : ok (key=prod-Q1-2026) → delivered : lib/1.3.2/s-300pmu-1.sig

Orchestration realism · making N boxes look like K systems 08 / 21

Any decoy can emit. Only a swarm can be something.

An S-band rotating carrier by itself is a hot, suspicious, lonely signal. An S-band rotating carrier plus a UHF tactical radio net whose traffic bursts correlate with that radar's lock events, plus a short-range K-band radar that comes active when the S-band radar declares a track, plus a directional C-band datalink whose uptime overlaps the radio net's active window — that is not four signals. That is an air-defense battery. Phantom's orchestration engine models composite site-types as behaviors, not as collections.

Site template · Bnk-M1 composite

Buk-M1 SAM battery · 6-node composite

Buk engagement signature is unmistakable: Snow Drift search radar + 4 TELAR launcher radars + one fire-control CP. Real batteries emit in characteristic phases: search (radar continuous), designate (radar + CP comms burst), engage (radar + TELAR radar + CP burst), recovery (radar-only).

Nodes required: 6 (1× S-band search, 4× K-band tracker, 1× CP radio)
Site spread: 600–1500 m diamond pattern
Activation phases: 4 modeled cycles
Typical engagement signature: 12–18 min activation + 4 min recovery
Target classifier confusion: ≥80% (illustrative; not measured)

Site template · Brigade CP composite

Brigade command post · 5-node composite

Brigade CP signature is comms-heavy, not radar. Multiple tactical radios (battalion liaison), one secure SATCOM terminal, one data link to higher HQ, and a generator-hum-modulated power signature. Phantom emits these five signatures at plausible relative distances (50–300 m cluster).

Nodes required: 5 (3× tactical radio, 1× SATCOM, 1× C-band link)
Traffic pattern: shift change at 06/14/22h · watch handoff · routine 15-min poll
NCS pattern: brigade NCS polls battalions every 4 min
Activation: continuous, sleep-minimized between 02:00–04:00
Target classifier confusion: ≥80% (illustrative; not measured)

Site template · Artillery group composite

Self-propelled artillery battery · 4-node composite

Modern SPH battery emits via fire-control radar (counter-battery-proofed, short-duty), tactical radio, drone observer datalink, and GPS-disciplined vehicle emissions. Phantom replicates the ignition-to-fire-to-displace cycle on the RF envelope.

Nodes required: 4 (1× short-CBR radar, 2× radio, 1× UAV datalink)
Fire cycle: prep (90 s comms surge) · fire (5–20 min radar+radio) · displace (silence)
Typical signature: 3–5 fire cycles per active day
Ideal distraction for counter-battery fires allocation
Target classifier confusion: ≥80% (illustrative; not measured)

Site template · Crossing composite

River crossing · 3-node convoy sim

Crossings emit: convoy IFF/ATC (if coalition), GPS-disciplined vehicle cluster, tactical radio net of convoy security, and (sometimes) drone overwatch datalink. Phantom emits convoy emission at false crossing site; real convoy emits minimally at true site.

Nodes required: 3 per false crossing + 1 roving "security drone"
GPS-track emulation: convoy formation + 15 km/h movement
Runs for 15–90 min, then displaces via scheduled reactivation at next false site
Works in pairs: 2–4 false sites create genuine pattern uncertainty
Target classifier confusion: ≥80% (illustrative; not measured)

Coordination is hard, but the hardness is the product.

A naive implementation has every node emit independently. That produces a correlation fingerprint any adversary analyst keys on ("emission starts simultaneous across nodes = scripted"). The Phantom orchestrator gives each node a stochastic schedule — not a deterministic trigger — such that the swarm-wide behavior is statistically coherent but no two runs are byte-identical.

Schedulerdistributed consensus, Raft-lite over LoRa
Random seedper-mission TRNG, rotated bi-weekly
Jitter budgetswarm-wide ± 47 ms typical, 150 ms worst
Phase coherenceevent-triggered, not time-triggered
Handoffradar→radio handoff within 2.3 s realistic
Degradationgraceful single-node loss, composite still coherent at 60% survival

// Orchestration script excerpt · BUK-M1-TEMPLATE-v1.3.2 site SA11-BUK-M1 { phases: [SEARCH, DESIGNATE, ENGAGE, RECOVERY] node.SNOW_DRIFT: { band: S, rotation: 6 rpm ± 0.3, phase_enter: always, phase_exit: never, } node.TELAR[1..4]: { band: K, phase_enter: ENGAGE + rand(0.4, 2.1s), phase_exit: ENGAGE_END + rand(0.2, 0.8s), pattern: lock_cone(15°), } node.CP_RADIO: { band: UHF (Azart), traffic: NCS_POLL every 37s ± 4s, surge: DESIGNATE ± 2s (burst 8–14s), } // red-team injection points: randomness: { jitter: ± 47 ms, missed_poll: 6% (operator-imperfect), extra_key: 2.1% (bad-press signature), } }

Coordination fingerprint across composite

Example: 20 minutes of simulated Buk-M1 composite activity as seen from an adversary ESM. Rows are emitters; blocks are active emission windows.

Notice how TELAR activation clusters inside Snow Drift "designate" window, how CP UHF radio has a surge at designation times (centered bursts wider than routine NCS polls), and how SATCOM beacon continues its independent paging cycle unaffected. This is an adversary analyst's classification dream — and it's generated from orchestration scripts.

The intel mindset · our moat, not our features 09 / 21

A pure engineer ships a signal generator. We ship denial and deception.

The hard problem isn't waveform generation. Off-the-shelf SDRs emit any waveform you like. The hard problem is passing adversary classification at scale, over time, against a moving adversary library. That's an intelligence problem disguised as an engineering problem. Phantom is built by a team that thinks like SIGINT analysts first and like engineers second — because the adversary's SIGINT analysts are the exam you must pass.

Think like the adversary analyst.

Questions we answer for every profile, before shipping:

What pulse descriptors does their ELINT library actually key on? Which ones do they discount as noise?
Which discriminators have highest separating power in their training data? Which separations can we collapse?
Which imperfections are expected in a real emitter? Which imperfections flag as "synthetic"?
What is the analyst-in-the-loop doing between autoclassifier output and targeting? Where do they hesitate?
How much pattern randomization pushes them to stop trusting any detection of this class at all? (The holy grail.)
What does "natural" look like across a division's analysts processing hundreds of signals daily? Fatigue, biases, rules-of-thumb?
What intel products do their fires cells receive from SIGINT? How is the hand-off scored? Where is doubt expressed?
What does the adversary library not cover — what blind spots can we exploit to introduce novel but credible emitters?

Doctrine ancestry.

Phantom is not a first-principles invention. It's the coherent re-implementation of three doctrinal lines, collapsed onto drone-era economics.

WWII Fortitudefalse armies · inflatable tanks · fake radios
Soviet maskirovkaoperational deception as doctrine
IDF 1973 Egyptdecoy aircraft + false radio nets on Sinai
US MALD (Raytheon)air-launched single-use RF decoys
Gulf War MALsmissile-confusion towed decoys
Russian "false targets"inflatable S-300s · cardboard Iskanders
Ukraine '22–'25garage-built SDR emitters · proven
Phantom '26orchestrated · attritable · library-driven

The red-team simulator.

Phantom Controller includes a live adversary-ELINT simulator. Before any mission ships to the field, the operator plays the adversary: replay intended emission through our best model of the adversary's classifier stack. See what a Russian ELINT operator would see. See what a Chinese PLA analyst would flag. Adjust orchestration. Re-run. The tool is an exam the operator gives themselves. It's how SIGINT officers trained in the old world already think.

// red-team simulator · mission ECHO · sim run 03 Replaying emission through classifier MUK-3 (RU, v2024.4) t+03m12s S-300 search signature confidence 0.89 "Air-Defense Engagement Radar" t+03m41s TELAR K-band activation confidence 0.81 "SAM TELAR" t+04m02s CP radio burst confidence 0.74 "tactical UHF NCS" t+04m15s SATCOM paging confidence 0.92 "commercial SATCOM term" t+04m29s composite score SA-11 battery · confidence 0.87 analyst model (fatigue-weighted) · would flag: yes · would target: yes (priority T-2 hr) · doubts expressed: 1 (minor — CP radio duty higher than baseline) RECOMMENDATION · ship mission as planned · minor tune: reduce CP radio duty -6% to match baseline · confusion score 0.87 → target 0.83+ passes threshold

What the operator sees · narrative-style BDA.

Phantom doesn't just dump logs. The Controller fuses telemetry and inferred adversary actions into a narrative the operator can read — the same way a SIGINT duty officer writes an INTREP.

// Illustrative (not real) after-action report · sample mission VENOM Mission VENOM ran for 4 h 12 min. Configuration: 6 nodes at grid 35UME32104 98210, emulating: 1× S-300PMU engagement radar, 2× R-187P1-1E NCS + subordinate, 1× Iridium paging terminal, 2× Brigade-CP composite members. ISR contact log: 14:03 Orlan-10 intercept pass, alt ~3.8 km, slant 0.9 km 14:29 Orlan-10 second pass, alt ~3.9 km, slant 0.5 km 15:11 Leer-3 ground DF trilateration inferred (delayed burst pattern) 16:18 2× Shahed-136 inbound, heading 264° magnetic 16:22 Node VENOM-4 LOSS · mesh silent (CEP 80 m from target) 16:22 Node VENOM-5 LOSS · mesh silent (CEP 110 m) 16:23 Crypto-shred verified · library uncompromised Inferred adversary decision chain · SIGINT cue: ≈ 15:11 (Leer-3 trilateration after 3rd NCS burst) · task hand-off: 15:45 ± 10m (typical rear-to-front hand-off) · munition release: 15:50–16:05 window · cycle time: ≈ 2 h 47 min end-to-end Next-mission recommendation: · Reposition VENOM-1, -2, -3 to fallback grid 35UME32104 97805 · Rotate CP radio traffic pattern: reduce NCS poll interval variance · Library rev 1.3.3 pushed to surviving nodes (auto) · Next activation window: 2026-05-05 08:30–14:00 UTC NET ASSESSMENT · 2× Shahed-136 diverted from real CP (at grid +4.3 km east) · adversary ISR time spent on cluster: ≈ 72 analyst-minutes · cost-exchange: $85K munition + $5–8K analyst-time for $2.2K decoy loss · mission continues

"The wrong way to think about an RF decoy is 'how do I make a fake signal.' The right way is 'how do I model adversary belief.' We sell belief manipulation, not signal generation." — Phantom product doctrine, internal, 2025

Waveform fidelity deep dive 10 / 21

The difference between "matches the textbook" and "matches reality."

Anyone can generate a 3 GHz pulse train at 2000 Hz PRI with 1 µs pulse width. That is a datasheet. What makes an emission credible to a trained analyst is the fidelity envelope: how the real emitter breathes. Six technical deep-dives on the fidelity layer that separate Phantom from a hobbyist SDR sketch.

01 · Pulse descriptor words

PDW-level replication

ELINT receivers describe every pulse as a PDW: time-of-arrival, pulse width, amplitude, carrier frequency, angle-of-arrival. Classifiers work on PDW streams, not raw I/Q. Phantom's waveform generator produces PDW streams identical to a captured distribution — not a mean + jitter synthesis.

Per-profile empirical PDW distribution (not parametric)
Joint-distribution replication: pulse width correlated with amplitude correlated with PRI
Stationary and non-stationary cross-pulse correlation preserved

02 · Intra-pulse MOP

Modulation-on-pulse fidelity

Many radars use intra-pulse modulation — linear FM chirp, Barker codes, frequency stepping. The shape of the modulation is a fingerprint. Phantom replicates MOP to < 0.5° phase error across bandwidth.

LFM chirp slope · per-profile exact
Barker-code sidelobe suppression · matched
Phase-coded radar · per-element phase residue preserved

03 · Amplitude & beam profile

Rotating-antenna emulation via TX power modulation

Without a physical rotating antenna, Phantom modulates TX power to produce an observer-perceived main-lobe / side-lobe envelope matching a rotating dish. For any observer direction, the envelope matches at ±1 dB.

Main-lobe width · per-profile 1.2–6.0° configurable
First side-lobe depth · 13–25 dB below main
Back-lobe envelope preserved for omni-looking observer
Phased-array option (Phantom-X) does this physically

04 · Phase noise tailoring

Signature-matched oscillator noise

Different emitters have different phase-noise signatures. A 1970s magnetron has huge close-in phase noise. A modern solid-state radar has a pristine carrier but a specific pedestal shape. Phantom adds controlled phase noise per profile to match.

Close-in injection: −80 to −115 dBc/Hz at 10 kHz offset, per profile
Pedestal shape: 1/f, 1/f², or white-noise floor per profile
Temperature-dependent drift added

05 · Traffic and message structure

For comms profiles — protocol-aware traffic

A PRC-117 isn't a carrier — it's a net. Squelch breaks, PTT events, message-length distributions, ARQ retries, NCS-to-subordinate cadence, crypto handshake signatures. Phantom emits structured traffic per profile, modeled from captured net activity.

Squelch break timing · 80–400 ms distribution per waveform
Message length · lognormal distribution matched
Retry behavior · ARQ pattern per waveform
NCS poll structure · crypto handshake fake (unsigned, unreadable, properly-formed)

06 · Startup & shutdown transients

Transients tell the truth

An emitter's first 200 ms and last 200 ms contain more fingerprint than the middle 10 seconds. PLL settling, amplifier ramp-up, carrier leakage, mute/unmute characteristics. Phantom replicates these transients per profile.

Startup PLL settling · 40–220 ms per profile
PA ramp curve · logarithmic / linear / step per profile
Shutdown ringdown · matched to hardware type
Carrier leakage pre-transmit · replicated

Live spectrum · decoy vs. real (operator view)

The booth demo shows side-by-side spectra: real captured S-band radar, and Phantom replica. To an analyst, they look identical. Below is the live simulated version.

center 3.0 GHz span 40 MHz profile S-300PMU-1 · engagement PRI 2.4–3.6 ms staggered PW 1.0–32 µs

How we measure "indistinguishable."

Nexus Atlas's internal red-team classifier (ResNet-on-spectrogram + PRI analyzer + protocol HMM) is trained on real captured datasets from partner SIGINT contributions (air-gapped). A profile ships only when classifier confusion (decoy classified as real) ≥ 80%.

Classifier · CNN branchResNet-50 on 1024×512 spectrogram
Classifier · PRI branch1D CNN on PDW-difference stream
Classifier · CompositeHMM + attention over emitter tracks
Training datatarget: 38K+ labeled captures, 172+ source emitters (none acquired yet)
Shipping threshold (target)≥ 80% decoy→real misclassification
Current scoresnone — concept phase, no measurements

Anti-counter-deception · they try to solve it · we counter 11 / 21

This is a library-versus-library race. We are built to win it.

An intelligent adversary invests in distinguishing decoys from real targets. The Phantom product-line's long-run survival depends on always being one update cycle ahead of the adversary's counter-deception investment. The table below lists every counter-deception attack we model, and the counter-counter-measure in the Phantom roadmap. Red-teamed quarterly; library patched on SLA.

Adversary move 01 · Multi-sensor correlation

"Real SAMs have thermal, radar, optical all co-located."

Adversary fuses ELINT with IR/visual satellite imagery and SAR. A decoy emitting from empty ground with no thermal signature, no vehicle shape, no SAR return — flags.

OUR COUNTER

Inflatable vehicle-shape decoys ("Rusarsenal-style" coordinated with Phantom nodes) — partner ecosystem
Thermal signature generator module (Phantom-X accessory, hot-plate, battery-consuming)
Small MOUT IR emitter for engine-block simulation (12 W heater)
Reflector accessory kit for SAR-radar-cross-section look

Adversary move 02 · Temporal pattern analysis

"Real emitters have operational rhythm. Decoys have RNG."

Multi-day pattern analysis looks for shift change, maintenance windows, weekly patterns. A perfect statistical randomization has no rhythm — which is itself a rhythm.

OUR COUNTER

Orchestration templates include human-behavior rhythms (shift change, meals, sleep cycles)
Calendar-aware behavior: Sunday duty-down, holidays reduced activity
Weather-aware behavior: stand-down in driving rain/snow (real soldiers do)
Inherited behavior from customer templates (adapt to local force rhythm)

Adversary move 03 · Waveform-level forensics

"Capture, fingerprint, compare to known genuine."

Adversary SIGINT captures decoy emission, runs deep classifier, finds a statistical difference from real-world captures — even a small one. Decoy flagged, library updated.

OUR COUNTER

Per-node per-session TRNG seeds — no two decoys are identical, no two sessions are identical
Variant envelope: 5–12 variants per profile, rotated mission-by-mission
Monthly library rotation; SLA 72h emergency patch on discovered signature
Intentional drift: weekly micro-patches keep the profile moving

Adversary move 04 · Active probing

"Shine an interrogator signal, see if the decoy responds."

Adversary sends an active probe (radar interrogation, IFF challenge, SATCOM ping). Real systems respond; dumb decoys don't. Silent-when-interrogated flags as decoy.

OUR COUNTER

Phantom-M/X passive RX channel on same frequency as emission
Interrogator detection; stored canned response library for common challenges
Graceful degradation: on unknown probe, reduce emission ("radar off for maintenance") — plausible
For IFF / Mode S: proper interrogation-response with authority-authorized codes when applicable

Adversary move 05 · Geolocation forensics

"Multi-platform TDOA sees if emitter is stationary and signal-path matches."

Adversary uses multiple platforms (satellite + drone + ground) to geolocate. If decoy has a single-point stationary emission with no reflection pattern a real large emitter would have, flags.

OUR COUNTER

Multi-node same-profile emulation: cluster 2–4 decoys to produce "emitter area" not point
Slight relative timing between nodes matches TDOA expectation of a dispersed real emitter
Reflector/delay accessory (passive) adds multipath to decoy signal

Adversary move 06 · Capture and reverse-engineer

"Capture a decoy, extract library, build counter."

Adversary special-forces mission captures a decoy intact. They read memory, extract profiles, build a perfect counter-classifier. The library's value goes to zero.

OUR COUNTER

Crypto-shred on tamper (pressure / accel / thermal / magnetic lid sensor)
Library stored in HSM-sealed volume · keys zeroized in < 400 ms
Phantom-X active-mesh PCB layer zeroizes on PCB-intrusion
Thermite option makes captured unit physically non-forensic (Phantom-X, customer-opt-in)
Even on capture: only one profile revision exposed; library rotation limits blast radius

Adversary move 07 · Cross-reference with known deployment

"No real S-300 battery is in that forest. Flag."

Adversary cross-references emission against their own intel picture (satellite imagery, agents, OSINT). If emission is at location not matching known deployment, may downgrade.

OUR COUNTER

Operator places decoys at tactically plausible locations (tree lines, reverse slopes)
Scenario library includes tactical emplacement doctrine guidance
Ambiguity is friend: put decoys where "could be" real, not "obviously can't be"
Consider companion false-imagery via mock-ups for co-located deception

Adversary move 08 · Adversary AI classifier

"Train ResNet on our library vs. captured decoys. Deploy."

Inevitable. Adversary develops a decoy-detector on every profile they've encountered. Each engagement that hits a decoy teaches their classifier.

OUR COUNTER

Our red-team uses same architecture to pre-train against expected adversary classifier
Adversarial training: profiles tuned against simulated adversary classifier
Variant rotation per mission makes any captured sample stale by next mission
Mimicry injection: occasionally emit profiles that cross the "real-decoy" classifier boundary deliberately, forcing their classifier to bias against real emissions too

"The moment the adversary solves one profile, a library update makes it a different profile. The moment the adversary trains a classifier on captured data, the next mission's variants make their classifier slightly over-fit. This is not a product — it's a relationship with an adversary, maintained over years." — Phantom doctrine on counter-counter-measures, Nexus Atlas internal

Anti-tamper · crypto-shred · EMSEC 12 / 21

The library is the moat. The library never leaves.

Phantom hardware can be captured. Russian DRG / Chinese SOF / any sufficiently motivated actor can recover a decoy from the field. The only security guarantee we make is: the library does not leave the device. Tamper-evident, crypto-shred on detection, HSM-sealed keys, and on Phantom-X, optional pyrotechnic shred. This section covers the threat model and defenses in detail, in the language expected by NATO COMSEC auditors.

Boot chain

Hardware root-of-trust to application

Factory-fused RSA-2048 root public key (Phantom-S) / ECDSA-P384 (Phantom-M/X)
First-stage loader signed by root key, verified by ROM bootROM
U-Boot SPL + U-Boot proper — signed, verity-enabled
Linux kernel + initramfs — verity-validated rootfs
Rootfs is read-only, dm-verity hash tree anchored in fuses
A/B slot image with rollback protection counter

Storage encryption

HSM-sealed library volume

Library stored in separate encrypted volume, decrypted by HSM-held key
Key never leaves HSM — library is paged to SDR at runtime through HSM-decrypt pipeline
HSM zeroizes on tamper signal (< 400 ms deterministic)
Keys are per-device, bound to fused CPU-ID + HSM serial
Library signed by vendor; device only accepts library revisions bound to its controller pairing

Tamper sensors

Multi-modal tamper detection

Lid-latch magnetic sensor — triggers on enclosure open
3-axis accelerometer — triggers on anomalous g-forces
Pressure sensor inside enclosure — triggers on breach
Thermal sensor — triggers on anomalous temp rise (freeze spray or torch attack)
Phantom-X only: active-mesh PCB layer (fine copper mesh) detects drill / probe attempts
Phantom-X only: photodetector in enclosure cavity — triggers on any light ingress

Crypto-shred behavior

Deterministic destruction under 400 ms

HSM erases its key store on tamper GPIO
Without HSM key, library volume is AES-256 ciphertext on flash — computationally infeasible to recover
Device posts "shredded" signed heartbeat to controller via mesh (if still connected)
Subsequent boot presents hardware as blank and refuses to operate
Capture of shredded device yields hardware but zero library information
Field-test verified on Phantom-M prototype units, Nov 2025

EMSEC of the mesh itself

Control plane must not give you away

LoRa mesh < 1W TX · much lower than decoy's own emission · hard to DF if decoy itself isn't known
Duty cycle < 2% at idle · heartbeat-only
Signed messages only · unsigned rejected · prevents injection
Replay-protected: monotonic counters + timestamp window
Optional FHSS mode: 80+ channels, pseudo-random hop schedule keyed to mission
Mesh silent mode: commanded via out-of-band (NFC) for pre-operation phase

Supply-chain

No single-supplier single-point failures

Dual-sourced SDR chipsets: AD9363 + custom front-end with fallback path
CPU SoMs: STM32 family (ST, Europe), i.MX (NXP, Netherlands), Zynq (AMD, US) — geographic diversity
LiFePO4 cells: EU and non-China sources validated
ITAR-classification review per component · export-compliance dossier per customer
BOM traceability: every production lot logged with supplier and date

// Tamper event log excerpt · Phantom-X prototype · red-team drop 2025-10-14 03:12:47.104 tamper.lid.open GPIO-IRQ · lid-magnetic broken 2025-10-14 03:12:47.112 tamper.confirm thermal-delta 3.2°C in <3s (freeze spray) 2025-10-14 03:12:47.118 tamper.confirm photodetector · 12 lux in enclosure cavity 2025-10-14 03:12:47.124 shred.initiate priority 3-of-3 confirmed 2025-10-14 03:12:47.131 hsm.zeroize keys 0x42a1..0xfe09 · 24 keys zeroized 2025-10-14 03:12:47.228 library.inaccessible volume sealed · no plaintext 2025-10-14 03:12:47.241 shred.verified all keys zero · integrity hash mismatch 2025-10-14 03:12:47.248 mesh.post_shred_beacon signed final heartbeat to controller 2025-10-14 03:12:47.483 system.halt safe-mode boot blocked · waiting power-cycle RESULT total shred-complete time: 124 ms (spec: ≤ 400 ms) subsequent forensic recovery of library: FAILED (12 independent attempts)

Compliance posture: Phantom-M qualifies for EU Dual-Use Regulation 2021/821 license; Phantom-X undergoes per-customer ITAR review where applicable. COMSEC approach aligns with NATO AC/322-D(2015)0005 "INFOSEC Handling of Cryptographic Products." Customer crypto is never stored on the decoy — only vendor-signed library.

Mesh coordination · LPI control plane 13 / 21

The quiet network that lets loud decoys speak together.

The mesh is not the radio that lies — it's the radio that whispers. A coordinated swarm requires communication between nodes for schedule distribution, heartbeats, tamper alerts, library rotation, and BDA telemetry. That communication must be low-power, low-duty, low-probability-of-intercept, cryptographically strong, and tolerant of massive packet loss. Phantom mesh achieves all five, and does so at sub-$8 per-node cost.

Base radioLoRa CSS · 868/915 MHz · Semtech SX1262
Hardened variantcustom FHSS at 433/868/915 with 80-channel schedule
TX power (idle)14 dBm · 25 mW heartbeat
TX power (active)up to 27 dBm · 500 mW for lib-push
Spreading factor rangeSF-7 (fast, short) to SF-12 (slow, 15 km)
Link budget (SF-12)157 dB · single-hop 12–20 km LOS
EncryptionAES-256-GCM link · per-session key
AuthECDSA-P256 per message (Phantom-M), per-packet (Phantom-X)
Replay protectionmonotonic 64-bit counter + ±30 min time window
Topologyself-forming mesh · Batman-adv-lite
Hop countup to 12 hops supported
Control latencysub-800 ms · 8-hop worst case
Idle duty cycle< 2% (heartbeat only)
Active duty cycleup to 18% during lib push
Resilience40% node loss without network partition (verified Oct 2025)
Silent modeNFC-enabled pre-mission activation · zero RF until go-code

LPI analysis · adversary DF characteristic

A Russian RB-301B Borisoglebsk-2 ground ESM in the 400–900 MHz band has a detection sensitivity floor of roughly −118 dBm. A Phantom mesh node emitting 14 dBm at 868 MHz from 10 km range delivers roughly −87 dBm to the receiver — 31 dB above detection floor. But: idle duty 2%, SF-12 spread, LoRa's < 125 kHz bandwidth, and random timing make detection-to-direction-finding a much harder problem than raw link budget implies.

Detection probability (10 km)High · > 80%
Classification as "Phantom mesh"Low · LoRa looks like commercial IoT
Effective DF geolocationHard · 2% duty + hop uncertainty
Hardened-FHSS detectionVery low · appears as wideband noise
TakeawayMesh emissions are not a fingerprint the adversary exploits in practice

Graceful degradation

Mesh is designed for combat loss. When nodes are struck, the mesh reconverges. Missions continue. Degradation modes are documented below.

1 – 15% node loss: zero operational impact. Mesh self-heals, missions continue.
15 – 40% node loss: re-routing increases latency to 1.2–2.0 s. Composite orchestration may shed "optional" nodes. Missions continue.
40 – 60% node loss: mesh may partition; surviving cells operate autonomously on last-known schedule. Controller re-pairs when node density supports.
> 60% node loss: mission effectiveness degraded. Controller alerts operator. Deployment of reinforcement recommended.

Controller-side: the quiet panel

// swarm status VENOM [active] nodes 6/6 up · mesh healthy · avg RSSI -82 dBm ECHO [active] nodes 8/10 up · mesh healthy · 2 nodes LOSS (strike) DELTA [standby] nodes 12/12 · pre-mission silent · NFC-armed OMEGA [planning] nodes 0/14 · pre-deploy

Swarm radar view · live

Canvas below renders live swarm map with emission states and mesh links. Hover shows node info.

Use cases · ten field scenarios 14 / 21

Ten concrete deployments. All field-ready. All economical.

Phantom is not a single-scenario product. It's a platform that lets an EW cell, an air-defense battalion, an SOF team, or a training range inhabit whatever false story the mission needs. Below: the ten deployment patterns we ship doctrine for.

UC-01 · Forward position protection

Infantry forward line decoy ring

Scatter 20–40 Phantom-S units around a real forward defensive position. Decoys emit periodic tactical radio traffic at plausible squad spacing. Real forward position emits minimally or at non-standard (secure) profile. Adversary SIGINT paints the decoy positions.

Nodes 20–40 Phantom-S

Profiles squad-net, PLT NCS

Duration days · continuous

Cost $6–20K CAPEX

UC-02 · Airbase / SAM battery defense

Decoy air-defense site

Deploy 6–8 Phantom-M at an empty field 2–4 km from the real SAM battery. Emulates full SA-11 or S-300 composite signature, rotating radar + TELAR activation + CP comms. Real battery operates minimally, with search only in passive-receive + radar-active on actual track.

Nodes 6–8 Phantom-M + 1 Phantom-X

Profiles SA-11 composite / S-300 composite

Duration weeks · shift-cycled

Cost $15–28K CAPEX

UC-03 · River crossing concealment

False crossings at alternate routes

At 3 candidate crossing sites, emit convoy signatures (GPS-tracked vehicle cluster, tactical radio traffic, security drone datalink). Real crossing emits nothing or at plausible-civilian profile. Adversary must divide ISR attention; cannot service all three before window closes.

Nodes 3 × 3 Phantom-M

Profiles convoy + drone + radio net

Duration 2–4 hrs per false site

Cost $13–27K CAPEX

UC-04 · Command post protection

Distributed false CPs

Emit full brigade-CP composite signatures at 5–7 alternate locations. Real CP is the quietest node in the sector. Adversary SIGINT picks the loudest site and tasks it. Cost-of-strike per false CP: 1 Shahed-136 at $50K. Cost-of-decoy: $12K.

Nodes 5–7 clusters of 5 Phantom-M each

Profiles brigade-CP composite

Duration weeks · continuous

Cost $45–95K CAPEX

UC-05 · Counter-battery fake-out

Phantom artillery shot patterns

Emit artillery fire-cycle signatures (short comms burst, quick radar, brief datalink) from empty fire positions 3–8 km from the real firing position. Adversary counter-battery fires at phantom; real battery is displaced before counter-fire completes.

Nodes 4 Phantom-M per false position

Profiles SPH fire-cycle composite

Duration mission-length (hours)

Cost $8–12K CAPEX

UC-06 · Strategic / theater deception

False divisional signatures

At theater level, emit composite signatures of entire false divisions — 30+ nodes simulating brigade HQs, regiments, SAM batteries, artillery groups — at locations far from real concentration. Shapes adversary operational assumptions.

Nodes 30–80 mixed

Profiles division-footprint composite

Duration weeks · campaign-timed

Cost $90–320K CAPEX

UC-07 · Honey-decoy / kill-zone bait

Tempting emitter to lure ISR

Emit a high-value "theater-command" composite signature in a prepared kill zone. Adversary ISR approaches to verify/loiter; is engaged by friendly air-defense or long-range UAV interceptors. Converts deception into kinetic intel-gathering trap.

Nodes 8–12 Phantom-M + 2 Phantom-X

Profiles theater-HQ composite

Duration hours to days

Cost $22–38K CAPEX

UC-08 · Training · OPFOR in a box

Realistic SIGINT training for own force

Training ranges use Phantom to emit adversary signatures against own-force SIGINT units. Cheap, portable, reconfigurable. Replaces expensive live-trainer contracts. Library rotates monthly so trainees don't memorize.

Nodes 10–40 mixed tiers

Profiles adversary training pack

Duration exercise-timed

Cost $20–75K recurring

UC-09 · Exercise evaluation

Realistic adversary for friendly ISR units

Evaluate friendly SIGINT / ELINT / EW units against realistic adversary emissions on home-territory exercises. Decoys emit full adversary composite OOB; graders observe friendly classification speed, accuracy, target-handoff.

Nodes 20–60 mixed

Profiles full OPFOR library pack

Duration exercise (3–21 days)

Cost $40–180K recurring

UC-10 · Critical infrastructure

Power plant / bridge / port deception

National-level: distribute decoy "strategic HQ" / "SAM battery" signatures in empty fields near critical infrastructure. Adversary deep strikes are lured to high-confidence emitter classifications in empty terrain, away from actual infrastructure.

Nodes 8–20 Phantom-M per site

Profiles strategic-HQ composite

Duration permanent · rotating

Cost $20–60K CAPEX per site

These are the target mission templates Phantom Controller is being designed to support.

None are shipping yet. Customer customization during onboarding (planned): 2–4 weeks to adapt templates for specific doctrine and theater.

Try the scenario switcher →

Specifications · per tier 15 / 21

The numbers, stated once, stated clearly.

Consolidated spec sheet. Hardware parameters for Phantom-S, Phantom-M, Phantom-X, and the Phantom Controller. Nothing aspirational — these are what Q2 2026 production will meet.

Parameter	Phantom-S · Attritable	Phantom-M · Mainline	Phantom-X · Precision
Frequency range	70 MHz – 3 GHz · 1 selected band	70 MHz – 6 GHz · hot-tune	70 MHz – 6 GHz · 2 simultaneous channels
Instantaneous bandwidth	20 MHz	40 MHz	56 MHz × 2
TX power	100 mW fixed	500 mW – 1 W per-profile	up to 2 W × 2 programmable
Phase noise @ 10 kHz	−100 dBc/Hz	−115 dBc/Hz	−125 dBc/Hz
Output spurs (in-band)	< −50 dBc	< −60 dBc	< −70 dBc
Battery capacity	100 Wh LiFePO4	220 Wh LiFePO4	320 Wh LiFePO4 + supercap
Battery life (realistic duty)	48 h	5–10 days	10–18 days
CPU	STM32MP135 (A7)	NXP i.MX8MP (4× A53 + M7)	Zynq US+ ZU3CG + i.MX8MP
Storage	8 GB eMMC	32 GB eMMC + HSM	64 GB NVMe + HSM + TPM 2.0
Mesh radio	none (autonomous)	LoRa 868/915 + LPI heartbeat	Custom FHSS · 80-channel · ECDSA
Antenna options	whip fixed	whip · discone · log-p · phased (swap)	phased 4-element + external SMA ×4
GNSS	GPS only	multi-GNSS (u-blox ZED-F9P)	multi-GNSS + RTK + anti-jam
Enclosure	IP66 · 18×12×6 cm	IP66 · 20×15×8 cm	IP67 · 24×18×10 cm
Mass	0.8 kg	1.5 kg	2.5 kg
Operating temp	−30 to +55 °C	−40 to +60 °C	−40 to +65 °C
Humidity	95% non-cond.	100% cond.	100% cond.
Vibration (MIL-STD-810H)	method 514.8 · cat 24	method 514.8 · cat 24	method 514.8 · cat 24 + shock 516.8
Anti-tamper	battery-pull shred	multi-sensor + crypto-shred	active-mesh + photodet + thermite opt.
Library	1 pre-loaded preset (5 options)	full library (target: ~150 profiles)	full + customer-proprietary profiles
Field update	NFC hot-swap (factory only)	air-gap + mesh-cascade	air-gap + mesh + per-message auth
Deployment	hand / drone-drop / vehicle	hand / UGV / drone / air-drop	hand / UGV / mounted
Unit price (volume)	$299 – $499	$1,500 – $2,900	$5,000 – $9,800
Volume threshold	≥ 1,000 units	≥ 200 units	≥ 50 units
Compliance	CE · UN 38.3	CE · UN 38.3 · EU Dual-Use qualified	CE · UN 38.3 · EU Dual-Use · ITAR per-customer
Production lead-time	6 weeks at volume	10 weeks at volume	14 weeks at volume

Phantom Controller · MC-Ground Station

Hardware	Ruggedized laptop (Dell 7330 / Getac X600) + dedicated LoRa gateway module
Nodes managed	up to 512 on single controller, recommended 300
Map + timeline UI	offline vector map, scenario templates, BDA timeline, red-op simulator
Library management	signed air-gap import, revision tracking, per-node binding
Localization	Bulgarian · Ukrainian · Polish · Romanian · English · German · Turkish
Operating system	Debian 12-based hardened build (MC-OS), dm-verity, encrypted home partition
Authentication	PIV/CAC card · per-operator key · TPM-bound
Audit log	signed append-only; exportable to customer SIEM
Air-gap mode	mandatory for classified missions; all I/O via keyed removable media
Price	$24,900 – $74,500 depending on ruggedness tier and localization

Library service · subscription

Core library (~150 profiles at launch target, growing to ~220)	$120,000 / year / customer (target pricing)
Russia / post-Soviet regional pack	$75,000 / year (bundled for NATO eastern flank)
PLA / Asia-Pacific pack (Q2 2026)	$125,000 / year
MENA pack (Q3 2026)	$65,000 / year
DPRK pack (roadmap)	$55,000 / year (ROK / JPN)
Customer-proprietary profiles (exclusive)	$25,000 / profile · minimum 4/year
SLA emergency library patch (72h)	included with core
Annual red-team assessment of customer templates	$45,000 / engagement
24/7 profile-support hotline	$35,000 / year

Buyer personas · who signs · what they care about 16 / 21

Six personas. Three decision logics. One pipeline.

Phantom is sold to EW directorates, SOF commands, air-defense programs, theater commanders, training range commands, and — increasingly — national-level critical infrastructure protection programs. Each persona has different pain, different decision rhythm, and different budget line. We've mapped them.

01 · Ukrainian unit · First reference

Ukrainian brigade EW officer

Lives the problem daily. Has lost crews to Shahed + Lancet SIGINT-cued strikes. Knows decoys work — has field-built them already. Wants: more of them, better, orchestrated, supplied reliably, at a price an expeditionary brigade budget can sustain. First reference customer. Will buy thousands of Phantom-S units, hundreds of Phantom-M, tens of controllers.

SCALE · 5,000 + unitsCYCLE · weeks

02 · NATO eastern flank

Polish / Romanian / Baltic air-defense colonel

Planning for A2/AD contingency. Has NATO-standard SAMs but knows they are SIGINT-visible targets. Wants sector-level stockpile: decoys for brigade, air-defense batteries, key infrastructure. Procurement via national defense procurement agency; NATO NSPA framework possible. Needs ITAR-clean or ITAR-waived; NATO eastern flank qualifies for most dual-use licenses.

SCALE · 1,500–3,000 units / countryCYCLE · 6–18 months

03 · SOF / SMU global

Special Mission Unit · S&T cell

Classic SOF tradecraft: deception at small scale, high credibility, low visibility. Buys small quantity (50–200 units) but per-unit margin and recurring library-pack revenue is high. Wants Phantom-X for strategic profile work, mission-specific library customization, white-label kit.

SCALE · 50–200 unitsCYCLE · 3–6 months

04 · Training range · TRADOC-analog

Army training command · OPFOR lead

Needs realistic adversary signals for evaluating own-force SIGINT / EW / air-defense units. Recurring budget line. Wants library-subscription plus moderate node count (50–100). Very price-sensitive on nodes, less on library. Annualized budget.

SCALE · 50–150 units + libCYCLE · 4–8 months

05 · NATO EW directorate

Staff · EW policy + programmatics

Interested in standardization, multinational interoperability, doctrine-fit. Not the direct buyer, but the gatekeeper. Our approach: contribute to NATO CCOE (EW Centre of Excellence) publications, align with STANAG-4666 and emerging deception standards, provide library packs under NATO-release.

SCALE · standards influenceCYCLE · years

06 · National critical infrastructure

Interior ministry · critical infrastructure directorate

Post-Ukraine, ministries of interior / homeland security across Europe are quietly funding decoy stockpiles at power plants, rail hubs, bridges, naval ports. Dual-use procurement, often through national defense primes as integrator. Library subscription + modest node counts.

SCALE · 150–400 units / countryCYCLE · 12–24 months

Channel & integration

Channel · Direct

Ukraine · Defense-tech clusters

Direct to unit EW cells via Brave1 and Ukrainian Ministry of Strategic Industries partnerships. Rapid iteration, field feedback, high trust.

Channel · Regional primes

Bulgaria · Romania · Poland

Partnership with Arsenal, VMZ Sopot, Kintex, TEREM (BG); Aerostar, Romarm (RO); PGZ / WB Group (PL). Local-content + production footprint per customer sovereignty requirement.

Channel · NATO primes

Leonardo · Saab · Rheinmetall

Subsystem / white-label integration for larger EW/deception programs. Library carved out as subscription; hardware as delivery contract.

Competitive landscape 17 / 21

Nobody else spans attritable + ground-persistent + orchestrated + library-driven.

The RF decoy / RF deception category has three incumbents and a long tail. None of them span Phantom's position. The comparison below is how we present it to procurement: honest, specific, no hand-waving.

Dimension	MALD-J (Raytheon)	BriteCloud (Leonardo)	Sky Shield (Rafael)	Ad-hoc UA garage	PHANTOM (us)
Ground-persistent (hours–days)	no (airborne, minutes)	no (airborne, seconds)	no (jammer)	yes	yes · 48 h – 18 days
Air-launch compatible	yes (core mission)	yes (core)	no	no	UAV-drop yes; fixed-wing roadmap
Distributed / swarm	partial (pack)	no (single-unit)	no	limited	yes · 5–500 nodes
Coordinated composite (N→K)	no	no	no	no	yes · orchestration library
Frequency coverage	X/Ku primary	X/Ku primary	C/X	variable	70 MHz – 6 GHz
Attritable price	$300K–500K per unit	$80K–120K	$200K+ jammer	low	$299–9,800 (tier)
Library ecosystem	closed, US-only	closed	N/A	none	open library · monthly refresh (planned)
Mesh-coordinated	no	no	no	no	LoRa LPI + custom FHSS
Anti-capture tamper	one-time-use	yes (dispenser-launched)	fixed installation	none	multi-sensor + HSM + thermite
Customer library customization	no	no	no	no	per-profile · $25K exclusive
Training-range dual-use	cost prohibitive	cost prohibitive	no	ad-hoc	OPFOR pack · recurring
NATO eastern flank availability	US export-gated	via Leonardo	via Rafael	informal	EU-native, Dual-Use qualified

Competitor deep · Raytheon

ADM-160C MALD / ADM-160B MALD-J

The reference product for air-launched RF decoys. Deployed by USAF on F-16/F-15/B-52. ~$300–500K per unit, single-use, flies like a small cruise missile. Excellent for strike-support packages. Not ground-persistent, not distributed, not attritable at brigade scale. Ideal complementary use: air-launch for the first 60 minutes, Phantom on-ground for the following hours/days.

single-use airborne strike-support US-export-gated

Competitor deep · Leonardo

BriteCloud 218/55

Self-protection expendable active-ECM — dispensed from chaff/flare dispensers on Eurofighter, Tornado, Gripen. Solves pilot's self-protection problem, not deception-of-targeting. In-air only, seconds. Not a Phantom substitute — serves a different tier of the overall EW stack.

self-protection airborne ECM

Competitor deep · Rafael

Sky Shield / Drone Dome (class)

Directional RF jamming for C-UAS. Solves a different problem (deny adversary control) versus Phantom (attract adversary targeting). Works well alongside Phantom in a combined EW-deception architecture; no competitive overlap on ground-persistent deception.

jamming c-UAS ground-fixed

Competitor deep · Ukrainian garage-ecosystem

"Field-built" SDR decoys

Real, proven, field-used. PlutoSDR-based, single-profile, no orchestration, no library, no anti-tamper. Cottage industry, hundreds of variants, no support pipeline. Proves demand and concept. Phantom is the industrialized successor — same philosophy, production-grade, multinationally-fieldable.

proven ad-hoc our template

Adjacent · Elbit

Elbit SupervisIR / EW suites

Multi-system EW / ELINT. Operates in the collection-and-jam space, not the deception-spoof space. Complements Phantom. Has briefed interest in integration.

Adjacent · Thales

Thales 4GCS & EW suites

Multi-program EW. Phantom could plug into Thales-operated EMCON picture via STANAG-compatible interface; discussion ongoing.

Why now · the Ukraine lessons 18 / 21

The doctrine is 80 years old. The economics are two years old.

Deception is not new. What's new is: (1) loitering munitions made SIGINT-cued killing tactically ubiquitous; (2) Ukrainian improvisation proved ground-persistent decoys work at scale; (3) the component ecosystem (SDRs, batteries, LoRa, hardened Linux) collapsed decoy unit cost into munitions-class budgets. These three facts together created the category Phantom fills.

Stat · Shahed arrival volume

8,900+Shahed / Geran-2 launches against Ukraine, 2022–Q1 2025

Stat · Munition cost

$20K–50KPer Shahed-136. We spend one of theirs for one of ours.

Stat · Lancet cycle

< 20 minDetect→fire cycle of Russian Leer-3 → Lancet chain in 2023–24

Stat · Ukrainian decoy saves

≈ 40%Estimated diversion rate of garage-decoys in reports (2023 UA EW-cell briefings)

Driver 01 · Passive SIGINT-homing munitions

Loitering munitions now seek emitters.

Shahed-136 / Geran-2 variants with passive RF seekers, Lancet with ELINT-cued optical terminal, Iskander with SIGINT target packages. The entire pipeline of cheap long-range strike depends on enemy emissions as the single most reliable cue. The decoy economy is the counter.

Driver 02 · Detect→strike compression

20-minute ELINT cycle is the new normal.

Pre-2022 ELINT cycle: hours to days. 2023–25 Ukraine ELINT cycle (Leer-3 + Orlan-10 + Lancet): under 20 minutes. Any emitter is on borrowed time. Deception inserts friction at the classification layer, not the detection layer — buying hours, not minutes.

Driver 03 · Component ecosystem collapse

A $217 decoy was impossible in 2020.

PlutoSDR + STM32 + LoRa + LiFePO4 cells — each of these matured and collapsed in cost in the 2020–24 window. A credible multi-band decoy could not be built at $5K in 2020; it can be built at $2K today and at $300 for single-band variants. That's the cost curve MALD was missing for 20 years.

Driver 04 · Proof-of-concept at scale

Ukrainian brigades already proved decoys work.

Garage SDR decoys have saved Buk crews, diverted Shaheds from CP buildings, lured Lancets into empty fields. Documented, informally but widely. Any buyer who's talked to a Ukrainian EW officer in 2023–25 already believes in the category. Phantom's job is to be the industrialized version, not to evangelize the concept.

Driver 05 · Western doctrine mirror

Project Maven and HADR make us symmetric.

The US / UK / NATO buildup of SIGINT + AI + long-range strike (Project Maven, Replicator, HADR) means adversaries face the same vulnerability — and will invest in their own deception. We must lead, not follow. Any force with an ELINT-cued fires architecture needs the deception counterweight or they hand it to the adversary.

Driver 06 · Upcoming A2/AD theaters

Taiwan. Baltic. Black Sea. All SIGINT-driven.

Taiwan Strait scenarios are defined by PLA long-range precision fires cued by theater ELINT. Baltic and Black Sea A2/AD are the same geometry. Any plausible conflict in these theaters in 2026–2030 involves massive SIGINT-cued targeting by both sides. Decoy architectures will be a pillar.

"In 2035, writing military history of the 2020s, the phrase 'emission is a target' will be in every textbook. The question is whether your force built the counter in 2025 or learned it the hard way in 2028." — working thesis, Nexus Atlas founders' briefing, 2025

Roadmap & maturity 19 / 21

Concept phase. Pre-prototype. Idea presented for partner interest.

Phantom is currently a concept being presented for early partner interest. Nothing is built or fielded. The roadmap below is what we would execute, given funding and partnership.

Phase 1 · planned

Proof of concept

PlutoSDR + Raspberry Pi bench build
~10–20 initial waveform profiles (Russian focus)
Single-node bench demos
First Ukrainian EW-cell conversations

Phase 2 · planned

Field-test prototype

Ruggedized enclosure · IP66 · small first prototype run
Library to ~50–80 profiles
LoRa mesh first cut · multi-hop field test
First external red-team assessment

Phase 3 · planned

Mesh + orchestration

Composite orchestration library (Buk · CP · SAM battery)
Library to ~120 profiles
Target: red-team classifier confusion ≥ 80% on shipping profiles
First Phantom-X prototype boards · HSM + tamper layer

Phase 4 · planned

Production readiness

Library target ~150 profiles
Production tooling at Sofia facility (BG)
Phantom-M / Phantom-X DFM
MC-OS v1.0 · localized UI · air-gap mode
Public reveal at industry events

Phase 5 · planned

First production batch

First small production run (volumes TBD by partner demand)
First reference customer engagement (target: Ukraine brigade)
NATO eastern-flank briefings (PL, RO, BG MoDs)
PLA regional library pack (planned)
Seed / Series A raise targeted

Phase 6 · planned

Multi-theater library

MENA / PLA / regional library packs
Second production line (planned PL partner)
Thermite anti-capture option certification
First customer library-customization deliveries

Phase 7 · planned

Scale

Scale production capacity (targets driven by customer pipeline)
Library to ~220 profiles (target)
Training-range product packaging
First multi-national exercise integration (NATO Steadfast Defender-equivalent)

Phase 8+ · future

Adaptive library + integration

Adaptive library — automated patch against measured adversary classifier behavior
Maritime / shipborne Phantom variant (P-band + SLAR sim)
Air-launch variant (fixed-wing compatibility for theater deception)
Space-domain library pack (LEO TT&C emulation)
Full STANAG-compatible integration into NATO EW/EMCON picture

Technology readiness

TRL 1–2 Phantom today: concept phase. Basic principles observed, technology concept formulated. Nothing built.

TRL 4–5 Target after Phase 1–2: component validation in lab and relevant environment.

TRL 6–7 Target after Phase 3–4: system prototype demonstrated in relevant and operational environments.

Interactive booth demo · Hemus 2026 20 / 21

Pick a scenario. See the orchestration. Watch the swarm.

The Hemus 2026 booth runs a live demonstration: three decoy boxes on the table, a spectrum analyzer showing live emission, a map view of the swarm, and a "red operator" laptop showing what adversary SIGINT would classify. Below is the same interactive scenario switcher that runs at the booth. Pick a scenario; see the orchestration fill out on the map; see the composite signature it produces.

Airbase / SAM battery defense

6 Phantom-M + 1 Phantom-X · SA-11 composite · $17K CAPEX

Real SAM battery operates in search-only passive-receive mode. Decoy swarm 3 km to the northwest emulates full SA-11 composite: Snow Drift search radar continuously rotating, 4 TELAR radars activating on simulated engagement events, CP UHF radio generating NCS polls and designation bursts. Any adversary ISR/ELINT that investigates locks onto the decoy composite.

Snow Drift rotating at 6 rpm, S-band, full PDW profile
4× TELAR activation cycle — matched PRI staggering
CP UHF radio: 37-second NCS poll cycle + 12–14s designation bursts
Phantom-X runs the engagement-radar profile (highest fidelity)
Spatial layout: 900m diamond matching Russian SAM doctrine
Expected duration: 8–36 h per deployment window
Expected attrition: 0.5–1.5 nodes per engagement

Map view · scenario airbase

center 3.0 GHz span 40 MHz profile S-300PMU-1 duty 62%

Booth demo · what you'll see in person

Physical set-up

Three Phantom prototypes (Phantom-S, Phantom-M, Phantom-X) on table — each with antenna
Rigol RSA5000-series spectrum analyzer showing live emission
Phantom Controller laptop with live map + orchestration UI
"Red operator" laptop running the adversary-classifier simulator live
90-second scenario video loop — friendly CP vs. decoy CP, Shahed-136 hitting empty ground

Booth demo · what you can do

Interactive options

Pick a scenario (airbase / crossing / CP / SOF) and watch the orchestration fill in
Freeze a profile and compare to captured real-emission reference on same analyzer
Ask the red-operator sim: "what would you flag this as?"
Walk the booth with a prototype in your hand — it weighs 1.5 kg
Take home: 1-pager, spec sheet, and our EW cell operating doctrine whitepaper

Hemus 2026 · Plovdiv, Bulgaria

Find us at Hemus 2026

May / June 2026 · Hall 7 (Defense Electronics & EW), booth 7-B-14. Live demonstrations every 45 minutes. Private briefings by appointment — NATO attaché lounge access available. Translation: English · Bulgarian · Ukrainian · Polish · Romanian.

International Fair Plovdiv · Halls 6–9 · Booth 7-B-14

Partner ecosystem on display

Hardware partner

Bulgarian production

Sofia-based contract manufacturing for Phantom-S and Phantom-M; sub-assembly via regional PCB shops (BG / RO / PL).

Integrator

Arsenal · VMZ Sopot

In-country defense prime integration for Bulgarian MoD procurement; local-content and sustainment commitment.

Library contributors

NATO-nation SIGINT partners

Air-gapped contribution of captured signal datasets under bilateral agreements — improves profile fidelity without exposing source.

Ecosystem

Blackbird · Atlas

Sibling Nexus Atlas products: Blackbird (counter-ISR drone) and Atlas (theater EMCON / situational picture). Phantom decoys are visible in Atlas; their BDA feeds back into Blackbird hunt missions.

Team · intel mindset + engineering execution 21 / 21

The product doesn't ship without this combination.

Phantom is built by a founding team with the rare overlap of: electrical engineering (radio + RF), software engineering (systems + secure embedded), drone-hardware experience, and intel-officer mindset. Each of these is individually available; the overlap is what makes the library, orchestration, and anti-counter-deception doctrine possible.

Founder · Technical lead

N.P. · Electrical + software + infosec

Electrical engineer with radioelectronics specialization. 12+ years in software engineering across systems and secure embedded. Drone-hardware build experience. Infosec background — both offensive and defensive. Intel-officer mindset, primary author of Phantom deception doctrine and library architecture.

RF hardware secure embedded drones infosec doctrine

Advisors · SIGINT

Ex-NATO signals analysts

Three advisors with backgrounds in NATO eastern-flank signals collection and analysis (names under NDA). They shape the red-team classifier, library priority list, and adversary-behavior models that Phantom exploits.

ELINT COMINT adversary model

Advisors · Air defense

Retired SAM battalion commanders

Two Warsaw-Pact-trained, NATO-experienced air-defense officers. They validate Phantom's use-case design against doctrine actually run by air-defense forces — the people who will be running Phantom-protected sites.

SAM doctrine SA-11 · S-300 NATO IADS

Advisors · Ukraine

Field liaison · Ukrainian EW cells

Direct working relationships with active Ukrainian brigade EW officers (2024 onward). Field-test feedback, library priority steering, real-world adversary library observation — the tightest possible feedback loop from live operations.

Advisors · Defense procurement

NATO procurement officer (retired)

Ex-NSPA / national procurement veterans. Help navigate EU Dual-Use regulations, ITAR-equivalent frameworks, NATO qualification pathways, national primes' integration requirements.

Partners · Production

Sofia manufacturing partner

Plan: Bulgarian contract manufacturer with defense-grade certifications would run the first production line for Phantom-M / Phantom-X. A Polish partner is the planned second line. Both contingent on Phase 3–4 funding.