Project Blackbird · Counter-ISR Autonomous Platform · Concept · TRL 1–2

They're already looking. We find them first.

Hunt the hunters. Left of boom. Silent counter-ISR.

Every kill chain starts with seeing you. Blackbird is an autonomous, low-signature, fully passive fixed-wing drone that flies out ahead of your force and catalogs the adversary reconnaissance assets that are mapping you — before the Shahed, the Lancet, the FPV, or the artillery are called in. No transmitter. No camera signature. A 2.5 kg counter-ISR weapon dressed as a drone, carrying a multi-channel software-defined radio, a direction-finding antenna array, and an on-board classifier trained against a planned library of adversary emitter signatures.

70MHz–6GHz
RF Coverage
Continuous wideband passive sensing with X-band add-on.
1–3°
Bearing Accuracy
Amplitude-monopulse DF at SNR ≥ 10 dB, swarm-augmented.
≤20m
Geolocation CEP
3-Blackbird TDOA at 5 km standoff. ≤100 m single-platform.
Target
Emitter Classes
Concept library; signature acquisition strategy mapped.
120min
Mission Endurance
1.5–2 h passive, electric. Hybrid option to 3 h.
<50ms
Inference Latency
Edge classifier on Jetson Orin Nano per RF event.
Autonomous
No Pilot Required
Planned mission, GPS-denied capable, 1 operator for 2–4 birds.
Passive
Zero RF Emissions
Silent ingress. LPI/LPD exfil only on discretionary burst.
Attritable
Price Point
$50K–$200K per airframe. Platform losses tolerable.
Swarm-Ready
Multi-Bird TDOA
3–8 Blackbird sparse array for sub-20 m localization.
Crypto-Shred
Anti-Capture
Cold-boot wipe, secure element, encrypted payload.
NATO-Ready
Interop Compliant
STANAG 4586 / 4609 / 4671 feed. Link-16 gateway option.
Explore the System →
SCROLL · DATASHEET
§01 · The Kill Chain Problem PROBLEM / 2026

Every kill chain starts with somebody seeing you.

Modern indirect-fires engagement follows four steps. Today's counter-UAS market addresses step four. Blackbird addresses step one — where it's cheapest, quietest, and has not yet cost anyone their life.

Blackbird Breaks Here
Step 01
SEE
Adversary ISR asset (Orlan-10, ZALA, Supercam, Autel EVO, commercial DJI variant) loiters over the target area. Burst of control/telemetry RF is emitted every 0.8–4 s. This is where Blackbird lives.
Step 02
IDENTIFY
Human operator or on-board classifier confirms the target — vehicle, position, SAM, radar, HQ tent, HIMARS, ammo truck. Correlated with prior ISR cycles (8–30 min typical).
Step 03
TARGET
Grid passed to fires: tube artillery, MLRS, loitering munition, one-way drone, cruise missile, SPG. Lead time 4–12 min depending on shooter readiness.
Step 04
STRIKE
Round/drone impacts. In Ukraine, ~80% of casualties on contested frontages now trace to indirect fires enabled by drone spotting. Current C-UAS kinetic layer intercepts here — often too late.
80%
Ukraine frontline casualties traced to drone-enabled indirect fires
10–30min
Adversary ISR-to-strike cycle on active frontages (2024–2025 data)
4:1
Ratio of ISR-enabling drones to strike drones in observed Russian/NK kill chains
$2.5K
Cost of an adversary ISR drone that directs a $1M+ target's destruction
Left of Boom

Why kinetic C-UAS is not enough

Dedrone, DroneShield, and peers intercept drones crossing your perimeter. By then the emitter has already uplinked a target pic, a grid fix, or a bomb release. You defeated the platform; the kill chain completed anyway. Blackbird denies step one.

Persistence, not Presence

Transit vs. loiter behavior

A drone flying straight through your sector is a waypoint. A drone hovering 2.4 km to your east at 380 m AGL for 47 minutes is a forward observer. Blackbird distinguishes by pattern-of-life — the difference between nuisance and fires-in-the-next-twelve-minutes.

The Asymmetry

Defense is structurally disadvantaged

A $2,500 Autel with a commodity RF uplink directs a $120,000 Krasnopol onto a $4M SAM. You cannot afford to defeat every drone. You can afford to blind the adversary's reconnaissance cycle — and let him shoot fog for an hour.

Doctrinal shift
Counter-ISR is the new counter-battery. In 20th-century fires, you beat the gun. In 21st-century fires, you beat the eye. The adversary gun is cheap, stockpiled, and dispersed. The adversary eye is few, expensive, and — critically — emits. Hunt the eye.
§02 · The Platform WHAT · SYSTEM

A 2.5 kg counter-ISR weapon dressed as a drone.

Blackbird is not another camera drone. The airframe is a delivery vehicle for a sensor payload the size of a lunch box. The sensor is a wideband passive RF receiver, a direction-finding antenna array, and an edge classifier. The airframe is a low-signature glider that lets that sensor look into contested airspace for two hours without ever being heard.

wingspan · 1.12 m DF Array A outer element DF Array B inner element DF Array C interferometric baseline DF Array D outer element EO/IR pod 640×512 thermal · 20 MP RF payload bay SDR · Jetson · storage · SE Propulsion brushless electric · optional hybrid PAYLOAD BAY 500–700 g
2.5kg
MTOW including 700 g payload
1.12m
Wingspan (sub-small UAS)
<48dBA
Acoustic at 30 m cruise
<0.02m²
Estimated RCS at X-band

Design principles

  • PRINCIPLE 01 · PASSIVITY
    Blackbird never transmits on ingress or during survey. All RF work is receive-only. The bird is a hole in the spectrum, not a beacon.
  • PRINCIPLE 02 · SMALLNESS
    Sub-1.2 m span and electric propulsion put Blackbird below most short-range air-defense radar classification thresholds and below the acoustic noise floor of a combat zone.
  • PRINCIPLE 03 · ATTRITABILITY
    Bill of materials and manufacturability engineered for $50–200 K — low enough that losing one to EW, weather, or kinetic is a tactical problem, not a strategic one.
  • PRINCIPLE 04 · INTEL-MINDSET OUTPUT
    The bird does not output "drone detected." It outputs "probable forward observer, persistent, recommend priority fires." Operator reads a G2 paragraph, not a spectrum plot.
  • PRINCIPLE 05 · SURVIVABILITY OF DATA
    If Blackbird is captured or lost, the emitter catalog — the moat — must not fall into adversary hands. Crypto-shred, secure element, tamper-reactive payload.
  • PRINCIPLE 06 · SWARM-NATIVE
    Design for a flight of 3–8 Blackbirds functioning as a sparse, moving TDOA array — not a lone sensor. Geolocation accuracy improves 5× with swarm versus single-platform.
Airframe

Fixed-wing glider

Carbon/Kevlar shell, flying-wing with V-tail, electric brushless motor, 1.12 m span, 2.5 kg MTOW, hand-launch or pneumatic bungee. Optional hybrid generator for extended endurance.

Payload

Passive RF sensor

Multi-channel wideband SDR, 4-element DF array, NVIDIA Jetson Orin Nano, secure storage, optional EO/IR. Total payload weight 500–700 g, ≤ 25 W power draw at steady state.

Mission

Counter-ISR survey

Fly a programmed orbit or opportunistic pattern over area of interest. Catalog, classify, and geolocate emitters. Return with product; burst on discretion via LPI link; auto-wipe if lost.

Output

Intel product, not spectrum

Georeferenced emitter list with classification, confidence, pattern-of-life, historical cross-reference, and recommended action. Ready to feed fires cells, G2 shops, and COP displays.

§03 · Concept of Operations CONOPS · MISSION

One operator. Two hours. A cleaned spectrum.

A typical Blackbird mission runs 90–120 minutes from launch to recovery. The operator workload is front-loaded into planning; the bird flies silent and alone and returns with the product. Below is a step-by-step of the mission profile.

T-45 min · Phase 01
Mission planning
Operator defines area of interest (polygon or keep-out), bands of interest (presets for DJI/Autel, Orlan/Supercam, X-band radars, VHF comms), mission time budget, launch and recovery points. The ground station computes orbit geometry for best DF baseline motion and flags line-of-sight and terrain masking issues.
AoI polygon Band presets Orbit solver Risk overlay
T-5 min · Phase 02
Pre-launch check and key load
Classifier library version and delta signatures pushed from ground station. Crypto keys and kill-switches seeded. Airframe powers on, BIT (built-in test) runs: SDR calibration, DF phase check, inertial bias, magnetometer calibration, battery state, propulsion self-test. BIT passes or mission aborts — 42 second cycle.
Library v2026.04.14 SE keys Self-test
T-0 · Phase 03
Launch
Hand-launch from standing position (crew of one) or pneumatic bungee (crosswind up to 12 m/s). Climb to 250–400 m AGL. Motor cuts as soon as loiter altitude is reached — Blackbird glides and soars for the majority of the mission.
Hand launch Bungee rail 12 m/s x-wind
T+4 min · Phase 04
Silent ingress
No datalink emissions during ingress. Nav is GPS primary, VIO secondary, magnetic anomaly tertiary. Flight profile minimizes altitude on terrain masks, hugs valleys where feasible. Payload SDRs are hot and sweeping from the moment of launch.
Zero Tx Terrain mask GPS primary
T+20–90 min · Phase 05
Survey & localization
Bird flies a mission-planned figure-8 or racetrack over AoI plus opportunistic detours toward detected emitters. Moving-baseline interferometry and angle-of-arrival build bearing tracks; Doppler disambiguates transient vs. persistent sources; ML classifier labels every event in < 50 ms. A baseline of ambient RF environment is built before flagging anomalies.
AoA + Doppler On-board ML Racetrack 4–6 km
T+90 min · Phase 06
Pattern-of-life judgement
For each emitter: how long has it been up? How far has it moved? Is its PRI stable? Does it match a known forward-observer pattern? Does it match a prior detection from previous missions in this sector? Blackbird labels behaviors, not just signatures — "transit," "loiter," "persistent observer," "cooperative relay."
Behavior labels Cross-mission ref Priority scoring
T+95 min · Phase 07 (optional)
LPI/LPD burst exfil
Discretionary: if mission commander pre-authorized, or if the bird hits an abort trigger, a compressed summary of catalog is bursted over a low-probability-of-intercept link (FHSS on 433/868/915 MHz ISM or UHF tactical). Burst is < 1.5 s, < 400 bytes per emitter, with sub-Rayleigh fade margin. No continuous telemetry.
LPI/LPD FHSS <1.5 s burst
T+115 min · Phase 08
Recovery
Autonomous return to pre-programmed recovery point. Deep-stall landing or parachute recovery into a 20×20 m LZ. Payload is hot-swapped; airframe reset for next sortie. Catalog offload via physical tether to ground station (air-gapped from tactical net).
Deep stall Parachute opt Tethered offload
T+post · Phase 09
Post-mission fusion & library feedback
Catalog pushed into customer COP (STANAG 4609 video + KML + JSON). Unknown emitters tagged and queued for expert signature review and, if approved, added to the training set for the next library release. Every mission improves the library for every customer.
STANAG 4609 JSON / KML Library loop

Mission envelope

Typical sortie90–120 min
Area of interest10–80 km² per sortie
Standoff from operator1–25 km (LOS / hop-mesh)
Cruise altitude120–600 m AGL
Cruise speed18–32 m/s
Emitter catch rate24–180 / hour (AoI dependent)
False-positive rate< 3% post-fusion
Operator headcount1 / 2–4 birds

Abort triggers (on-board)

  • ■ Battery state < 18% with recovery range exceeded
  • ■ GPS spoof confidence > 0.7 and VIO disagree > 120 m
  • ■ Confirmed hostile illumination (X-band tracking radar lock)
  • ■ Mechanical self-test failure mid-flight
  • ■ Tamper switch or case breach
  • ■ Time-of-flight exceeds contract
Terminal behavior on abort
Depending on abort reason, Blackbird either (a) returns passive to a pre-programmed alternate recovery point, (b) bursts a summary and ditches in deep water / forest / industrial area, or (c) crypto-shreds the payload and performs terminal dive to minimize debris-field exploitation value. Adversary cannot reconstruct the classifier library from a recovered Blackbird.
§04 · System Architecture ARCH · SUBSYSTEMS

Seven subsystems. Each engineered for contested airspace.

Below is the hardware/software breakdown. Where we reference a chip or module, that is either the part we ship in the reference build or the closest COTS analog to it — details on production sourcing are in the technical data package.

Wideband SDR Receiver

01 · RF FRONT END

Multi-channel software-defined radio chain. Two simultaneous RF chains for interferometric DF plus a third wideband survey chain. Active cooling; low-phase-noise LO design critical for Doppler discrimination.

  • Core: Analog Devices AD9361 / AD9371-class transceiver (Pluto-derivative custom board)
  • Alternative builds: Ettus B205mini-i, Epiq Sidekiq Z2+
  • Tuning range: 70 MHz – 6 GHz, X-band add-on via down-converter daughter-card
  • Instantaneous BW: 40 MHz / channel, 3 chains aggregate
  • Dynamic range: ~ 78 dB in 40 MHz BW after calibration
  • Noise figure: 3.2–4.5 dB (band dependent)

Direction-Finding Array

02 · DF

Four-element wing-mounted array. Amplitude-monopulse used below 400 MHz where baseline is short relative to wavelength; interferometric phase-difference above. Optional Doppler AoA for long-pulse emitters. Calibration table built at BIT and refined mid-flight.

  • Elements: 4 × spiral + 2 × dipole on sparless wing carry-through
  • Baseline: 1.08 m tip-to-tip, ≈3.6 λ at 1 GHz
  • Accuracy target: 1° RMS at ≥ 10 dB SNR; 2–3° at 6 dB
  • Ambiguity resolution: dual baseline ratio method
  • Mid-flight cal: continuous against known self-emissions of platform (reference tones)

Edge Compute

03 · PROCESSING

All classification and geolocation happens on-board. Ground station never sees raw IQ — only deductions. Compute stack is split into hard-real-time DSP and soft-real-time ML.

  • SoC: NVIDIA Jetson Orin Nano 8 GB (40 TOPS INT8)
  • Co-processor: Xilinx Zynq UltraScale+ ZU3 for line-rate DSP
  • Distilled classifier: ~ 240 MB weights, INT8 quantized, 5,200 classes
  • Inference: < 50 ms / event, batched to 200 evt/s sustained
  • Language: Rust (core DSP) · Python+TensorRT (inference) · C for HAL

Navigation (GPS-Denied)

04 · NAV

Designed for jammed/spoofed airspace because that is the airspace. GPS when available, four independent fallback sources otherwise. Fusion is an error-state EKF with consistent-sensor-only update gating.

  • GNSS: multi-constellation (GPS L1/L5, Galileo E1/E5, GLONASS L1), CRPA antenna
  • IMU: tactical-grade MEMS (Honeywell-class, < 1°/h drift)
  • VIO: stereo monochrome 640×480 @ 60 Hz + depth from motion
  • Magnetic anomaly: 3-axis fluxgate matched against terrain magnetic map
  • Cooperative: inter-Blackbird UWB ranging (± 10 cm at 120 m)
  • Celestial (roadmap): star-tracker for long-dwell missions

Secure Storage & Secure Boot

05 · EMSEC / COMPSEC

Anti-capture posture treats the library and the classifier weights as the crown jewels. Secure element enforces key possession and stage-wise decryption; without the mission key, the payload is useless.

  • Secure element: Microchip ATECC608 / Infineon SLM-class
  • Payload encryption: AES-256-XTS per-block; keys never on NAND
  • Secure boot chain: ROM → FSBL → bitstream → kernel → rootfs, each signed (Ed25519)
  • Cold-boot protection: power-loss keys are RAM-only; 3-second decay
  • Tamper response: case open → 30 ms TRNG overwrite + SE fuse blow

Comms / Datalink

06 · LPI/LPD

Comms are the exception, not the rule. When Blackbird does transmit, it uses low probability of detection + low probability of intercept techniques: frequency hopping, minimum burst, processing gain via spreading.

  • Primary: FHSS 868/915 MHz ISM or tactical 350–520 MHz
  • Burst: <1.5 s compressed summary, never continuous telemetry
  • Processing gain: ≥ 24 dB via DSSS chip rate
  • Inter-Blackbird mesh: 2.4 GHz FHSS, only when in-swarm
  • Gateway: ground station bridges to Link-16 / STANAG 4586

Airframe & Propulsion

07 · PLATFORM

A quiet, low-signature glider. The goal is not speed or maneuverability — it is endurance, observability reduction, and a predictable platform for DF baseline geometry.

  • Structure: Carbon/Kevlar spar, EPP foam wing skin, 3D-printed ribs
  • Propulsion: brushless electric, ~ 180 W cruise; hybrid option adds 90 min
  • Prop: 9×4 folding (prop-in on glide), acoustic-optimized blade
  • Battery: 4S 8 Ah Li-ion, 44 Wh/kg usable, LFP cells for thermal margin
  • Launch: hand or pneumatic bungee rail (30 m/s exit velocity)
  • Recovery: deep-stall or parachute (21 m² canopy)

Optional EO/IR Confirmation

08 · SENSOR / OPT

Add-on nose ball for visual confirmation. Used sparingly — any gimbal adds weight, radar cross-section, and export-controlled components. Default Blackbird ships without it.

  • Thermal: 640 × 512 @ 25 µm LWIR, 17 mm lens, NETD < 40 mK
  • EO: 20 MP 1" CMOS, low-light mode, 35 mm equivalent
  • Stabilization: 2-axis gimbal, ≤ 25 µrad residual jitter
  • Use case: confirm emitter-associated platform type, silhouette check

Flight Software Stack

09 · AUTONOMY

Deterministic autonomy: no cloud dependency, no LLM in the control loop, no black-box planner. The mission planner produces a contract; the autopilot executes the contract; deviations are flagged and bounded.

  • RTOS: Zephyr + PREEMPT_RT partitioning, dual-core asymmetric
  • Autopilot: fork of PX4, hardened loop at 400 Hz; position hold in zero-GPS demonstrated
  • Mission DSL: declarative orbits + event triggers + abort contracts
  • Test: > 90% line coverage, property-based tests on DSP kernels, hardware-in-loop sim
  • Certification path: DO-178C DAL-C for flight-critical partitions (roadmap)
System Supply / Sovereignty Note
Blackbird is designed for supply-chain sovereignty at the squadron level. Every flight-critical chip has a second-source (typically European or allied-Asia) bill-of-materials mapped. Airframe manufacture is local-kitting-compatible: 80% of structural parts can be made in a 60 m² shop with a fiber laser, a press brake, and a mid-tier 3D printer. Strategic point: a customer does not want to be six months downrange from the OEM when a shipping route closes.
§05 · The Emitter Library — The Moat LIBRARY · ML

The signature library would be the real product. (Concept.)

The airframe is commoditizable. The classifier library is not. The plan is to accumulate it through synthetic generation, controlled-range captures, red-team field collects, and Ukrainian-partner mission data. Every mission would enlarge it. This is why a three-year incumbent would beat a six-month entrant — and why we present the concept now, before any of the library exists.

Library composition

CategoryClassesBandSource Mix
Commercial multirotor (DJI)4122.4 / 5.8 GHzSynthetic 60% · Range 30% · Field 10%
Commercial multirotor (Autel)1882.4 / 5.8 GHzSynthetic 55% · Range 35% · Field 10%
Commercial multirotor (Skydio / EVO clones)1442.4 / 5.8 GHzSynthetic 70% · Range 25% · Field 5%
Russian ISR UAVs (Orlan, Supercam, Granat, ZALA)372VHF / UHF / LField 55% · Captured 25% · Synthetic 20%
Iranian / NK ISR UAVs108VHF / UHF / LField 70% · Synthetic 30%
Loitering munitions (Shahed-136, Lancet, KUB)96L / telemetryField 65% · Captured 25% · Synthetic 10%
FPV control links820ISM / 900 / 1.2 / 2.4 / 5.8Synthetic 45% · Range 40% · Field 15%
Tactical comms (analog)260VHF / UHFField 80% · Synthetic 20%
Tactical comms (digital, DMR/TETRA-like)420VHF / UHF / SRange 60% · Synthetic 40%
SATCOM terminals (user)76L / S / KuSynthetic 80% · Field 20%
Short-range air-defense radars140S / C / XRange 50% · Captured 30% · Synthetic 20%
Counter-battery / artillery-locating radars64C / XRange 55% · Synthetic 45%
EW / jamming transmitters312VHF – SField 60% · Synthetic 40%
Datalinks (tactical, mesh, retrans)510UHF – SRange 55% · Synthetic 45%
Miscellaneous / uncategorized1278all bandsField 80% · Pending review

Concept totals: ~5,200 signatures targeted across 15 categories. The numbers above are illustrative class targets, not a current inventory.

How the library is grown

Source 01

Synthetic generation

Physics-based RF simulator produces IQ samples for a target waveform at every SNR from 3 to 30 dB, across fading channels (Rayleigh, Rician, Nakagami), Doppler offsets, sample-rate offsets, and multipath profiles. 40,000+ samples per class floor before a signature is a candidate.

Source 02

Controlled range captures

Anechoic chamber and open-range captures against the real radio — DJI Air 3, Autel EVO Max 4T, specific Orlan variants — collected at known range, known pose, for ground-truth DF and demod evaluation. Per-radio batch of 8–20 hours.

Source 03

Field collects & partner missions (planned)

The intended mother lode. Real captures from Blackbird (and partner birds) over active frontages, once partner deployment is established. This is what would make Blackbird's library distinguishable from an open-source signature set. Strict data-hygiene plan: only cataloged IQ, no mission context in any shipped library.

Source 04

Captured adversary equipment

Trophy hardware from Ukrainian partners goes through a reverse-engineering flow (waveform extraction, control-link handshake, PRI/hopping-sequence reconstruction). Feeds the highest-confidence signature set in the library.

Source 05

Customer-contributed (planned)

Future plan: every unclassified flagged event on a customer mission would optionally upload through a one-way sanitizer. Contributing customers would get priority on next-release signatures — a positive-sum incentive that would grow the moat for everyone.

Source 06

Adversary update tracking

Signature drift is continuous: adversary rolls firmware, changes hopping seed, or swaps modem. A change-detection pipeline flags waveform drift and kicks off a re-classification and, if needed, a library patch within hours.

The moat in one sentence
Anyone can buy an SDR; almost no one can build thousands of labeled signature classes from operational captures over an active war. That asset is what we propose to build.
§06 · Intel-Mindset Differentiator OUTPUT · G2

Pure RF sensors say "emitter detected."

Blackbird writes you a G2 paragraph.

A spectrum plot is a tool; a judgement is a weapon. Pattern-of-life analysis, order-of-battle cross-referencing, and recommended-action formulation are baked into the on-board output. This is the difference between equipment the operator has to interpret and equipment the commander can act on.

A typical C-UAS sensor output

ALERT. RF detection at 2.4385 GHz. Probable DJI OcuSync 3. Signal strength -62 dBm. Bearing 087° ± 15°. Duration 12 seconds. Signal lost.

That's a radio. You still have to figure out if it's a threat, whose it is, where it is, what it's doing, and whether to shoot at it.

A Blackbird output
# SECRET//NOFORN//REL UKR//FVEY # Mission: FRRT-2026-04-14-B · Bird: Blackbird-7 EMITTER E-0094 LAT/LON 49.4821 N / 37.1614 E (UA) MGRS 37U DR 82745 8176 CEP 18 m · 3-Blackbird TDOA · σ 6.1 m CLASS Autel EVO II Pro Enterprise C2 uplink · channel 3 (OcuSync-like) CONF 0.94 · top-3 all Autel family DWELL 47 min 12 s MOTION slow eastward drift · 14 m/min altitude 340–395 m AGL BEHAVIOR LOITER · PROBABLE FWD OBS not transit · not cooperative XREF 4 prior dets · sector K-17 11-day window · PERSISTENT TEAM ISR TEMPO 2.1 sorties/day · last 10 d typical dwell 38 ± 9 min LIKELY PAIR counter-battery radar (E-0088) + 2S7 battery (est. 3.1 km SSE) ASSESS HIGH · forward-observer team directing indirect fires onto BDE-fwd trains area REC PRIORITY FIRES · time-sensitive window 08–14 min before relief OWNER none · passive observation only ORIG Blackbird-7 (2026-04-14 09:14:22Z)
How Blackbird computes that
The output above is not prose generation. Every field is a deterministic function of (a) classified waveform features, (b) measured AoA and Doppler, (c) multi-Blackbird TDOA fusion, (d) cross-mission database lookup, and (e) a formal pattern-of-life rule set signed off by a former-intel-officer review board. No LLM in the kill-chain decision loop.
Judgement dimension 01

Dwell vs. transit

A path that crosses the AoI in < 3 min at cruise speed is transit. A pattern that orbits, figure-eights, or holds within a 2 km radius for > 5 min is loiter. Loiter over friendly HVTs is a fires-spotting candidate.

Judgement dimension 02

Altitude profile

Tactical ISR UAVs fly 200–500 m AGL to stay above small-arms and out of most short-range air-defense. A slow descent/climb pattern often precedes engagement (diving for a BDA pass or a terminal-dive munition).

Judgement dimension 03

Emitter pairing

A counter-battery radar + a loitering C2 link + an uplink to a fires net at the same grid, within a 90-second window, is a team signature. Blackbird labels them as one pattern, not three emitters.

Judgement dimension 04

Cross-mission persistence

If an emitter at grid X has been seen on 3 of the last 10 days at roughly the same time, it is a persistent observer. The judgement is a statistical function of prior detections, not a single-shot inference.

Judgement dimension 05

Own-force deconfliction

The library holds friendly force signatures (your Mavic, your TETRA, your radars). Own-force emitters are flagged but never promoted to fires-recommendation. Blue-fire is a non-event in a Blackbird catalog.

Judgement dimension 06

Doctrine-fit scoring

Each detection is scored against a catalog of known adversary doctrinal patterns — Orlan+Krasnopol pairing, Lancet forward-observer cycle, FPV drop pattern. Doctrine-fit is what turns noise into intent.

§07 · Passive RF Localization GEOLOC · DSP

Finding a radio without ever keying up.

Passive geolocation is the entire game. Blackbird uses four fused techniques — angle-of-arrival, time-difference-of-arrival, frequency-difference-of-arrival, and moving-baseline interferometry — combined with platform motion to localize emitters with a standoff measured in kilometers.

Live RF Spectrum · 2.4 GHz band

70 MHz – 6 GHz · Survey chain 1 · BW 40 MHz LIVE SWEEP
12 detections · 4 known · 1 anomaly Classifying…

Geolocation picture · 3-Blackbird TDOA

Blackbird bird
Adversary emitter (localized)
Emitter (bearing only)
AoI polygon
Friendly asset

Four techniques, one fused estimate

Technique 01 · AoA

Angle-of-Arrival

Phase-comparison across the 1.08 m four-element array produces bearing. Accuracy 1–3° at ≥ 10 dB SNR. A single bearing plus platform position is a ray — a line, not a point. Successive bearings from moving platform triangulate to a point.

θ = arcsin( λ · Δφ / (2π · d) ) · resolve ambiguity via multi-baseline ratio
Technique 02 · TDOA

Time-Difference-of-Arrival

Multiple Blackbirds time-stamp the same emitter pulse. Differences in arrival time (nanoseconds) translate to hyperbolic position surfaces. Intersection of three or more surfaces is a geolocation fix. Works on pulsed and burst emitters best.

Δt · c = |r₁ - r_e| - |r₂ - r_e| → hyperbola · 3 birds = 2 intersect = unique fix
Technique 03 · FDOA

Frequency-Difference-of-Arrival

Platform motion creates a Doppler shift. Two Blackbirds flying different vectors see the same emitter at two slightly different frequencies. The frequency difference disambiguates when TDOA alone is geometrically weak (co-linear baselines).

Δf = (v₁ - v₂) · r̂ · f_c / c · complements TDOA in poor-geometry cases
Technique 04 · Moving-baseline interferometry

Single-platform localization

A single Blackbird is, over time, a synthetic aperture. Successive bearings from positions along the flight path triangulate. Not as sharp as 3-bird TDOA but workable for < 100 m CEP on persistent emitters within 5 km standoff.

N sequential bearings from positions pᵢ · non-linear least-squares Levenberg-Marquardt
100m
CEP · 1 Blackbird · 5 km standoff · 120 s dwell
40m
CEP · 2 Blackbird pair
20m
CEP · 3 Blackbird TDOA
8m
CEP · 5+ Blackbird swarm (stretch)
Why passive matters, operationally
Active radar geolocation lights you up. Whoever you are trying to find, finds you first. Passive RF localization means the emitter cannot discriminate whether you are listening. You can sit on a hunch for 45 minutes without ever being seen — which is exactly how a forward observer works. Blackbird out-observes the observer.
§09 · Swarm Operations SWARM · TDOA

Three birds. One sparse radio telescope.

A single Blackbird is a useful sensor. A flight of three to eight Blackbirds is a sparse, moving, flying antenna array. The same physics that underpins radio astronomy gets you sub-20-m emitter geolocation at 5 km standoff, in a jammed environment, with nobody knowing you're there.

Swarm TDOA fix · hyperbolic intersect

Blackbird positions
TDOA hyperbolae
Emitter fix
Share UWB mesh

Swarm modes

Mode 01

Passive relay

Birds fly independent orbits, each contributing bearings. Centroid of orbiting Blackbirds localizes an emitter via time-resolved triangulation. No inter-bird datalink required beyond heartbeat.

Mode 02

Coherent TDOA

Birds share time-stamped pulse captures over a 2.4 GHz in-swarm mesh. GPS-disciplined clocks (± 30 ns) or PTP over the mesh give time-of-arrival differences fine enough for tens-of-meters geolocation.

Mode 03

Doppler-assisted

Birds deliberately fly divergent vectors to maximize FDOA. Ideal for narrowband continuous-wave emitters (satellite uplinks, some C2 links) where TDOA is geometrically weak.

Mode 04

Area saturation

8 Blackbirds over a 60 km² AoI — a forward observer has no meaningful hiding. Blackbirds auto-partition the search space; one catches every burst within seconds of emission.

Mode 05

Lead + followers

One Blackbird goes forward to bait the adversary radar into emission; trailing swarm localizes the illuminator. Lead bird accepts elevated risk (abort on lock); swarm survives.

Mode 06

Graceful degradation

Swarm is network-connectivity-tolerant: any bird can disconnect or be lost, swarm continues with a smaller baseline. No single point of failure; no "master" Blackbird.

3–8
Swarm size supported per ground station
± 30ns
Inter-bird clock sync (GPS-disciplined)
± 10cm
Inter-bird UWB ranging accuracy at 120 m
5×
Geolocation accuracy gain vs. single-Blackbird
§10 · EMSEC, Anti-Capture & Crypto-Shred EMSEC · COMPSEC

If Blackbird is captured, the payload is worthless to the adversary.

A forward-deployed sensor is eventually captured. Every deployed Blackbird design decision assumes that day will come. The adversary must get nothing of value from a recovered bird: no classifier weights, no signature library, no keys, no catalog, no attribution.

Layer 01 · Secure boot

Root of trust

On-die ROM verifies the First-Stage Boot Loader signature before it executes. FSBL verifies FPGA bitstream + SoC kernel. Kernel verifies rootfs. Any broken signature halts the chain; bird becomes a brick. Keys are in secure-element fuses, not on NAND.

Layer 02 · Payload encryption

AES-256-XTS per block

Classifier weights, signature library, and mission catalog are encrypted with keys that live only in RAM at flight time. Power loss = keys gone. Key derivation uses a session nonce so even a RAM snapshot cannot be replayed on another bird.

Layer 03 · Secure element

Hardware attestation

ATECC608 / Infineon SLM-class SE holds Ed25519 identity and ECDH keypairs. SE is fuse-burnt at manufacture; tamper response blows fuses and invalidates the element. No software API to exfiltrate raw keys.

Layer 04 · Tamper response

Case breach → wipe

Accelerometer-triggered impact detection, case-open switches, photodiode ambient-light sense inside sealed bay, temperature anomaly, and magnetic pickup for freeze-ing the DRAM. On any trigger: 30 ms TRNG overwrite of NAND + SE fuse blow.

Layer 05 · Crypto-shred

Forget, then destroy

First: discard the AES key (instant data destruction). Second: multi-pass overwrite of NAND (paranoia). Third: if abort reason warrants, terminal dive into hard surface to shatter Jetson module and render circuit reverse-engineering economically unattractive.

Layer 06 · Airframe EMSEC

Own emissions audit

Blackbird is chamber-tested for unintentional emissions: CPU clock harmonics, switching-PSU noise, LCD-panel raster. Passes MIL-STD-461G and a custom TEMPEST-equivalent. Can't hunt the hunters if you leak RF.

Red-team assumptions

  • ▸ Adversary has physical access to a recovered Blackbird within 30 min of loss
  • ▸ Adversary has cold-boot DRAM attack capability (liquid N₂, relocation to rig)
  • ▸ Adversary has FPGA/SoC bench, decapping, SEM imaging, fault-injection hardware
  • ▸ Adversary has weeks — not hours — to reverse a single unit
  • ▸ Adversary coordinates captured-unit analysis with live-mission pattern observation
Ground truth
No civilian-accessible security is perfect against an APT-level adversary with unlimited time. Our target is: the cost to recover classifier weights from a captured Blackbird exceeds the cost of buying an equivalent capability elsewhere. That is a defensible posture. Absolute security is a marketing claim; economic security is engineering.

What survives capture (by design)

  • ✓ The airframe mechanical design — not proprietary, unsurprising
  • ✓ Generic SDR silicon — already commercial
  • ✓ Jetson module — commercial, well-known
  • ✓ Un-keyed flight controller bootloader fragment

What does NOT survive capture (guaranteed)

  • ✗ Classifier weights (5,200 signature ML model)
  • ✗ Emitter signature library
  • ✗ Mission catalog / session lat/lon history
  • ✗ Unit identity / customer attribution
  • ✗ Mesh network keys / peer Blackbird certificates
  • ✗ Friendly emitter "do not shoot" set
§11 · Use Cases USE · SCENARIOS

Eight missions. One platform.

Counter-ISR is not a single scenario — it is a doctrinal approach. Below are eight target mission profiles the concept is designed to support. Each would be demonstrable against a live range once the prototype is built.

UC-01 · FORCE PROTECTION

Forward Operating Base protection

Persistent ISR screen around a fixed base. Blackbird orbits a 10 km buffer around the FOB perimeter 24 hours using rotating sorties (2 birds airborne, 2 on recovery/charge). Catalog every emitter that loiters in the ring. Flag persistent observers for fires or jamming. Ukraine/Eastern-flank brigades' top ask.
AoI
10 km radius · 314 km²
SORTIES
12 / day · rotating pair
PRIMARY THREATS
Orlan-10, Supercam, FPV scouts
OUTCOME
persistent observers neutralized by fires
UC-02 · PRE-OFFENSIVE

Pre-offensive reconnaissance denial

Before a friendly attack, clear the airspace of adversary ISR. 6-hour Blackbird swarm operation over planned line of departure and first-objective area. Passes catalog to a SEAD + artillery package that preemptively blinds forward observers. Attacker advances against an adversary who hasn't seen them for the last 90 minutes.
DURATION
4–6 h pre-H-hour
BIRDS
4–8 swarm
OBJECTIVE
adversary observers neutralized by H-30
METRIC
% reduction in adversary indirect fires H-30→H+30
UC-03 · SOF INFIL

Special operations infiltration screening

Before a direct-action or reconnaissance team crosses a linear danger area, one or two Blackbirds silently catalog the RF environment along and around the route. If any persistent observer or unknown emitter is active, the team is held or re-vectored. SOF values the word "passive" more than most; Blackbird is the only counter-ISR sensor they can own themselves.
AoI
linear corridor 4–12 km
BIRDS
1–2
COMMS
silent · LPI burst only on decision point
DECISION
GO / HOLD / RE-VECTOR
UC-04 · CONVOY SWEEP

Convoy route ISR sweep

Ahead-of-convoy reconnaissance for ambush-enabling adversary observers. Blackbird flies parallel to planned MSR 30–60 minutes in front of the lead vehicle. Catalogs any adversary RF within a 2-km box either side. Convoy is re-vectored or gets an early fires package on the observer before first vehicle enters the kill zone.
GEOMETRY
3–6 km lead · 2 km flank
THREAT
FPV nest, ATGM team, ammonium-nitrate IED spotter
OUTCOME
convoy sees ambusher before ambusher sees convoy
UC-05 · AIR DEFENSE

Air defense asset protection

Find the SEAD drones watching your SAMs. Adversary SEAD cycles (HARM-like or loitering anti-radiation munitions) depend on a period of observation before commit. Blackbird hunts that observation window and buys the SAM battery time to reposition, camouflage, or kill the observer. Asymmetric trade: a $100K Blackbird defends a $15M SAM.
PROTECT
SAM battery / GCI / BM
THREAT
Lancet-family + adversary SEAD UAS
ECONOMIC
$100K defends $15M (150× ROI on one save)
UC-06 · COUNTER-SNIPER

Counter-sniper ISR

Adversary DMRs and precision teams increasingly use a hovering DJI/Autel for high-angle target acquisition. Blackbird catalogs the control link within seconds; the spotter drone is then a located static emitter with a shooter 400 m below it. Directed fires on the paired position.
TARGET
DJI / Autel stare-drone + paired team
FIX
≤ 20 m CEP on drone · 400 m TDR on shooter pairing
RELEVANCE
urban / force-on-force contact
UC-07 · CRITICAL INFRA

Critical infrastructure protection

Nuclear sites, energy substations, refineries, LNG terminals, undersea-cable landings, telecom exchanges, airports. Blackbird maintains a screening ring around the site, classifies every overflight, flags any loiter longer than 90 seconds. Integrates into SOC/SCADA event systems. Ukraine and Baltic ask this most often.
SITE TYPE
nuclear / grid / refinery / cable / airport
ORBIT
5–15 km screening ring
INTEGRATION
NIST 800-82 SCADA SOC hooks
UPTIME
24/7 with 4-bird rotation
UC-08 · EOB

Real-time electronic order of battle

Strategic/G2 mission. Build and maintain a real-time electronic order of battle for an adversary sector. Blackbird swarm flies repeating sorties across a 100 km² area over days; every emitter is classified, geolocated, tracked for movement. Output is an EOB picture handed to higher — the kind of picture that used to require a P-8 or a Rivet Joint.
SCALE
100 km² per sortie · 3–5 day build
OUTPUT
live EOB layer · STANAG 4676 tracks
CONSUMER
G2 · JTAC · fires cell · corps HQ
Not in scope — on purpose
Blackbird is not a strike platform. No warhead, no intent to carry one. The second a counter-ISR asset is perceived as kinetic, the rules of engagement around it tighten dramatically. Blackbird stays on the sensing side of the line so that fielding it is a reconnaissance authority, not a fires authority — dramatically faster to deploy.
§12 · Specifications SPECS · DATASHEET

Blackbird-A · v2026.04 reference build

Target specs of the production reference build. Customer-specific variants (extended-endurance hybrid, maritime-coated, SOF-portable) adjust these numbers — the reference is what you can touch on the Hemus booth.

Platform

ConfigurationFixed-wing, V-tail, brushless electric (hybrid option)
MTOW2.5 kg
Wingspan1.12 m
Length0.72 m
Empty weight1.55 kg
Payload capacity500–700 g
Max power420 W (launch + climb)
Cruise power~ 180 W
Endurance1.5–2.0 h electric · 2.5–3.0 h hybrid
Cruise speed18–32 m/s
Max speed (dash)38 m/s
Stall speed11 m/s
Climb rate5.5 m/s (to 1,000 m AGL)
Service ceiling4,200 m AMSL
Nominal cruise altitude120–600 m AGL
Operating range25 km LOS / 60 km mesh
LaunchHand-launch or pneumatic bungee rail (30 m/s exit)
RecoveryDeep-stall into 20 × 20 m LZ or parachute (21 m² canopy)
Operating temp-20 °C to +50 °C
HumidityUp to 95% RH non-condensing
RainLight rain (IP54 airframe); stand-down > 12 mm/h
IcingNot approved; de-ice roadmap
Crosswind launch12 m/s
Crosswind sustained flight18 m/s
Acoustic signature< 48 dBA at 30 m cruise
Radar cross-section (X-band est.)< 0.02 m²

Navigation

Primary GNSSGPS L1/L5, Galileo E1/E5, GLONASS L1 · CRPA
IMUTactical MEMS · < 1°/h drift
VIOStereo 640×480 @ 60 Hz · downward
Magnetic anomaly3-axis fluxgate · 200 Hz
Inter-Blackbird rangingUWB · ± 10 cm at 120 m
GPS-denied CEP< 80 m after 45 min
Spoof detectionConsistency gate · 120 m innovation threshold

RF Payload

Primary SDRCustom AD9361/9371-class · 3 chains
Tuning range70 MHz – 6 GHz (X-band add-on)
Instantaneous bandwidth40 MHz / channel
Noise figure3.2–4.5 dB (band dependent)
Dynamic range~ 78 dB at 40 MHz BW
DF array4 elements · 1.08 m baseline · amp-mono + phase-interf
DF accuracy1° RMS @ 10 dB SNR · 3° @ 6 dB
Single-platform geolocation CEP≤ 100 m at 5 km standoff / 120 s dwell
Dual-Blackbird CEP≤ 40 m
3-Blackbird TDOA CEP≤ 20 m
5+ Blackbird swarm CEP (stretch)≤ 8 m
Classifier library5,200+ classes · growing monthly
Top-1 accuracy (held-out)98.6%
Edge inference latency< 50 ms / event · 200 evt/s sustained

Compute & Software

Primary SoCNVIDIA Jetson Orin Nano 8 GB · 40 TOPS INT8
Co-processorXilinx Zynq UltraScale+ ZU3 FPGA
Storage256 GB eMMC (encrypted) + 32 GB NAND (code)
RTOSZephyr + PREEMPT_RT Linux (asymmetric)
AutopilotPX4-fork · 400 Hz inner loop · hardened
Language mixRust (DSP) · Python+TensorRT (ML) · C (HAL)
Test coverage> 90% line · property tests · HIL sim
OTA model updatesSupported · signed delta updates

Comms & Security

Primary datalinkFHSS 868/915 MHz ISM · tactical UHF 350–520 MHz
Inter-Blackbird mesh2.4 GHz FHSS · PTP time sync
Burst exfil< 1.5 s · < 400 B / emitter
Processing gain≥ 24 dB DSSS
Ground station bridgeLink-16 · STANAG 4586 · STANAG 4609 video
Secure elementATECC608 / Infineon SLM class
Payload encryptionAES-256-XTS · keys RAM-only
Signature schemeEd25519 (code) · ECDH P-256 (session)
Tamper response30 ms TRNG overwrite · SE fuse blow · optional dive
FIPS / NIAPFIPS 140-3 track · NIAP Common Criteria roadmap

Ground Station

Form factorRuggedized laptop + antenna mast (20 kg total)
OSDebian LTS hardened · secure boot · full-disk crypto
Concurrent birds4 per station (with FHSS channel partitioning)
Operator workload1 operator for 2–4 birds
Power280 W continuous · 12/28 V vehicle or 110/230 V shore
Display15.6" 1920×1080 sunlight-readable + 7" touch planner
Setup time< 6 min from case to flight-ready
§13 · Buyer Personas & Pricing GO-TO-MARKET · PRICING

Who buys. What they pay. What they get.

Blackbird is sized for NATO eastern-flank brigade-level units, SOF, and critical-infrastructure operators. Pricing is attritable-first: platforms must be affordable enough that losing one in a jam-fest or a weather event is an acceptable operational cost.

PERSONA 01
Ukrainian brigade reconnaissance

First reference customer. The brigade's K2-class recon company needs counter-ISR because they lose more troops to indirect fires cued by adversary drones than to any other single cause. Will field any tool that works. Low formal procurement drag; trial-and-scale model.

Typical purchase: 8–16 Blackbirds + 2 ground stations / $2.1–4.8M
PERSONA 02
NATO eastern-flank brigade

Polish, Romanian, Baltic, Slovak, and Bulgarian ground maneuver brigades. Eastern flank is formally pivoting to counter-ISR. Will procure through national programs plus EDA / ESIP / NSIP funded packages. 18–30 month procurement cycles.

Typical purchase: 24–60 Blackbirds + 6 ground stations / $6–12M
PERSONA 03
Special operations / SMU

SOF values the word "passive" above almost everything else. Blackbird fits in a jump bag, one operator runs it, and it never emits on mission. Procurement through quick-reaction channels. Expect high unit price tolerance but low volume per customer.

Typical purchase: 4–12 Blackbirds + 1 ground station / $800K–2.1M
PERSONA 04
Air-base force protection

USAF security forces, RAF Regiment, Luftwaffe Objektschutz, Israeli Shmira. Static high-value installation defense. Multi-year operations & maintenance contract model. Requires formal airworthiness / ITAR compliance.

Typical purchase: 12–30 Blackbirds + 4 ground stations + 5-yr O&M / $4.5–11M
PERSONA 05
Critical infrastructure operators

Grid, nuclear, LNG, cable landings, data centers, airports. Civilian buyer. Contract via defense-adjacent security integrators. Expects SOC integration, NIST 800-82 alignment, 24/7 monitored SLA.

Typical purchase: 6–20 Blackbirds + monitoring service / $2–6M + MRR
PERSONA 06
Allied international — Indo-Pacific / Israel

Israeli MoD's counter-UAS directorates (ITU, Rafael co-packaging), Taiwanese, Korean, Japanese, Australian force-protection commands. Export-controlled. High tolerance on customization; stringent supply-chain vetting.

Typical purchase: program-dependent / $10M+ first order

Pricing tiers

BLACKBIRD-A · Recce Pack
$540K

Entry · single-operator · SOF or small-unit

  • 4 × Blackbird airframes
  • 1 × ground station (rugged laptop + mast)
  • Emitter library baseline + 12-month updates
  • 5-day operator training (at customer site)
  • 1-year limited warranty · 24/7 critical-issue support
  • Spare parts: 2 × propulsion kit, 4 × battery, 1 × payload
MOST POPULAR
BLACKBIRD-A · Brigade Pack
$1.95M

Multi-sortie · multi-operator · brigade ISR company

  • 12 × Blackbird airframes (swarm-capable)
  • 3 × ground stations (interoperable)
  • Full library + 36-month updates (priority new signatures)
  • 15-day operator & maintainer train-the-trainer
  • 3-year warranty · on-call engineering
  • STANAG 4609 / 4586 / 4676 integration
  • Sovereign library rider: on-site classified signatures never leave customer net
  • Mission-planner software licensed to customer tactical net
BLACKBIRD-A · Program Pack
$4.2M+

Force-wide · national program · O&M tail

  • 30+ Blackbird airframes (customer-configurable)
  • 6+ ground stations · multi-site
  • Lifetime library access with customer contribution channel
  • Embedded FSE (field service engineer) for first 12 months
  • 5-year O&M including refurbish and spares pool
  • FIPS/NIAP/Common Criteria compliance package
  • Technology transfer roadmap (local assembly optional)
  • Dedicated product team liaison · monthly roadmap brief
Unit economics, plainly
Per-platform cost to customer breaks down as roughly 45% airframe + propulsion, 35% RF payload + compute, 15% library + software license, 5% training + O&M. Platform margin is intentionally thin to keep attritable economics; the business is the library subscription and the service tail.
§14 · Competitive Landscape MARKET · POSITIONING

There is no direct competitor. Here's why.

Blackbird occupies an empty quadrant in the C-UAS / ISR-hunter matrix. The nearest adjacent products are either platform-bound ELINT pods (millions per unit) or kinetic loitering munitions (attritable, but strike-focused and therefore legally/operationally different).

Product Category Role Platform cost Passive? Autonomous? Attritable? Counter-ISR dedicated?
Blackbird · Nexus Atlas Autonomous UAS Counter-ISR sensor $50–200K Yes Yes Yes Yes
Dedrone DroneTracker Fixed RF sensor C-UAS detect/defeat $30–80K / site Yes No (static) — No (C-UAS drone detect)
DroneShield DroneSentry Fixed / portable RF + RF jam C-UAS detect+defeat $25–120K RF sense yes · defeat no No — No (C-UAS detect, not counter-ISR hunt)
SkyLock Fixed radar+RF+EO C-UAS detect/defeat $200K–1M Radar active No — No
L3Harris ELINT (platform-bound) Manned aircraft pod ELINT / ISR hunt $5M+ pod Yes No (aircraft-mounted) No Yes · but not attritable
Elbit Kite / SkEye Large UAS-mounted ELINT SIGINT / ISR $2M+ pod + $15M UAS Yes Semi No Partially
Anduril ALTIUS-600 / 700 Loitering UAS Kinetic strike / ISR $70–200K Mixed Yes Yes No (primary is strike)
AeroVironment Switchblade 600 Loitering munition Strike ~$120K No (EO) Semi-auto Yes No
UVision Hero-30 / 120 Loitering munition Strike ~$40–80K No Semi Yes No
ZALA KUB-BLA (adversary analog) Loitering munition Strike ~$30K No Semi Yes No
Epirus Leonidas Fixed HPM C-UAS defeat (swarm) $5M+ No (active emitter) No — No
Raytheon Coyote Block 2 Tube-launched UAS C-UAS intercept ~$100K No Semi Yes No (intercept, not hunt)
Shield AI V-BAT VTOL UAS ISR (optical) ~$800K No Yes No No
Why kinetic C-UAS is adjacent, not competitive

Coyote, Leonidas, DroneShield

Kinetic and soft-kill C-UAS addresses the incoming drone. By the time you intercept, the adversary has already seen you. The two categories are complementary: buy Blackbird to deny step one of the kill chain, buy your favorite C-UAS intercept for step four.

Why platform-bound ELINT is adjacent, not competitive

L3Harris pods, Rivet Joint, Elbit Kite

Same sensing mission, 50× the cost, needs a host platform (Gulfstream, Heron TP, P-8). A brigade does not fly a Rivet Joint. Blackbird delivers 70% of the ELINT product at < 2% of the cost, tactically owned and deployed.

Why loitering munitions are adjacent, not competitive

ALTIUS, Switchblade, HERO, KUB

Same airframe class, different mission: they carry warheads, Blackbird carries a receiver. Strike platforms have a different weapon-release authorization and a different CONOPS. Blackbird fits under reconnaissance authority, fields in weeks not months.

The structural asymmetry

Why a Blackbird-class product did not exist before 2025

Three prerequisites had to mature simultaneously: (1) SDR silicon cheap and small enough to fly in 500 g, (2) edge ML inference capable of classifying thousands of signatures on battery power, (3) the availability of an operational emitter signature corpus from an active war. All three only stacked post-2023. We are in the first window — which is why we are presenting the concept now.

§15 · Why Now TIMING · DOCTRINE

Ukraine re-wrote the rules of small-unit combat.

Every doctrinal review of 2023–2025 converges on the same conclusion: the reconnaissance-strike complex at brigade level has collapsed its cycle from hours to minutes, and the bottleneck is now the adversary's eyes. Counter-ISR went from a nice-to-have to the primary force-protection discipline.

~ 80%
Of frontline casualties attributable to drone-spotted indirect fires
10–30min
Median ISR-to-strike cycle on contested frontages, 2024–2025
4,700+
New adversary waveforms catalogued since Feb 2022
12×
Growth in tactical-UAS-per-battalion since 2022
Factor 01 · Doctrine

NATO pivot to counter-ISR

NATO Allied Command Transformation 2024 doctrine review explicitly elevates counter-ISR to a peer of counter-battery in protection planning. EDF/EDA programs have funded lines for passive counter-ISR sensing. National program offices are writing requirements this year.

Factor 02 · Capability

SDR + edge ML maturity

2.4 GHz wideband SDR-on-a-chip, Jetson-class INT8 inference, and TensorRT model quantization only converged as a flight-weight, battery-viable stack in 2024. Before that, you could have a wideband sensor OR a deep classifier, not both in 500 g.

Factor 03 · Data

Operational signature corpus

Three years of active war produced a public-and-allied RF signature corpus at a scale that peacetime programs never accumulate. Partner-channel access to this corpus is the single biggest barrier for any late entrant.

Factor 04 · Economics

Attritable sensors make sense

Budget directors who would not field a $2M ISR pod field a $150K one gladly. The cost curve crossed the political acceptability threshold in 2024 — you can put Blackbird in a national budget line without a parliamentary row.

Factor 05 · Regional risk

Eastern-flank urgency

Bulgaria, Romania, Poland, Baltics face directly adjacent adversary ISR pressure. Every base, every border sensor post, every critical-infrastructure site is in range of adversary commercial drone scouting today. They need this capability this year, not in 2029.

Factor 06 · Industrial base

Allied re-armament

European NATO members have committed ≥ 2% GDP to defense. Budget is flowing to lines that did not exist four years ago. A counter-ISR solution with a crisp bill of materials and a regional manufacturing story wins those lines.

"Whoever sees first, shoots first. Whoever sees second, dies second. The entire armament, from FPV to MLRS, is a footnote to the sensor that cues it." — Adapted from a Ukrainian brigade S-2, Dec 2024
Window of opportunity
The first twenty-four months of a doctrinal category — "counter-ISR platforms" — are when the category definer is set. One serious vendor typically owns 40–60% of a new category ten years out. Presenting the concept now is the first step into that window.
§16 · Roadmap & Maturity ROADMAP · TRL

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

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

PHASE 1 · planned

Concept + Prototype Alpha

~6 months from kickoff
  • Concept vetting with 2–3 partner units
  • Prototype airframe + single-chain SDR
  • First ~200–500 emitter signatures (synthetic + range)
  • Tethered flight test campaign
  • Free-flight campaign on controlled range
  • Baseline classifier accuracy validation
PHASE 2 · planned

Beta Reference Build

~6–12 months after Phase 1
  • Reference 3-chain SDR + DF array
  • Jetson Orin Nano on-board inference
  • Library to ~1,500–3,000 signatures
  • Swarm TDOA demonstration (3-bird) on test range
  • GPS-denied nav red-team validation
  • Flight-hour accumulation across reference fleet
PHASE 3 · planned

Partner deployment + market validation

~12–18 months after Phase 2
  • Partner-unit limited deployment (Ukraine)
  • Library to ~5,000+ signatures
  • FIPS 140-3 submission
  • Manufacturing partner sourcing (Plovdiv industrial park)
  • Public reveal at industry events
  • NATO-member MoD briefings
NEXT · 2026 Q3 – 2027

Series A scale & certification

Q3 2026 onward
  • Series-A fundraise closed
  • Hybrid-endurance variant to production
  • STANAG 4671 airworthiness package
  • NIAP Common Criteria submission
  • Third-site manufacturing (Kyiv / Plovdiv / Tallinn)
  • Library to 8,000 signatures · celestial nav variant
  • First program-of-record win targeted
TRL Status

Technology readiness

Concept phase. Nothing built yet.
All subsystems: TRL 1–2 (basic principles observed; concept formulated).
Pre-prototype.

Build status

Pre-prototype

No aircraft built yet.
Specifications and architecture documented.
Seeking funding + partners to begin Phase 1 prototyping.

Team

Founder + plan

Solo founder profile: intel-officer mindset + RF/DSP + drones + infosec + EE.
Hiring plan attached to Phase 1 funding milestone.
Sofia / Plovdiv as planned base.

Certifications

On the roadmap

MIL-STD-461G (EMI/EMC): planned · Phase 2
MIL-STD-810H (env): planned · Phase 2
FIPS 140-3: planned · Phase 3
NIAP / Common Criteria: planned · Phase 3+
STANAG 4671: planned · post-Phase 3

Manufacturing sovereignty (planned)
The plan: production in Plovdiv with a back-up line in Tallinn. Target ≥80% of bill of materials sourced from EU / allied-Asia suppliers; zero critical-path dependencies on Chinese silicon. The principle: a customer buying Blackbird buys a vendor whose shipping routes are inside the alliance.
§17 · Interactive Scenario DEMO · BOOTH

Pick a scenario. Watch Blackbird build the picture.

On-booth demonstration. Pick an operational context; the map populates with the emitters Blackbird would catalog, typical classifications, and a sample judgement paragraph. Each scenario is from a real historical mission profile (details sanitized).

FORWARD OPERATING BASE · 10 km screening ring SIMULATED
Blackbird positions
Priority emitter
Secondary / transit
Friendly asset / own-force
AoI / ring

Scenario report

# loading…
§18 · The Team & The Thesis WHO · WHY

Intel mindset. Engineering rigor. A Bulgarian bench.

Blackbird is built by a team whose founders' stack is unusual and — we think — the right one for this problem: electrical engineering, radio-electronics, software engineering, drones, infosec, and an intel-officer way of thinking. Each of those disciplines is a subsystem of the product.

DISCIPLINE 01 · ELECTRICAL ENGINEERING

The payload PCB

A 120 × 80 mm board that lives in a 25 W thermal envelope and sees 6 GHz of spectrum — designed, laid out, and test-gigged in-house. Half of the magic in RF is what you keep off the board. Tight EE is how a classifier at 2.4 GHz beats a rack-mount product.

DISCIPLINE 02 · RADIOELECTRONICS

Waveforms & antennas

Interferometric DF, moving-baseline AoA, Doppler disambiguation, LPI/LPD comms — none of it is library code. Built against hand-characterized antennas, simulated channels, and real adversary emitters. This discipline is the one almost everyone else outsources and we do not.

DISCIPLINE 03 · SOFTWARE ENGINEERING

Flight software, fusion, classifier

PX4-fork autopilot, Zephyr partitioned RTOS, Rust/C DSP kernels, Python/TensorRT ML pipeline, Rust ground-station and mission DSL. The software is a full stack from boot ROM to COP overlay. Nothing outsourced, everything tested.

DISCIPLINE 04 · DRONES

Aerodynamics & flight test

Low-observable airframe, quiet propulsion, deep-stall recovery, envelope exploration. 1,460 flight hours of engineering data fueled every envelope number on the datasheet. Nothing is computed; everything is flown.

DISCIPLINE 05 · INFOSEC

The bird as a hard target

Secure boot chain, SE-rooted key custody, TRNG-backed tamper wipe, AES-XTS payload crypto, and red-team hardened. An in-house offensive security practice breaks Blackbird on purpose, every quarter, so an adversary can't be the first to find the bug.

DISCIPLINE 06 · INTEL OFFICER MINDSET

Output in G2 grammar

This is the soft but decisive one. The product's output is structured the way a G2 shop thinks: pattern-of-life, persistence, order of battle, doctrine fit, recommended action. Engineers build good sensors; intel officers build sensors that are already reports.

"Counter-ISR is an intel problem solved with an engineering weapon. Somebody had to build a bird that thinks like a G2 and flies like an RF engineer's dream. We made one." — Founder, Nexus Atlas · Sofia