Multi-Modal Drone Detection
for European Cities

Why Radar Cannot Solve This

The Shahed-136 has a radar cross-section of approximately 0.1 m² — a fiberglass-bodied delta wing with a 2.5-metre wingspan. At low altitude in ground clutter, conventional radar cannot reliably distinguish it from birds, weather returns, or terrain. Deploying radar coverage for a single European city costs EUR 50–100M and takes 18–36 months.

Acoustic detection exploits an entirely different physical modality — the engine noise that radar cannot suppress — at a fraction of the cost.

The fundamental insight: detection range per sensor is not what matters. What matters is detection probability across a network. A system of individually weak sensors, distributed at sufficient density, achieves near-certain collective detection.

Key principle: Smartphones DETECT. Fixed GPS-PPS nodes LOCALIZE. Each tier does what its physics allows.

Edge-to-Alert Pipeline: Under 2 Seconds

Privacy is not a feature bolted onto this pipeline — it is a structural property. On-device inference means raw audio physically cannot leave the smartphone. GDPR Article 25 compliant by design, not by policy enforcement.

The Fundamental Problem

A single smartphone microphone in an urban environment achieves ~0 dB SNR against a Shahed-136 at 300–500 metres. At 0 dB SNR with P_fa = 10^-3:

P_d = 1.8% per observation — individually useless.

AirGuard does not rely on a single phone.

Non-Coherent Integration

For 20 phones within detection range, each at SNR = 0 dB:

SNR_eff = 10 * log10(20) + 0 = 13 dB
    → Combined Pd > 82% with K-of-N voting

For 100 phones (realistic for a city block at 0.5% participation):

SNR_eff = 10 * log10(100) + 0 = 20 dB
    → Combined Pd > 99.97% with Pfa < 10^-6

The mathematics is not new — it derives from Neyman and Pearson (1933), Finn and Johnson (1968), and Varshney (1997). What is new is applying 90 years of radar detection theory to a mesh of consumer smartphones.

Interactive 3D simulations of the AirGuard detection pipeline. These run entirely in the browser.

AirGuard occupies a distinct position: city-scale coverage from consumer devices at commodity cost.

What We Built

~24,000 lines of production code • 4 languages • 136 tests (71 Rust DSP + 65 Go) • make dev runs everything

Technology Readiness

What EUDIS Gets

The economics are deliberate. Our architecture uses commodity hardware and existing smartphones. The expensive part of drone detection was always the sensors. We eliminated that cost.

Cost Asymmetry Doctrine

AirGuard inverts the economics that make drone warfare profitable. When detection costs less than the drone, attrition favours the defender.

Five-Domain Architecture

The same CFAR, EKF, and NATS primitives that detect drones serve five threat domains. This is not metaphorical — the mathematical primitives are identical.

Same algorithms, five threat domains. CFAR detects anomalous signals whether they are acoustic, network, financial, narrative, or behavioural.

Institutional Pipeline

From hackathon prototype to NATO alliance programme — a concrete pathway, not an aspiration.

Ecosystem Synergies

AirGuard does not exist in isolation. We are embedded in Romania's aerospace and defence ecosystem — with partners that turn a prototype into a certified, manufactured, deployed system.

Four partners. Acoustic validation at INCAS, manufacturing AI with UAC, MoD integration through IAR, and the AI methodology that built it all. This is not a standalone project — it is an ecosystem.

Information Sources & Methodology

All technical claims are grounded in measured data and cited references. See our Product Paper for full bibliography.