Technology

Separation detection.
Voice inference. Web simulation.

Three custom-built systems working together to produce the most rigorous automated ATC candidate assessment available today.

Separation Integrity Engine

Predicting conflicts before they happen.

The SIE is Falcon's core safety analysis module. It runs continuously during each simulation, monitoring every aircraft pair for potential separation violations.

Rather than simply checking current positions, the SIE projects each aircraft's future trajectory using predicted intercept vectors (PIV) — the same mathematical principle underlying real Short Term Conflict Alert (STCA) systems used in operational ATC centres.

The result is a detection rate exceeding 99% for all horizontal, vertical, and terminal-area separation violations — with sub-second latency.

SIE — Detection Coverage

Horizontal radar separation✓ Covered
Vertical separation✓ Covered
Terminal area incursions✓ Covered
Speed control violations✓ Covered
Predicted future conflicts✓ Covered
Detection is performed using mathematical trajectory analysis — not rules-based string matching. The SIE has no false-positive rate threshold and requires no manual calibration per scenario.

Voice Recognition & Inference

Understanding what controllers actually say.

Speech Transcription
Candidate speech is transcribed in real time using a robust ASR (automatic speech recognition) model tuned for aviation vocabulary. The model handles accents, radio-effect audio, and cross-talk without requiring speaker-specific calibration.
LLM Inference
Transcribed text passes to a large language model trained to understand ATC phraseology. The model resolves ambiguous transmissions, corrects transcription errors using contextual understanding (e.g. inferring "flight level two four zero" from "flight level to forty"), and classifies each transmission by intent and correctness.
Phraseology Scoring
Each transmission is scored against the ICAO Doc 4444 phraseology standard for the relevant action type. Non-standard language is flagged and quantified — the model distinguishes between harmless informality and genuinely ambiguous or unsafe phrasing.
Readback Monitoring
The simulator occasionally plants deliberate readback errors — where the simulated aircraft reads back an incorrect value. The voice layer tracks whether candidates catch and correct these errors, producing a readback vigilance score that is strongly predictive of safety performance in real operations.
Rate of Speech
The model measures words per minute across all transmissions. ICAO recommends ≤100 words per minute for standard transmissions. Candidates scoring above thresholds receive progressively lower grades — mirroring the communication standards controllers are assessed against in real environments.
Command Redundancy
The system tracks whether a candidate issues instructions an aircraft is already executing — a common sign of poor situational awareness or excessive workload. These "repeated command" events are flagged and scored accordingly.

Web-Based Simulation

A study-level terminal radar — in any browser.

The simulation environment is built in Unity and compiled to WebGL, allowing full deployment to any modern browser without plugins, extensions, or local installs.

The environment renders a full terminal area radar display with:

  • Realistic aircraft position updates at radar scan intervals
  • Flight data blocks with altitude, heading, and speed readouts
  • Electronic flight strips for clearance tracking
  • ATIS broadcast and weather overlay
  • Configurable traffic density and scenario complexity
  • Escalating difficulty across successive runs

The simulation does not simplify the operational environment — it models the full cognitive load of terminal radar control, which is why assessment results are predictive of real-world training outcomes.

Technical Stack

Simulation engine Unity · WebGL export
Separation detection Custom PIV engine (C#)
Voice transcription ASR · Real-time
Phraseology inference Large language model
Assessment dashboard Next.js · React
Infrastructure AWS · CloudFront · Edge-deployed

System Requirements (Candidate)

  • Any modern browser (Chrome, Firefox, Edge, Safari)
  • Standard USB or Bluetooth headset with microphone
  • Stable internet connection (≥5 Mbps)
  • No downloads, installs, or special hardware required

Technical Writing

From our engineering blog.

Separation integrity engine

January 2, 2026

Air traffic control voice recognition and inference

December 9, 2025

Bringing study-level air traffic control simulation to web with Unity

October 25, 2025

Questions about the technology?

We're happy to go deep on the SIE, the voice inference model, or the simulation fidelity standards we use.

desk@falconatc.com