Aircraft Noise Tracker

1Project Goals

Aircraft Noise Tracker is an iOS application designed to produce community-grade acoustic measurements of aircraft overflights that are defensible, reproducible, and linked to specific flight operations. The core thesis is that aviation-noise regulation depends almost entirely on modeled, averaged L_eq metrics (DNL, CNEL) that systematically undercount event exposure. The app addresses this by generating per-event, per-aircraft observational records correlated with ADS-B flight data.

The specific objectives are:

• Event-level measurement: capture peak dBA for every aircraft that passes within a defined proximity zone, linked to the aircraft's callsign, type, altitude, and geometry at the moment of measurement.
• Psychoacoustic characterization: decompose the noise into perceptual dimensions (loudness, sharpness, annoyance) that map to specific health outcomes, going beyond what dBA alone conveys.
• Evidentiary data quality: produce records with known uncertainty bounds, validated against a calibrated Type 2 sound-level meter, suitable for regulatory submission.
• Community scale: provide a low-cost instrument deployable at multiple receptor sites simultaneously, enabling spatial coverage that fixed airport monitors cannot provide.
• Subjective-effects correlation: link measured acoustic events to real-time annoyance reports from observers, connecting the physical stimulus to the experienced burden.

2Code Architecture

The app is written in Swift using the SwiftUI framework for iOS. It follows an ObservableObject architecture with six principal components that communicate through shared state:

3Audio Processing¹

Signal Chain

The app uses AVAudioEngine with a tap installed on the audio input node. The tap delivers 1024-frame PCM buffers continuously at the device's native sample rate (typically 44,100 Hz or 48,000 Hz). The signal is processed in real time, with peak values accumulated over 2-second windows before being recorded as a single SPL sample.

A-Weighting (IEC 61672)

A-weighting is implemented as three cascaded biquad IIR filter sections, derived from the analog IEC 61672 prototype via bilinear transform. The poles correspond to the four standard A-weighting characteristic frequencies (f₁ = 20.6 Hz, f₂ = 107.7 Hz, f₃ = 737.9 Hz, f₄ = 12,194 Hz). The cascade is normalized so the gain at 1 kHz equals 0 dB — the digital equivalent of a hardware A-weighting network, computed sample-by-sample in the audio callback.

C-weighting uses the same architecture but with only two biquad sections (the f₁ and f₄ poles, omitting the mid-band section), giving the characteristic flat response with low- and high-frequency rolloff.

SPL Computation

For each buffer, the RMS amplitude is computed across the frame:

The 2-second peak window captures the maximum instantaneous reading across all buffers within the window. This is conservative — it represents the highest sound level heard during the window rather than an energy average. The C–A delta (difference between C-weighted and A-weighted readings) is tracked simultaneously as a wind-contamination indicator.

Psychoacoustic Analysis²

Independently of the SPL chain, raw PCM samples accumulate in a 4096-sample buffer. When full, the buffer is dispatched to a background analysis queue running PsychoacousticAnalyzer:

• 1/3-Octave Band Analysis: a Hann-windowed FFT (vDSP framework, radix-2) decomposes the signal into 28 bands from 25 Hz to 12.5 kHz, corresponding to the Bark critical-band centers. Band power is summed from FFT bins within each 1/3-octave boundary and converted to dB SPL using the calibration offset.
• Zwicker Loudness (ISO 532-1): each band level is compared to the threshold in quiet (ISO-standardized per band). A simplified excitation model converts band levels above threshold to specific loudness in sone per Bark; total loudness integrates across the Bark scale. Output in sone (perceptual loudness) and phon (equivalent loudness level) — the most physiologically grounded single-number descriptor of perceived noise.
• Sharpness (DIN 45692): specific loudness is weighted by a frequency-dependent factor emphasizing the upper Bark bands (above ~16 Bark), capturing the "piercing" or "bright" character of turbine noise. Output in acum. Turbine whine scores significantly higher than broadband propeller noise at the same dBA — the 737 MAX vs. NG difference is detectable in this metric.
• Psychoacoustic Annoyance (Zwicker & Fastl): a composite index combining total loudness and sharpness; sharpness contributes a multiplicative penalty when it exceeds 1.75 acum.
• Onset Rate: computed from the last five 2-second SPL samples as the maximum rate of increase in dB/s. Captures the startle character of rapid noise onset.

4Interface Overview

The app is organized into four tabs plus settings:

• Monitor tab: live dBA readout with A-weighting and psychoacoustic indicators (loudness, sharpness, annoyance, onset rate), a scrolling SPL time-series chart, and a live aircraft list showing nearby traffic with geometry (callsign, type, slant range, altitude, bearing, flight phase). Monitoring is started and stopped explicitly; the 2-second peak window updates continuously while active.
• Radar / Map view: a compass-oriented display showing aircraft positions relative to the observer's GPS location. Aircraft are sized and labeled by proximity.
• Assess tab (SEAT): when an aircraft enters the 1.5-mile proximity gate, a Low / Medium / High annoyance tap becomes active, confirmed with a second tap. When the aircraft exits the gate, a 3-question ISO 15666-aligned survey appears — the activity disrupted, degree of interference, and any physical effects experienced. Responses are logged to a separate SEAT CSV and uploaded to the backend.
• Log tab: chronological list of all observations with per-aircraft summary statistics — peak dBA, slant range, altitude, psychoacoustic readings, flight phase. Includes a session summary showing N65/N70/N80 event counts, event density, inter-event interval statistics, and a recovery-deficit count (intervals shorter than 15 minutes — the cortisol recovery threshold).
• Settings: microphone-profile selection (internal / external / custom), windshield correction (none / foam / fur), measurement type and position description, calibration interface, monitoring schedule, and slant-range gate.

5Calibration and Offsets

The app applies a single SPL offset (dBA) to all readings, converting raw ADC-scaled dB to approximate absolute SPL. The offset is stored per microphone profile in UserDefaults and applied throughout the signal chain.

Manual Calibration Offset

The offset is set by comparing the app reading against a known source (calibrated tone, reference meter, or known aircraft). The user enters the target SPL and the app computes the correction. An initial +96 dBA offset was established for the Pop lavalier microphone, derived from known-level test signals and validated against overflights.

Auto-Calibration (Two-Phase Transfer)

Auto-calibration uses the iPhone's built-in microphone as a stable transfer reference to characterize an external microphone's offset relative to it. Phase 1 (90 s) captures the internal mic's LA50 in ambient conditions; Phase 2 (60 s, after switching to the external mic in the same acoustic environment) captures the external mic's LA50. The difference determines the offset correction. This procedure needs no calibration source; it assumes the internal mic's response is consistent across sessions, confirmed across multiple test sessions.

Windshield Correction

Three windshield types are supported — none, foam ball, and fur windscreen — each applying an additive correction to logged values, compensating for the acoustic insertion loss of the material. This is logged to every CSV row for traceability.

C–A Delta Wind-Contamination Detection

The C–A delta (dBC − dBA) is computed in real time alongside the A-weighted reading. High C–A values indicate disproportionate low-frequency energy — the acoustic signature of wind turbulence at the diaphragm rather than aircraft noise. The app flags observations as "possible" or "likely" wind-contaminated when dBA ≥ 55 and C–A ≥ 15 or ≥ 25 respectively. Flags are written to the CSV, but values are not auto-excluded; the observer makes the final call.

Altitude Reference

Aircraft altitude drives the slant-range geometry, so its vertical datum matters. As of 2026-10-20 the app resolves altitude to height above mean sea level: GNSS altitude from the position feed is corrected to MSL using the geoid offset the iPhone measures at the observer's location (removing a roughly 100 ft error near BWI), and where only barometric altitude is available it is corrected with the local air pressure (QNH) from Open-Meteo. A new Altitude Source field records the method per row — GNSS, Baro+QNH, Baro (uncorrected), or GNSS (no geoid corr.) when the device lacks a good altitude fix.

Measurement Typology

Every session is tagged with a measurement type from a five-category framework:

The type and a free-text position description are logged to every CSV row, allowing downstream analysis to group or weight observations by receptor methodology.

Field Validation

Multiple validation sessions against a calibrated AZ8930 Type 2 sound-level meter:

• 11-aircraft post-correction comparison: SR-position mean residual −0.1 dB, Façade mean −0.78 dB (both conservative).
• 33-departure session (2026-07-31): mean residual −0.09 dB, SD 1.76 dB after the NG/MAX aircraft-type split was applied.
• Single-point check (2026-08-06): B767 at 0.6 mi / 2,490 ft — app 75.8 dBA, freshly calibrated SLM 76.0 dBA (delta 0.2 dB).

6Data Sharing and Upload

On-Device Persistence

Observations are stored in two formats simultaneously:

• JSON (aircraft_observations.json): structured array with the full field set, used for in-app display and reimport.
• CSV (aircraft_log_YYYY-MM-DD.csv): 41-column flat file, one row per observation, plus an appended summary-statistics section.

The CSV includes: timestamp, observation type (entry/track), dBA level, all psychoacoustic metrics with plain-language health labels, complete aircraft geometry (callsign, ICAO24, type code, type name, registration, operator, flight phase, ground distance, slant range, altitude, bearing, elevation angle, speed, vertical rate), observer GPS coordinates, windshield configuration and correction, measurement type, position description, C–A delta, wind speed and direction, temperature, relative humidity, altitude source, and an exclusion flag.

A companion reference file (aircraft_noise_reference.txt) is generated alongside the CSV with column definitions, health-scale descriptions, and source citations — suitable for submission alongside data to regulators or researchers.

Ground-Traffic Log

At near-field sites within about 1.5 mi of the airport, taxiing and parked aircraft are logged separately to aircraft_ground_log_YYYY-MM-DD.csv — one row per second while an airborne aircraft is being tracked, each carrying the measured dBA at that instant, so a ground-source peak (a taxiing jet) can be told apart from an overflight peak. It is isolated: it is not uploaded to truenoise.org, and it does not affect the on-screen aircraft list, SEAT prompts, the N65/N70/N80 counts, the charts, or the website data. At receptors farther from the field (such as the porch monitor) the file is normally not created. It is included in Export when present and removed by Clear Log along with the main log.

Upload to truenoise.org

The user initiates upload from the app. UploadService packages the current day's CSV as a multipart/form-data POST to the truenoise.org API (/api/v1/upload-session), authenticated with an API key. Before upload, a preprocessing step:

1. Strips the summary-statistics section (sends only observation rows).
2. Removes any rows flagged Excluded=Yes.

The backend responds with a summary: session ID, observation count inserted, N65/N70/N80 event counts, recovery-deficit count, and unique aircraft count. Failed uploads are queued locally and retried on the next attempt.

SEAT Data Upload

When a user completes a subjective-effects survey, SeatLogger passes the completed event to UploadService.uploadSeatEvent(), which POSTs a JSON payload to /api/v1/seat/assessment. The payload includes source identifier (ios_app), timestamp, annoyance level, callsign, observer GPS coordinates, and the three survey responses. Upload is fire-and-forget — a failure is logged but does not surface to the user, since the primary record is already in the local SEAT CSV.

External Data Sources

• Aircraft positions: OpenSky Network REST API — /states/all bounding box around the observer, OAuth2 client-credentials auth, ~10–15 s poll, state vectors as positional metric arrays converted on device. Migrated from airplanes.live in August 2026 after that service closed its free API.
• Aircraft details (type code, type name, registration, operator): hexdb.io (per-aircraft lookup keyed on ICAO24, cached in memory). The type code now comes from this lookup, since the OpenSky state vector carries only a coarse category.
• Weather & pressure (wind speed/direction, temperature, humidity, and the local air pressure used to correct barometric altitude): Open-Meteo (free, no key required).

7Summary of Measurement Capabilities