Skip to documentation content

Acoustic Survey Methodology — Notes & References

Working notes on how professional interior-noise surveys are run, distilled from a real TfL passenger-car survey and cross-referenced against the governing standards. The goal is to reuse the systematic parts of this practice when we design our own Tube-noise data collection, rather than reinventing (or accidentally weakening) an established method.

Source document that prompted these notes: TfL Technical Services Report N&V – R3291 v1.1, "Acoustic Survey in Passenger Car of 1992 Tube Stock" (Central line, measured 8 July 2023, issued 15 Sept 2023). Released via FOI (FOI-2746-2324). The report is public to read but not open-licensed — treat it as a methodology reference, not a redistributable dataset.


1. What the TfL report does "properly" (and why)

Reading R3291 as a template, the disciplined parts of the method are:

1.1 Traceable, class-rated instrumentation

  • Uses Class 1 sound level meters (SLMs) and a Class 1 acoustic calibrator.
  • Every instrument is listed with manufacturer, type, serial number, calibration certificate number, and calibration date (their Table 1). Nothing anonymous.
  • SLMs calibrated to traceable standards within the previous 2 years; the acoustic calibrator within the previous 12 months.
  • Conformance is explicitly claimed against named standards (BS EN 61672-1:2013, BS EN 61260-3:2016, BS EN 60942:2018).

Why it matters: the measurement is only as trustworthy as the chain back to a national reference. Serial numbers + cert dates make the run auditable and repeatable.

1.2 Field calibration before and after

  • A 94 dB @ 1000 Hz calibration tone applied on-site before and after the survey, confirming the meters didn't drift during the run ("remained stable throughout").

Why it matters: a pre/post check catches drift. If the post-check disagrees with the pre-check, the whole run is suspect — you find out before trusting the data.

1.3 Well-defined measurement positions

  • Two fixed heights measured simultaneously:
    • Standing: 1.6 m above floor
    • Seated: 1.2 m above floor
  • These correspond to ear height for standing vs seated passengers — the positions the standard prescribes (see §2.2).

Why it matters: "how loud is it" is meaningless without "measured where". Fixed, justified heights make results comparable between runs and between vehicles.

1.4 Controlled, documented conditions

  • Passenger car unoccupied for the run.
  • PA announcements disabled throughout (removes a non-representative source).
  • Operating conditions representative of normal service.
  • Deviations recorded: the westbound Wanstead→Leytonstone leg where the train stood still mid-journey was excluded, and the exclusion is stated in the table footnote with the actual transit time.

Why it matters: controlling and disclosing conditions is what separates a survey from an anecdote. The reader can see exactly what was and wasn't included.

1.5 Consistent spatial unit: station-to-station

  • Every value is reported per station-to-station section, in both directions (eastbound Table 2 / westbound Table 3), each with the transit duration (mm:ss).

Why it matters: a stable, meaningful segment (platform to platform) is directly mappable and comparable. Duration is reported so an Leq is never quoted without its averaging time.

1.6 Multiple, layered metrics — not a single number

  • Primary: L<sub>Aeq,T</sub> (A-weighted equivalent continuous level).
  • Also: L<sub>Ceq,T</sub>, L<sub>Cpeak</sub>, and full one-third-octave band L<sub>eq</sub> from 20 Hz to 20 kHz in the appendix.

Why it matters: A-weighting tracks perceived loudness; C-weighting + peaks + octave bands preserve the low-frequency and impulsive content that A-weighting hides. Keeping them separate (never averaged together) lets you diagnose why a section is loud.

1.7 Reproducible analysis + provenance

  • Names the analysis software and version (01dB Trait 6.3.1 build 1).
  • Records the train unit/reference (775/1005134), date, author, reviewer.
  • Cites its normative references explicitly.

Why it matters: version + unit + reviewer means someone else could rerun the analysis and get the same numbers.


2. The governing standards (what "proper" is measured against)

2.1 ISO 3381:2021 — Railway acoustics, interior noise

Railway applications — Acoustics — Noise measurement inside railbound vehicles. The core standard for measuring noise inside trains. Key points:

  • Specifies methods for stationary, constant speed, and accelerating/decelerating running, plus driver's cab cases.
  • Divides the vehicle into acoustic areas; areas are split into segments of max 5 m, each with ≥1 measurement position.
  • Prescribes measurement heights: 1.2 m at seated positions, 1.6 m in standing areas (exactly what R3291 uses).
  • Requires Class 1 instrumentation per IEC 61672-1, IEC 60942, IEC 61260-1.
  • Grades results per ISO 12001:
    • Constant-speed / stationary type tests → engineering grade (grade 2).
    • Accelerate/decelerate → survey grade (grade 3).
    • Relaxed / in-service monitoring is explicitly below engineering grade — important: it means our in-service Tube runs are, by definition, survey-grade at best, and we should label them that way.
  • Weather/background rules: no rain/hail/snow, external wind < ~5 m/s, background at least ~10 dB below the measured signal, reported before/after each series.
  • https://www.iso.org/standard/77368.html

2.2 Instrumentation standards

  • IEC/BS EN 61672-1 — sound level meter specifications (defines Class 1 vs 2). <https://www.iso.org/standard/... (IEC 61672-1)>
  • IEC/BS EN 61260 — octave / fractional-octave band filters (the 1/3-octave bands).
  • IEC/BS EN 60942 — sound calibrators (the 94 dB / 1 kHz reference source).

Class 1 ≈ ±0.7 dB tolerance (lab/precision); Class 2 ≈ ±1.0–1.5 dB (general field). Most smartphones, at best, approximate Class 2 and only after calibration (see §3).

2.3 ISO 1996-1:2016 — Environmental noise descriptors

Defines the quantities and how to express them. Relevant definitions:

  • L<sub>eq,T</sub> — equivalent continuous level: the energy-average over a stated duration T. Always quote with a weighting and a duration (e.g. LAeq,5min).
  • A-weighting (dBA) — approximates the ear at moderate levels; the default for loudness/exposure. Attenuates lows heavily (~-40 dB at 63 Hz).
  • C-weighting (dBC) — near-flat 31.5 Hz–8 kHz; used for peaks and low-frequency content. L_Ceq − L_Aeq > ~10 dB flags significant low-frequency energy.
  • L<sub>Cpeak</sub> — the maximum instantaneous C-weighted pressure (not time-averaged). The regulatory peak metric.
  • https://www.iso.org/standard/59765.html

2.4 Occupational context (UK)

  • Control of Noise at Work Regulations 2005 — uses L<sub>EX,8h</sub> (daily A-weighted exposure) for action/limit values and L<sub>Cpeak</sub> for peaks. Useful framing for why cab/driver surveys report dose, not just section Leq.

3. Applying this to our own (crowdsourced / smartphone) collection

We won't have Class 1 meters on a volunteer's phone. The research is clear about what that means and how to stay honest:

3.1 Smartphone accuracy — what's realistic

  • iOS is far more trustworthy than Android: uniform hardware lets apps be verified. The NIOSH SLM app (iOS) is validated to ±2 dBA in lab conditions and shows strong correlation with professional SLMs. Android hardware is too fragmented for NIOSH to certify.
  • Uncalibrated apps tend to under-report — the dangerous direction for a health message. From a public-health view, err slightly high, not low.
  • An external calibrated microphone can bring a phone to ~±1 dB; without it, expect ~2 dB variability at best.
  • Apps are an adjunct/screening tool, not a replacement for regulatory measurement.

References:

3.2 Participatory-mapping protocol (borrow from NoiseCapture / Noise-Planet)

Practical rules that keep crowd data usable:

  • Phone in hand, microphone unobstructed — never in a pocket/bag.
  • Don't handle/tap the phone mid-measurement (handling noise).
  • Wait ~4 s after starting for GPS to stabilise before recording.
  • Keep each session to a homogeneous sound environment; start a new session when the environment changes (e.g. platform → moving train). ~10–15 min max per session to avoid data loss from calls/GPS drop/battery.
  • Avoid rain and strong wind (>~5 m/s), matching ISO 3381.
  • Cross-calibrate each device against a reference (Class 1 SLM or an already-calibrated phone) and store the per-device correction factor.
  • Record hardware metadata (device model, OS, app version) so we can do a-posteriori bias correction and filtering.
  • GPS/tunnel caveat: underground GPS is unreliable — expect to snap to station-pair geometry (we already have OSM track geometry) rather than trust raw fixes, and filter points with poor reported accuracy.

References:

3.3 Minimum metadata to capture per measurement (our schema starter)

Mirror what makes R3291 auditable:

FieldExampleWhy (from the pro method)
line + directioncentral / eastboundcomparable spatial unit (§1.5)
from / to stationStratfordLeytonstation-to-station section
metricLAeqnever mix metrics (§1.6)
weightingA / Crequired with every level (§2.3)
value + unit98.2 dBA
duration T143 sLeq is meaningless without T
device model / OSiPhone 13 / iOS 18a-posteriori correction (§3.1)
app + versionNIOSH SLM 1.2.6reproducibility (§1.7)
calibration offset+1.4 dBtraceability substitute (§1.1)
position/heightstanding, held ~1.5 mmeasurement position (§1.3)
conditions/exclusionscrowded; door chime includeddisclose deviations (§1.4)
gradesurvey (grade 3)honest quality label (§2.1)
gps accuracy (m)snapped / 12 mfilter bad fixes (§3.2)

3.4 Honesty rules (non-negotiable)

  • Label everything crowd-collected as survey grade at best — never imply engineering-grade parity with the FOI surveys.
  • Keep A / C / peak / octave data in separate fields; don't blend.
  • Store the correction factor, don't silently bake it in — keep raw + adjusted.
  • Show coverage and uncertainty, not just a colour.

4. TL;DR — the transferable checklist

  1. Class-rated (or best-available, characterised) instrument, with a stated tolerance.
  2. Calibrate before and after; discard runs that fail the post-check.
  3. Fixed, justified measurement positions/heights.
  4. Controlled + documented conditions; disclose every exclusion.
  5. Consistent spatial unit = station-to-station, with duration.
  6. Layered metrics (LAeq + LCeq + LCpeak + octaves), never averaged together.
  7. Full provenance: device, software version, operator, date, correction factor.
  8. Grade the result honestly (engineering vs survey) per ISO 12001 / ISO 3381.

5. Source index