# Aircraft noise: Effective Perceived Noise Level

> One page of the phonometry documentation, as Markdown. Index: https://docs.jmrp.io/phonometry/llms.txt

Canonical: https://docs.jmrp.io/phonometry/aircraft/aircraft-noise/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/aircraft/aircraft-noise/index.md
Version: phonometry 4.0.0
Install: pip install "phonometry[full]"
Author: José Manuel Requena Plens (https://orcid.org/0000-0003-1250-6212)
Published: 2026-07-12
Updated: 2026-11-16T21:36:13Z
Standards: ICAO Annex 16, Vol. I, 8th ed.; Doc 9501, 3rd ed.; IEC 61265:1995; SAE ARP 5534, reaffirmed 2021
Verified: 7 conformance checks (ICAO Annex 16; Doc 9501; IEC 61265:1995; SAE ARP 5534) https://docs.jmrp.io/phonometry/reference/conformance/#aircraft-noise-icao-annex-16-iec-61265
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: ICAO Annex 16 Vol. I Appendix 2 EPNL: perceived noisiness, PNL, tone correction by the slope method, duration correction and the IEC 61265 verifier.

---

The **Effective Perceived Noise Level (EPNL)** is the noise-certification metric
for transport-category aircraft. It condenses a half-second one-third-octave
spectral time history of a flyover into a single number, in EPNdB, through five
steps of **ICAO Annex 16, Vol. I, Appendix 2**. This page covers the four
primitives that build the metric and the IEC 61265 measurement-system verifier.
Each quantity is validated against the worked examples of the ICAO Doc 9501
Environmental Technical Manual (ETM) Vol. I.

Every certification level on this page is asked for at one of the three
reference points Annex 16 fixes around the runway, so it helps to see them
before the mathematics.

![The three aircraft noise certification points of ICAO Annex 16 Chapter 3, in plan and side view sharing the same distance scale: a lateral reference line parallel to the runway at 450 m where take-off noise is greatest, with a symmetric point on the other side; the flyover reference point on the extended centre line 6 500 m from the start of roll, under the climb-out; and the approach reference point 2 000 m from the threshold, 120 m below the 3 degree glide path that meets the ground 300 m beyond the threshold; microphones sit 1.2 m above the ground and the metric at all three points is EPNL in EPNdB](https://docs.jmrp.io/phonometry/media/diagram_aircraft_certification.svg)

Three points, because a transport aeroplane is loud in three different ways.
The **lateral** point, on a line 450 m to the side of the runway and placed
where the take-off noise is greatest (§3.3.1 a), catches full take-off thrust at
a shallow elevation angle, where ground effect and engine installation matter
most. The **flyover** point, 6.5 km from the start of roll on the extended
centre line, catches the aeroplane after thrust cutback, which is the community
case. The **approach** point, 2 000 m before the threshold and 120 m below the
3° path, catches an airframe- and fan-dominated condition at low power. All
three are scored with the EPNL of this page, but the limits differ and are not
single numbers: Chapter 3 §3.4 sets them as lines rising with maximum
certificated take-off mass: 101 EPNdB at flyover for a twin at 385 000 kg and
above, 104 for a trijet and 106 for four engines or more, against 103 lateral
and 105 on approach at the top of their own mass ranges. The same EPNL can
pass at one point and fail at another. The microphone is 1.2 m above the ground
at every point, which is why the measurement section below cares so much about
grazing incidence.

## Perceived noisiness and PNL

Each of the 24 one-third-octave-band levels (50 Hz–10 kHz) is converted to a
perceived **noisiness** in noys by the analytic piecewise law of Table A2-3,
then combined into the total noisiness $N = 0.85\,n_{\max} + 0.15\sum n$ and
the perceived noise level $\mathrm{PNL} = 40 + (10/\log_{10} 2)\,\log_{10} N$.

A noy is a linear scale of perceived noisiness: two noys are meant to be twice
as noisy as one, which is what lets the bands be combined arithmetically instead
of energetically. The per-band conversion is piecewise-linear in level with
slopes that differ from band to band, so the same 70 dB is worth far more noys
at 3 kHz than at 63 Hz. The law carries the ear's sensitivity in the same way
an equal-loudness contour does, but calibrated on judgements of *noisiness*
rather than loudness. The combination rule is a masking statement, not an energy
sum: the loudest band alone accounts for 85 % of the total, and everything else
together is credited with 15 %, which is why a single fan tone can dominate a
spectrum that is broadband on paper. The logarithm then buys a level-like
quantity: with $\mathrm{PNL} = 40 + 33.22\lg N$, one noy is 40 PNdB and every
doubling of total noisiness adds exactly 10 PNdB. Two limits come with it. The
law is tabulated only for the 24 bands from 50 Hz to 10 kHz, so energy outside
that range is invisible to EPNL by construction; and typical transport flyovers
at the certification points land between about 90 and 110 PNdB, which is the
scale a printed PNL should be read against.

```python
from phonometry import aircraft

noys = aircraft.perceived_noisiness(spl)      # per-band noys (spl = 24 band levels, dB)
pnl = aircraft.perceived_noise_level(spl)      # PNdB
```

## Tone correction

A pure tone is judged much noisier than the same energy spread across its
one-third-octave band, so spectral irregularities (fan and turbine tones) are
penalised by a **tone correction** $C$. The slope ("encircling") method
reconstructs what the spectrum would look like *without* the tone by smoothing
the band-to-band slopes into a background $\mathrm{SPL}''$, so the excess
$F = \mathrm{SPL} - \mathrm{SPL}''$ is a tone-only quantity. Above 1.5 dB it is
mapped to a correction factor by the rule of Table A2-2, and the maximum over
bands is taken:

| Band | $F < 3$ dB | $3 \le F < 20$ dB | $F \ge 20$ dB |
|---|---|---|---|
| 50 Hz-400 Hz and 6.3-10 kHz | $F/3 - 0.5$ | $F/6$ | $3\tfrac{1}{3}$ dB |
| 500 Hz-5 kHz | $2F/3 - 1$ | $F/3$ | $6\tfrac{2}{3}$ dB |

The mid-band branch penalises a tone twice as hard, which is why a 2500 Hz fan
tone dominates the correction of a turbofan spectrum. The implementation
reproduces the ICAO Doc 9501 ETM Vol. I Table 3-7 turbofan example exactly
($C = 2.0\ \text{dB}$ at 2500 Hz).

Two traps come with the method. A rippled but tone-free spectrum can produce a
correction, which the standard itself allows you to reject after a narrow-band
re-analysis (§4.3.1 Step 10). And a tone falling between two bands splits its
excess, so the raw correction comes out too small. The **bandsharing
adjustment** $\Delta_\mathrm{B}$, the shortfall of $C$ at the $\mathrm{PNLTM}$ record
against the mean of the records within one second of it, adds it back before
the 10 dB-down window is located. Annex 16 states it in TPNdB, the unit of the
maximum PNLT it is added to (Nomenclature, I-7), and the results do too.

```python
from phonometry import aircraft

c = aircraft.tone_correction(spl)              # dB; added to PNL to give PNLT
# start_band=2 (80 Hz) is the aeroplane start of analysis; a helicopter
# procedure starts at 50 Hz, which is what procedure="helicopter" selects in
# effective_perceived_noise_level below.
```

## Measuring the flyover

What goes into the chain is a time history of **unweighted one-third-octave
band sound pressure levels**: 24 bands with nominal midband frequencies from
50 Hz to 10 kHz, one spectrum every 500 ms ± 5 ms (App. 2 §3.7.2). That cadence
is prescribed, and the 10 s reference duration of the metric below is fixed
independently of it, which is where the 13 dB constant of the uniform case comes
from. A reader arriving with A-weighted data, or with a 1 s cadence, gets a
plausible-looking number that is not an EPNL.

![A noise certification measurement station of ICAO Annex 16 Appendix 2. The site elevation shows the aircraft on its reference flight path with the slant path drawn down to a microphone, an obstruction-free cone of 80 degree half-angle about the vertical drawn as two dashed rays with one tree clear of it and a taller one breaking into it marked site rejected, an independent tracking sensor sighting the aircraft, and a 10 metre meteorological mast carrying an anemometer and a temperature and humidity sensor, noted as within 2000 m of the station. A larger-scale detail draws the microphone itself: the sensing element 1.2 m above local ground inside a windscreen whose insertion loss stays within plus or minus 1.5 dB and is corrected for, with the arriving ray. A plan inset shows the capsule axis perpendicular to the plane of the flight path so every ray arrives at 90 degrees, grazing, beside a box reading 24 one-third-octave bands from 50 Hz to 10 kHz, one sample every 500 ms plus or minus 5 ms. The footer gives the aeroplane test window: no precipitation, minus 10 to 35 degrees Celsius and 20 to 95 per cent relative humidity over the path above 10 m, 8 kHz attenuation not more than 12 dB per 100 m, wind not more than 6.2 metres per second average and 7.7 peak, crosswind not more than 3.6 and 5.1; the helicopter window of 5.1 and 2.6 metres per second; and at least six valid runs per measurement point with a 90 per cent confidence limit not exceeding plus or minus 1.5 EPNdB](https://docs.jmrp.io/phonometry/media/diagram_aircraft_noise_station.svg)

The station that produces those spectra is fixed clause by clause.

- **The microphone.** The sensing element sits 1.2 m above the local ground and
  is oriented for grazing incidence, with the element substantially in the plane
  defined by the predicted reference flight path and the station (§3.5.2). In
  practice the capsule axis points across the track, so sound arrives at 90° to
  it throughout the flyover. Its free-field sensitivity level must stay within
  ±1.0 dB of the calibration-check frequency from 50 Hz to 5 kHz and within
  ±2.0 dB at 6.3, 8 and 10 kHz (§3.5.3), and its directional response must meet
  Table A2-1 (§3.5.4). The windscreen's insertion loss must stay inside ±1.5 dB
  over the same range at grazing incidence (§3.4) and is corrected for.
- **The site.** Relatively flat terrain with no excessive sound absorption (no
  thick, matted or tall grass, shrubs or wooded areas), and no obstruction that
  significantly influences the sound field inside a cone of 80° half-angle
  about the vertical through the microphone (§2.2.1). The people taking the
  measurement count as obstructions.
- **The window.** For aeroplanes: no precipitation; −10 °C to 35 °C and 20 % to
  95 % relative humidity over the propagation path between 10 m and the
  aircraft; the 8 kHz attenuation coefficient no more than 12 dB/100 m over that
  path; average wind at 10 m no more than 6.2 m/s with a maximum of 7.7 m/s, and
  average crosswind no more than 3.6 m/s with a maximum of 5.1 m/s (§2.2.2.4.1).
  For helicopters the wind limits tighten to 5.1 m/s and 2.6 m/s and the
  temperature and humidity limits apply at 10 m only. Temperature, humidity,
  wind speed and direction are measured at 10 m above the ground within 2 000 m
  of the noise measurement point (§2.2.2.2).
- **The tracking.** Aircraft position is determined by a method independent of
  the cockpit flight instrumentation and time-synchronised with the acoustic
  record over at least the interval in which the noise is within 10 dB of
  $\mathrm{PNLTM}$ (§2.3.1-2.3.2), the window the metric integrates over.
- **The repetition.** At least six valid runs per measurement point, and the
  sample must be large enough that the 90 % confidence limit on the average does
  not exceed ±1.5 EPNdB (§5.4.2).

## EPNL

Over the flyover, $\mathrm{PNLT} = \mathrm{PNL} + C$ and its maximum is
$\mathrm{PNLTM}$. The metric converts the tone-corrected level back to energy,
sums it over the records between the two 10 dB-down points and divides by a
fixed reference duration $\mathrm{t}_0 = 10\ \text{s}$ (App. 2 §4.6),

$$
\mathrm{EPNL} = 10\lg\!\left[\frac{1}{\mathrm{t}_0}\sum_k 10^{\mathrm{PNLT}(k)/10}\,
\Delta t_k\right],
$$

so EPNL is the level of a hypothetical 10 s exposure carrying the same
noisiness-weighted energy as the real event. For the uniform half-second cadence
this collapses to $10\lg\sum 10^{\mathrm{PNLT}(k)/10} - 13$, the 13 dB being
$-10\lg(0.5/10)$. The fixed reference duration is what makes
$D = \mathrm{EPNL} - \mathrm{PNLTM}$ a *correction* rather than an integral: a
flyover that lingers near its peak for much longer than 10 s scores above
$\mathrm{PNLTM}$, and one that sweeps past scores below it. In the ETM Vol. I
Table 4-4 example the peak is $\mathrm{PNLTM} = 97.40$ TPNdB, the 10 dB-down
window runs from record 4 to record 28 of 31 and spans 10.2 s of increments, and
the result is $D = -4.78$ dB and $\mathrm{EPNL} = 92.62$ EPNdB, just below
$\mathrm{PNLTM}$ because the window is barely longer than the reference
duration.

The two 10 dB-down points are the records whose PNLT is nearest to
$\mathrm{PNLTM} - 10$, the first counted from the start and the last from the
end, so a dip below that threshold inside a history with more than one lobe does
not cut the window short: §4.5 asks for the longest duration such a history
allows.

That chain has an order to it, and the order is the part a finished time
history cannot show. The tone correction is measured on each spectrum as it
arrives, one record at a time, by fitting a background under the spectrum and
reading the excess of the tone over it; the 10 dB-down window, by contrast,
cannot be placed until the whole pass is over, because its two edges are
defined relative to a peak that is not known until the last record has been
taken. The clip runs a synthetic flyover in that order:

**Video:** [A certification flyover assembled record by record, from spectra to a single EPNL](https://docs.jmrp.io/phonometry/media/anim_epnl_flyover.webm): Forty-one one-third-octave spectra arrive at half-second intervals as an aircraft passes a microphone. On each spectrum a dashed background is fitted under the 2500 hertz fan tone and a double arrow measures the tone's excess over it, which at the peak reaches a correction of 3.97 decibels. Beside it two level histories are written point by point: the perceived noise level and, shaded above it, the tone-corrected level that exceeds it by that correction, rising to a peak of 120.57 PNdB as the aircraft reaches the microphone. Only once the pass has ended is the peak known; a dotted line then appears 10 decibels below it, the integration window fills between records 16 and 24, and the duration correction of -6.56 decibels and the final EPNL of 114.01 EPNdB appear last.

EPNL is not a maximum level, and it is not computed in the order it is written. The tone correction is a per-record measurement made while the aircraft is still passing, but the 10 dB-down window is retrospective: its edges are defined from a peak that only exists once the event is over, which is why a certification result cannot be read off a meter in real time.

![Aircraft-flyover perceived-noise-level time history: PNL and the tone-corrected PNLT versus time, with the maximum PNLTM marked and the 10 dB-down integration window shaded, annotated with the resulting EPNL and duration correction](https://docs.jmrp.io/phonometry/media/epnl.svg)

*Two corrections, both visible. The gap between PNL and PNLT is the tone
correction, worth 3.97 dB at the peak here because a 2500 Hz fan tone stands
above its neighbours, which is why $\mathrm{PNLTM}$ = 120.57 TPNdB sits above
the 116.61 PNdB the plain PNL reaches. The shaded 10 dB-down window spans
records 16 to 24, about 4 s, well short of the fixed 10 s reference, so the
duration correction is $D = -6.56$ dB and $\mathrm{EPNL}$ = 114.01 EPNdB lands
*below* the peak. A slower flyover over the same peak would score above it.*

```python
import numpy as np
from phonometry import aircraft

# spectra: a (K, 24) array of one-third-octave band levels sampled every dt s,
# here the synthetic flyover of the figure above
k, dt = 41, 0.5
idx = np.arange(k)
shape = 15.0 * np.exp(-((np.log10(aircraft.NOY_BANDS_HZ) - np.log10(400.0)) ** 2) / 0.5)
gain = 30.0 * np.exp(-((idx - 20.0) ** 2) / (2 * 5.0**2)) - 5.0
spectra = (55.0 + shape)[None, :] + gain[:, None]
spectra[:, 17] += 12.0 * np.exp(-((idx - 20.0) ** 2) / (2 * 6.0**2))  # 2500 Hz fan tone

res = aircraft.effective_perceived_noise_level(spectra, dt=0.5)
print(res.epnl, res.pnltm, res.duration_correction, res.band_limits)
res.plot()   # PNL/PNLT time history (needs matplotlib)
```

`effective_perceived_noise_level` returns an `EPNLResult` bundling the per-record
`pnl`, `tone_correction`, `pnlt`, the peak `pnltm`, the `bandsharing_adjustment`,
the `duration_correction`, the `epnl` and the 10 dB-down `band_limits`. Its
`procedure=` argument starts the analysis at 80 Hz for an aeroplane (the
default) and at 50 Hz for a helicopter, and `dt` may be a per-record array
rather than a scalar, which is how the variable increments of the integrated
method enter. The reference-condition integrated-method example of ETM Vol. I
Table 4-4 (a different, real certification time history, with increments
between 0.395 s and 0.431 s) is reproduced as
$\mathrm{EPNL} = 92.6\ \text{EPNdB}$.

When the PNLT history is already in hand, `epnl_from_pnlt` runs the duration
and window machinery on its own: it takes the PNLT series and its `dt` (and,
optionally, the per-record tone corrections, which switch on the bandsharing
adjustment) and returns an `EPNLIntegration` record that names `epnl`, `pnltm`,
`bandsharing_adjustment`, `band_limits` and `duration_correction`.

**Show the code for this figure**

```python
aircraft.effective_perceived_noise_level(spectra, dt).plot()
```

### ICAO EPNL report (`.report()`)

`EPNLResult.report(path)` renders a one-page PDF fiche laid out like an
aircraft-noise-certification data sheet: a standard-basis line (ICAO Annex 16
Vol. I Appendix 2), an optional TCDSN-style metadata header (aircraft,
manufacturer / type-certificate holder, applicant, measurement point), a metrics
table of the informational intermediate quantities (the peak $\mathrm{PNLTM}$,
the duration correction $D$, the 10 dB-down record window and, when non-zero,
the bandsharing adjustment) above the full-width landscape
$\mathrm{PNLT}$-versus-time plot (the result's own `.plot()`), the boxed
$\mathrm{EPNL} = X\ \text{EPNdB}$ single number, a
Level | Limit | Margin verdict row when a certification limit is supplied, a
static reference-conditions strip (25 °C, 70 % RH, sea level, zero wind, ISA) and
a footer with the fixed disclaimer. It uses the same `ReportMetadata` container
and rendering engine as the [ISO 717 insulation
fiche](https://docs.jmrp.io/phonometry/buildings/insulation/insulation-ratings/#iso-717-report-report); a supplied
`requirement` is read as the certification EPNL limit in EPNdB (the EPNL passes
at or below it), and `metadata=None` produces a lightweight prediction fiche with
no verdict row. Rendering needs reportlab and, for the figure the fiche embeds,
matplotlib (`pip install "phonometry[report,plot]"`); only `engine="reportlab"`
is supported. The fiche renders in English by default; pass `language="es"` for
a Spanish fiche (translated fixed strings and a comma decimal separator), e.g.
`res.report("epnl_fiche_es.pdf", language="es")`. The fiche is a computational
EPNL result and is not an official State noise certificate; it does not
reproduce any TCDSN.

```python
from phonometry import ReportMetadata, aircraft

# spectra: a (K, 24) array of one-third-octave band levels sampled every dt s
res = aircraft.effective_perceived_noise_level(spectra, dt=0.5)
res.report(
    "epnl_fiche.pdf",
    metadata=ReportMetadata(
        specimen="Example twin-turbofan transport",
        manufacturer="Example Aircraft Company",
        measurement_standard="ICAO Annex 16 Vol I Amendment 14 Chapter 4",
        laboratory="Phonometry Reference Laboratory",
        requirement=101.0,          # certification EPNL limit (EPNdB)
    ),
)                                   # EPNL (EPNdB) with PNLTM and D
```

The example fiche is regenerated with `make reports` and kept rendered in the
repository; click the preview to open the PDF. It is rendered from its own,
quieter synthetic flyover (EPNL 98.3 EPNdB), not from the spectra of the figure
above: run those through the same call and the fiche reports 114.0 EPNdB and a
FAIL against the same 101 EPNdB limit, which is the verdict row doing its job.

**Report:** [ICAO Annex 16 EPNL example report (PDF)](https://docs.jmrp.io/phonometry/media/icao_epnl_example.pdf): One-page aircraft-noise-certification fiche: a metadata header, a reference-conditions strip, a metrics table with the peak PNLTM and the duration correction D and the 10 dB-down record window, the full-width PNL/PNLT time-history plot with the marked PNLTM and the shaded integration window, the boxed EPNL = 98.3 EPNdB single-number result and a PASS verdict against a 101 EPNdB certification limit.

ICAO Annex 16 EPNL certification fiche (EPNLResult.report), EPNL in EPNdB with PNLTM and the duration correction D.

## From a measured flyover to a certification level

What the section above computed is the EPNL of the flight *as flown, on the day
it was flown*. A certification level is the same quantity transported to a fixed
reference day (sea-level pressure of 1 013.25 hPa, 25 °C at sea level with the
ICAO Standard Atmosphere lapse rate, 70 % relative humidity, zero wind, per
Chapter 3 §3.6.1.5) and to a fixed reference flight path. Appendix 2 §8 is
that transport, and it is the missing half of the chain.

The transport is written in two sound propagation paths (§8.1.3.1,
Figure A2-10). $\mathrm{K}$ is the noise measurement point and $\mathrm{Q}$
the position of the aircraft on the measured flight path when it emitted the
sound that $\mathrm{K}$ observed as $\mathrm{PNLTM}$; the line
$\mathrm{QK}$ makes the sound emission angle $\theta$ with the direction of
flight. $\mathrm{K}_r$ is the reference measurement point and
$\mathrm{Q}_r$ the position on the reference flight path from which
$\mathrm{Q}_r\mathrm{K}_r$ makes the same angle $\theta$. $\mathrm{QK}$
and $\mathrm{Q}_r\mathrm{K}_r$ are the lengths of the two paths, in
metres: the first is travelled through the air of the test day, with the
absorption $\alpha(i)$, and the second through that of the reference day, with
$\alpha_\mathrm{R}(i)$.

![Two elevations on one scale and, below them, the four terms the two lengths enter. (a) Measured flight path, test day: the path from X to Y climbing over a microphone K on the ground, the aircraft at Q, past the microphone, and the slant line QK down to it, labelled with the test-day absorption alpha(i); the angle theta is marked at Q between the direction of flight and QK. Q is where the aircraft was when it emitted the sound heard as PNLTM at K. (b) Reference flight path, reference day: a lower path from Xr to Yr over the reference point Kr, with Qr the point that sees Kr under the same angle theta, and a shorter slant line QrKr labelled with the reference-day absorption alphaR(i). The terms: 0.01 \[alpha(i) minus alphaR(i)\] QK, the two days' absorption over the measured sound path; 0.01 alphaR(i) (QK minus QrKr), the reference absorption over the change of length; 20 lg (QK/QrKr), spherical spreading over the change of length, all three from 8.3.2.1; and minus 7.5 lg (QK/QrKr) plus 10 lg (VG/VGR), the duration term Delta 2 of 8.3.4.2](https://docs.jmrp.io/phonometry/media/diagram_flight_path_adjustment.svg)

*The measured and the reference sound propagation paths of Figure A2-10, in
elevation for a microphone beneath the flight path, and the terms the two
lengths enter. The print draws them in perspective, the measured ground track
apart from the extended runway centre line, and §8.1.3.3 has $\theta$
established in three dimensions; a microphone to the side of the track is
Figure A2-11, which adds the elevation angle $\psi$ of $\mathrm{QK}$ above
the ground.*

In the simplified method (§8.3) each band of the $\mathrm{PNLTM}$ spectrum is
moved to the reference day by

$$
\mathrm{SPL}_\mathrm{R}(i) = \mathrm{SPL}(i)
 + 0.01\,[\alpha(i) - \alpha_\mathrm{R}(i)]\,\mathrm{QK}
 + 0.01\,\alpha_\mathrm{R}(i)\,(\mathrm{QK} - \mathrm{Q}_r\mathrm{K}_r)
 + 20\lg\frac{\mathrm{QK}}{\mathrm{Q}_r\mathrm{K}_r},
$$

which swaps the test-day absorption coefficient for the reference one over the
measured path, re-absorbs over the corrected path length, and applies an
inverse-square term for the change in path length (§8.3.2.1). The adjusted
spectrum is re-run through the noy law and the tone correction, the bandsharing
adjustment of the test day is added back, and the difference
$\Delta_1 = \mathrm{PNLTM}_\mathrm{R} - \mathrm{PNLTM}$ is one of four terms added to the
measured EPNL:

$$
\mathrm{EPNL}_\mathrm{R} = \mathrm{EPNL} + \Delta_1 + \Delta_{\text{peak}}
 + \Delta_2 + \Delta_3
$$

(§8.3.6). $\Delta_{\text{peak}}$ handles a secondary peak within 2 dB of
$\mathrm{PNLTM}$ that overtakes the adjusted maximum (§8.3.3);
$\Delta_2 = -7.5\lg(\mathrm{QK}/\mathrm{Q}_r\mathrm{K}_r) + 10\lg(\mathrm{V}_\mathrm{G}/\mathrm{V}_\mathrm{GR})$ is the change in duration
from the path length and the ground speed (§8.3.4); and $\Delta_3$ corrects for
the difference between the propulsion operating parameters flown and those
specified for the reference condition, from manufacturer sensitivity curves
(§8.3.5). §8.2.3 fixes when the simplified method is not enough and the
integrated method must be used instead: a flyover disagreement above 8 EPNdB, an
approach disagreement above 4 EPNdB, or a result within 1 EPNdB of the Chapter 3
limit.

The library supplies the ingredients and not the bookkeeping. `air_attenuation`
and `band_attenuation_arp5534` below give the $\alpha(i)$ and $\alpha_\mathrm{R}(i)$
of the first two terms, though not in their unit: §8.3.2.1 takes the
coefficients in dB/100 m with $\mathrm{QK}$ in metres, which is what the 0.01 converts,
while `air_attenuation` returns dB/m (the $0.01\,\alpha(i)$ of those terms) and
`band_attenuation_arp5534` a band attenuation over a given path, in dB (the whole
$0.01\,\alpha(i)\,\mathrm{QK}$ for the path $\mathrm{QK}$). The EPNL machinery re-runs an
adjusted spectrum as readily as a measured one; the flight-path geometry
$\mathrm{QK}/\mathrm{Q}_r\mathrm{K}_r$ and the four $\Delta$ terms themselves are not
implemented. Adjust first, then report: the
EPNL this page returns is a test-day value, and a `requirement` verdict is only
a certification verdict once §8 has been applied.

## Measurement-system verification (IEC 61265)

`verify_aircraft_noise_system` checks measured performance against the
IEC 61265:1995 tolerances: the microphone directional-response limits (Table 1,
identical to Annex 16 Table A2-1) and the scalar frequency-response, linearity
and resolution limits.

The arguments are measured data, not settings. The outer key of `directional`
is a one-third-octave midband frequency in hertz; the inner key is the sound
incidence angle in degrees from the microphone's principal axis; and the value
is the measured **free-field sensitivity-level difference in decibels relative
to 0°, normal incidence**, for sinusoidal waves in a free field. At 4 kHz the
limits are 0.5 / 1.0 / 2.0 / 2.5 / 2.5 dB at 30 / 60 / 90 / 120 / 150°, so the
example below passes with margin. An angle between two tabulated ones is held
to the limit of the greater (§4.4.2). That check matters because the aeroplane
sweeps through a wide range of incidence angles while the microphone stays
fixed, and the whole station is built around sound arriving within a few degrees
of grazing. `frequency_response` values are deviations in decibels from the
level at the calibration-check frequency, in the reference direction.

The data comes from an anechoic free-field turn of the microphone with its
windscreen fitted, or from an actuator response plus the supplier's free-field
corrections; the electrical half comes from inserting a low-distortion sinusoid
within 5 dB of the calibration level and stepping it in 5 dB intervals, refined
to 1 dB around the ends of the linear range and the overload point. Run and
report the periodic test in a controlled environment (19-25 °C, 100-102 kPa,
40-70 % relative humidity) with a calibrator traceable to national standards.
Read `report.passes` as a verdict on data you supplied: a failed check
invalidates the certification data, it is not a correction to apply. Band-filter
conformance is a separate matter: Annex 16 §3.7.3 requires IEC 61260 class 1,
with class 2 admissible only if the certificating authority allows it, and the
library's own filter verification, `verify_filter_class` of
[Filter class verification](https://docs.jmrp.io/phonometry/signals/filters/filter-compliance/), is the
place that is checked.

```python
from phonometry import aircraft

report = aircraft.verify_aircraft_noise_system(
    directional={4000.0: {30: 0.4, 60: 0.9, 90: 1.9, 120: 2.4, 150: 2.4}},
    frequency_response={1000.0: 1.2},
)
print(report.passes, report.checks)
```

## Atmospheric absorption (SAE ARP 5534)

Correcting a measured flyover to reference atmospheric conditions needs the
one-third-octave-band attenuation over the path. The pure-tone coefficient is
the one of ISO 9613-1, whose terms ARP 5534 §3.1 repeats in its Eqs. 1-3, with
the saturation vapour pressure that ARP 5534 prints in Eqs. 5-6 (the longer
form of ANSI S1.26, not the Annex B formula of `air_attenuation`); on top of it
`band_attenuation_arp5534` adds the **SAE Method** (ARP 5534 §3.2.2) mapping the
pure-tone mid-band path attenuation $\delta_\mathrm{t} = \alpha\,s$ to the band
attenuation $\delta_\mathrm{B}$, consistent with the Exact Method well beyond the 50 dB
Approximate-Method limit.

![Aircraft atmospheric absorption versus frequency for two path lengths; the SAE-Method band attenuation stays below the pure-tone mid-band value at high absorption](https://docs.jmrp.io/phonometry/media/aircraft_atmospheric_absorption.svg)

*Solid is the band, dashed the pure tone at the band centre, and the two only
coincide where the absorption is small. At 1 km the gap is 0.18 dB in the 4 kHz
band and 10.78 dB at 10 kHz; at 7620 m it is 29.95 dB at 4 kHz and 167.97 dB at
10 kHz, against a mid-band value of 753.92 dB. Both are the same 25 °C, 70 % RH
atmosphere: the difference is that a band keeps the energy of its low edge long
after its centre frequency has gone.*

A third-octave band is not a tone. The absorption coefficient rises steeply with
frequency, so the low edge of a band is attenuated far less than its midband
frequency, and as the path lengthens the surviving band energy is carried almost
entirely by that lower half. The band attenuation therefore falls progressively
behind the pure-tone value, which is what the two curves show: at 1 km they are
indistinguishable below a few kilohertz and separate only at the top of the
range, while at 7620 m the gap at high frequency is large. This is exactly the
regime the older 50 dB-limited Approximate Method of ARP 866B could not
follow. The
consequence for §8 is direct: correcting a flyover to reference conditions with
the pure-tone coefficient overstates the high-frequency correction, and with it
the reference-day EPNL.

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np
from phonometry import aircraft

freqs = 1000.0 * 10.0 ** (np.arange(-13, 11) / 10.0)   # 50 Hz-10 kHz thirds
fig, ax = plt.subplots()
# solid: SAE band attenuation, dashed: pure-tone mid-band
for s in (1000.0, 7620.0):
    att = aircraft.band_attenuation_arp5534(
        freqs, s, temperature_c=25.0, relative_humidity_percent=70.0
    )
    line, = ax.semilogx(att.frequencies, att.band_attenuation, marker="o",
                        markersize=3, label=f"SAE band ({s:.0f} m)")
    ax.semilogx(att.frequencies, att.midband_attenuation, "--", alpha=0.6,
                color=line.get_color())
ax.set(xlabel="Frequency [Hz]", ylabel="Attenuation [dB]",
       title="Aircraft atmospheric absorption at 25 °C, 70% RH")
ax.grid(True, which="both", alpha=0.3)
ax.legend()
plt.show()
```

```python
import numpy as np
from phonometry import aircraft

freqs = 1000.0 * 10.0 ** (np.arange(-13, 11) / 10.0)   # 50 Hz–10 kHz thirds
att = aircraft.band_attenuation_arp5534(freqs, path_length_m=7620.0,
                              temperature_c=25.0, relative_humidity_percent=70.0)
att.plot()   # band vs pure-tone mid-band (needs matplotlib)
```

The result is an `AircraftBandAttenuation` holding the band attenuation
$\delta_\mathrm{B}$ as `band_attenuation`, the pure-tone mid-band value
$\delta_\mathrm{t} = \alpha\,s$ as `midband_attenuation` and the pure-tone
coefficient $\alpha$ itself, in dB/m, as `coefficient_db_per_m`, one value per
band each.

The regression was fitted for roughly 6-32 °C and 20-95 % relative humidity
(the 14 CFR Part 36 test window) and for path lengths up to 7620 m
(25 000 ft). The function's defaults (25 °C and 70 % RH) are the ARP 5534
reference point and the Annex 16 reference day. The mapping is reciprocal, so
the same call serves both directions of the reference-condition adjustment:
removing the test-day absorption from a measured record, and re-applying the
reference-day absorption to it.

The certification chain ends here. Turning these aeroplanes into noise
around an airport, the noise-power-distance tables, the per-segment
corrections of a flight path and the ground contour of a single event, is
the ECAC Doc 29 method of
[Airport noise](https://docs.jmrp.io/phonometry/aircraft/airport-noise/).

## What this guide covers

- **Covered.** ICAO Annex 16, Vol. I, Appendix 2: perceived noisiness and PNL from Table A2-3
  (`perceived_noisiness`, `perceived_noise_level`), the slope- method tone
  correction (`tone_correction`, matching the ETM Vol. I Table 3-7 turbofan
  example) and the full EPNL chain (`effective_perceived_noise_level`, matching
  the ETM Vol. I Table 4-4 example). IEC 61265:1995 measurement-system
  tolerances via `verify_aircraft_noise_system`. SAE ARP 5534 band atmospheric
  absorption via `band_attenuation_arp5534`.

- **Not covered.** The Appendix 2 §8 adjustment of a measured EPNL to reference conditions is not
  implemented, in either the simplified or the integrated form: the returned
  EPNL is the as-measured value, so adjust first and then report against a
  limit. `verify_aircraft_noise_system` checks the IEC 61265:1995 tolerances,
  not the superseding 2018 edition. The ECAC Doc 29 airport-noise chain has
  moved to [Airport noise](https://docs.jmrp.io/phonometry/aircraft/airport-noise/).

## See also

- [Airport noise (ECAC Doc 29)](https://docs.jmrp.io/phonometry/aircraft/airport-noise/): the NPD
  engine and the single-event contour chain around an airport.
- [Rotorcraft noise: the hemisphere method](https://docs.jmrp.io/phonometry/aircraft/rotorcraft-noise/):
  the ECAC Doc 32 method, whose single-event EPNL is this same Appendix 2 chain
  driven by a noise hemisphere instead of a certification flyover.
- [The ANP fleet database](https://docs.jmrp.io/phonometry/aircraft/anp-fleet/): the measured
  noise-power-distance tables for real types, which is where the certification
  data ends up for modelling work.
- API reference: [`aircraft.certification`](https://docs.jmrp.io/phonometry/reference/api/aeroacoustics/certification/) and [`aircraft.atmospheric_absorption`](https://docs.jmrp.io/phonometry/reference/api/aeroacoustics/atmospheric-absorption/).

## Quick answers

### Why can EPNL come out below the peak tone-corrected level?

EPNL is not the peak: `aircraft.effective_perceived_noise_level` sums the
tone-corrected energy of the records around it and divides that energy by a
fixed reference duration, `reference_time`, whose default is the one Appendix 2
sets and the [EPNL](#epnl) section gives. What decides the sign is the
tone-corrected energy of the whole 10 dB-down window against PNLTM held for
`reference_time`, not how long the peak lasts: a window whose energy falls short
of that scores below its PNLTM, and one whose energy exceeds it, a brief peak
over a long shoulder included, scores above it.
`res.duration_correction` is that difference, `res.epnl` minus the peak
`res.pnltm`, and `res.band_limits` names the first and last records the sum
took in.

## References

- **International Civil Aviation Organization. (2017).** [*Annex 16 to the Convention on International Civil Aviation: Environmental protection — Volume I: Aircraft noise*](https://store.icao.int/en/annex-16-environmental-protection-volume-i-aircraft-noise) (ICAO Annex 16, Vol. I, 8th ed.).
  The normative Appendix 2 EPNL procedure implemented by the noisiness, tone correction and EPNL sections.
- **International Civil Aviation Organization. (2018).** [*Environmental technical manual — Volume I: Procedures for the noise certification of aircraft*](https://store.icao.int/en/environmental-technical-manual-volume-1-procedures-for-the-noise-certification-of-aircraft-doc-9501-1) (Doc 9501, 3rd ed.).
  The worked examples (Table 3-7 tone correction, Table 4-4 integrated-method EPNL) used as the numeric oracles.
- **International Electrotechnical Commission. (1995).** [*Electroacoustics — Instruments for measurement of aircraft noise — Performance requirements for systems to measure one-third-octave-band sound pressure levels in noise certification of transport-category aeroplanes*](https://webstore.iec.ch/en/publication/5076) (IEC 61265:1995).
  The measurement-system tolerances checked by the verifier. Since revised as IEC 61265:2018 (https://webstore.iec.ch/en/publication/32635); the 1995 edition is the implemented one.
- **SAE International. (2012).** [*Standard values of atmospheric absorption as a function of temperature and humidity*](https://www.sae.org/standards/arp866b-standard-values-atmospheric-absorption-a-function-temperature-humidity/) (SAE ARP 866B, stabilized 2012).
  The predecessor SAE atmospheric-absorption practice, source of the 50 dB-limited Approximate Method the SAE Method is contrasted with.
- **SAE International. (2013).** [*Application of pure-tone atmospheric absorption losses to one-third octave-band data*](https://www.sae.org/standards/arp5534-application-pure-tone-atmospheric-absorption-losses-one-third-octave-band-data/) (SAE ARP 5534, reaffirmed 2021).
  The SAE-Method band attenuation of the atmospheric-absorption section, with the pure-tone coefficient of Eqs. 1-6: the ISO 9613-1 terms with the saturation vapour pressure of Eqs. 5-6.
