What a Weighting Network Is Doing to Your Reading

Why this matters

An office tenant reports a rumble. The tech measures 46.4 dB A-weighted, which is quieter than a lot of libraries, and reports that there is nothing wrong. The tenant is not imagining it: in the 63 Hz octave band, unweighted, that same signal is 68 dB, and the A-weighting network threw 26 dB of it away before the meter ever produced a number. Nobody made an arithmetic error. A weighting network is a filter that runs ahead of the level calculation, and choosing the wrong one does not give you a slightly wrong answer, it gives you a correct answer to a question you did not ask.

A filter, applied before the number exists

Weighting is not a correction applied to a finished level. It is a frequency-dependent gain applied to the signal, band by band, and the overall figure is then computed from what survives the filter. That ordering is the whole subject: once the bands have been weighted and summed into one number, the individual bands are gone and no arithmetic recovers them.

So an A-weighted overall level is not "the level, adjusted". It is the level of a different signal, one from which most low-frequency energy has been deliberately removed.

The three networks and what each is for

A-weighting rolls off low frequencies hard, is essentially flat through the speech range, and adds a slight boost around 2 kHz to 4 kHz. It was derived from equal-loudness contours at low sound levels, so it approximates how a person weights a quiet sound. That is not why occupational work uses it, which is worth being clear about: 29 CFR 1910.95 is written in A-weighted sound pressure level because A-weighting correlates usefully with the risk of noise-induced hearing loss, not because it represents loudness at 100 dB. Applying a low-level loudness curve to a loud signal is a known approximation, accepted for that purpose.

C-weighting is nearly flat from about 63 Hz to 2 kHz and rolls off gently at both ends. It was derived from a high-level equal-loudness contour. Its field value is that it keeps the low-frequency content A discards, which makes it the reference against which A's discarding becomes visible.

Z-weighting, formerly called linear or unweighted, is flat within the instrument's stated frequency range as defined by the sound level meter standard. It is what you want when you intend to do band arithmetic yourself, or when a test method calls for unweighted band levels.

Say which one produced any number you write down. A figure of 46 dB re 20 micropascals is not a measurement. A figure of 46 dB A-weighted, overall, 1-hour equivalent-continuous, re 20 micropascals, at a stated position, is.

The weighting values, and the condition they were derived under

These are the nominal frequency weightings at octave band centre frequencies, as defined in the sound level meter standard IEC 61672-1 and its adopted national equivalents, which bind you only through the test method, purchase specification or ordinance that names them and in the edition that document names.

The condition these hold under: they are nominal values at exact band centre frequencies, for hand calculation from band levels. A real instrument's response has a tolerance around them set by its class, and that tolerance widens toward the frequency extremes, which is exactly where A-weighting is doing the most work. That is the meter class card's territory, and it means a hand-computed A-weighted total from band levels and a meter's own A-weighted reading of the same signal will not agree exactly.

Worked example: one spectrum, three overall numbers

Same tenant, same corner office, same hour. Octave band levels measured unweighted, 1-hour equivalent-continuous, re 20 micropascals, microphone at 1.5 m at the complaint position.

Combining each column by energy, using the addition method the sibling card owns:

• Unweighted overall: 68.4 dB Leq,1h re 20 micropascals
• C-weighted overall: 67.7 dB LCeq,1h re 20 micropascals
• A-weighted overall: 46.4 dB LAeq,1h re 20 micropascals

Three legitimate numbers, one signal, a 22.0 dB spread between the largest and the smallest. Every one of them is correct and only one of them will be quoted in the complaint file.

Where the A-weighted total comes from. The 63 Hz band alone contributes 34.8% of the total A-weighted energy, and 63 Hz plus 125 Hz together contribute 70.4% of it. Even after A-weighting has cut 26.2 dB off the 63 Hz band and 16.1 dB off the 125 Hz band, those two bands still own most of the A-weighted answer. That is a useful surprise: A-weighting did not make this a mid-frequency problem, it made a low-frequency problem look small.

What each number is good for here. The 46.4 dB A-weighted figure is the one to compare against an A-weighted criterion, and only against an A-weighted criterion. The band levels are the only figures that can drive a treatment decision, because every acoustic product's performance is published per band. The 67.7 dB C-weighted figure is not a treatment input at all; its job is the comparison in the next section.

The failure mode, concretely. A tech who reports only 46.4 dB A-weighted has reported a number that sounds fine, closed the ticket, and left a tenant with a real 68 dB 63 Hz band. A second tech, sent back later, buys mid-frequency absorptive treatment because the overall A-weighted figure gave no reason to look lower. The treatment has almost no absorption at 63 Hz, the tenant hears no change, and the shop is now two visits and one product into a problem it has not yet characterised.

C minus A, as a one-line field diagnostic

Subtract the A-weighted overall from the C-weighted overall of the same signal, on the same time basis, at the same position. The gap is a direct read on how much of the energy sits low.

In this example: 67.7 minus 46.4 = 21.3 dB. A gap that large says the signal is overwhelmingly low-frequency, and it says it in two readings without a spectrum analyser. A signal dominated by mid and high frequencies gives a gap of only a few decibels, because C and A are close above 500 Hz.

Two qualifiers, attached rather than left downstream. First, this is a screen, not a spectrum: it tells you to go get band levels, it does not tell you which band. Second, both readings must be the same time basis over the same period, ideally logged simultaneously by an instrument that reports both, because a C reading taken ten minutes after an A reading on a varying source is comparing two different hours.

Why you cannot un-weight a number

If you know the spectrum, you can compute the A-weighted total from it, as the table above did. The reverse is not possible. An A-weighted overall is one number produced by summing seven or more weighted bands, and there are infinitely many spectra that produce the same total. Nothing in the number records which one you had.

This is why a specification, an ordinance or a submittal that gives only an A-weighted overall has withheld the information needed to select any treatment, and why the standing request is for octave band data plus the overall, not the overall alone.

The practical corollary for records: never store a level without its weighting, and never assume an unlabelled level is A-weighted because most levels are. A C-weighted 67.7 filed as an unlabelled 67.7 will be read next year as an A-weighted figure and will misrepresent that room by more than 20 dB.

Verifying you got this right

Recompute one band by hand. Take any band level from your instrument's unweighted output, apply the table value, and check it against the instrument's own A-weighted band level. They should agree closely; a large disagreement means one of the two outputs is not what you think it is.

Check the overall against the bands. Sum your weighted band levels by energy and compare with the meter's overall on the same weighting. Expect close agreement, not identity, because the instrument's real filter response carries its class tolerance rather than the nominal table values. A gap of several decibels is not tolerance, it is a basis mismatch somewhere, most often a different time period.

Confirm the weighting is on every stored figure, including the ones in the photograph of the display, the ones in the email, and the ones in the quotation. This is the field where the loss happens, because it is the one nobody types.

Where verifying sends you to the plant rather than the office: hearing protection in any space where you must raise your voice to be heard at arm's length, and it does not change what the microphone on the stand reads. Place the stand outside the plane of any rotating component with guards in place per 29 CFR 1910.212(a)(1), never reaching past a guard. If reaching the source means going onto a roof, name the Part before you go, because the fall protection trigger is 4 feet under 29 CFR 1910.28(b)(1) in general industry and 6 feet under 29 CFR 1926.501 in construction.

References

• 29 CFR 1910.95, Occupational noise exposure, which is written in A-weighted sound pressure level
• IEC 61672-1 and its adopted national equivalents for frequency weightings and meter specifications, binding through the test method, specification or ordinance that names them, in the edition named there
• 29 CFR 1910.212(a)(1) for machine guarding, and 29 CFR 1910.28(b)(1) or 29 CFR 1926.501 for fall protection depending on the Part covering the work
• See related: Why Decibels Do Not Add the Way Numbers Do; What a Sound Level Meter Class Actually Buys You