What Octave Bands Tell You That an Overall Level Cannot

Why this matters

Every acoustic control you can buy works over a frequency range and does nothing outside it. Absorption needs thickness measured against wavelength. Mass works better as frequency rises. A gasket fixes a leak that only matters above a few hundred hertz. So the question a control purchase actually asks is "where does the energy sit", and an overall level answers a different question entirely. Two rooms can report the identical overall A-weighted level and need opposite products, and a shop specifying from the overall number is guessing between them.

What a band is

An octave band spans a doubling of frequency. The standard nominal centers are 31.5, 63, 125, 250, 500, 1000, 2000, 4000 and 8000 Hz, and each band runs from its center divided by the square root of two up to its center multiplied by the square root of two. Third-octave bands cut each of those into three.

Band levels are still sound pressure levels, dB re 20 uPa, and they still carry a weighting and a time basis. Collect them unweighted (Z) whenever you can, because unweighted band data can be A-weighted or C-weighted afterwards, and A-weighted data can never be un-weighted. That is a one-way door and it is the most common wasted site visit in this work.

Bands sum back to the overall by the same logarithmic rule any two levels follow, so the band set and the overall level are not two measurements, they are one measurement at two resolutions.

The weighting tables you will actually use

To convert an unweighted octave-band level to its A-weighted or C-weighted contribution, add the value for that band.

Read the 31.5 Hz row as the whole reason this article exists. A-weighting discards nearly 40 dB there. A rumble that dominates the room can arrive at the report as a rounding error.

The gate: which bands your control has to reach

Here is the rule the rest of this card runs against. A-weight every band, find the highest A-weighted band, and list every band within 5 dB of it. Those bands, and only those, are what a control must reach to change the overall A-weighted level. Anything 10 dB or more below the peak band is contributing under half a decibel and is not worth buying a product for.

Alongside it, one screening number worth taking on every complaint: the overall C-weighted level minus the overall A-weighted level, at the same position and the same time basis. Under about 5 dB and the energy is mid and high frequency. Above about 15 dB the problem is low-frequency dominated, and the controls that work on it are a different family.

Getting band data without an acoustics budget

Band data is cheaper to collect than most shops assume, and there are three real decisions in it.

Instrument class. A class 1 sound level meter with an octave-band filter set is what a contested measurement needs, because a class 2 instrument carries wider tolerances at the ends of the range and those ends are exactly where a low-frequency argument lives. For your own diagnostic work, a class 2 meter with real filters is usually enough to choose a control family. A phone application with the handset's own microphone is not an instrument at either class: the microphone rolls off at low frequency, which is precisely the information you came for, so a phone reading that says "no low-frequency problem" has told you nothing. Use it to record a clip for triage and nothing else.

Integration time. Report Leq over a stated period rather than a spot reading, and make the period long enough to contain a whole cycle of whatever the machine does. On equipment that stages, that means covering a full stage-up and stage-down, or the number is a level of one operating state described as if it were the plant.

Position. Take band data at the position the complaint is made from, at seated or standing ear height, and record the position well enough to return to it. Then take a second set close to the suspected source, wearing hearing protection for that set if the plant area runs anywhere near the 85 dB(A) eight-hour action level that 29 CFR 1910.95(c) sets. The difference between the two sets, band by band, is the first real evidence you have about the path, and it costs one extra minute.

Worked example: same overall level, opposite purchases

Two complaints, both LAeq,10min at the complainant's position, dB re 20 uPa, measured with a class 1 meter calibrated before and after. Neither reading required background correction: in both cases the residual with the plant off was more than 10 dB below the total. The 8 kHz band was more than 20 dB below the peak band in both spectra and contributes under 0.1 dB, so it is dropped after the check.

Case 1, office under a rooftop unit. Unweighted octave-band levels, and the A-weighted contribution of each.

Summed: 53.6 dB(A). Summed with C-weighting instead: 73.1 dB(C). C minus A is 19.5 dB, which is over the 15 dB screen, so this is low-frequency dominated before you look at a single band.

Run the gate. Peak A-weighted band is 125 Hz at 49.9. Within 5 dB: 250 Hz at 46.4 and 63 Hz at 45.8. The 500 Hz band at 43.8 is 6.1 dB down and falls outside. The control must work at 63, 125 and 250 Hz.

Case 2, tenant space beside a fan room. Same instrument, same basis.

Summed: 53.6 dB(A), the same number to a tenth. Summed with C-weighting: 54.9 dB(C). C minus A is 1.3 dB, under the 5 dB screen.

Run the same gate. Peak A-weighted band is 1000 Hz at 49.5. Within 5 dB: 2000 Hz at 48.7, 4000 Hz at 44.5 (5.0 down, in), and 500 Hz at 44.8 (4.7 down, in). The control must work from 500 Hz to 4 kHz.

What the two answers buy

Case 2 is the easy one, and the products that fix it are the ones everybody already stocks. Porous absorption of a couple of inches is doing real work by 500 Hz and is fully effective above it. A sealed, modest-mass barrier performs well across that range. Duct liner, a lined elbow, a gasketed door and a sealed penetration all act squarely where the energy is. Expect the overall to move roughly as much as the control's own rated insertion loss over 500 Hz to 4 kHz, because there is no low-frequency tail waiting underneath to floor the result.

Case 1 buys none of that. Absorption is a thickness problem measured against wavelength, and at 63 Hz the wavelength in air is on the order of 18 ft, so a panel that is doing real absorbing there is measured in feet, not inches. Hanging that absorption in the office also treats the wrong quantity: it reduces the reverberant build-up in the receiving room, not the energy arriving through the structure. The controls with a mechanism at 63 to 250 Hz are source-side (blade-pass frequency, fan speed, balance), path-side (structural decoupling, isolator selection, breaking a rigid tie) and mass at the boundary, and each of those has its own failure conditions. Why the whole low-frequency family behaves this way is a separate subject and a sibling card owns it.

The failure mode here is specific and it is expensive: a shop reads 53.6 dB(A), specifies the case-2 package because that is the package it knows, installs it in the case-1 building, and measures 53.4 dB(A) afterwards. Nothing was defective. The product's own insertion loss below 250 Hz was never claimed by its maker, and nobody looked.

Sibling-rule check. Every level above carries its quantity, weighting, reference, bandwidth and time basis. Band sums use the logarithmic rule stated in the summation card rather than an arithmetic average. The low-frequency conclusion routes to the low-frequency card instead of re-deriving it. No corrected figure is compared against an uncorrected one, because neither case needed a background correction and that is stated rather than assumed.

Octave or third-octave

Octave bands are enough to choose a control family. Move to third-octaves when you suspect a tone, because a tone that stands 10 dB above its neighbours in third-octave resolution can be almost invisible in octaves: the octave band containing it is three third-octaves wide, so the tone's energy gets reported alongside two bands of ordinary broadband noise and the peak flattens. If the complaint language is "whine", "hum" or "drone" rather than "rumble" or "hiss", collect third-octaves on the first visit. What a tone does to the judgement of the noise is its own subject and a sibling card owns it.

How to verify you got this right

Sum your own A-weighted band levels and check the result against the overall A-weighted level the meter reported at the same time. Agreement inside about 1 dB confirms the band set is complete and nothing was clipped or filtered out. A band set that sums BELOW the reported overall means energy outside the bands you captured, usually below 31.5 Hz.

Then check the direction of your own conclusion. If your report recommends a control and the control's published data sheet gives no performance figure in your peak band, the recommendation is unsupported no matter how strong the product is elsewhere. Ask the manufacturer for band data, and if the answer is a single overall number, you have been handed the same problem this card started with.

References

• ANSI/ASA S1.11 and IEC 61260, octave-band and fractional-octave-band filter specifications, in the edition your specifier names; band center frequencies and edges come from these
• ANSI/ASA S1.42 and IEC 61672, frequency weighting characteristics, which are the source of the A and C values tabulated above
• 29 CFR 1910.95, occupational noise exposure, for the hearing-conservation action level that governs standing beside running plant to collect a source-side band set
• Manufacturer octave-band insertion-loss or absorption data for any control you specify, with the test method and mounting stated on the report
• See related: Why a Single Number Hides the Fault; Why Low-Frequency Noise Is the Hard Problem; What Tonal Noise Is and Why It Is Judged Differently