Octave band analysis: what the frequencies tell you

This page explains what the rules and common practice call for. Every site is different, and what a survey or a filing needs at your facility is something we work out with you after we see it.

A single decibel number tells you how loud it is. It cannot tell you what to do about it.

That is the whole case for octave band analysis, which breaks the noise down by frequency, from low pitch to high.

A normal noise reading is given in decibels (dBA), which compresses every frequency into one number. That number is useful for compliance, because OSHA's permissible exposure limit (PEL) is written in it.

For planning a control, however, one number is no help. Two rooms can both read 95 dBA and need completely different controls, and the control that suits one will not work in the other.

• One room is mostly a low rumble from a fan, at 125 hertz (Hz), and it needs mass (heavy material) and isolation.
• The other room is mostly a high hiss from compressed air, at 4 kilohertz (kHz), and it needs a nozzle change.

Order barriers for the second room, and you will spend a capital budget on a problem a nozzle change would have solved. The overall dBA reading is the same in both rooms, so it gives you no way to tell them apart.

When you actually need it

Before you select any engineering control. Sound-absorbing material, barriers, enclosures and silencers all perform differently at different frequencies, and sometimes the gap is huge. A material that absorbs sound very well at 2 kHz may do almost nothing at 125 Hz. Selecting controls from a single dBA reading is guessing with a budget attached.

To select hearing protectors by the NIOSH long method. The Noise Reduction Rating (NRR) printed on the package is a single number, and it comes from laboratory conditions.

The long method works one octave band at a time, setting the protector's attenuation (how much noise it blocks) in each band against the noise measured in your plant. It is much more accurate, especially in low-frequency noise, where many protectors fall well short of their NRR. If your workers are in low-frequency noise and you selected protection by the NRR, you may be over-protected on paper and under-protected in fact.

When the noise has one clear tone. A single tone such as a whine, a hum or a squeal is more annoying and does more harm than broadband noise at the same level, and it can often be traced to one component. Octave band analysis finds it, or a finer narrowband analysis does where that is needed, while an overall dBA reading hides it completely.

When you are tracking down a source. In a plant with 6 machines running, the frequency spectrum is often what shows which one is really the problem. The frequencies produced by meshing gears, passing blades and bearings can be calculated, and matching them against the measured noise will often identify the culprit before you shut anything down.

When exposure over the PEL calls for controls. Where workers are exposed above Table G-16, the table of permissible noise exposures in OSHA's noise standard, the standard requires feasible engineering or administrative controls. Engineering controls are built into the job, and administrative controls change how the work is run. Whether a control is "feasible" is an engineering judgment, and you cannot make that judgment, or defend it later, without frequency data.

What we measure

A survey normally covers the standard octave bands from 31.5 Hz through 8 kHz, with third-octave bands where locating a source needs finer detail.

Results should be reported both unweighted and A-weighted. A-weighting is the adjustment built into a normal decibel reading (the A in dBA), and seeing both shows you what A-weighting discards, which in low-frequency noise is a great deal.

A useful report gives you three things:

• The results by frequency band, called the spectrum.
• What those results mean for selecting controls.
• Where it applies, a long-method calculation for your hearing protectors, set against the noise measured in your plant and not against the NRR.

What the NRR is really worth in your plant

The NRR comes from a laboratory fit, and real-world protection is lower. There are three separate adjustments to know about before you rely on a protector, and the trade term for marking a rating down is derating.

The Appendix B adjustment. Appendix B of OSHA's noise standard covers noise measured in A-weighted decibels, which is how a noise dosimeter reports it. In that case, the standard requires you to subtract 7 decibels from the NRR before you use it. Set against an A-weighted reading, an NRR of 29 starts life as 22.

The real-world derating. OSHA guidance derates the NRR by a flat 50% for all types of protector. NIOSH derates by type, because how reliably each type fits differs sharply:

The spread is the point. A pre-molded plug and a muff with the same NRR do not give the same protection, so the type of protector your plant issues changes the answer as much as the number on the box does.

Plugs and muffs together. Wearing plugs and muffs together does not add the two ratings. The accepted approach is to take the higher of the two ratings and add 5 decibels.

Octave band data lets you skip most of this estimating for the protectors that matter. The long method works out the protection band by band against your measured noise, and it does not rely on adjusting a single label number.

Compliance data is not the same as noise control data

A noise survey that reports dBA values satisfies the standard, and that is what most clients ask for. Octave band analysis takes extra field time and produces data nobody asked for. That changes at the point where someone wants the noise reduced, because a dBA-only survey cannot guide a single decision about controls.

If your goal is a record that shows compliance, you may not need this. If your goal is a quieter plant, you cannot move ahead without it.