Audio Noise Measurement And Weighting Filters
Understanding A-Weighted, C-Weighted and ITU-R 468 Specifications
Oct 28, 2009
Audio reproduction systems usually strive for a flat frequency response, transmitting all audible sounds without favouring any parts of the frequency range over any others. However, the human ear detects some frequencies much easier than others.
Not surprisingly the ear is particularly sensitive to frequencies present in human speech, and it is less sensitive to very high frequencies and very low frequencies.
Noise Weighting - Simulating the Frequency Response of the Human Ear to Unwanted Sounds
The uneven frequency response of the ear causes a problem when trying to evaluate the annoyance of unwanted sounds, something that often has to be judged. Whether it is the noise in a transmission path, or trying to establish how annoying aircraft noise is compared to lawnmowers and all the other sonic tribulations of the urban environment, some steps have to be taken to allow for the greater sensitivity of the ear to noise at mid-frequencies between 3-6 kHz relative to to other frequencies.
To do this the noise is passed through a filter that approximates the frequency sensitivity of the ear to noise. That filter is the weighting filter. There is no universal agreement as to the response of this filter, but three different response curves are in common use: A, C and ITU-R 468 weighting filters.
A-Weighted Noise Measurements
A-weighting is the most commonly used weighting curve, with a broad peak between 1kHz and 6kHz but very strongly discriminating against low frequencies. The original rationale for this is said to be based on the 40-phon Fletcher-Munson equal loudness contours for pure tones, and this is widely used in sound level meters for assessing environmental noise.
A-weighting is more applicable to noise at low SPLs below 55dB (Ref 1), which makes its use questionable when applied to SPLs that would give rise to hearing damage. It is better suited for SPLs at which unwanted sound begins to become annoying rather than those which are already at high annoyance volumes.
C-Weighted Noise Measurement
The strong discrimination against low frequencies in A weighting means that low frequency annoyances like mains hum are not adequately represented at normal listening levels. C-weighting is better suited for SPLs above 85dB, which is a better match for music playing systems than A weighting.
ITU-R 468 (was CCIR 468) Noise Weighting Curve
This weighting curve was developed in the 1960s and 1970s where analogue systems that relied of high-frequency pre-emphasis were prevalent - the Compact Cassette and FM broadcasting. It is very different from A weighting, with a large peak of 12dB at 6.3kHz.
This reflects studies done by the BBC (ref 2) based on earlier work from Deutsche Bundespost Telekom. These found that the subjective effect of unwanted noise was higher in this region than the equal loudness curves would lead one to expect. It is used more widely outside the United States, particularly in Europe.
Noise Weighting - Conclusion
Noise weighting is needed to reflect the fact that the human ear is more sensitive to mid-frequency noise between 1 to 6 kHz than to low and high frequency noise. Fortunately, A-weighting matches reasonably well the annoyance of noises that are low level and just becoming annoying, hence the widespread use in environmental noise measurements.
The variation in noise weighting curves used shows that establishing the annoyance value of noise is an inexact science - to do it better would probably require a frequency response varying with sound pressure level. That would be a closer match to the way the ear's frequency response varies with sound level, but would be prohibitively expensive to implement and difficult to calibrate and verify.
Sources
- The science and applications of acoustics, Daniel R. Raichel, Springer, 2006, ISBN 0387260625 p53
- BBC research report EL-17, The Assessment of Noise in Audio-Frequency Circuits, W.K.E.Geddes, 1968
 |
 |
 |
 |
