The ITU-R 468-weighting curve (originally defined in CCIR recommendation 468) is widely used when measuring noise in audio systems, especially in the UK, Europe, and former countries of the British Empire such as Australia and South Africa. It is less well known in the USA where A-Weighting has always been used.
While most audio engineers are familiar with the A-weighting curve, which is said to reflect the ‘equal-loudness contours’ derived initially by Fletcher and Munson (1933) and later Robinson and Dadson (1956), few seem to realise that these curves relate only to the subjective loudness of pure tones, not noise. Furthermore, recent experimental work casts doubt on their accuracy (see entry for A-weighting and revised ISO 226 :2003)
In fact the human ear responds quite differently to noise, and it is this difference that gave rise to the 468-weighting, which arguably is the only valid weighting to be used for all noise measurements, whether on audio equipment or in the assessment of low-level environmental noise.
History and Original research
Developments in the 1960’s, in particular the spread of FM Broadcasting and the development of the compact cassette with Dolby-B Noise Reduction, alerted engineers to the need for a proper weighting curve, and the fact that A-weighting was not giving consistent results. Pre-emphasis of high frequencies in all these systems was resulting in increased noise readings that did not correlate with subjective effect, and it was possible for a cassette deck to measure worse and yet sound better!
Experiments in the BBC led to BBC Research Department Report EL-17, ”The Assessment of Noise in Audio Frequency Circuits”, in which experiments on numerous test subjects were reported, using a variety of noises ranging from clicks to tone-bursts to ‘pink’ noise. Subjects were asked to compare these with a 1 kHz tone, and final scores were then compared with measured noise levels using various combinations of weighting filter and quasi-peak detector then in existence (such as a German DIN standards).
CCIR Recommendation 468-1 was published soon after this report, and appears to have been based on the BBC work. Later versions up to CCIR468-4 differed only in minor changes to permitted tolerances. This standard was then incorporated into many other national and international standards (IEC, BSI, JIS, ITU) and adopted widely as the standard method for measuring noise, in broadcasting, professional audio, and ‘Hi-Fi’ specifications throughout the 1970’s. When the CCIR ceased to exist, the standard was officially taken over by the ITU (International Telecom Union).
The CCIR curve differs greatly from A-Weighting in the 5 to 8 kHz region where it peaks to +12.2 dB at 6.3 kHz, the region in which we appear to be extremely sensitive to noise. While it has been said (incorrectly) that the difference is due to a requirement for assessing noise intrusiveness in the presence of programme material, rather than just loudness, the BBC report makes clear the fact that this was not the basis of the experiments. The real reason for the difference probably relates to the way in which our ears analyse sounds in terms of spectral content along the cochlea. This behaves like a set of closely spaced filters, which, if they had constant ‘Q’ would have bandwidths proportional to their centre frequencies. High frequency hair-cells would therefore be sensitive to a greater proportion of the total energy in noise than low frequency hair cells. Though hair-cell responses are not constant Q, and matters are further complicated by the way in which the brain integrates adjacent hair-cell outputs, the resultant effect appears roughly as a tilt centred on 1 kHz imposed on the A-weighting.
Dependant on spectral content, 468-Weighted measurements of noise are generally about 11 dB higher than A-weighted , and this is probably a factor in the recent trend away from 468-weighting in equipment specifications as cassette tape use declines.
It is important to realise that the 468 specification covers both weighted and ‘unweighted’ (using a 22 Hz to 22 kHz 18 dB/octave bandpass filter) measurement and that both use a very special quasi-peak rectifier with carefully devised dynamics (A-weighting uses RMS detection for no particular reason). Rather than having a simple ‘integration time’ this detector requires implementation with two cascaded ‘peak followers’ each with different attack time-constants carefully chosen to control the response to both single and repeating tone-bursts of various durations. This ensures that measurements on impulsive noise take proper account of our reduced hearing sensitivity to short bursts.
This was once important because outside broadcasts were carried over ‘music circuits’ that used telephone lines, with clicks from Strowger exchanges. It now finds fresh relevance in the measurement of noise on computer ‘Audio Cards’ which commonly suffer clicks as drives start and stop.
Engineers in the USA never ‘caught on’ to 468-weighting, probably because for many decades they were part of a strong independent manufacturing economy that tended to import little from abroad. For the same reason they never adopted the PPM (Peak programme meter), which also came out BBC Research. Nevertheless, 468-weighting is still demanded by the BBC and many other broadcasters, and knowledge of its existence and validity needs to spread. It is superior in allowing fair comparison of specifications for all types of equipment, which A-weighting cannot do because of differing noise characteristics.
Present Usage of 468-Weighting
468-weighting is also used in weighted distortion measurement at 1 kHz. Weighting the distortion residue after removal of the fundamental emphasises high-order harmonics, but only up to 10 kHz or so where the ears response falls off. This results in a single measurement (which Lindos refer to as Distortion Residue measurement) which corresponds well with subjective effect even for power amplifiers where crossover distortion is known to be far more audible than normal THD (Total harmonic distortion) measurements would suggest.
Measurements of microphone noise are easier using 468-weighting because it emphasises the audible noise more in comparison to low-frequency noise. A-weighted microphone measurements require quieter conditions to avoid the effects of slow pressure variations caused by wind and air conditioning.