Universal Audio WebZine
Volume 3, Number 6, August 2005
[UA Universe] [Ask the Doctors] [Artist Interview] [Analog Obsession]
[Support Report] [The Channel] [Plug-In Power] [Playback] [Featured Promotion]
[Graphic-Rich WebZine]
[Back Issues] [UA Home]

Dr. David Berners (left) is the Universal Audio Director of Algorithm Development; Dr. Jonathan Abel is the co-founder and CTO
Ask the Doctors: What is Multiband Compression?
by Dr. Dave Berners

In this month's column, we delve into what may seem like a fairly simple question: What is multiband compression, and what is it for?

Multiband compression is a process by which a signal is separated into different spectral components by means of a filterbank, and each component is passed through a compressor. The compressed outputs are then recombined. Usually, the filterbank is designed to be spectrally flat; in the absence of compression, the output signal will have the same equalization as the input. This is called a perfect magnitude reconstruction filterbank.

For multiband applications, it is desirable that the compressors themselves have a wide frequency response to avoid phase-related artifacts when the bands are recombined.

One of the earliest uses of multiband compression was Dolby A-type noise reduction, introduced in 1966. Compression was used to reduce the noise floor when users were recording to magnetic tape, and separate bands were used so that the time-varying compressor gain would not modulate out-of-band noise, which might be audible. Dolby A had four separate bands of compression, with transition frequencies at 80 Hz, 3 kHz and 9 kHz.

The compression used in the Dolby scheme was as shown in Fig. 2. A parallel, "dry" signal path was mixed in with the compressed output, so that the gain reduction could not exceed a preset amount. This amount was set to 10-15 dB for Dolby A, depending upon the band. The "parallel compression" scheme is also popular in modern multiband compressors.

Figure 1. Flowchart for Dolby A Compression Scheme
Multiband units with a parallel dry path will often have a control labeled Range, which is used to select the amount of dry signal that is mixed in with the compressed output, and thus determines the maximum amount of compression that can be achieved for the band.

Multiband compression is valuable for mastering applications, where dynamic adjustment m ay be desired for a spectrally localized event that is already imbedded in a mix. An example of this is de-essing; compression applied to a tight band somewhere between 3 and 6 kHz can be used to de-ess a mixed track without causing full-spectrum dropouts during compression. Multiband compression can also be used to artistically treat one spectral band of a mix, e.g. bass. Other applications include attempts to increase the RMS level of a mix. By separately compressing different bands of a mix, it is sometimes possible to achieve overall RMS levels that would be unattainable with a single-band limiter. When using large amounts of multiband compression, a mix can be coerced to match a preselected "target EQ" curve by using a high compression ratio in each band, and setting the output levels of the bands to the target levels. The "target EQ" method is used when the desired spectral distribution is known for a track, but the equalization of the existing track is unknown; or when the equalization of the existing track changes during the course of the track. The drawback of this method is that large amounts of compression must be used in all bands for the processed signal to resemble the target spectrally, and it is difficult to avoid artifacts from the (necessarily) heavy amounts of compression.

Compression algorithms used for multiband compression can be similar to single-band compression algorithms. For multiband applications, it is desirable that the compressors themselves have a wide frequency response to avoid phase-related artifacts when the bands are recombined.
Figure 2. Parallel Compression Scheme
Program dependence is less of an advantage for multiband applications, because dropouts associated with a slow release are limited to a subband, rather than being full-spectrum.

With the advent of digital processing, it is possible to create subband filters for multiband applications which have linear phase. With linear phase, filterbanks can be made to be perfect reconstruction rather than perfect magnitude reconstruction. Perfect reconstruction filterbanks are entirely transparent in both frequency and magnitude. Because of this, linear-phase filters are especially valuable when multiband devices are used to fix isolated problems in a mix; except when one or more bands are being compressed, the output signal will be identical to the input signal, provided the filterbank is perfect reconstruction. For applications where one or more bands are in compression much of the time, the advantage of linear phase becomes more nebulous. In any case, the difference between linear-phase filterbanks and the more cost-effective minimum-phase filterbanks is most apparent for very low transition frequencies.

Multiband compression is a valuable tool, especially for the mastering engineer who does not have access to individual components of a mix. It is a convenient way to fix problems, and to perform semi-automatic spectral shaping. In many cases, multiband compression can be used more transparently than single-band compression. However, multiband compression is a powerful technique that must be used with care and taste.

Do you have a question for the Doctors?