Multi-band compression is one of the most powerful processing techniques available to the recording engineer. However, if you don't know how to apply it effectively it can just as easily ruin the sound of your entire production.
Most of the compressors we use in music production operate on what is known as the full-band principle, where the entire audio signal is processed via a single gain-control element. When gain reduction occurs, the whole signal level is reduced, just like turning down a volume control. In other words, whenever a loud peak occurs (whatever its frequency content) that forces the compressor to act and the level of the whole signal is reduced until the loud event has passed.
A common problem arising from this method of compression is that a loud kick drum (which produces mainly low frequencies), will trigger the compressor and consequently pull down the gain of everything else that happens to be passing through the compressor at the time, even though those other sounds might not need compressing. If the compressor is acting on a solo kick drum track, then clearly this isn't a problem, but if the whole drum kit is being compressed, the high-frequency sounds, such as hi-hats and cymbals (which carry relatively little acoustic energy), will be compressed along with the kick drum, the outcome of which is often a subjective dulling of the sound during times of heavy compression.
The situation worsens when you're compressing a whole mix, because the low-frequency sounds in the mix determine the compression applied to everything else, so what tends to happen is that the kick drum and bass line dictate how the mix will be compressed. One way to disguise this aspect of full-band compression is to set a slightly longer attack time so that transients can pass through cleanly before the gain reduction takes place, but clearly this does little to control peak levels, which is important when you're working with digital systems that can't tolerate overloads, however brief.
In some situations these side effects can be musically desirable, and in some forms of rock and pop music a hint of gain pumping (audible compression due to rapid gain changes) can add energy and excitement, providing it isn't overdone. However, for mastering and other critical applications, multi-band compressors provide far greater flexibility.
In essence, a multi-band compressor comprises a set of filters (similar to a PA system's active crossover) that splits the audio signal into two or more frequency bands. Three- or four-band compressors are perhaps the best compromise between versatility and ease of setting up. After passing through the filters, each frequency band is fed into its own compressor, after which the signals are recombined. The main advantage of multi-band compression is that a loud event in one frequency band won't trigger gain reduction in the other bands, so when that loud kick drum comes along, instead of pulling the whole mix down with it, only the low-frequency sounds (kick and bass instrument) will be compressed, leaving the mid-range and high frequencies untroubled. Similarly, a loud event in the mid-range won't affect the extreme high or low frequencies for the same reason. There are however, other advantages.
With a full-band compressor, whatever settings you apply affect the full frequency range, but a multi-band compressor can use different compressor settings in each band if necessary — different attack and recovery time constants, for example, or different ratio and threshold parameters. Furthermore, if the effects of compression change the overall tonal balance unfavourably, you can restore it by changing the levels of the three compressor bands relative to each other.
With such a system, it is quite possible to fully control the signal in each band but still end up with a signal that is too high in level when the bands are combined. One solution is to meter the combined level so that the user has a chance to reduce the overall gain if problems occur, but a good belt-and-braces approach is to follow the compressor with a fast-acting peak limiter to prevent overshoots or excess levels — again very important in digital systems. Another consideration, which admittedly affects the designer more than the user, is that any form of filter circuit introduces phase shifts, so great care has to be taken to ensure that, when the three bands are recombined, the optimal phase balance is restored. Failure to do this can result in loss of clarity, in much the same way as you might experience when using a very cheap equaliser.
Commercial multi-band compressors often allow the user to adjust the crossover points, so if we take a three-band compressor as an example, where is the best place to set them? The answer is that it depends on the type of material being processed and on what adjustments are needed. OK, I know that doesn't seem very helpful, so let's narrow it down a bit and assume we're processing a full mix of a typical pop song. Setting a crossover point in the middle of the vocal range can mess up the vocal sound, especially if you use radically different compression settings on each side of the crossover frequency.
In my experience, it's best to set the low crossover point below the vocal frequency range and set the high crossover point no lower than 2.5kHz. A sensible low-frequency setting might be 120Hz, as this is below the vocal range, but above most of the deep bass and kick drum elements going on in the rhythm section. These are only suggested starting points and, as a rule, I'd suggest moving the high crossover point higher than this if you plan to do more than very subtle high-end processing, as it's easy to over-emphasise the 2-4kHz area of the spectrum, which is where things can start to sound harsh. By contrast, if you restrict heavy processing to above 6kHz, you can significantly enhance the sense of detail and air, without affecting the crucial mid-range and disturbing the natural reproduction of vocals and many acoustic instruments.
The next decision is how to set up the compressor parameters in each of the frequency bands. There are two very different ways of using compressors — and, of course, all the settings in between! Very often we use compressors to control only those signals that are too loud, in which case we set the threshold at the 'enough already!' level, then set a compression ratio so that gain reduction is applied to anything exceeding this level. The compression ratio is defined as the number of decibels the input level must rise (once the threshold has been reached) for the output to rise by just 1dB, so for a compression ratio of 5:1, a 5dB increase at the input would be needed to cause the output to increase by 1dB. Below the threshold, no gain reduction takes place, so the signal passes through unchanged. If you need to tame vocal peaks or other occasional excesses, this is the usual way to set the compressor.
The other way is to set a low threshold (say, -30 to -40dB) with a very low ratio (typically less than 1.2:1). Now, instead of applying a lot of compression to signal peaks, you'll find that a little compression is being applied to all but the quietest signals. This is a good way of adding density and energy to a mix that's already well balanced, but it won't help with controlling peaks in poorly balanced material.
These two strategies, and any settings between, can be applied to setting up a multi-band compressor, but what makes this approach so flexible is that you can use different compression settings for each band. For example, if I'm working on a track where I feel the bass end is lacking in weight, I may use a higher threshold and a higher ratio to bring up the average bass level, but still use subtle low-ratio settings for the mid-range and high end. The make-up gain (which is usually available independently for each band), may then be used to restore any spectral imbalance caused by the difference in compression settings. Similarly, by applying heavier compression to frequencies above, say, 6kHz it is possible to increase the high-frequency density to create an effect similar to that of using an enhancer.
Do bear in mind, though, that the more make-up gain you apply, the more you will be bringing up the background noise — tape hiss, spill, mechanical playing noises and so forth. Again, careful selection of crossover points and ratio/threshold settings can help to mitigate any adverse effects, relying on noise-masking in the different frequency bands in much the same way that Ray Dolby discovered with his noise-reduction systems.
That leaves the mid-range, and in a mastering situation that's the area I tend to process least. Often a ratio of 1.1:1 and a threshold of -35dB is enough to knit everything together, and when this is combined with the other two bands (which may be more heavily processed if they need it), the whole mix sounds more vibrant, clear and punchy, but without sounding obviously processed. The 'louder at all costs' brigade may then want to process the finished mix via a peak limiter set to trim around 6dB off the loudest peaks so that the overall level can be increased by 6dB without clipping.
Because synthesizers are electronically generated sources that can be controlled in level via MIDI velocity, it may seem that there's little need to compress them, but there are occasions on which compression can be of benefit. Furthermore, there are times when multi-band compression will yield better results than full-band compression. For example, take the case of a sound based around a resonant filter sweep. If you apply full-band compression to tame the (often significant) increase in peak level at the sweep's centre frequency, the whole sound is reduced in level, sometimes dramatically. In this instance, multi-band compression may yield more transparent results by keeping the peak under control, but without changing the level in the other frequency bands. In the case of a three-band compressor, this means that heavy compression will only ever be occurring in one of the three bands at any one time.
Level peaking may also occur when two sounds are beating with each other or where a flange effect is being applied (due to the peaks and troughs caused by comb filtering), and though full-band compression will control the level, a multi-band compressor will do so much more transparently. A similar situation exists when you have a bass synth sound that also has a lot of higher harmonics. By using a multi-band compressor, the bass end (which is where most of the energy resides) can be kept at an even level without the level of the higher frequencies pumping up and down in sympathy.
So much for ratios and thresholds, but what of the attack and release settings, where these are independently adjustable? Here I tend to judge each band on its merits, where I might use a moderately fast attack time on the bass end, because very low frequencies have no fast transients to compromise. This brings the level under control reasonably quickly, and I'll set the release time as short as I can get it without any audible gain pumping being evident. As a rule, the busier the music, the faster the release time you need to ensure the compressor's gain resets itself between notes, but low frequencies tend to hang on longer than high frequencies, so you may need a longer release time at the bass end than in the mid-range. As a starting point, setting twice the release time as for the mid-band might be reasonable, though if you prefer to keep it simple, you'll probably get reasonable results by starting off with the attack and release times for all three bands set the same.
The mid-band covers the majority of what we perceive in a musical mix, so this can be set up much as you'd set a full-range compressor. In other words, ease up the attack time if you need to enhance transients, and use a fairly short release time, but not so short that you hear pumping.
What you do in the high band depends on the result you're after. If the plan is to increase the density of the high end, use a higher ratio setting to push up the energy and possibly a faster attack if you need to prevent the transients from becoming too pronounced. High transients tend to decay faster than low frequencies, so you may be able to use a faster release time than you did in the mid-range, but listen carefully for any pumping effects or other unnatural gain changes, and increase the release time as necessary to get rid of them. If you're using a compressor that allows you to solo each of the frequency bands, setting the release time may be easier, but as a 'ball park' figure, use half the mid-range release time as a starting point.
In radio broadcasting, and a few television services now too, it is standard practice to use multi-band compression. Most broadcast systems provide five bands, but some offer even more. Originally, multi-band processing was performed in the analogue domain, but most modern systems are entirely digital, and many incorporate other aspects of the broadcast chain including advanced peak limiting and stereo multiplexing functions. Whereas most pop and rock music stations employ the full five-band mega-squash (check out the almost total absence of dynamic range from the UK's BBC Radio 1, for example), more refined music stations tend to opt for a more subtle approach. The UK classical music channel BBC Radio 3, for example, tends only to employ three-band processing with relatively modest settings, and then normally only for the few hours around the rush hour 'drive time'.
Multi-band compression in broadcasting serves several purposes. Firstly, by increasing the overall level of programme material it makes it more audible in noisy environments — in the car, in the factory, and so on. It also improves the useful coverage area for a given transmitter power, maximising the potential audience numbers and justifying higher advertising rates.
Multi-band compression also allows a 'sonic signature' to be created, allowing each station to have a distinct and recognisable sound character. This is achieved by adjusting the amount of compression in each band and the relative levels of bands to create a certain spectral balance. Try tuning through the top end of the FM spectrum and listen to a variety of local commercial radio stations — some have very distinctive characters which quickly become recognisable in their own right.
It's not all good news though — there are some drawbacks to using multi-band compression in this environment. Firstly, multi-band compression generally means that you can compress far more heavily than with a full-band device. However, in high-quality listening conditions, such a heavily compressed signal can be very fatiguing to listen to. So whereas it can increase the 'listenability' in outlying coverage areas and in noisy environments, it can turn off more serious listeners in good monitoring situations.
It is also the case that multi-band compression generally takes place automatically after a radio programme has been mixed, and the studio engineer (Studio Manager in BBC parlance) may have little idea how the compression will change the balance, unless he or she can listen live off air as the programme is being produced. Consequently, a broadcast mix may end up sounding completely different from that originally created, which can be a little demoralising to the studio manager concerned, and hardly makes best use of the technology.
As far as the rebroadcast of commercial music is concerned, to try to overcome the potential problem of multi-band compression changing the sonic character and mix balance, a lot of commercial music is now being mastered with truly huge amounts of multi-band compression to try to ensure that the broadcast station's own processors have nothing to do, and therefore cannot change the sound. However, this approach strikes me as a case of the production companies cutting off their nose to spite their face, since although the material comes through the broadcast system relatively unscathed, it also tends to sound pretty poor on decent CD replay systems at home!
So far I've talked mostly about compressing complete mixes at the mastering stage, and the same criteria apply to mixing if you're processing subgroups of two or more instruments. However, there's no reason not to use multi-band compression on individual tracks if you want to. I must admit that I seldom use multi-band compression on single tracks, because the side effects of full-band compression are often musically flattering to single sounds, but where you need gain control without obvious compression, a multi-band model is more likely to deliver the results you need. In this case, I'd tend to start with identical settings for all three bands and set the values in a similar way as when using a full-band model. The main difference is that you probably won't have to set an artificially long attack time to keep transients intact. Complex instruments, such as acoustic pianos or drum kits miked with a limited number of microphones, probably benefit most from multi-band compression, but I know engineers who also use them on vocals or sources like guitar and wind instruments.
While compressing mixes, we tend to keep the crossover point away from the mid-band, but different instruments may require different settings. For example, a bass guitar might benefit from a low crossover point set to around 50 or 60Hz so you can control the really deep bass more precisely. At the same time, the upper crossover point could be reduced to just a few hundred hertz as the 50Hz to 800Hz band is quite critical to the way a bass guitar sounds. Above that, the energy tails off quite significantly, so the high band (800Hz and up) is quite capable of looking after the rest. If you feel the sound has got a boxy coloration that needs dealing with, then you can move the bands in closer to bracket the offending area and treat it differently to the bands either side, either by pulling its level down, or by applying less compression so as to avoid increasing its energy content. As with any conventional compressor, the more compression you apply, the more make-up gain you need to get the peak level back to where it was, but you have the option of using less make-up gain in a specific band if you want to suppress the contribution of that band to the overall sound.
Conventional full-band compressors designed to be used in stereo invariably have two audio channels plus a stereo link control. The link control sums the two side-chain signals and then uses that to control both compressor channels so that the same amount of gain reduction is always applied to both channels. At the same time, control of the compressor parameters is often handed over to just one set of the front panel controls, to avoid the necessity of setting up two channels identically. Without the combined side-chain signal there could be audible stereo image shift when one channel is (even occasionally) significantly louder than the other. A multi-band compressor works in a similar way, except that each band has its own side-chain, so a three-band stereo compressor would have three stereo-linked sections: one each for high frequencies, mid-range frequencies and low frequencies, for example.
Most routine vocal compression can be done using a full-band compressor and, though that may not be as transparent as a multi-band device, it's the way we are used to hearing pop vocals, so it's what usually sounds 'right'. Even so, there are instances where using a multi-band compressor on voice can be useful, not so much for out-and-out level control, but more for controlling or shaping timbre. For example, applying a higher ratio of compression to the high band can give sizzle and definition to an otherwise bland voice, while heavier compression in the bass band can add to the thickness and richness of the voice.
Some experimentation will be necessary to find the best crossover frequencies, but I'd suggest starting with around 150Hz and 4kHz and then adjusting by ear.
You may also find that you want to accentuate the rasp of a voice, in which case, try to identify the frequency area you're interested in, set the crossover points fairly close on either side, and then apply more compression to the mid band. Adjusting the levels of the various bands will also provide the opportunity to do some tonal shaping without resorting to EQ.