This story was originally published in the April 1996 edition of Studio Sound Magazine. The technique described here is more commonly known as parallel or New York compression. A later version of this was published in 2001.
IN AN IDEAL WORLD each stage of the recording / transmission / reproduction chain would have an equally wide dynamic range capability. In reality the dynamic range of each subsequent stage is generally less than that of the preceding stage. The average analogue console has a usable dynamic range of about 110 dB. Plain 16 bit digital audio has typically less than 96 dB. FM radio has around 55 dB and the worst case listening environment, the car, can fall to 20 dB or less! The dynamic range capability of the human ear is about 120 dB although there may actually be a shifting range of around 60 dB if other factors are considered. In this article I will look at a way of reducing dynamic range to suit FM radio and particularly how to implement this during recording and post-production. The system does have other uses which will also be examined.
The difficulty in broadcasting is that there is a strict limit on the transmittable dynamic range. This can be broadly defined as the difference between the transmission systems noise floor and the point of maximum modulation. In FM radio it is around 55 dB. Some classical music can exceed this figure and so it becomes desirable to reduce its dynamic range. The actual usable dynamic range is ultimately dictated by the listener’s environment. Control of the station’s dynamic profile to suit this environment is really the domain of station processing. Later I will briefly discuss the possibility of using this compression system in that application.
BEFORE COMPRESSION WAS AVAILABLE engineers had to tailor dynamic range by hand. This involved following a score (or working from memory) and making level changes in anticipation of the peaks and troughs in the music. These changes would be made as slowly as possible and resulted in very transparent dynamic control. I am sure many readers will remember doing this in the old ‘AM days’. Interestingly, this method mimics approximately the apparent dynamic response of the ear1. For subtlety, you still can’t beat a sensitive engineer with a good set of ears.
As technology advanced, the range and quality of electronic compression devices improved drastically. Even so, one of the main objections to the normal compressor configuration (fig. 1, at left) is that it reduces the impact of loud passages by ‘holding them back’. This is particularly noticeable when these passages contain transient material such as percussion. It is the departure from a 1:1 gain law at high levels and the effect this has on the impact of the music that is most noticeable. This, combined with the apparent lifting of low-level passages caused by slow release times and gain make-up after the compressor, has probably generated more letters to the editor than any other broadcast topic!
Some manufacturers have tried to address these problems with multi-stage or multi-band techniques and programme-dependent attack and release times. On the whole these are an improvement, especially where there is time to fine tune the device to suit the audio being processed. The BBC has even developed a system called ‘DRACULA’ that uses a three second look ahead to try and better place level alterations in the correct dynamic context2. However, there will always be a section of the public and industry that finds the action of these devices disturbing. I think I can safely say that the listeners most likely to notice the effects of dynamic controllers tend to be the ones that listen to the types of music that defy transparent use of those controllers! The following compression system is subtle enough to go largely unnoticed yet still provide ‘real’ amounts of control.
SOME YEARS AGO IN Studio Sound magazine3 a colleague found a brief description of the method presented here. It has turned out to be such a useful tool that I am presenting it again, this time in more detail and with some refinements. In addition I will examine other uses and implementations of the system. For those wanting a fuller understanding there is a mathematical treatment.
This system relies on a secondary compressed signal path running in parallel with the main signal path (see fig 2, at right). This basic idea is not new. It has been around since at least the early sixties and dozens of companies have used it as the basis of a wide range of products. The most well known user of the technique is Dolby Labs, with a string of patents as long as your arm. A colleague dubbed our implementation ‘Side Chain Compression’ or simply ‘Side Chain’. These terms should not be confused with the internal control side chain of the compressor itself. This name has stuck and I will use it here, although in retrospect ‘dynamic scaling’ might have been better.
Setting up the system is fairly easy. In my case I use a secondary stereo buss to feed the compressor (fig 3, at left). Any audio to be processed is sent to the main and secondary busses. A stereo channel is used to return the compressor’s output to the desk which in turn is sent to the main buss only. The return must be in phase with the original signal.
Calibration is also straightforward and enables the user to achieve consistent results at different locations. 1 kHz tone at reference level ( 0dBr ) is fed to the Side Chain buss only. This would normally be 0 VU or PPM 4. The compressor is set for 16 dB of gain reduction at a 2:1 ratio. Fast attack and slow release times should be used and in my case these are 2.5 mS and 800 mS respectively. Release times as short as 25 mS can be used depending on the type of music and the effect required. Experiments with digital compressors within ProTools have shown that times shorter than these don’t really work because they follow the waveform too closely.
The level at the compressor’s output is made up to 0, in this case requiring 16 dB of gain. The compressor’s return channel is trimmed so that a setting of 0 on the fader gives a reading of 0 (dBr) on the main meters. In order to get a consistent indication of the amount of Side Chain the compressor’s return channel can be marked with a chinagraph (grease) pencil as in figure 4 (the fader, at left). To get these points, put tone back on both busses and add enough Side Chain level to increase the level on the main buss by 0.25, 0.5, 0.75, 1, 2, and 3 dB.
Figure 5 (on the right) shows the effects of ‘1 dB’ of Side Chain compression (dBsc). Through the middle is a 1:1 gain law representing the main signal path. Below is the output of the compressor at 1 dBsc (-18 dB set-point). Above is the resultant slope – a ratio of 1.14:1. As can be seen there is approximately 5 dB of lift at the -32 dBr threshold. A linear transfer curve (1:1 offset by 5dB) is maintained below this point.
Figure 6 (right) shows the actual plots of the system I use. The main difference is the curve at the threshold due to the compressor’s ‘soft knee’. Because the system has 1 dB of ‘gain’ this can be corrected at the master faders. This shifts the slope down by 1 dB giving +4 dB at -32 dBr. In practice it is only necessary to reduce the masters when the amount of Side Chain exceeds 1 dBsc. The calibrated scale outlined above will show exactly how much this reduction should be.
The original 1977 description of this method called for 20 dB gain reduction. In our case we were using Neve 2254s which have only 16 dB of metered GR available and this became our ‘standard’.
The table below shows a comparison of the two. It is possible to implement the system on some digital audio workstations.
Figure 7 at right shows Side Chain at 1 dBsc compared with 2:1 compression. As can be seen, the low level lift at the Side Chain’s threshold of -32 dBr is the same in both cases. Both methods achieve 4 dB of gain reduction. In the case of standard compression this is achieved by scaling the top 8 dB at 2:1. The Side Chain method spreads this gain change over 32 dB, giving a ratio of 1.14:1. This is achieved by adding a highly compressed version of the signal at low level with itself. At high input levels the effect of the compressed path is small – nil if you reduce the master faders as detailed earlier. As the level drops, the additive effect of the Side Chain path increases and this is what gives such a gentle slope.
There are two main reasons why Side Chain compression sounds better. Firstly the gain law is gentler when it departs from 1:1. In fact, the threshold is so low that the system is in ‘gain reduction’ most of the time. Secondly, the method of implementation takes the compressor out of the main signal path. Standard compression derives its control directly from the signal and this control is applied via a feedback loop to the signal itself. With Side Chain, the main signal still ‘controls’ the amount of gain change but the gain law is derived indirectly by signal addition. Purists should still expect some smearing of transients due to non-linear group-delay in the compressor and return channel but the effect of this is insignificant when compared with a compressor in the main signal path.
Like a standard compressor, the correct level must be fed into the system for it to be effective. The audio should be peaking to the correct level on the main meters before the Side Chain is added. Be warned that over-driving the system is likely to push the compressor beyond its gain reduction range and could result in distortion. Under-driving it increases the effect of the Side Chain by effectively raising the threshold. This is the opposite of the effect from standard compression under the same circumstances. As a rule, if the maximum peaks hit PPM 6 (0 VU) the rest of the dynamic should take care of itself.
Provided these steps have been followed the amount of Side Chain to be added is up to the user. For classical music I have found the aesthetic maximum to be about 1 dBsc but it also depends a lot on the instrumentation. For example, solo harpsichord never needs Side Chain because its highly transient nature makes it sound loud even at low levels. A full orchestra performing dynamic repertoire can accommodate 1 dBsc or more. If there are extended low level passages I prefer to ride the master faders rather than adding more Side Chain since noise can become obtrusive at higher settings.
The great thing about Side Chain compression is that you can alter the amount you add in real time and those alterations can be very subtle. This is useful when recording or broadcasting public concerts, as you can remove the Side Chain in between movements so that audience noise is not increased. On the other hand, it can be used to create a ‘cushion’ of atmosphere which sits nicely under announcements. The other advantage of using it is that low level passages can be lifted without the risk that unexpectedly loud passages will get out of hand. In the case of live broadcasts I don’t use it on the announcer – standard compression set for a medium attack and a slow release is more appropriate if it’s needed at all.
As a general rule I also remove the Side Chain at the ends of pieces unless the audience sounds a bit thin. In that case I use it to increase the density of the applause. Because I use a secondary buss to feed the Side Chain it is possible to pick exactly what is processed. I don’t send reverb or ambience mics to it but that’s only a personal preference. At the suggested setting the audible side-effects are virtually nil. Like any form of compression though, it increases the density of the sound and this is particularly noticeable in the reverb. It can also appear to change the balance between a soloist and accompanying ensemble. This subtle lifting of low level information is very difficult to detect in isolation so I would suggest that you include its use in your session notes. Ultimately, because the quiet material is lifted, the average level increases and the ear perceives that the sound is louder and somehow ‘better’4.
It is possible to use other compression ratios and different amounts of gain reduction. Even devices that have soft knees or varying ratios can be used. The only proviso is that attack and release times should be fast enough to closely track the programme’s dynamics. See the side bar for a full mathematical treatment.
WHILE I USE Side Chain compression primarily on classical music there are some other areas where it could be useful. The system could be used as the main on-air processor for FM radio with a couple of caveats. Because it reduces the dynamic where the peak level has already been correctly set it is not suitable as a gain-riding processor. It is best used where the levels are already being properly set such as on a dual operation type station with a separate engineer and announcer. Because the effect is so subtle the result is very listenable, even when set as high as 3 dBsc. At this level care is needed setting the release time. If a fixed amount of Side Chain were used all the time the station could offer a ‘decoder’ of some sort to reverse the compression for critical listeners. As with any over-all station processing, the level of vocal (spoken) material may have to be reduced slightly to maintain the correct on-air balance between it and music.
This mode of compression works well on non-classical music too – both on whole mixes and individual tracks. I have found that 2-3 dB of Side Chain compression applied to a mix destined for AM radio creates a denser sounding track without affecting the transients too much. The result of this ‘pre-processing’ is a track which is less likely to be mangled by the station’s own on-air processing. 3 dBsc or more can be used to make really dense sounding tracks and yet the ratio is still only 1.36:1. It may be necessary to slow the release time a little depending on the desired effect.
Standard compression can be used quite creatively when tracking5. It is also relatively transparent at this stage as each source is processed individually. Side Chain can be used where very transparent control is needed. Transients are retained while the dynamic can be precisely controlled to make it sit in the mix. While it works well with most instruments it doesn’t work so well with singers. If you want to try it for vocal work, it would be advisable to use a peak limiter prior to tape to catch any big peaks. The vocal application in which it really excels is the recording of speech. I have used it for documentaries and the like to reduce the progressive reduction in level that can happen during long reads. This gives a more natural and open sound during lengthy passages of speech.
I have also found it a useful mastering tool – particularly for classical releases destined for cassette. The whole dynamic can be scaled to fit comfortably on cassette without audio being buried in tape noise a lot of the time. I also subtly peak limit to avoid saturation. I then let the cassette duplicator know where the biggest peaks are, thus guaranteeing a hot print on that medium. The CD version is printed without it of course. I have also occasionally used it when mastering pop/rock albums where the effect seems appropriate. Ultimately, it depends on what you want to achieve and what sounds best6.
In pure mastering applications it is best to remove the Side Chain as the track fades or dies away. This is because it causes a slight lift in the noise floor as the audio disappears.
As with any compression system there are side effects which can be useful in themselves. In one case I used the system to repair the balance of a classical concert with solo vocalist. It was the kind of recording that post-production engineers have nightmares about. Individually the orchestral balance was fine as was the sound of the vocalist. Unfortunately the vocalist was far too loud and the recording was very dry. I added 3 dB of Side Chain compression with a fast release time. This had the effect of lifting the orchestra in between vocal passages. Two reverbs were used. One send was EQed to have mostly voice and the other mostly orchestra. The orchestral reverb was set in such a way that it carried the sound through the start of vocal lines – hiding the fact that the overall level had ducked. This relatively severe treatment was acceptable for a one-off broadcast but would have been too much for a CD release. The untouched audio was broadcast with pictures elsewhere and listeners to radio were surprised at how much better ‘our’ sound was.
THERE ARE A COUPLE of other things you might like to try. Putting a bass cut in the Side Chain return reduces the lifting of low frequency noise. In some cases I have EQed other areas for various effects. Boosting at 100 Hz and 10 kHz (both peaking) can give you a subtle contouring effect at low levels. You could also use a multi-band compressor like MDT* in ProTools to create subtle dynamic equalisation effects. More severe EQ can be used to lift instruments in a mix.
One commercial production operator I knew, used to EQ the feed to the compressor. He would boost the speech intelligibility frequencies only and this gave immense HF density without resorting to lots of EQ in the main signal path.
The other thing that you can try is to send from the Side Chain channel to your reverb. The send-level can even be set slightly higher than those of your audio source/s. This creates a cushion of reverb that seems to ‘stick’ better – especially with cheaper reverb units. It also stops recordings ‘drying up’ as the level decreases. I also insert a compressor in the send to the reverb to flatten quick peaks. This reduces excessive reverb hang-over during repetitious staccato passages
It would also be possible to process the sum and difference signals rather than the discrete left and right signals. The effects of fixed processing of sum and difference signals have been covered in Studio Sound several times [7, 8]. The idea with processing the sum and difference signals would be to enhance or alter the image width. It’s probably more useful as a special effect than for serious work.
This system of compression provides an alternative to regular compression where subtlety is required and is a useful tool to have in one’s ‘box of tricks’. I would welcome feedback from anyone who tries this system themselves or has any suggestions.
Richard Hulse is a senior recording engineer with Radio New Zealand Limited.
The mathematics are actually not too difficult. I worked it out by plotting the standard set up outlined (16dB GR @ 2:1 +1dBsc) and then worked out the relationships on a calculator. Firstly we work out the system threshold. For this example I will assume the standard setup above.
System Threshold = [Gain Reduction x Ratio] / Ratio -1
= 16 x 2(:1)/1
Next we calculate the set point for +1dB of system gain (dBsc). The set-point is how much the secondary path is offset relative to the main path.
Set-point = [(antilog (system gain increase / 20) – 1)] log20
= [(antilog (1 / 20) – 1)] log20
= – 18.27dB
From the set-point and threshold we derive the level added to the main signal path at the threshold.
Compressor’s effective output at threshold = (threshold / ratio) + set -point
= (3212) + (- 18.27)
Adding the input threshold and the compressor’s effective output at that threshold gives us the total system output at the threshold.
System output at threshold = [(antilog (output at threshold / 20)) + (antilog (threshold / 20))] log20
= [(antilog 34.27 / 20) + (antilog 32 / 20)] log 20
The difference between the system threshold and the output at that threshold is the amount of lift.
threshold = system output threshold
= (27) ( -32)
We work out the resultant gain law as follows.
Ratio = threshold / (system output at threshold -system gain)
= -32 / ((- 27)-( +1))
Some general patterns emerge if multiple settings are plotted against each other.
Figure 8 shows how altering different parameters affects the resultant transfer characteristic. The general ‘rules’ are as follows:
- Increasing gain reduction lowers the threshold and increases the lift at that point.
- Decreasing gain reduction raises the threshold and decreases the lift at that point.
- Tighter ratios on the compressor with the same gain reduction and set -point give the same threshold with a greater amount of lift.
- Given the same GR and set -point the steeper compression ratios give progressively less boost at the threshold.
ProTools 2.5 TDM and Side Chain Compression
It is possible to generate the system entirely in software using ProTools with TDM. The main advantage is that the DSP compressors add far less noise and distortion than their analogue counterparts. The only disadvantage is that limited gain reduction is available on the gentler slopes.
Two mono tracks are used as a stereo pair and are fed to two stereo auxiliaries. Because the compressor takes a finite amount of time to implement in DSP, one must be inserted into each of these auxilaries to ensure there is no time delay between them.
Setting the faders is very simple. Set the attack and release at fast and slow release times, respectively, the threshold at -56 and the ratio at 2:1. The output is set at +8. This means that the dBsc scale is 8 dB higher than in the analogue implementation. The aux channel fader’s set-point for 1 dB of Side Chain compression is -10 dB. The master is set at -1 dB. These settings give 4 dB of boost at a -32 dB threshold. The main signal path can be limited or compressed if you need to control the absolute peak level. It is also possible to feed input to output through the two aux channels. This allows real time processing of digital or analogue sources. The effect as set up appears in the output path and can be monitored during recording to assess its effect and then altered later if required.
- Ted Fletcher, ‘A Personal View Of Psychoacoustics – Overload’, Studio Sound, May 1983, pages 64 – 65.
- N.H.C. Gilchrist, ‘DRACULA: Dynamic range control for broadcasting and other applications’, BBC RD 1994/13
- Mike Bevelle, ‘Compressors And Limiters’, Studio Sound, October 1977 page 32 (also reprinted June 1988).
- Ted Fletcher, ‘A Personal View Of Psychoacoustics – The Transparent Compressor’, Studio Sound, November 1983, pages 70 – 71.
- Denis Degher, ‘An Engineers Guide To Compression And Limiting’, Recording Engineer/Producer, October 1986, pages 56 – 62.
- Rick Clark, ‘Applications of Dynamics Signal Processing’, Mix Magazine, April 1994, pages 32 – 43.
- Michael Gerzon, ‘Stereo Shuffling : New Approach – Old Technique’, Studio Sound, July 1986 pages 122 -130.
- Michael Gerzon, ‘Fixing it outside the mix’, Studio Sound, Sept 1990, pages 78 – 93.
* Multiband Dynamics Tool by Antares Systems, PO Box 697, Applegate, CA, 95703-0697
- Richard C. Cabot, ‘Dynamic Range Modification : Limiters, Compressors, Expanders.’, Recording Engineer/Producer, October 1986, pages 44 -54.
- Michael Morgan, ‘Analogue Signal Processing for Digital Recording’, Recording Engineer/Producer, April 1989 pages 32 – 38.
- Marvin Caesar, ‘Digital Audio, Dynamic Range and the Real World’, as 2. pages 26 – 30.
Dedication: To my friend Adrian Prowse, who passed away in 2012, and who created the illustrations for this story.