Understanding Compression 4

In the first instalment of this four-part series we introduced important concepts such as gain reduction (GR), threshold, ΔInput, ΔOutput and ratio. We learnt how to use the threshold and ratio controls to achieve the desired gain reduction, and we saw the simple relationship that ties all of the above-mentioned concepts together:
Output Signal Level = Input Signal Level – Gain Reduction
In the second instalment we introduced the knee control and saw how it allows us to ‘fade in’ the ratio, applying lower ratios for lower values of ΔInput and higher ratios for higher values of ΔInput. We also examined the inscrutable inconsistencies of attack and release time specifications among different compressor manufacturers, and explored the RC time constant in an effort to understand why those attack and release times are so hard to standardise.
In the third instalment we saw how the attack and release time settings alter the sound’s envelope, giving it more or less attack while also enhancing its impact and increasing its consistency from note to note or chord to chord. We introduced the concept of leftover gain and how it affects the following envelope(s), and we also learnt when and why to use make-up gain.






From this knowledge we can derive a strategic approach to solving any downward compression problem. If we know the highest level in the signal and the level we want to reduce it to then we can calculate the threshold, ΔInput and the required ratio. This mathematical approach provides an excellent starting point, but the amount of gain reduction we settle on will ultimately be affected by things we cannot derive mathematically and therefore cannot solve mathematically – specifically the settings of the knee, attack and release controls. We have to adjust these settings by ear and by eye because there are no standards for how they are measured, and yet they are the controls that determine how the applied gain reduction affects the sound aesthetically. Nobody cares about the theoretically correct GR if it sounds bad.
Everything the compressor is doing to our signal (except for Output Gain) is reflected in the GR meter. It shows us how much gain reduction is being applied or removed, when it is being applied or removed, and how fast it is being applied or removed. We correlate what we’re seeing on the GR meter with the changes we’re hearing in the signal, and make strategic adjustments based on what we’re trying to achieve. It’s an iterative process; as we make the knee softer and/or make the attack and release times slower, we will usually get less gain reduction. If we like how it is sounding and how the GR meter is moving in relation to the timing of the music or the dynamics of the sound but we need more gain reduction, we can lower the threshold and/or increase the ratio. After repeating this process a few times we should find the right combination of settings that work for the sound, the music and our ears, thereby achieving our target gain reduction (or a compromise thereof) in the most sonically acceptable way.
When dealing with subtle or hard-to-hear compression adjustments it is always worth keeping an eye on the GR meter because it allows us to see what the compressor is doing, thereby making it easier to hear what the compressor is doing. An understanding of how the compressor’s controls interact, and how those interactions are shown on the GR meter, allows us to replace most of the maths shown in the previous instalments and get straight to work with the GR meter.
The examples below show us how to set up four common compression situations, each taking advantage of the GR meter to minimise the maths and simplify the process.
PEAK-LIMITING
This type of corrective compression reduces the signal’s dynamic range by reducing the levels of unwanted or unnecessary peaks, allowing us to raise the overall signal level without clipping.
For peak-limiting we need to set up our compressor with peak sensing and the lowest threshold, along with the fastest attack time possible and a hard knee to ensure that the gain reduction is applied at the required ratio as soon as the signal exceeds the threshold level. We also need the fastest release time possible to avoid ‘leftover’ gain reduction affecting the signal after the level falls below the threshold – as demonstrated in the third instalment of this series.



Nobody cares about the theoretically correct GR if it sounds bad.









To apply peak-limiting strategically we need to know two things about the signal we’re processing: the highest peak level and the acceptable peak level. Our goal is to reduce the highest peak level to the acceptable peak level. We start by determining how much gain reduction we need to apply. Let’s say the difference between the highest peak level and the acceptable peak level was 6dB; this means our target GR is 6dB.
The most unfaltering way to find the starting points for our threshold and ratio controls is to create a loop containing a short excerpt of the signal (e.g. three or four seconds) that contains acceptable peak levels and a short excerpt of the signal that contains the highest peak level.
With the compressor set up as described above and with the monitoring muted, we run the loop and slowly raise the threshold until we’re seeing very little to no GR during the acceptable peaks, but maximum GR during the highest peaks. At this point the maximum gain reduction will be equal to the target gain reduction (or close enough), but that gain reduction is being applied at a much higher ratio than necessary because we are using the compressor’s highest ratio. As shown in the illustration below, the threshold has been set above the acceptable peak levels so that the compression only affects the unacceptable peak levels.













The goal now is to find the lowest ratio that provides the target GR. Monitor the GR meter while looping, and slowly reduce the ratio until just before the point where the GR starts becoming less than the target GR at the highest unacceptable peaks. At this point we have found the lowest ratio required to achieve the target GR.
We can now unmute the monitoring and take a listen. If it sounds good then the job is done; this is a likely outcome in situations where the acceptable peak levels are relatively low but the signal contains a few unexpectedly high and fast peaks. If it doesn’t sound good then we have to fix it. We’ll start by simultaneously lowering the threshold while reducing the ratio correspondingly to maintain the same target GR; the basic idea is to spread the GR over a wider dynamic range so it has a less extreme effect on the envelope. Lowering the threshold means the limiter will also be affecting the acceptable peaks in the signal that we didn’t want to affect, but lowering the ratio correspondingly to maintain the same target GR on the unacceptable peaks means the acceptable peaks will not be compressed too heavily.
If the compressor has a variable knee control we can choose to leave the ratio alone and make the knee ever-so-slightly less hard so that it has less effect on the peaks that we don’t want to compress, but is still providing the full gain reduction on the peaks we need to limit. This is something we will see on the GR meter, of course.













In this situation we need to lower the threshold to make sure the softer knee reaches the correct ratio at the correct level to apply sufficient gain reduction to control the peaks – which we can confirm with the GR meter. The illustration above shows a peak-limiting situation where the desired threshold of -12dBFS and ratio of 8:1 controlled the unwanted peak levels, but the audible effect was too abrupt. To remedy this we softened the knee and lowered the threshold level accordingly (to -20dBFS) to ensure the ratio reached 8:1 whenever the input signal level exceeded -12dBFS. The softer knee and lower threshold essentially ‘faded in’ the ratio from 1:1 at -20dBFS to 8:1 at -12dBFS, resulting in an audibly more acceptable result while achieving the same overall gain reduction and peak limiting. This ‘softer knee/lower threshold’ approach is illustrated below.













If the result sounds like clipping it means we are in the high threshold/high ratio situation described in the first instalment of this series (see ‘An Extremely Common Problem’), whereby the peak-limiter is ‘flat topping’ the peak, creating a problem that looks and sounds like clipping but is not indicated as clipping on any meters. If we cannot prevent this by lowering the threshold and ratio and/or softening the knee a little as described above, we need to change tact and adopt the ‘low threshold/low ratio’ solution shown in the first instalment (see ‘A Smarter Solution’), or consider chaining two peak-limiters together – one with a higher threshold to grab the worst parts of the peaks and bring them down a little, followed by one configured as above.
Having solved the excessive peaks, we can now apply Output Gain to raise the overall level without clipping the peaks – which is usually the goal of peak-limiting. We set the Output Gain to be equivalent to the maximum GR we have achieved. So if we have achieved a maximum GR of 4dB then it is appropriate to add 4dB of Output Gain. The result will be an overall level increase of 4dB without clipping the excessive peaks.
MATCHING DYNAMICS
This type of corrective compression aims to put a sound into the same dynamic perspective as other sounds in the mix, so that it sits comfortably alongside them without relying on extreme level automation and risking the inherent expression inversion that comes with it. Unlike peak-limiting, the goal here is to spread the gain reduction over as much of the signal’s dynamic range as possible to minimise its effect at any given point of the envelope. We do this by using low thresholds that, in turn, allow low ratios.
For matching dynamics we need to set up our compressor with RMS sensing because we are trying to match the perceived levels of one sound with the perceived levels of another sound. We will start with the lowest threshold, the highest ratio, a hard knee and the fastest attack and release times possible. After we get the threshold and ratio right, we will adjust the other controls by ear and by eye, but for now we need to minimise the influence of those other controls otherwise they’ll confuse things – as we know, every control affects the gain reduction, and if we don’t approach those controls methodically we can be easily lead astray.













We start by finding the lowest level of interest in the signal – this could be the softest played note in a musical performance, the softest spoken word in dialogue, or the softest sound of interest in a field recording (e.g. the background rain in a thunderstorm recording). Select an excerpt of the audio that includes the lowest level of interest along with a few seconds of the audio before and after it. As with the previous example, we might have to edit together a temporary ‘compressor set-up track’ – in this case containing excerpts of the lowest level of interest alongside excerpts of the highest levels of interest that will definitely require some gain reduction.
Loop the selection and start raising the threshold. The very high ratio ensures that we will see obvious changes on the GR meter whenever the signal level exceeds the threshold level. The goal here is to set the threshold to a point where we are seeing no GR movement during the lowest level of interest, but GR is obvious at all other times. Setting the threshold below the lowest level of interest will serve no benefit other than adding impact and authority to background noises – increasing their audibility and also reducing their dynamic separation from the desired parts of the sound. The process of determining the required threshold and ratio is described in the third instalment of this series (see ‘Reducing Dynamic Range’).













With the threshold determined we now loop over excerpts that contain the highest levels of interest and reduce the ratio until we are seeing the gain reduction required to rein in the sound’s musical dynamic range appropriately, putting it in the same dynamic perspective as the sounds it has to sit comfortably alongside of in the mix.
Assuming we have set our threshold and ratio correctly, it is time to adjust the other controls. Unlike peak-limiting, which relies on a hard knee with very fast attack and release times, matching dynamics requires careful tuning of the knee and the attack and release times to suit the dynamics of the sound. Keeping an eye on the GR meter, we first reduce the attack time and then reduce the release time until the GR meter appears to be moving with the sound’s perceived level, as if we have tuned the ballistics of the GR meter to match the dynamics of the signal. The GR meter is now essentially another version of the input signal meter that is working the wrong way: starting at 0dB and moving lower. This type of ‘GR motion’ indicates that all of the controls are at approximately the right settings; from here, making small changes to any of the settings should make clearly audible differences that are definitely either ‘good’ or ‘bad’ without much room for ‘maybe’.













Softening the knee and/or slowing down the attack time will reduce the overall gain reduction. We compensate for this by increasing the ratio – which, in turn, might require re-tuning of the other settings. As we have learnt in this four-part series, each control has an effect on the gain reduction; therefore changing one control is likely to require changing others. For matching dynamics, it will typically take a few iterations to fine-tune the compressor for the desired musical or aesthetic result. If we cannot get the signal to the desired dynamic perspective but can get within a couple of dB of it, we can use automation to solve the rest of the problem with little risk of expression inversion.
If necessary, we can now use the Output gain to make-up for any loss in the sound’s overall level.








ADDING PUNCH
The examples given above, for peak-limiting and matching dynamics, are both forms of corrective compression, and can be approached analytically if we know the existing signal level and the desired signal level.
Envelope shaping is a form of enhancing compression. It is commonly used to give a sound more punch or impact by using the gain reduction to alter the shape of its envelope, as we saw in the examples given earlier with envelopes A to D. It is a subjective process that cannot be approached as analytically as corrective compression. However, it is not based on guesswork: a strategic approach will always get us there faster than knob-twiddling and wish-casting.
Envelope shaping is primarily used on sounds with percussive envelopes that tend to feature one note or chord at a time, such as drum hits and plucked strings where each note is a clearly defined event and the shape of each note’s envelope does not change significantly from one note to the next, as shown below:













The envelope begins with a fast attack transient (drum stick hitting drum skin, plectrum/finger plucking guitar string, etc.) that quickly decays to a level that is briefly sustained by the resonance of the instrument’s body (drum shell, acoustic guitar body, etc.) before fading away (the envelope’s release).
The goal here is to spread the gain reduction over the envelope’s attack, decay and sustain, reducing the level of the transient while also increasing the consistency of the sustain. Reducing the level of the transient allows us to increase the overall level and give the sound more impact, while compressing and maintaining the level of the sustain reduces perceived level differences from note to note to create a more consistent performance.
For punchy envelope shaping we need to set up our compressor with peak sensing, and start with a high threshold, a high ratio, a hard knee, and the fastest attack and release times possible. These exaggerated settings will allow us to see the GR being applied as soon as the envelope’s level goes above the threshold, and see it being removed when the envelope’s level goes below the threshold.













We start by selecting an excerpt of the audio, long enough to contain four or five typical examples of the note/envelope that we want to shape. Looping the selection, we lower the threshold until we start seeing movement on the GR meter. The goal is to set the threshold low enough to affect the sustain part of each note’s envelope without affecting the envelope’s release, to ensure the compressor is accessing the parts of the envelope that we want to alter but not the rest of it. This means we should see the GR being instantly removed when the sound’s envelope has entered its release – in other words, we should see no gain reduction during the final moments of the sound’s envelope.













With the threshold set, we lower the ratio to 1:1 (so no GR is showing) and then slowly increase the ratio until we hear the individual notes becoming more consistent and impactful; this will typically occur at ratios of 3:1 and higher. We choose a ratio that sounds appropriately impactful while being mindful of overdoing it; the history of novice mixes is littered with super-compressed kick drums that nothing else in the mix can match in terms of impact, meaning they are either way too strong or way too weak in the mix with no happy medium.
With the ratio set, we slow down the compressor’s attack time to allow some of the envelope’s initial transient to pass through unaffected. This will bring back some definition and restore the performer’s articulation from note to note (i.e. expression). For most percussive/plucked sounds we will find that a relatively slow attack time of 5ms to 10ms is ideal. (Note that it is relatively slow compared to the duration of the percussive transient). For this type of envelope shaping a very fast release is usually most effective; it punctuates the end of a note quickly, instantly removing the gain reduction and thereby dramatizing the sense of impact.













The illustration above shows the process of giving a percussive envelope more punch. After determining the threshold and applying the compression, the final step is adding make-up gain. We can see that the first and third envelopes have the same peak level, but the third envelope has a significantly larger surface area on the screen – which ultimately translates to a more impactful version of the same sound. It also looks like it would be delicious if it were ice cream or licorice…
LEVELING
The primary purpose of leveling compression is to keep a sound at or around a certain perceived level (hence ‘leveling’). It is typically used for non-percussive sounds that have a small musical dynamic range (i.e. a small difference between the quietest note or chord and the loudest note or chord), and is very effective for sounds that are intended to add texture rather than play a featuring role in the music – such as string pads, drones, strummed acoustic guitar, harmony vocals, etc. The effect is like subtly riding a fader up and down a dB or two throughout the mix to prevent a sound from becoming too quiet or too loud. Most situations that require leveling can be solved with automation, but, unlike fader riding, leveling has the added benefit of minimising expression inversion.
For leveling we need to set up our compressor with RMS sensing because we are essentially manipulating the perceived level of a sound within the mix. We will start with the lowest threshold, the highest ratio, a hard knee and the fastest attack and release times possible. As with the previous examples, after we get the threshold and ratio right we will adjust the other controls by ear and by eye.













Theoretically we would set the threshold to the signal’s median level throughout the mix, and adjust the compressor to increase levels that fall below the threshold and decrease levels that rise above the threshold – just as we would do if riding a fader. However, this series focuses on downward compression, which can only apply gain reduction, i.e. it can reduce levels that are above the threshold but it cannot increase levels that are below the threshold. So we have to find another way to achieve a similar result. How? We’ll use a lower threshold and a soft knee to create a similar result by placing the signal’s median level somewhere in the middle of the knee, where it is already getting some gain reduction. Levels that fall below the median will get less gain reduction and sound as if they are being turned up, and levels that rise above the median will get more gain reduction and sound as if they are being turned down.
The process for finding the threshold is essentially the same as we used for matching dynamics. We start by selecting an excerpt of the audio that contains the lowest level of interest, i.e. a part of the performance that would otherwise need to be turned up in the mix to maintain a consistent level. Looping over the excerpt with the threshold at its lowest, we should see high levels of GR. Now we slowly raise the threshold until the GR stops or decreases significantly as the loop passes through the lowest level of interest. The threshold should now be sitting just above the lowest level of interest. Ideally, we will see no GR for signal levels below this threshold, but significant GR for signal levels above it.
To set the ratio we need to select an excerpt of the audio that includes the loudest part of interest; i.e. a part of the performance that would otherwise need to be turned down in the mix to maintain a consistent level. Looping over the excerpt we now lower the ratio until we are seeing the appropriate amount of gain reduction at the loudest part of interest. Most sounds that benefit from leveling typically only need a few dB of gain reduction to achieve the desired result, so let’s use that as a starting point: we’ll adjust the ratio so that we are seeing 3dB on the GR meter as the loop passes through the loudest part of interest.
With the threshold and ratio set, we now play through the sound while watching the GR meter, slowing down the attack and release times considerably so that the GR meter is mostly sitting around the middle of our desired range; for this example we were aiming for 3dB of GR, therefore we’d like our GR meter sitting mostly between 1dB and 2dB, subtly increasing to 3dB for the loudest parts and decreasing to 0dB for the softest parts. All of the changes in gain reduction should be subtle, happening at about the same speed as a finger riding a fader in response to the sound. To get a tactile feel for this, we can place a finger tip on a flat surface and move it up and down as if it was moving a fader during those moments; the movements of the GR meter should hopefully mimic the movements of our finger tip. If so, we’ve got our settings about right.
We now tweak the knee control from hard to soft, listening for the most appropriate result while watching the GR meter. We may need to do some minor tweaking to the attack, release, ratio and knee controls, but by this point we have a good idea of what changes to make based on what we’re seeing on the GR meter and what we’re hearing with the sound.
Note that because we are using downward compression there will be an enhancement of the sound’s impact at the higher levels of gain reduction, as discussed earlier in this instalment. However, these ‘impact enhancements’ should be very subtle and help to push the performer’s expression. There is a reason why the performer played a bit louder in some parts of the performance and a bit softer in other parts, and the subtle enhancement of impact offered by the leveling maintains that ‘lean in/lean out’ expression without allowing the sound to get too loud or too soft – thereby helping the mix to serve the music, rather than forcing the music to serve the mix.
GET ON WITH IT…
Hopefully, this brief introduction to compression has brought you to the threshold of attacking any dynamic range problem and releasing some truly great mixes. For further reading and application tips, be sure to read the fifth instalment of the Mixing With Headphones [coming soon] series, where the use of dynamic processing to match the dynamics of individual sounds and create an appropriate dynamic perspective for the mix is explored in depth…



Return to first instalment…


OPTICAL LEVELING
The opening illustration used throughout this series is based on the iconic Teletronix LA-2A Leveling Amplifier from 1962, a refined version of the earlier LA-2 and LA-1 and predecessor to the LA-3A, LA-4 and LA-5. Note that, on the front panel at least, it differs to the VCA compressors discussed in this series because it has none of the familiar controls we’ve discussed. There is no threshold, ratio, knee, attack or release controls. What’s the deal?
Dynamics processing was relatively new to the audio industry in the early ‘60s, and the LA-2A hit the market before naming conventions had standardised for the different types of dynamics processors or their controls. The LA-2A was designed for leveling so Teletronix sensibly called it a leveling amplifier, hence the ‘LA’ designation.













In today’s terminology the LA-2A is an optical compressor, which means it uses a light source and a photocell to provide the gain reduction. Essentially, the light source is equivalent to the Envelope Follower used in the examples throughout this series, and the photocell is equivalent to the VCA. The input signal passes through the photocell to the output, but it also controls the brightness of the light source. Increasing the brightness of the light source alters the photocell’s impedance in a way that results in more gain reduction. Decreasing the brightness of the light source alters the photocell’s impedance in a way that results in less gain reduction.













It’s also worth noting that the LA-2A’s front panel controls don’t follow the typical left-to-right input-to-output signal flow placement. The Gain control on the left is equivalent to the Output Gain control discussed in this series, and offers up to 40dB (!) of make-up gain to restore the signal’s level after compression. The Peak Reduction control on the right is essentially the threshold control; rotating clockwise results in more gain reduction by sending more of the input signal to the light source. A rotary switch in the upper right corner selects what the VU meter shows: the centre position selects gain reduction, while the options either side of centre show the output signal level relative to different reference levels.
There are no attack, release or knee controls because the LA-2A’s application of gain reduction is dependent on the inherent reaction times of the light source and the photocell. The attack time is determined by how fast the optical circuit responds to increases of light level, while the release time is determined by how fast the optical circuit responds to decreases of light level. Neither of these responses is linear; in fact, it is their inherent non-linearities that make the LA series of leveling amplifiers so popular for musical applications.




The post Understanding Compression 4 appeared first on AudioTechnology.