You are here

Compressor Topology

Feed-forward Or Feedback? By Paul White
Published February 2023

Back & ForthPhoto: Universal Audio

Feedback and feed‑forward compressors can sound very different. Which type is right for which job?

Most of us have a pretty good idea of what a compressor does, but a less widely understood difference between devices is the choice of a feed‑forward control circuit or a feedback one. To help you understand the pros and cons of each approach, I’ll recap the basic building blocks of analogue compressors before offering some thoughts specifically on feedback and feed‑forward compression.

Compressor Basics

At the heart of any analogue compressor is some type of gain‑control element. You can visualise this as a kind of automated volume knob, though in reality there are no moving parts. (Technically the electrons move but you get my drift!) During the evolution of compressors, we’ve seen various types of gain‑control circuitry, including variable‑mu valves, photocell‑and‑lamp arrangements, diode bridges, FETs (Field Effect Transistors) and solid‑state VCAs (Voltage Controlled Amplifiers). Not all of these offer linear gain control, meaning the output gain may not exactly follow changes in the control signal. (These non‑linearities can contribute in a subjectively positive way but that’s another story.) The most linear of the gain‑control elements mentioned above is the VCA, which is why so many commercial compressors use them; their behaviour is predictable. In the software world it’s possible to create very linear compressors too, or to emulate the behaviour of VCAs or that of less linear hardware. But to avoid having to keep repeating ‘gain control element’, I’ll use the term ‘VCA’ from here on.

The part of the circuitry that instructs the VCA when to bring down the gain is called the side‑chain. This takes an audio signal as its source, and generates a signal that follows the envelope of the audio waveform. Usually, the source of the audio used to feed the side‑chain comes from somewhere within the compressor itself — it’s based on the source you’re processing — but when setting up things like ducking it can be taken from an external source, assuming the compressor has a dedicated external side‑chain input.

Further circuitry is used to set a threshold level above which processing will take place, and also to adjust the compressor’s attack and release times, which determine the speed of its response to changes in the signal level. A ratio controls adjusts the strength of the compression, with higher ratios leading to more compression. For example, a 5:1 compression ratio means that for every 5dB that the input signal level increases above the threshold, the output level increases by just 1dB. There may also be further processing to change the linearity of the processing or to achieve soft‑knee compression, whereby the intensity of compression (the ratio) increases as the signal level increases, rather than a fixed amount of gain reduction being applied as soon as signals cross the threshold.

Because low frequencies in a musical signal are generally much higher in level than high frequencies, compressors can often appear to dull the sound: the low end tends to dictate where gain reduction is applied. Engineers will often set a slightly longer attack time to allow the high‑frequency detail within transient sounds to pass before significant attenuation occurs, or might filter low frequencies out of the side‑chain signal. Over the years, there have been various other design strategies to get around this issue, including the Drawmer DL241 in which a small amount of high‑frequency signal was allowed to bypass the VCA stage so as to help preserve high‑frequency detail during heavy compression, without resorting to multiband compression.

Ear & Now

As far as this article is concerned, the main point of interest is from where in the compressor’s internal signal path the side‑chain signal should be ‘tapped’. There are two options: the side‑chain can look at the signal’s level before it enters the VCA stage, or after it leaves it. If taken from before the VCA, the topography is said to be ‘feed‑forward’ and if from after the VCA it’s ‘feedback’.

To help you appreciate the difference, imagine a time before compressors. An engineer operating a console fader would listen to the audio coming from the monitors/headphones and move the fader to compensate for excessively loud or excessively quiet sections. Because a human can’t react as quickly as electronic circuitry, this is a somewhat imperfect way of controlling short signal peaks but it can work well enough for controlling average levels over a longer timescale. The important point isn’t that human delay (which isn’t present in a compressor responding to an electrical signal) but that the engineer is reacting to a signal level they’ve already altered using the fader: this is a ‘feedback’ approach.

In a feedback compressor, the control signal is taken from after the gain‑reduction stage...In a feedback compressor, the control signal is taken from after the gain‑reduction stage...

To set up a 'feed‑forward' system using the same engineer, we’d need the monitors to play a signal taken from before the fader. In this setup, the engineer would be able to make adjustments based on the input signal level, but they’d have no way of knowing what the output signal was actually doing. That being the case, they might underreact and not turn the loud parts down enough, or may overreact and turn them down too much. They’d have no way of hearing the effect of the level changes until they played back the mix that was recorded to tape.

...whereas in a feed‑forward design, the control signal is taken from before the gain‑reduction stage, so the compressor is responding to a different signal....whereas in a feed‑forward design, the control signal is taken from before the gain‑reduction stage, so the compressor is responding to a different signal.

Feed For Thought

Returning to the world of electronics, it’s usually easier for a user to set up feedback compressors because they’re essentially corrective in nature: as the unit is monitoring the output, it automatically corrects any level errors, including those caused by non‑linearities in the gain‑reduction element’s response to the control signal.

The feed‑forward arrangement might strike you as being rather hit and miss, but many very successful compressor designs work this way.

The feed‑forward arrangement may strike you as being more hit and miss, yet many successful compressor designs work this way, and this raises a question: how can compressors be effective if they don’t get any feedback from the actual output signal? The answer has two parts. First, that they need very precise electronics to make the control signal level linearly proportional to the gain‑reduction level. (That’s why VCA‑based compressors are well suited to this topology, and early compressors were less so.) Second, the control loop does include a form of feedback, since, as with the finger‑on‑fader example, the engineer’s judgements about the compressor settings are based on listening to the compressed signal. To make that possible, a good feed‑forward compressor really needs to offer the engineer a comprehensive set of controls. But by carefully adjusting the attack, release, threshold and ratio of a feed‑forward compressor, the engineer can get anything from gentle compression to really fast and assertive gain reduction. As the control signal is derived from the input, the attack and release controls behave very predictably. And the feed‑forward arrangement also makes it possible to create a so‑called ‘look‑ahead’ compressor/limiter, in which the side‑chain gets a sneak preview of what’s coming up, giving it a millisecond or two to anticipate incoming peaks and so act earlier where required.

In theory, a feedback compressor’s attack and release times are less precise, though it’s not a major issue in practice. It’s perhaps worth mentioning, though, that if you were to mult out the input signal and route the duplicate to a feedback compressor’s side‑chain input, it would act as if it were a feed‑forward type: the external side‑chain doesn’t form part of any feedback loop. A more practical issue with feedback compressors can be long release times leading to very brief periods of excess level (relative to the rest of the signal) not exceeding the threshold, and thus ‘sneaking past’ the gain reduction.

In subjective terms, feedback compression tends to sound smoother and more musically satisfying. Indeed, many classic FET, valve and opto compressors were feedback types and are still prized for their ‘musicality’. While they may not be as good when you need fast and snappy compression, and can’t be used as brickwall limiters, for general gain control you know that your audio is in a safe pair of hands. Feed‑forward designs have different advantages: they tend to sound tighter and more punchy, so can often be better choices for hard drum and bass guitar compression. In fact, in the hands of a good engineer, a feed‑forward compressor can be a formidable tool, particularly when you need to catch fast transients or to impose hard limiting. This means they’re also well suited to classic parallel compression, in which very heavy compression of the peaks/transients is required.  

Popular Feedback & Feed‑forward Designs

There are more feedback compressors than I care to count, but popular models include almost all vari‑mu and optical designs, the SSL G‑series bus compressor, the UREI 1176 and the classic Neve models such as the 2254. Though less common, there are some very popular feed‑forward compressors too, including the dbx 160 (and its descendants) and Elysia’s MPressor. If you can’t decide which type of compression is best for a specific job, some models, notably from API and Rupert Neve Designs, which can be switched between feed‑forward and feedback operation. In the software world, of course, there’s even more choice...