Available compressor circuits?

[fryingpan]

I'm looking to design my own take on a compressor, inspired by my own mixing practices on bass. Namely, it is common to clip (soft or hard) after compressing with not too short of an attack to avoid completely sucking the life off a track, with the light clipping of the peaks shaving off another few dBs without sounding obvious and actually enhancing the attack itself and maintaining playing dynamics without them being overbearing. Now, I've seen quite a few designs, but none seem to meet my basic requirements for the compressor per se. I'm not interested in the finished product, merely the compressor stage itself.

Most pedal forms are built around:

- a vactrol of some sort (maybe home built): these might have the right attack times (usually around 6-10ms, which in a feedback configuration is kinda slow) but release times are rather high (1.5-2s) while I'd like the option of much shorter release times (let's say, starting from 250ms)

- an OTA (which should allow for the right attack and release times, but at the expense of lots of distortion and noise; placing them in the feedback path of an opamp may cure this issue, but almost all of the OTA-based compressors I've seen still have long release times)

- a VCA (which of course is the more versatile and also "clean" of the technologies)

- a FET working as a variable resistor of sorts (like in an Orange Squeezer, I think?), although I don't know what kind of attack and release times this topology can afford.

Could you point me to some designs that may help me out?
Again, my priorities are:

- low noise
- fast release
- the option of a moderate ratio (the OTA-based comps I've seen are all limiters, while I'd be more interested in the 3:1-10:1 range)*
- generally speaking, uncoloured frequency response (many compressors cut bass a lot and push the treble)

Slowing down the attack is quite easy, usually (there is some lowpassing involved in the sidechain, generally), so I imagine this won't be a problem. Really, I'd love an optical compressor if only it could be made a bit faster.

*: I'm thinking that a limiter could be made into a milder compressor if the sidechain were soft-clipped above the threshold?

[ElectricDruid]

The thing about optical compressors, at least the classic ones, is that the time constants you get are...well, they're the ones you get! You don't get any *choice* about it beyond what LDR you use. In the old days, studio compressor manufacturers could get something made to order, but that's probably beyond our means, so we're left with whatever we can find.

If you can find something that suits you and your playing style, it'll be great. If not, forget it. What you won't get is a lot of adjustability on the Attack/Decay.

[Fancy Lime]

For low noise: FET
For variable attack and release from instantaneous to slow: FET
For slow attack and/or variable ratio: feed forward side chain
For uncolored frequency response: well, just don't color the frequency response and don't overdo it with the compression.

I have had good experiences with a J112 in the ground leg of a non-inverting opamp stage. That is pretty much the least noisy variable gain stage possible. I used a feedback design but there is no reason why this would not work in feed forward. Make sure to use a full wave rectifier when using a FET as the gain control element and thorough side chain filtering to avoid rippling. These things are MHz-fast.

HTH,
Andy

[andy-h-h]

Are you familiar with the THAT range?  https://thatcorp.com/design-notes/ there's a lot of info in the design notes, all for their products of course, but they do make very good products.

Maybe a bridge diode compressor, care of Jonny Reckless https://www.diystompboxes.com/smfforum/index.php?topic=120686.0.  A few tweaks and it would be fine for bass.  I like the sound of these on guitar, never tried it on bass.

[Mark Hammer]

Historically, we have run the gamut of all forms of control element in compressors.  We have paid lip service to multiband compression over the years, but we have never really explored it to the extent that we have explored many other categories of devices.  For instance, is multiband "better" for some kinds of control elements than others? (e.g., optical vs OTA).  How many bands?  Where should the frequency spectrum be divided up?  How steep should filtering be?  What are the rules of thumb for time constants?  And so on.

Of course, all of that is subject to variation, depending on purpose.  What can make a singer sound powerful, yet intimate, can be very different than what gives a snare punch, or a guitar sound smoother.

Some folks here have deeper knowledge about all this, and I hope they speak up.

[Rob Strand]

Historically, we have run the gamut of all forms of control element in compressors.  We have paid lip service to multiband compression over the years, but we have never really explored it to the extent that we have explored many other categories of devices.  For instance, is multiband "better" for some kinds of control elements than others? (e.g., optical vs OTA).  How many bands?  Where should the frequency spectrum be divided up?  How steep should filtering be?  What are the rules of thumb for time constants?  And so on.

Some of the Trace Elliot bass amps had multi-band compression.   As I recall the two bands don't sum to unity even under steady-state input.

Another compressor variation is dual time constants (on a single band).   This was touched on in a fairly recent thread started by Vivek, re the Rockman compressor (and others).

[merlinb]

- an OTA (...but almost all of the OTA-based compressors I've seen still have long release times)

http://valvewizard.co.uk/engineersthumb4.html
You can make the ratio and release times anywhere from nothing to infinity.

[amz-fx]

I tested a bunch of compressors... start reading here with Part1

There are 3 pages of tests, and in the articles, I cover at least 1 model of each of the types of compressors: ota, fet and opto.

Best regards, Jack

[fryingpan]

Thanks for all your replies. I have a few questions:

- the ET appears to be a limiter. While I do like limiters, I would like the option of a shallower compression (3:1 would be good). The ratio control on the ET is really a blend control, which is OK (parallel compression is often useful) but not exactly the same thing. Also, it appears that the recommended minimum release on the ET is somewhere around half a second? As is, the release is around 2s, IIRC.
- switching the sidechain between feed-forward and feed-back modes shouldn't be too hard, right? Essentially all that matters is where you're taking the signal from, with at most some added amplification for feed-forward? In software, where everything is timed to the clock, race conditions can be a problem, but I don't think it would be a problem in audio even if the feed-forward sidechain were to lag (or predate) the actual compression?
- placing the OTA in the feedback loop of an opamp minimises its distortion and noise contribution to the signal; but wouldn't that noise and distortion become significant if you set your compressor to essentially be "always on", as is common with lower ratio compressors?
- the datasheet for the VTL5C3 vactrol states rise and fall times around 3-6ms and 35-100ms respectively; could I slow down the release like you do, for instance, with an AM demodulator, or a rectifier for that matter?

[Fancy Lime]

Given your conditions of (1) slow(ish) attack and (2) true compression instead of limiting, you definitely want to stay away from feedback side chains. The cannot do (2) without some extra hoops (parallel compression) and suffer from dynamic overcompression (aka pumping) because of (1). This latter effect is kind of the closest equivalent to a race condition. The envelope detector needs to be slowed down to allow a long attack and then it overcompensates and then the signal gets to compressed but the detector reacts to late and so on. And that sounds like pumping. This is not a problem with feed forward designs because there is no feedback (duh!) to overcompensate back into itself. With feed forward you have to be careful with the sensitivity though, lest you get static overcompression (negative correlation between output and input amplitude). However, with sanely chosen thresholds and ratios, this is not normally a big problem.

[fryingpan]

Given your conditions of (1) slow(ish) attack and (2) true compression instead of limiting, you definitely want to stay away from feedback side chains. The cannot do (2) without some extra hoops (parallel compression) and suffer from dynamic overcompression (aka pumping) because of (1). This latter effect is kind of the closest equivalent to a race condition. The envelope detector needs to be slowed down to allow a long attack and then it overcompensates and then the signal gets to compressed but the detector reacts to late and so on. And that sounds like pumping. This is not a problem with feed forward designs because there is no feedback (duh!) to overcompensate back into itself. With feed forward you have to be careful with the sensitivity though, lest you get static overcompression (negative correlation between output and input amplitude). However, with sanely chosen thresholds and ratios, this is not normally a big problem.

The idea would be that, by switching the compressor to its feedback mode, the attack becomes faster (or, by switching it to feedforward mode, the attack becomes slower, I presume adding a lowpass somewhere in the chain?). Feedback compressors are kind of slow anyway even if the modulating element is fast.

Feedback compressors don't necessarily have to be limiters (eg. LA-2A, 1176). Most vintage studio units allowed for a variety of ratios and were all feedback compressors without employing parallel compression. (In fact, feedback compressors can never be true limiters, because the true maximum ratio is somewhere around 5:1, something more with gain in the sidechain).

[Rob Strand]

(In fact, feedback compressors can never be true limiters, because the true maximum ratio is somewhere around 5:1, something more with gain in the sidechain).

It's possible to get over 10 with ideal rectifiers and hard thresholds but soon as you add a diode drop (or VBE junction) in the detector the CR tends to be 5 to 10 depending on the softness of the diode.

The old Philips NE570/NE571 notes had a limiter with the comparator at the output.  This produces a true limiter because it essentially has infinite gain.  See, figure 11.
https://www.onsemi.com/pub/Collateral/AND8227-D.PDF

I dislike comparators for transient signals because when the feedback overloads during a transient the gain control is driven very hard and that makes the feedthrough as bad as you could possibly make it.

Generally feedback designs have more risk of something overloading.  Once overload happens you get a whole lot of side effects and things don't get back to normal until the feedback is back in control.   The feedforward designs tend to avoid that.

The idea would be that, by switching the compressor to its feedback mode, the attack becomes faster (or, by switching it to feedforward mode, the attack becomes slower, I presume adding a lowpass somewhere in the chain?). Feedback compressors are kind of slow anyway even if the modulating element is fast.

Attack time is increased decreased and Release time is extended.   You need to set the actual time constants differently to get the same observed performance as a feedforward.

[merlinb]

Thanks for all your replies. I have a few questions:

- the ET appears to be a limiter. The ratio control on the ET is really a blend control, which is not exactly the same thing.

Compression can be as soft as you like, just use small values of ratio control (e.g. zero to 20k).

Also, it appears that the recommended minimum release on the ET is somewhere around half a second?

Release in most compressors is just an RC, so you can make it as fast or slow as you like. You could make it instantaneous if you reduce the time constant to zero! The recommended minimum is simply a reflection of the fact that when the time constant approaches the time period of the audio itself, you move away from compression and into distortion. That's just physics.

- switching the sidechain between feed-forward and feed-back modes shouldn't be too hard, right?

I don't see why not.

- placing the OTA in the feedback loop of an opamp minimises its distortion and noise contribution to the signal; but wouldn't that noise and distortion become significant if you set your compressor to essentially be "always on"

Not sure what you mean by this.

[fryingpan]

Thanks for all your replies. I have a few questions:

- the ET appears to be a limiter. The ratio control on the ET is really a blend control, which is not exactly the same thing.

Compression can be as soft as you like, just use small values of ratio control (e.g. zero to 20k).

If the ratio is really a dry/wet control, you are simply blending the limiter's action with the clean signal. Parallel compression, which sounds a bit different to low ratio compression. (Namely, you are superimposing the compressor's envelope to the clean signal instead of "highlighting" the character of the compression).

Also, it appears that the recommended minimum release on the ET is somewhere around half a second?

Release in most compressors is just an RC, so you can make it as fast or slow as you like. You could make it instantaneous if you reduce the time constant to zero! The recommended minimum is simply a reflection of the fact that when the time constant approaches the time period of the audio itself, you move away from compression and into distortion. That's just physics.

Oh, OK. Fine.

- switching the sidechain between feed-forward and feed-back modes shouldn't be too hard, right?

I don't see why not.

- placing the OTA in the feedback loop of an opamp minimises its distortion and noise contribution to the signal; but wouldn't that noise and distortion become significant if you set your compressor to essentially be "always on"

Not sure what you mean by this.

If you set the compressor so that it is on constant gain reduction, then the distortion and noise of the OTA would become apparent?

[merlinb]

Parallel compression, which sounds a bit different to low ratio compression.

Maybe you can hear the difference, I'm not sure I can   

If you set the compressor so that it is on constant gain reduction, then the distortion and noise of the OTA would become apparent?

It only exists while the signal is above threshold and actually being compressed, and you want the signal to compress hence why you are building a compressor, so I still don't follow. In any case the noise and distortion are extremely low for a guitar compressor pedal.

[PRR]

> If the ratio is really a dry/wet control

It's not. If it was, the limited output would rise at high input level. (I know mix-dudes who do that for effect, specially for Bass, but it is technically wrong.)

All the parameters are continuums. Light compression to heavy limiting and even reverse limiting (more in makes less out). Time constant of zero is, as said, blunt distortion, but a single RC makes a useful filter, dual-TC filters may fool the ear better, exponential-domain filters do it different, and there is no need to use a simple filter.

Fer instance: Set a comparator 1dB below signal-chain clipping. Set a CPU to read it every milllisecond. Every "over", the CPU cuts gain 1dB. Even quite large overs will be tamed before the ear gets real annoyed (or a 2nd-order ratchet can goose for serious overs). We don't just want a gain ducker though. Say every 200mS we add-back 1dB. Or if the post transient level goes below -30dB, do NOT add-back gain right away (noise pump).

Niggling about the gain-control elements just means you have the wrong technology. For most of the 20th century we didn't! Blesser resorted to a PWM scheme as the best of a bad lot. After early iterations, dBx/THAT VCA are the bees-knees of analog. Even before that, I quit fooling with analog and did all my mooshing in the DAW. (Not always cool for live tracking.)

Blesser, B. (1969). Audio dynamic range compression for minimum perceived distortion. IEEE Transactions on Audio and Electroacoustics 17:22-32.

Bäder, K. and Blesser, B. (1968). Ein Kompressoren mit variablen Eigenshaften und Pulsedauermodulation. Radio Mentor Electronics, 9:33-34.

Blesser, B. and Kent, A. (1968) Analysis of a feedback-controlled limiter using a logarithmic measuring scale. IEEE Transaction of Audio and Electroacoustics 16:481-485.

[fryingpan]

> If the ratio is really a dry/wet control

It's not. If it was, the limited output would rise at high input level. (I know mix-dudes who do that for effect, specially for Bass, but it is technically wrong.)

I'm a recording engineer myself. Still, I dont know what you mean with "the limited output would rise at high input level". I think that's what the idea is about blending dry and wet: instead of having a limited output, transients can pop over the threshold. Parallel compression (as I said). I'm not a great fan of parallel compression, I prefer using compressors 100% wet and set them right/choose the right kind, but the good thing with parallel compression is, essentially, that you can put any bad compressor to good use, just overcompress and add in just a smidge over the dry signal, and there you get a nice fat yet dynamic track (folks do this a lot on drums). The Engineer's Thumb, according to its creator, like all other (OTA?) compressors is really a limiter and there is no real way to reduce the ratio. I'm not familiar with OTAs (they are a somewhat specialist part, they're noisy, they have very little range due to the fact that they distort following an arctan curve for input differences over 50mV or so). If I get it right, they can be used as a variable resistor. Therefore, the ET essentially has a ratio control setting the amplification of the opamp when the OTA is in its high Z state (maybe?) and it becomes some sort of unity gain buffer when the OTA switches to a low Z state, since it's parallelled with the former. If this is so, I frankly don't get why it should be a blend control, since the ratio is effectively controlling... the ratio of amplification when the opamp *isn't* a unity buffer.

All the parameters are continuums. Light compression to heavy limiting and even reverse limiting (more in makes less out). Time constant of zero is, as said, blunt distortion, but a single RC makes a useful filter, dual-TC filters may fool the ear better, exponential-domain filters do it different, and there is no need to use a simple filter.

Fer instance: Set a comparator 1dB below signal-chain clipping. Set a CPU to read it every milllisecond. Every "over", the CPU cuts gain 1dB. Even quite large overs will be tamed before the ear gets real annoyed (or a 2nd-order ratchet can goose for serious overs). We don't just want a gain ducker though. Say every 200mS we add-back 1dB. Or if the post transient level goes below -30dB, do NOT add-back gain right away (noise pump).

Niggling about the gain-control elements just means you have the wrong technology. For most of the 20th century we didn't! Blesser resorted to a PWM scheme as the best of a bad lot. After early iterations, dBx/THAT VCA are the bees-knees of analog. Even before that, I quit fooling with analog and did all my mooshing in the DAW. (Not always cool for live tracking.)

Blesser, B. (1969). Audio dynamic range compression for minimum perceived distortion. IEEE Transactions on Audio and Electroacoustics 17:22-32.

Bäder, K. and Blesser, B. (1968). Ein Kompressoren mit variablen Eigenshaften und Pulsedauermodulation. Radio Mentor Electronics, 9:33-34.

Blesser, B. and Kent, A. (1968) Analysis of a feedback-controlled limiter using a logarithmic measuring scale. IEEE Transaction of Audio and Electroacoustics 16:481-485.

I fail to see the point. If the idea is that digital technology is much more flexible, no argument from me there. Of course it is. If you're mixing, you also get the benefit that you can, potentially, ask the compressor to "think carefully" about what to do, based on what's about to come, before doing anything (which is how you get extremely effective yet transparent limiters, for instance). If we really want to nitpick, this doesn't even really require digital technology. Analog computers are a thing (what's a compressor if not a very simple one? But you can make better ones. In certain domains, analog computers are all we have really).

Still, funnily enough, all the rage today is modelling precisely old analog technology, with essentially all software developers in the field competing for the latest and greatest <insert coveted vintage piece of gear>. People these days are adding noise and distortion to tracks that were an order of magnitude clearer and cleaner than anything you could manage to capture on tape. Do I approve of this? Not necessarily. But sometimes imperfections mean character.

[PRR]

> I'm not familiar with OTAs (they are a somewhat specialist part, they're noisy, they have very little range due to the fact that they distort following an arctan curve for input differences over 50mV or so).

Merlin uses his OTA "backward". Many of the objections get different. The OTA does not add to idle-hiss, for example.

Analog computers inspired my father--- my sis just found one of his early articles as she was moving the folks to a new home. I maintained an early synth which was just an analog computer. I yield to your obviously superior experience on the use of the machines. But I do know more than a bit about the back-ends.

Really: find and digest the main Blesser paper. Everything from noise-gate to brick-wall is the same "machine" with different settings.

[Rob Strand]

I'm a recording engineer myself. Still, I dont know what you mean with "the limited output would rise at high input level". I think that's what the idea is about blending dry and wet: instead of having a limited output, transients can pop over the threshold. Parallel compression (as I said). I'm not a great fan of parallel compression, I prefer using compressors 100% wet and set them right/choose the right kind, but the good thing with parallel compression is, essentially, that you can put any bad compressor to good use, just overcompress and add in just a smidge over the dry signal, and there you get a nice fat yet dynamic track (folks do this a lot on drums).

If you blend some clean through it basically tilts the gain curve upward and reduces the CR at higher input levels.   

Professional limiter/compressors have a true CR control but for simple compressors the shape of the CR profile is largely fixed.  The compression adjustment is more like moving the CR profile left and right (largely variable threshold).   Also, the ultimate compression ratio for higher levels is fairly high and is pre-determined by the simple circuit.  By sliding the CR profile left/right you kind of tune the effective CR since only the high level signal makes it into the high CR zone.   With these types of compressors a clean blend lets you get more control of the CR as the high level signals are no longer subject to the inherent high CR.

One advantage of using clean blend to reduce the CR is under high levels the clean path stays clean so even if the compression path clips under high inputs some clean always gets through.   Studio compressors use high supply rails so it's not such a big deal, although you can overload OTAs, but small 9V pedals have limited headroom.

The Engineer's Thumb, according to its creator, like all other (OTA?) compressors is really a limiter and there is no real way to reduce the ratio. I'm not familiar with OTAs (they are a somewhat specialist part, they're noisy, they have very little range due to the fact that they distort following an arctan curve for input differences over 50mV or so).

The Engineer's Thumb actually has an implicit blend due to the non-inverting gain configuration.  No matter how much feedback is applied the input signal must leak through with a gain of unity.

As PRR mentioned the flip trick on the Engineers thumb can help reduce the perceived noise.   Modern OTAs are pretty much at the limit of semiconductor physics for noise.

FYI: this is a good paper on compressors

https://ajoliveira.com/ajoliveira/gen/pdf/preprints/paris88.pdf

It was also published in the main AES journal, in a neater format.

[Eb7+9]

was hoping one day to see real measurements, preferably yours ...

and, golly ... what do we see ??!


looks like ALC transfer curves to me



now, the idea of a soft knee shouldn't be mistaken with the profile of low ratio ALC curves // them's two different things ...
technically speaking

- switching the sidechain between feed-forward and feed-back modes shouldn't be too hard, right?

I don't see why not.

**puff of smoke suddenly appearing // just like magic ...


in feedback mode you will NOW get the so-called Compressor mode curves (response)


at this point you might have a compelling reason to actually/finally call it a compressor, not that it matters that much
but, again, technically speaking etc ...


now, maybe you can plot out that FBK response set for us then ...


hint: it should look something like what EBK posted here:

https://www.diystompboxes.com/smfforum/index.php?topic=128654.msg1240074#msg1240074

---

some of you will still be asking what ALC circuits are used for in general electronics ...
looking back at my posts you will see a dozen or so references to the ALC topology in the context of this not-uncommon idea

(consider it a quiz question)

which, btw this is basically the same overall structure as Figs. 6 and 7 in Oliveiri's paper Robb referred to above
but actually more like the SE/NE-570/571/572 datasheet compressor and ALC circuit examples

anyway, at least now we have a set of FFD curves ... let's hope for its complimentary set one day