Improving my compressor design

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

vixo

Member
Joined
Jan 23, 2018
Messages
5
I was playing around with FET compression circuits recently, after I read this article - VCAs (section 7). The FET circuit does indeed work but I was only using a fairly general purpose 2N3819.

I was pointed to this forum as a good resource, does anyone know any resources I could use to further my knowledge specifically in FET - or opto - compressor design? i am interested in finding how to select FETs for great performance (even though I realise they are now hard to find) and design principles for making great sounding circuits. I'm fairly competent with analogue electronics but new to compressor design. if anyone could link past threads that would be great, i haven't had much luck finding them for some reason

I noticed in particular these designs are quite noisy and the switching can be a bit rough, it would be good to know how great FET compressors get around this

any help much appreciated
 
To keep distortion low, it is generally needed to keep the voltage over the FET low, so you will need relatively much amplification to bring the output level up to line level. The low level over the FET and the needed amplification after the gain reduction circuit will lower the signal to noise ratio.
If you are looking for replacement FET's, it is very important to match the FETs, if you want good stereo tracking.
If there is a second FET in the gain reduction metering circuit, you will even need a 'quad' matched FETs! (For stereo.)
 
vixo said:
does anyone know any resources I could use to further my knowledge specifically in FET - or opto - compressor design?
Study the schematics of all the revered units, 1176, LA2/3, Dyna-Mite, ...
You may also want to read the THAT papers; although they are mainly centered around their VCA and RMS chips, there is a wealth of information at the bend of the path.
Also read the wiki articles about the FET.
Unfortunately, the FET manufacturers litt is quite poor about distortion reduction.
 
I've been on a similar kick so I welcome the discussion.

It seems to me the sidechain is not so many parts..

threshold control
rectifier
time constant controls
DC level shift / biasing of control voltage
some way to introduce 1/2 of the drain voltage on the gate in addition to the control voltage to reduce distortion

Is this about right?
 
dogears said:
I've been on a similar kick so I welcome the discussion.

It seems to me the sidechain is not so many parts..

threshold control
rectifier
time constant controls
DC level shift / biasing of control voltage
some way to introduce 1/2 of the drain voltage on the gate in addition to the control voltage to reduce distortion

Is this about right?
The side chain is where the magic happens... In my judgement the dominant defining characteristic of a dynamics processor is the side chain. The control voltage effectively multiplies the audio signal through manipulation of the gain control element. Of course if the gain element is bad enough we can hear the noise/distortion added by it alone, but in general the gain element impact is secondary, to the side chain manipulations. 

That said there are many ways to skin that cat (side chain design)and no single "best" approach because the results of dynamics processing is subjective.  YMMV

JR
 
Yes, John - Sidechain IS the compressor.

But in terms of anatomy, is the list I put together correct? Granted it may be like describing an NFL football player as skeleton, muscles, helmet, and shoes.  :p
 
dogears said:
Yes, John - Sidechain IS the compressor.

But in terms of anatomy, is the list I put together correct? Granted it may be like describing an NFL football player as skeleton, muscles, helmet, and shoes.  :p
it seems "about" right...  that is why we breadboard designs... kids today sim them...

JR
 
So, when you look at the 1176 sidechain for example...

Where's the magic? The attack and release controls are boilerplate... attack resistor series, cap, release resistor parallel.

Ratio is a front end threshold combined with a bias just before the rectifier diodes. That may be magic?

Is the rectifier special/magic?
 
dogears said:
So, when you look at the 1176 sidechain for example...

Where's the magic? The attack and release controls are boilerplate... attack resistor series, cap, release resistor parallel.

Ratio is a front end threshold combined with a bias just before the rectifier diodes. That may be magic?

Is the rectifier special/magic?
Indeed, the 1176 is almost a textbook design. As is the Fairchild 670. The magic of the 670 is in the fact that no corners were cut. If you compare it to a contemporary StaLevel or UA176, you realize how many corners they cut.
The 1176 offered performance similar (not identical) to a 670 at a much lesser cost.
 
dogears said:
I’d like to read that textbook.  ::)
There are no such design textbooks AFAIK... (I wish there was..... but there are some old tube manuals, and ARRL handbooks with sundry basic designs). In later recent decades there are some design "cookbooks" written by popular technical writers.

Today is so much easier with google and IC manufacturer app notes.

The best we could do back in the day was digest every schematic we could find and try to grok why they did what they did... An ugly (not so secret) secret is that old school design engineers did not have the rich pallet of canned solutions like we do these days. They did the best job they could with the limited tools available to them... Sometimes magic happens like the intrinsic time constants of a luminescent light panel combined with a light dependent resistor that euphoniously work together to make musical compression/limiting that does not suck (an accident of time constants defining attack and release well). The design engineers only wish they did it on purpose****. 

JR

*** I copied one such serendipitous lucky combination back in the 90s when Peavey made a tube comp/limiter that didn't suck (VCL).
 
Alright, here's another one then. In the grand tradition of grokking schematics.

We see here a threshold amplifier - since this takes the AC input drain signal before the makeup amp, it gets bumped up a bit. Then a transistor to provide both phases, and diodes to release into the sidechain. Simple attack and release (though not the usual shared cap - we have a release cap and pot followed by an attack cap and pot) followed by the interesting bit to me - germanium diode into a 1meg pot for ratio.

When I look at this I'm left to wonder - if people had ridiculously cheap op amps in 1979, would they have ditched the threshold amp and transistor and just gone with a simple two-op amp full wave rectifier with gain?  Is there any practical difference?

And - with super-cheap op amps, how do you improve the ratio setup?
 

Attachments

  • COMP - Trident CB9146.jpg
    COMP - Trident CB9146.jpg
    162.5 KB · Views: 123
dogears said:
Simple attack and release (though not the usual shared cap - we have a release cap and pot followed by an attack cap and pot)
Where do you see that?  What schemo are you referring to?
Actually several designs use two cascaded time-constant. The goal is to tune the side-chain's transient response. The standard rectifier is a 1st-order filter. It is tempting to increase the order of the filter since higher order filters can be tuned to present a faster rise-time (faster Attack) for a given characteristic frequency (Release). One must be aware that this can result in unstability in a feedback compressor. Another issue is that it makes continuous variability almost impossible in practice.

When I look at this I'm left to wonder - if people had ridiculously cheap op amps in 1979, would they have ditched the threshold amp and transistor and just gone with a simple two-op amp full wave rectifier with gain?  Is there any practical difference?
One advantage of using opamps for the rectifier is that is allows building an almost perfect, threshold-less rectifier. Is it a welcome feature? It depends. Since a threshold has to be included in a compressor, it may as well be in the rectifier, anyway there is nothing substantially wrong with having a threshold there. An additional variable threshold may be added somewhere for adjustability. That's the case in the 1176, although the threshold is operationally controlled by the Input pot. However, the ratio selector is also a threshold control.
In the 90's I designed a compressor for which I chose to use a threhold-less rectifier. The schemo is in the LA AUDIO section of the Technical documents https://groupdiy.com/index.php?topic=49634.msg762968#msg762968
Look around U5/6/7 &10.
The reason I did that is because I had two rectifiers working jointly; one the standard RMS dbx 2252, the other a conventional two-pot one cap "peak" rectifier. The second 2252 was ther only for providing the decilinear converesio needed for proper drive of the VCA's control port. Had I used a conventional rectifier-with-threshold, the compressor's threshold and ratio controls would have bee all over the place when switching between them - you may note that in fact it was not actually switching, but varying continually between them via a pot.

And - with super-cheap op amps, how do you improve the ratio setup?
What parameter do you want to "improve"?
 
abbey road d enfer said:
Study the schematics of all the revered units, 1176, LA2/3, Dyna-Mite, ...

I'm looking at the 1176 schematic (http://www.gyraf.dk/gy_pd/1176/1176sch.gif), there are a few points i want to make sure i understand - the overall function of the compressor is fairly clear, the input, signal pre-amp, GR control amp controlling the voltage on the BF245A.  A little signal from the Q2 and Q4 seems like it's mixed in there to the gate CV, via the series R21, C5, R9/C4 - is there some magic happening here? I'm not sure the reason for the exact choice of those values and the feedback of both signal from Q2 and Q4 .. can someone could enlighten me?

also if there is some detailed online analysis of this circuit or any others, please share!
 
vixo said:
I'm looking at the 1176 schematic (http://www.gyraf.dk/gy_pd/1176/1176sch.gif), there are a few points i want to make sure i understand - the overall function of the compressor is fairly clear, the input, signal pre-amp, GR control amp controlling the voltage on the BF245A.  A little signal from the Q2 and Q4 seems like it's mixed in there to the gate CV, via the series R21, C5, R9/C4 - is there some magic happening here? I'm not sure the reason for the exact choice of those values and the feedback of both signal from Q2 and Q4 .. can someone could enlighten me?

also if there is some detailed online analysis of this circuit or any others, please share!
It's been hinted many times. In order to reduce THD, the gate must be half-way between source and drain AC-wise.
There are several ways of doing this, it can be done the 1176 way, which has a detrimental effect of delaying the control voltage, but perhaps that's what gives the 1176 it's "magic"?
It's often done by superposing a fraction of the output voltage to the side-chain.
 
abbey road d enfer said:
In the 90's I designed a compressor for which I chose to use a threhold-less rectifier. The schemo is in the LA AUDIO section of the Technical documents https://groupdiy.com/index.php?topic=49634.msg762968#msg762968
Look around U5/6/7 &10.
The reason I did that is because I had two rectifiers working jointly; one the standard RMS dbx 2252, the other a conventional two-pot one cap "peak" rectifier. The second 2252 was ther only for providing the decilinear converesio needed for proper drive of the VCA's control port. Had I used a conventional rectifier-with-threshold, the compressor's threshold and ratio controls would have bee all over the place when switching between them - you may note that in fact it was not actually switching, but varying continually between them via a pot.

Hello Abbey

The peak/RMS blender is a nice feature !
Could we consider this as a knee contol ?

Side question, why including the attack pot (RC) in the release network (R//C)
22k wont affect that much in series with 1M+47k with long release setting
But with shortest release, attack form min to max will change release time about 30%
Is that on purpose ? for some kind of "adaptative" TC ?
More you faster the att more you faster the release too

Best
Zam
 
zamproject said:
Hello Abbey

The peak/RMS blender is a nice feature !
Could we consider this as a knee contol ?
It's really a timing control that gives two different flavors. In one of the prototypes, I had a hard/soft knee control. decided to ditch it after critical listening.

Side question, why including the attack pot (RC) in the release network (R//C)
22k wont affect that much in series with 1M+47k with long release setting
But with shortest release, attack form min to max will change release time about 30%
I prefered that because short attack and release is a receipe for bad sound. Prevents a stupid operator to complain about distortion.
 
abbey road d enfer said:
I prefered that because short attack and release is a receipe for bad sound. Prevents a stupid operator to complain about distortion.

Ok, make sense  :)

Best
Zam
 
abbey road d enfer said:
Where do you see that?  What schemo are you referring to?
Actually several designs use two cascaded time-constant. The goal is to tune the side-chain's transient response. The standard rectifier is a 1st-order filter. It is tempting to increase the order of the filter since higher order filters can be tuned to present a faster rise-time (faster Attack) for a given characteristic frequency (Release). One must be aware that this can result in unstability in a feedback compressor. Another issue is that it makes continuous variability almost impossible in practice.
One advantage of using opamps for the rectifier is that is allows building an almost perfect, threshold-less rectifier. Is it a welcome feature? It depends. Since a threshold has to be included in a compressor, it may as well be in the rectifier, anyway there is nothing substantially wrong with having a threshold there. An additional variable threshold may be added somewhere for adjustability. That's the case in the 1176, although the threshold is operationally controlled by the Input pot. However, the ratio selector is also a threshold control.
In the 90's I designed a compressor for which I chose to use a threhold-less rectifier. The schemo is in the LA AUDIO section of the Technical documents https://groupdiy.com/index.php?topic=49634.msg762968#msg762968
Look around U5/6/7 &10.
The reason I did that is because I had two rectifiers working jointly; one the standard RMS dbx 2252, the other a conventional two-pot one cap "peak" rectifier. The second 2252 was ther only for providing the decilinear converesio needed for proper drive of the VCA's control port. Had I used a conventional rectifier-with-threshold, the compressor's threshold and ratio controls would have bee all over the place when switching between them - you may note that in fact it was not actually switching, but varying continually between them via a pot.
What parameter do you want to "improve"?
I was looking at the one attached to the post. 

I don't really understand the ratio control, beyond the fact that it's a germanium diode giving it some sort of nonlinear effect.
 
dogears said:
I was looking at the one attached to the post. 
Aah, OK, the Trident.
There's a couple of clever tricks in there. The two-cap release circuit is aimed at speeding release when there is a sharp reduction of the input signal. Diode D12 helps discharging C22.

I don't really understand the ratio control, beyond the fact that it's a germanium diode giving it some sort of nonlinear effect.
In case of a sudden blank, voltage at D11's anode becomes low enough for the diode to be less conductive, which reduces the shunting effect on the rectified voltage thus increasing ratio. It results in downward expansion, thus reducing noises that usually appear when compression ceases.
Apparently the person who designed it gave it a lot of thought, I don't know how it translates in actual performance.
Maybe someone would chime?...
 
Back
Top