auto-gain time constant

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

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

dogears

Well-known member
Joined
Nov 15, 2017
Messages
547
Location
Texas
Anyone ever played with makeup gains in a compressor sidechain?

I'm fooling around with a design that takes the "final" sidechain control voltage, splits it into a second very slow RC integrator using an inverting op amp, then sums that signal back to the  control voltage.

kDkOIhZ.png


I know the time constant of the integrator needs to be "very" long - longer than the release time of the circuit, obviously.. but I don't have a feel for where to start at "very". 2 times the longest release? 4? 10? Any help?
 
JohnRoberts said:
That is why we build prototypes...

Listen for how long it needs to be.

JR

I struggle so much with this answer. I’ve found to my chagrin I have crap for ears. And I don’t trust my own subjective opinions at all.

I at least like to start from a rule of thumb or some kind of calculated point or reference to guard a bit.
 
dogears said:
I struggle so much with this answer. I’ve found to my chagrin I have crap for ears. And I don’t trust my own subjective opinions at all.
Sorry this just does not sound like that simple of a modification.
I at least like to start from a rule of thumb or some kind of calculated point or reference to guard a bit.
===
Anyone ever played with makeup gains in a compressor sidechain?
Commonly done to mitigate loss of output level due to above threshold compressions/limiting "only" reducing gain. (Several decades ago I designed a full range compressor at Loft that both boosted and cut to truly compress an audio signal. It was an odd bird and didn't sell well.)
I'm fooling around with a design that takes the "final" sidechain control voltage, splits it into a second very slow RC integrator using an inverting op amp, then sums that signal back to the  control voltage.
This sounds like it would cancel out the slow moving longer term gain reduction, while allowing short term level changes to take effect. The long term correction will keep correcting after the signal subsides, so will have a release time associated with it. The slower you make it the longer it will take to clear. A common problem with compressors is noise floor bloom or apparent boosting of the noise floor. I have even added downward expanders on top of compression to reduce this undesirable noise.

Revisiting your question, effectively how slow to make this. Pick a number and see if it does what you think it does, then make it longer or shorter as needed. I just cannot imagine designing any kind of dynamics processor without listening to it.

JR
 
Thank you for the detailed response. I was looking to do auto gain makeup, but I don’t think what I initially put would actually do what I need. It’s just imposing a long release average correction - and I didn’t consider it could even cause the level to come up after the initial compression action had cleared.

I think to do what I was initially thinking correctly you’d have to have a second feedback loop that took an RMS average of the post compression output level, and subtract that from the input level.
 
I did a lot of dynamic processor design back in the 70s-80s (compressor/limiter/gate, tape noise reductions, etc).

It involved a lot of bench work listening to how different experiments worked. Back then we were dealing with lower quality gain cells and other problems.  I even rolled my own variant tone burst  gate, with a bypass feed around so I could artificially juice up the transients and crest factor of pre-recorded sources.

Have fun but the final arbiter is that it must sound good or why bother?

JR
 
dogears said:
Anyone ever played with makeup gains in a compressor sidechain?
What type of program? What are you aiming at?
For example, if you want to compensate for sudden jumps in level, you need a rather fast attack, but if your goal is compensating level discrepancies between programs, you need a slower attack and an even much slower release.
See how many respected SE's use several compressors in cascade, all with different characteristics, making them complementary, rather than fighting each other.
 
abbey road d enfer said:
What type of program? What are you aiming at?
For example, if you want to compensate for sudden jumps in level, you need a rather fast attack, but if your goal is compensating level discrepancies between programs, you need a slower attack and an even much slower release.
See how many respected SE's use several compressors in cascade, all with different characteristics, making them complementary, rather than fighting each other.
The idea was to simply have automatic makeup gain.

Put this way - input is level A. After gain reduction, you have a new signal at level B, albeit at with a different waveform. If you're adding compression, B < A. The idea was to automatically add gain to the final signal so that the final output signal is the waveform of B, at the average level of A.

I've included provision in the sidechain for manual makeup gain. That's simple enough. I was just trying to do it automatically.

One way to do it could be a second VCA stage that compares the first stage's and second stage level using feedback from the output stage. In a way that is what you're describing.

.. something like this:


 

Attachments

  • 1.png
    1.png
    17.7 KB · Views: 15
dogears said:
The idea was to simply have automatic makeup gain.

Put this way - input is level A. After gain reduction, you have a new signal at level B, albeit at with a different waveform. If you're adding compression, B < A. The idea was to automatically add gain to the final signal so that the final output signal is the waveform of B, at the average level of A.

I've included provision in the sidechain for manual makeup gain. That's simple enough. I was just trying to do it automatically.
OK, got it. You realize that these two side-chains will fight each other, I guess... One will try to compensate what the other does.
The usual approach is to make the leveller's side-chain much slower than the compressor's. The obvious fault is that the leveller will be late catching up the initial level drop and will continue to increase gain long after the signal has ceased, which results in the well-known noise-floor lift.
Many designers have tried to find a solution, that often relies on sampling the Gain Reduction at intervals and holding it in between. I would think Bob Orban is the specialist. Go to the Orban's website and download the 8100A manuals appendices F-M; you will see it's not a simple issue.
 
Abbey, I was thinking the same thing!  I haven't worked in broadcast systems for years, and Orban offered the brilliant solution(s) decades ago.

I recall the CBS Audimax/Volumax boxes 40 years ago..  I don't recall which was which, but one box was a "gain rider"  to correct for the errant on-air DJ's, while the other was a limiter to avoid transmitter overmodulation.

Bri
 
abbey road d enfer said:
OK, got it. You realize that these two side-chains will fight each other, I guess... One will try to compensate what the other does.
The usual approach is to make the leveller's side-chain much slower than the compressor's. The obvious fault is that the leveller will be late catching up the initial level drop and will continue to increase gain long after the signal has ceased, which results in the well-known noise-floor lift.
Many designers have tried to find a solution, that often relies on sampling the Gain Reduction at intervals and holding it in between. I would think Bob Orban is the specialist. Go to the Orban's website and download the 8100A manuals appendices F-M; you will see it's not a simple issue.

Maybe I’m missing it but the two control loops I drew are decoupled. They shouldn’t fight at all. If one reduces 2 dB the other adds 2 back, because the makeup sidechain is looking at the difference of the two rectified levels. They’re independent.
 
dogears said:
... If one reduces 2 dB the other adds 2 back...
Which is exactly what fighting each other is. We may disagree on words, but the end result is that you have to find a way to make these two automated processes work together, not against each other. Think of how you set up make-up gain manually. It's a process that involves choices based on experience, hindsight, prediction and iterative readjustment.
 
Yeah... I suppose with completely identical time constants you get no change whatever except for more distortion. I keep trying to think of it as two separate things but in the time domain if both are acting at the same speed it will undo itself.

Which puts me back at my original premise. How long is a good starting point for the time constant of the makeup gain? I honestly don’t even know where to start.

Intuition actually says to add a parallel  RMS with a fixed 10-15 ms time constant to do this work on both ends, and leave the variable timing for compression duties. That way it’s kind of average to average, even if your compression activity is working on peaks.
 
dogears said:
Intuition actually says to add a parallel  RMS with a fixed 10-15 ms time constant to do this work on both ends, and leave the variable timing for compression duties. That way it’s kind of average to average, even if your compression activity is working on peaks.
Why don't you try it and find for yourself if it works for you? I know it doesn't for me, but each one his own...
 
I never likes any auto-makeup. But someone has to or it wouldn't exist.

For me it's always been kind of a loss of control since you never really know how much gain is applied.

But anyway, i really think 10-15ms time constant is way too short. I would try first with a time constant over 100ms. But maybe implement a min/max gain. Ie between 2-3min / 6-8 max would be enough for most applications.

Thomas
 
totoxraymond said:
I never likes any auto-makeup. But someone has to or it wouldn't exist.
It's almost always to make things easier for untrained people. Some designers have made the gain make-up a function of the threshold and ratio controls; it's never perfect but good enough.
It may be because the dsigners wanted as few controls as they could, on the (false) premises that the less knobs, the less errors to be made
dbx were quite successful with their one-knob compressors. Tascam digital mixers have a choice of auto-make-up based on the same paradigm: the lower the threshold and the higher the ratio, the higher make-up gain is cranked.
IMO, it's a better solution than the moving target resulting from input/output comparison.


For me it's always been kind of a loss of control since you never really know how much gain is applied.
I agree; I don't like it.
 
I’m going to breadboard and play with it but I scratched it from the pcb layout for now. Thinking to start at 5x the release.
 
I've never seen a knob labeled makeup gain on a compressor that wasn't a static value of x dB. I've seen makeup gain tied to the Threshold value. When the Threshold reduced to a lower level the makeup gain increases by the same amount. This is to take advantage of the increase in headroom.

What you are describing sounds more like an expander.
 
Gold said:
I've never seen a knob labeled makeup gain on a compressor that wasn't a static value of x dB. I've seen makeup gain tied to the Threshold value. When the Threshold reduced to a lower level the makeup gain increases by the same amount. This is to take advantage of the increase in headroom.

What you are describing sounds more like an expander.
Yes, it would be compressing and expanding.  But static makeup gain is expanding as well, just without a time constant.

If you consider what the engineer is doing when he's applying makeup gain it's usually something like a peak-to-peak or VU-to-VU comparison of in and out levels, then adjusting until the post-compression level is the same as the input level.  But makeup gain is applied without regard to timing. So you'd need some kind of sample and hold, perhaps, to do this automatically.

I'm giving up on the idea for now, too many other projects going on, and I don't think it's a great feature to begin with.
 
dogears said:
Yes, it would be compressing and expanding.  But static makeup gain is expanding as well, just without a time constant.

That's a difference with a rather large distinction. Static gain doesn't alter wave shape of the source audio.

and I don't think it's a great feature to begin with.

I agree.
 
Back
Top