The GPa... At last another compressor!

[PRR]

...To be very clear : ...

It was clear to me. Maybe you are not reading what you are typing.

[Fancy Lime]

So...

I had a bit more if a think and finally saw the error in my ways. I was going to explain it but Jack beat me to it by a mile or two.

Seems like a major redesign is in order. I also listened to it again and found that indeed long held loud notes on high compression settings exhibit the two knee behavior with a plateau in between. Not good.

I have some ideas for fixing it but I need to test them first.

Btw, Vivek, the inverted half of the rectifier is not there for extra current but to make the pumping signal twice as fast. In the slow recovery setting, this indeed makes no difference but it allows to have twice as fast recovery times without getting audible rippling on slowly decaying notes

Cheers and thanks for all the input,
Andy

[Vivek]

Andy, Thanks for pointing out possible usefulness of inverting stage and discharge transistor for ripple removal in fast mode

So far, I did all analysis in slow mode with both caps in the circuit, so I missed out this point.

[Fancy Lime]

At T1 the fwr transistors start to turn on. The knee is very small here and the output quickly goes into limiting (as I mentioned in a previous post). There is limiting up to T2 where the fet has pinched off and can no longer adjust gain. When the fet is fully off, the circuit becomes a unity gain buffer, as would be expected from a non-inverting op amp.

For signals with a Vp of 0.5 or greater, it is only acting as a buffer.

This is typical of non-inverting based compressors when the control element reaches its limit to restrict gain.

I wonder how it would respond if you increased the gain of the op amp from 6x max to a much high amount? like 60x for example

Best regards, Jack

So I had another think. Turns out, I'm an idiot and most of what I thought and wrote about how this thing works was quite wrong. But hey, at least I am a stubborn idiot and I think I found a way to salvage this mess. I think...

So first up, it *is* a limiter after all. As Jack pointed out in Vivek's graph, it has three "regions of operation": 1) uncompressed gain up to T1, 2) limiting up to T2, and 3) buffer. T1 is the point where input voltage times gain equals detector threshold and T2 is where input times 1 equals detector threshold. So all we need to do to extend the useful input voltage range is increase the threshold. To make up for that we may also need to increase the gain to get a good compression feel. The easiest way to manipulate the threshold, would be to insert a 1M pot wired as a variable resistor between the right sides of R18 and R19. But maybe a fixed 100k resistor would do as well. This should set the threshold high enough to get a decent response on most guitar out pot levels. A pot instead of R14 may be worth exploring. Between gain and threshold, one should probably be a switch and one a pot. With a gain pot, it would indeed be very close to a JFET version of the DOD 280. A FET 280, if you will.

Anyway, this is all at the "untested brain fart" stage as of now. I'll breadboard it as soon as I can. Whether it works or not: huge THANKS to Vivek, Jack and Paul for your help and patience in putting up with my nonsense!

Andy

[Fancy Lime]

Hi there,

based on the discussion here, I made some changes that should improve the usability of this thing:



The Compression control is now a gain control, making it even more of a JFET version of the DOD 280. There is also a Recovery control, which is exactly what the Boss CO2 and many others somewhat misleadingly call "Attack". The recovery times are very short with a 1u cap as C17. In the shortest settings, it gets very ripply on low notes, really more of a fat overdrive, which I like a lot. The longest setting corresponds to the shortest one on the CS2 and most other Dyna Comp clones, which is really more than slow enough for my taste. R19 and R20 are chosen such that if Q1 has the maximum Vgs(off) according to the datasheet, the envelope detector can still turn it all the way of. Choosing these to match the Vgs(off) of the actual Q1 JFET exactly would largely eliminate the risk of overcompression but as long as we are using short compression times, ball park is close enough. The most important change, which I would never have figured out was necessary without Vivek's help and insatiable curiosity, is the addition of R17. Thanks for that! R17 attenuates the signals of both paths going into the envelope detector by about 12db. This means that the whole shebang should not go into buffer mode unless the input signal is higher than about 4.4V peak to peak, at which point D2 and 3 will already have clipped it. I also added a dry out to allow parallel processing of the uncompressed signal. You could run this through a distortion or an eq and mix it back with the compressed signal. And/or you could use the dry out to feed the control in of an envelope filter somewhere down the chain. Or have a compressed (or rather limited) signal on stage but send an uncompressed one to the FOH desk for further shenanigans. Or you could be boring and use it as a tuner out. Your choice. That meant I had an opamp half left over, which I used to buffer the output and I made it a NE5532 for the higher current capabilities compared to the TL072 but pretty much any opamp will do fine in this position.

Caveat: some more testing and tuning will occur as soon as I find the time.

Cheers,
Andy

[iainpunk]

hey, seems im late to the party here.
looking through the posts here is very interesting, although im not really into compressors, i like the technology, design elements and considerations behind the design very much.

thanks for posting
cheers

[Vivek]

I entered GPA 02 into LTSPICE



I am sure I must have made a mistake somewhere because this is what I currently see :

X axis is time


Conditions:
Input signal of 500mvP, 1Khz
Signal starts at 100ms and lasts for 1 second

Green is output voltage
Blue is control voltage on the C17 integration cap

Compression pot at 50%
Recovery Pot at 50%

The output is very distorted too. second harmonic is 19 dB down from fundamental.


Let me spend some more time on it and find out my error.

( PS It does remind me of graphs in Chemical Engineering when a control system is not properly damped, and it overshoots and rings till it finds equilibrium. Maybe the transistors need collector resistors ? )

[Fancy Lime]

Hi Vivek,

No, I don't think you made a mistake. This looks about right. We do see ringing of an underdampened feedback system here. The thing is that in this kind of design, the recovery time setting controls how quickly the system settles and the optimum recovery time depends on the frequency.

What happens is that the initial uncompressed peak pushes the envelope into turning Q1 completely off. The device goes into the buffer region. We could make the recovery from this overshoot faster by matching R19 and 20 better to the actual Vgs(off) of Q1. Could you try what happens if you swap the values of those two or make them both 2M2?

But yes, the overshoot is an inherent flaw of feedback compressors that we can only mitigate but never really eliminate. I can definitely hear it on the breadboard. It is only audible when the compression is set too high in relation to the input signal when the recovery is set slow. By adjusting compression and recovery, you can dial the "compressor breathing" in and out.

There are some other ways to dampen the ringing and reduce the overshoot, like increasing attack time and clipping the signal before it goes into the envelope detector. But these have their own drawbacks and need some testing.

The distortion probably has to do withe the asymmetry of the output, which is probably caused by the C13, R16 arrangement. I'll have to see what I can do about that. I'll try if there is more or less distortion without those or with different values.

Cheers,
Andy

[Fancy Lime]

Hi all,

finally, the long promised new version:

Not many points of note here, except the "duck and swell suppressor arrangement C13/R21 after the holding cap C16. Works quite effectively by providing a very quick initial recovery if there is overcompression on the first transient. With proper setting of the Compression and Recovery pots, this thing works quite well now as an unobtrusive but fattening limiter. Also note that I took out the resistor in series with the gain-regulating JFET Q1. This helped greatly in reducing distortion on high compression settings but has the drawback that the "no compression gain" is now directly dependent on the on-resistance of Q1. No biggie with the fairly tight specs of the J112 but it means that the other gain controlling resistors and caps need to be adjusted if a drastically different JFET is used. The Brightness switch is also back and does a good enough job to compensate the treble loss on high compression settings for my liking. Good bread'n'butter limiter now (which we call compressor because we call most limiters compressors) with decent signal level handling for guitar and bass purposes and very good noise floor and no fancy-pants odd parts.

Cheers,
Andy

[sergiomr706]

Hallo Andy, very interesting projects you have going, justo would like to know if this would be possible to add a variable hpf to the detection parte of the circuit, not to the audio, between some of the actual components. Thanks a lot

[PRR]

> add a variable hpf to the detection parte of the circuit, not to the audio

[sergiomr706]

Thank you Paul, I ll try something there.

[Vivek]

Would cap in parallel to R12 become HPF on control signal path?

[antonis]

Would cap in parallel to R12 become HPF on control signal path?

In series, perhaps..

[Fancy Lime]

Yes, what Paul said. I would try inserting a 100n cap there, with a variable resistor to Vref (4.5 V). If you use a 3.3k resistor in series with a logarithmic 100k pot (with lug 1 being one end and lugs 2+3 being the other end) as the variable resistor, you get a range of about 15-500 Hz for the high pass cutoff, which I would assume to be a reasonable starting point for further experimentation.

Andy

[Vivek]

Would cap in parallel to R12 become HPF on control signal path?

In series, perhaps..

Thanks Anthony.

Please help me to understand this.

I feel an HPF will be created with cap in parallel to R12

[antonis]

I feel an HPF will be created with cap in parallel to R12

A HPF needs a series cap and a shunt resistor..
(see Paul's and Andy's posts above..)

You could consider R15 (actually half of its value due to -1 gain) as a shunt resistor for the cap in parallel to R12 but it should alter U1B unity gain..

It's more safe to implemenet the standard AC inverting amp configuration..

[PRR]

Would cap in parallel to R12 become HPF on control signal path?

That sends HF gain to infinity on one half cycle. Why??

It is not that complicated. Say 10K to VR (to provide bias and cap-load), and a cap of your choice. (Nearest common value will be fine.)

[Fancy Lime]

I just realized that I wrote "resistor to ground" in my last post, which is of course nonsense. That would mess up the just like Paul warned. I have corrected it to "resistor to Vref".