THAT (was Fet) Compressors

bluebird said:

Hmmm, The diodes labeled attk and release on the schematic are reversed, I just cut and pasted that schematic together.  John had them like that because he wasn't planing the inverting threshold stage between the RMS and his circuit. Regardless, its just a matter of adjusting the resistors to the opposite diodes for attack and release right? Or does the CV polarity change the way the SVF works in general?

Click to expand...

The trick fast attack circuit depends on the direction of side chain voltage moving for transients. The 2M pull up resistor creates a current threshold that the normal fast attack tracks when current (through D12 and R5) is below that threshold. When that current exceeds the 2M current, the op amp U1A disconnects from the 0.1uF and becomes 10x faster controlled by the 0.01uF cap. During this transient duration the 0.1uF  charges through the 2M, but not much and after the transient subsides the side chain has not been pumped by the transient.  The gain reduction happened but the side chain was not distorted like an actual fast attack would.

If you flip the polarity in front, you needed to flip the uber-fast attack too. 

Would this circuit work after the threshold op amp, or right after the RMS detector? I will definitely breadboard this!

Ok I will try that. For some reason (when putting Johns circuit directly after the RMS detector)I am having trouble with over all offset voltage reaching the VCA and skewing the overall gain extremely. 

Click to expand...

As drawn my SVF smoothing circuit should be unity gain for DC voltage.

So the schem below should work correctly? I didn't try the threshold circuit after like that.

Click to expand...

If that is boilerplate limiter from THAT app notes it should be OK... I never made a THAT limiter.  The smoothing circuit is non inverting unity gain, so order of the two circuit blocks in series should only change dynamic behavior, not functional operation.

JR

I made some headway, Put the threshold after the smoothing circuit, things are starting to sound good now. 

In order to get a decent signal into the smoothing circuit from the RMS detector, I had to get rid of the 200K input resistor. Its 5K now. It is definitely catching those transients better now.

John, is that 2M in between op amps to mix in a little super fast RMS detector voltage in with the attack and release signal?

bluebird said:

In order to get a decent signal into the smoothing circuit from the RMS detector, I had to get rid of the 200K input resistor. Its 5K now.

Click to expand...

So it's not unity-gain anymore; the compression ratio now is about 20 times more.

It is definitely catching those transients better now.

Click to expand...

Maybe increasing the compression ratio was what you wanted.

  is that 2M in between op amps to mix in a little super fast RMS detector voltage in with the attack and release signal?

Click to expand...

Both JR and I told you earlier that the diode effect is active on one transition only. If you have your polarity right, it will act on release only.
[/quote]

bluebird said:

I made some headway, Put the threshold after the smoothing circuit, things are starting to sound good now. 

Click to expand...

that was plan

In order to get a decent signal into the smoothing circuit from the RMS detector, I had to get rid of the 200K input resistor. Its 5K now. It is definitely catching those transients better now.

Click to expand...

That is a huge change 200/5 =40x 

The design concept is for my smoothing circuit block to be non-inverting unity gain...  just with a different dynamic response as the signal changes.

The voltage at the output of the  smoothing block should be similar to voltage at the original RMS port.

The ratio of R20 to R4 allows you to scale up the voltage gain inside the loop to hit the steering diodes harder or softer.

John, is that 2M in between op amps to mix in a little super fast RMS detector voltage in with the attack and release signal?

Click to expand...

The 2M  R18, is the baseline "slow" time constant R for steady state signals that do not engage the att/release steering diodes for faster attack or faster release.

The 2M  R7 is the uber fast attack current threshold... for signal attacks pulling less current the integrator is smoothed by the 0.1uF. For attacks drawing more current , the 0.1uF runs out of current and the 0.01uF integrator cap takes over momentarily, until the transient passes and things settle back down. 

Perhaps you need to twiddle the resistor coming from the output of the smoothing circuit to your op amp threshold comparator.. No value shown on my copy of your schematic.

JR

JohnRoberts said:

That is a huge change 200/5 =40x 

Click to expand...

I know thats what I was thinking but the smoothing circuit just wasn't responding to the RMS voltage with any bite.  I know the circuit is unity and why it is that way with 200K on the input, and why the pad works. But it really doesn't seem like there is 40x gain going on there now that the 200K is a lot smaller. I suppose its impossible to be otherwise. I don't know.

JohnRoberts said:

The voltage at the output of the  smoothing block should be similar to voltage at the original RMS port.

Click to expand...

yes I know but when the RMS port was hooked up through the 200K there wasn't much movement at all in the smoothing circuit. I could see the RMS ripple where the TC diodes meet and that looked normal but there was nothing going on at the output of the next stage. Its possible I had something screwed up there... I'll check it again.

JohnRoberts said:

The 2M  R18, is the baseline "slow" time constant R for steady state signals that do not engage the att/release steering diodes for faster attack or faster release.

Click to expand...

Got it.

JohnRoberts said:

The 2M  R7 is the uber fast attack current threshold... for signal attacks pulling less current the integrator is smoothed by the 0.1uF. For attacks drawing more current , the 0.1uF runs out of current and the 0.01uF integrator cap takes over momentarily, until the transient passes and things settle back down. 

Click to expand...

Yes I hear this working as I scale the 2M to 4M to 6M and change the value ot the .1uF.  As you said there is a lot of adjustment to be made.

JohnRoberts said:

Perhaps you need to twiddle the resistor coming from the output of the smoothing circuit to your op amp threshold comparator.. No value shown on my copy of your schematic.

Click to expand...

Yes its 30K now and It does seem to affect the smoothing circuit.
Thanks John, I'll keep chipping away at it.

Ok I have everything functioning properly now and using the correct 200K input resistor to the smoothing circuit from the RMS detector. There is plenty of quickness now. It very sensitive to all the components, more so than before. Its going to take me another month to adjust this to something usable.  There is a balance that is needed between stopping transients and not distorting lower frequencies.

Which has me wondering about Abby's circuit. Abby would your circuit be better suited for a more traditional attack and release circuit before it? Would just a big timing cap on the RMS tc port be enough? When looking at the scope at the output of Johns smoothing circuit the transient voltage bump above steady state is only about 50mv.  Do you think Johns TC circuit would be compatible with yours or would it be too much?

bluebird said:

Which has me wondering about Abby's circuit. Abby would your circuit be better suited for a more traditional attack and release circuit before it?

Click to expand...

As I mentioned, I have never actually tested this, but my intention was to add this to a standard THAT-style NLC  http://www.thatcorp.com/datashts/dn114.pdf

Would just a big timing cap on the RMS tc port be enough?

Click to expand...

I don't think so. The NLC works very well for me.

When looking at the scope at the output of Johns smoothing circuit the transient voltage bump above steady state is only about 50mv.

Click to expand...

  Well, 50mV is about 8dB gain change. It's up to you to decide it's acceptable or not.

Do you think Johns TC circuit would be compatible with yours or would it be too much?

Click to expand...

In principle, it should be, but I'm not sure the final result is to your liking. It"s akin to chaining several compressors; they may be complementing each other or fighting against each other.

ok, I have your circuit made up on the breadboard. Where the schematic is marked V+5 and V-5 I'm assuming thats where +/-5 volts go? Or should it be relative to the supply voltage?  I'm running +/-18.
I've messed with the NLC circuit and never got something I was satisfied with but I will try it again in conjunction with your circuit. I will put the threshold amp after your circuit and see what gives...

bluebird said:

ok, I have your circuit made up on the breadboard. Where the schematic is marked V+5 and V-5 I'm assuming thats where +/-5 volts go? Or should it be relative to the supply voltage?  I'm running +/-18.

Click to expand...

It just happens to be the fifth set of rails in the design; nominally +/-17.

I've messed with the NLC circuit and never got something I was satisfied with but I will try it again in conjunction with your circuit. I will put the threshold amp after your circuit and see what gives...

Click to expand...

I would think it's better to put it between the threshold circuit and the ratio control; that's how I had originally planned it. Please see attachment. Note the threshold circuit has 11x gain.
There is a choice of RMS+NLC detection or conventional "peak" detection, where the RMS detector is just used as a log converter+level shifter. In all cases, diode D5 makes the onset of compression somewhat progressive over a range of about 10dB.

Wow, nice. So you have a choice to use the deadband circuit or not.

What advantages does that precision rectifier have over a more standard one?

You are mixing the standard time constants with the NLC circuit before the deadband circuit. I'm going for more limiting and apparent loudness over compression. Which would you say works best for that, tuning the attack and release circuit, or using the NLC circuit? Which path?

thanks!

bluebird said:

Wow, nice. So you have a choice to use the deadband circuit or not.

What advantages does that precision rectifier have over a more standard one?

Click to expand...

As it is, it allows driving the transistors more adequately.

You are mixing the standard time constants with the NLC circuit before the deadband circuit.

Click to expand...

They're not really mixed. The CMOS switches really select one or te other. The 10k resistors are there just to make sure the input is never left floating. I guess they could be 100k or 1Meg it wouldn't make much difference. BTW I experienced with mixing CV's from RMS and quasi-peak
http://www.laaudio.co.uk/product_bcl20_1378.aspx
it turns out it's not that useful and rather confusing for some.

I'm going for more limiting and apparent loudness over compression. Which would you ay works best for that, tuning the attack and release circuit, or using the NLC circuit? Which path?

Click to expand...

There is not so much difference after all. RMS is quicker/easier to dial, quasi-peak is more tunable. Ideal would be a tunable rms, not too dissimilar to what dbx done with the 165. However, their take, which was basically attaching TC's to the  output of an rms rectifier with a small timing cap (1uF), did not benefit from the NLC circuit (which had not yet been discovered/implemented by dbx's team). I believe Valley People have explored these possibilities (Dyna-Mite).

bluebird said:

Ok I have everything functioning properly now and using the correct 200K input resistor to the smoothing circuit from the RMS detector. There is plenty of quickness now. It very sensitive to all the components, more so than before. Its going to take me another month to adjust this to something usable.  There is a balance that is needed between stopping transients and not distorting lower frequencies.

Click to expand...

distorting low frequencies is from too fast of a release time...  You have independent control of this with the release steering diode (D10), but note that this is in addition (parallel) to the 2M allways in R18, so maybe make that a bunch larger too.  The LF distortion is mainly a quirk of processing sine waves which aren't that common, but too fast of a release time can be audible on held bass notes.  In an ideal world, the ripple inside the loop for a steady level sine wave is below +/- a diode drop but this may be impractical for very low frequency sine waves.

I pulled all those initial values out of my butt, so polishing is required.  I used a 0.1uF main timing cap because that was a reasonable value mylar, but maybe it should be larger (scale the 0.01uF transient cap too, while the 1/10 ratio is also arbitrary). . 

Which has me wondering about Abby's circuit. Abby would your circuit be better suited for a more traditional attack and release circuit before it? Would just a big timing cap on the RMS tc port be enough? When looking at the scope at the output of Johns smoothing circuit the transient voltage bump above steady state is only about 50mv.  Do you think Johns TC circuit would be compatible with yours or would it be too much?

Click to expand...

You shouldn't need two side chain smoothing circuits, maybe try both separately and select the one that sounds better to you.

JR

JohnRoberts said:

distorting low frequencies is from too fast of a release time...  You have independent control of this with the release steering diode (D10), but note that this is in addition (parallel) to the 2M allways in R18, so maybe make that a bunch larger too.  The LF distortion is mainly a quirk of processing sine waves which aren't that common, but too fast of a release time can be audible on held bass notes.  In an ideal world, the ripple inside the loop for a steady level sine wave is below +/- a diode drop but this may be impractical for very low frequency sine waves.

I pulled all those initial values out of my butt, so polishing is required.  I used a 0.1uF main timing cap because that was a reasonable value mylar, but maybe it should be larger (scale the 0.01uF transient cap too, while the 1/10 ratio is also arbitrary).

Click to expand...

Yes I've been experimenting with 0.1uf and 1uF. I can get louder with faster release times but there is more distortion. Also I have played with R18... At a certain point it becomes a subjective choice. Just wanted to make sure I could get the maximum performance out of the given circuit which I think I have now. I really appreciate all the help with this John!

abbey road d enfer said:

They're not really mixed. The CMOS switches really select one or the other. The 10k resistors are there just to make sure the input is never left floating.

Click to expand...

Ahh got it, sorry missed that in the previous post.

abbey road d enfer said:

There is not so much difference after all. RMS is quicker/easier to dial, quasi-peak is more tunable. Ideal would be a tunable rms, not too dissimilar to what dbx done with the 165. However, their take, which was basically attaching TC's to the  output of an rms rectifier with a small timing cap (1uF), did not benefit from the NLC circuit (which had not yet been discovered/implemented by dbx's team). I believe Valley People have explored these possibilities (Dyna-Mite).

Click to expand...

Ok I will play around with this and report back. Thanks!

Ian

bluebird said:

Yes I've been experimenting with 0.1uf and 1uF.

Click to expand...

You can make it both faster and slower... increasing the cap size scales everything slower (except for trick fast attack)  so normal fast attack parts must be rescaled .

I can get louder with faster release times but there is more distortion.

Click to expand...

Just to be clear the purpose of the steering diodes especially for release time is so you can make it slow enough to not distort steady state sine waves, but when true amplitude releases occur engage the fast release steering diode for a faster release time constant. This allows the loudness increase from fast release times without the distortion on steady LF tones...  (I might be tempted to use two diodes in series for the release to make that no-distortion zone a little wider). This is all tweakable on the bench with proper test waveforms.

I did a lot of work (decades ago) optimizing these trick circuits to make transparent tape noise reduction. The goal was to have fast release on compression encode so you don't hear tape noise during the quiet parts right after a loud part, but of course you can't add distortion so it has to always sound clean. There are a number of psychoacoustic factors about human audition and noise masking.  We all know you can't hear quiet noises when loud noises are present, but there are also time intervals after a loud noise stops, before we can hear quiet noises again. Good noise reduction takes advantage of these factors to work before the ear catches up. Likewise we can't perceive very brief distortion events, so some transient clipping is not very audible.

Dynamic processing, when you are trying to change the sound is indeed subjective and a transparent compressor that you can't hear working might not be an obvious goal, or even benefit.  8)

JR

Also I have played with R18... At a certain point it becomes a subjective choice. Just wanted to make sure I could get the maximum performance out of the given circuit which I think I have now. I really appreciate all the help with this John!
Ahh got it, sorry missed that in the previous post.

Ok I will play around with this and report back. Thanks!

Ian

Click to expand...