Transformerless Vari-Mu Compressor

Help Support GroupDIY:

Heikki

Well-known member
Joined
Jul 31, 2008
Messages
230
Location
Finland
abbey road d enfer said:
Looks like we may have a different understanding of what thump means.
I described CV pass-through, when you describe pumping (punch-in).

Thump is CV pass-trough. I'm was not trying to describe pumping. My ability to write coherent sentences is probably lacking. The sidechain will handle CV pass-trough like any other signal and in a worst case it results in motorboating. In a feed forward compressor we wouldn't have this problem.
 

abbey road d enfer

Well-known member
Moderator
Joined
Jan 22, 2008
Messages
12,618
Location
Marcelland
Heikki said:
Thump is CV pass-trough. I'm was not trying to describe pumping. My ability to write coherent sentences is probably lacking. The sidechain will handle CV pass-trough like any other signal and in a worst case it results in motorboating. In a feed forward compressor we wouldn't have this problem.
Then how would a HPF in the side-chain reduce thumping? I have an idea but I'd like to know how you see it.
 

Heikki

Well-known member
Joined
Jul 31, 2008
Messages
230
Location
Finland
abbey road d enfer said:
Then how would a HPF in the side-chain reduce thumping? I have an idea but I'd like to know how you see it.

I don't think I said it would reduce thumping. I was trying to say that the thump might cause overcompression (if that is the right term) that would not happen if there were no control voltage feed-trough. If I have 5ms attack time and because of the tube imbalance it causes a 33Hz transient that is very large. The sidechain will create a control voltage from that 33Hz transient . Now if I cut some of the low frequencies at the sidechain input like -3dB at 50Hz, that 33Hz will create slightly smaller control voltage. In a feed forward compressor the CV pass-trough never gets back to the sidechain and it avoids such problems completely.
 

Script

Well-known member
Joined
Nov 26, 2008
Messages
1,357
Location
Japan
abbey road d enfer said:
No. The cause of thump is a sudden displacement of the anode voltage when the gain cell is dubmitted to the Control Voltage. It has nothing to do with the signal frequency. Thump frequently appears on transients.
There are two phenomenons at work:
  • The two tubes do not match precisely, which results in different voltage variations between them. So there are both a common-mode signal and a differential-mode signal. the CM signal should be rejected by the subsequent stages, but the differential signal is impossible to remove.
  • The subsequent stages may not have enough CMRR to completely ignore the CM signal.
In Heikki's build, these subsequent stages have a very large CMRR, so teh only remaining source of thump is tube unbalance. It's dynamic unbalance, so static balance adjustment has a limited action.
Indeed lowering the attack speed rejects the thump spectrum towards infrasonics, but the usual challenge is to get a fast attack and minimum thump.
I love your condensed explanations, as always. The more often I read about this phenomenon, the more I learn, including the CMRR aspect.

I was looking at it from a test point of view. Feeding a series of sharp (squarewave), very low frequency impulses into a vari-mu set to shortest attack can be used to reveal tube imbalances -- it's artificially driving it into extreme territory, which makes thumping more pronounced / audible. Transients, yes. DC slip-through.

With my limited tube knowledge, the thought above was to avoid sharp transients to reduce thumping, which I think I understood to be most pronounced in the bass fq region. One approach is longer attack times, yes, which I just learned pushes the thump into subsonic land :) -- however, this lets more bass in the audio signal pass through. The other approach is HP filtering, which too I think I understand makes the unit react less to high-amplitude bass - with less high energy transients in DC signal, ergo smoother, ergo less attenuation of bass in audio signal.

Both approaches do not eliminate thump -- they try to avoid it from happening. Thump can still and probably will happen when driven harder and/or with sharp transients (say a lound, high-pitched tabla hit e.g.). But hopefully the thump (1) is shorter and thus less noticeable, (2) is hopefully reduced enough to be masked by 'program material, and/or (3) it is looked at not as a fault, but a feature ;).

Either way, both approaches yield overall less attenuation triggered by the 'problematic' bass fq region, I'd think -- so effectively, it's a vari-mu 'full-bandwith' limiter not anymore but some kind of compressor.

Quick dirty (listening) test on PRR Vari-mu:
Driving the input pretty hard with a setting of longest attack time and no HP filtering sounds pretty close to a setting of fastest attack and a 700 400 Hz HP filter feeding the sidechain.
 

abbey road d enfer

Well-known member
Moderator
Joined
Jan 22, 2008
Messages
12,618
Location
Marcelland
Heikki said:
I don't think I said it would reduce thumping. I was trying to say that the thump might cause overcompression (if that is the right term) that would not happen if there were no control voltage feed-trough. If I have 5ms attack time and because of the tube imbalance it causes a 33Hz transient that is very large. The sidechain will create a control voltage from that 33Hz transient . Now if I cut some of the low frequencies at the sidechain input like -3dB at 50Hz, that 33Hz will create slightly smaller control voltage. In a feed forward compressor the CV pass-trough never gets back to the sidechain and it avoids such problems completely.
OK, that's more or less what I thought. I wouldn't have thought the LF components of the CV would play a role.
 

JohnRoberts

Well-known member
Staff member
Moderator
Joined
Nov 30, 2006
Messages
20,480
Location
Hickory, MS
I am not the tube guy here but know a little about dynamics processors.

Any gain control device that gets modulated in response to signal levels is multiplying the dry audio times that side chain gain change signal. Any dramatic gain changes (like step functions) can be audible. Further gain elements with DC errors that get modulated by the gain control changes will introduce "thumpy artifacts".

Freedom from control voltage feedthrough is a quality factor for gain elements, while steps and sudden voltage changes also need to be filtered out of side chain paths generally using low pass filters. Of course this is an oversimplification, side chains often utilize different strategies to reduce audible artifacts.

JR

 

5v333

Well-known member
Joined
Jun 30, 2013
Messages
477
Location
Gothenburg
Ive been studying the effect a little bit theese last days and following this cool thread.

Its my understanding that the effect is not primarily because of low common mode rejection but the fact that the change in quesent current forces a voltage change with the DC load at the plates. and this effect is not bound to LF, although, most HF ad MF is filtered out by the time constant components leaving a sluggish CV.

Looking at load lines (DC line) would prob give a better picture dont you think?

I also think as Abbey suggested that a synthetic inductor, with a DC voltage incorporated, can give a DC load line that is more or less straight (can the the emitter resistor be super low?) which would prevent voltage generation at the plate when current is changed.
 

abbey road d enfer

Well-known member
Moderator
Joined
Jan 22, 2008
Messages
12,618
Location
Marcelland
5v333 said:
Its my understanding that the effect is not primarily because of low common mode rejection but the fact that the change in quesent current forces a voltage change with the DC load at the plates.
If the tubes were perfectly identical, the voltage change on both sides would be equal, so there would be no differential voltage there to start with. So I maintain that thumps are due to two factors:
differential thump due to gain cell unbalance, and possible unsufficient CMRR of teh subsequent stages.
 

5v333

Well-known member
Joined
Jun 30, 2013
Messages
477
Location
Gothenburg
I think i understand what you are saying. I still think differential stages are a little bit mysterious...
 

JohnRoberts

Well-known member
Staff member
Moderator
Joined
Nov 30, 2006
Messages
20,480
Location
Hickory, MS
5v333 said:
I think i understand what you are saying. I still think differential stages are a little bit mysterious...
A differential stage is just math, in fact early analog computers did addition and subtraction using antique op amps. 

A differential audio stage with + and - inputs will add the + input and subtract the - input. CMR (common mode rejection) means that if you put 2V into the + input and 2 volts into the - input,  +2V -2V =0V.

CMRR (common mode rejection ratio) is a measure of how well a differential stage rejects common mode (same in both inputs) signal. Of course this is a simplification and cancellations can be degraded by polarity, phase errors, component precision, and more.

JR
 

Script

Well-known member
Joined
Nov 26, 2008
Messages
1,357
Location
Japan
Heikki needs some 'bad' tubes. Unfortunately I don't have and the following is just another veer-off.

In a solid-state vari-mu we have no transformers to potentially shave off erratic transient spikes. We don't necessarily want a comp to react to such spikes (1ms, but also longer ones depending on chosen attack time, as well as strong signal onsets where we had to back off on the attack). So provided such spikes and signals made it through the tube stage without slamming against a transformer that could smear or swallow them, they would appear at the output. 

What would be a smart solution here to get closer to classic vari-mu WITH TX behaviour?
(A) I'd guess driving the tubes harder (spikes dissolve into 'pleasing' distortion)
(B) some kind of more or less crude diode clipper (diode clipping can sound unobtrusive when used sparingly but can easily just as well get plain stinky).

So what could be a solution that is NOT diode clipping here ? Something inductor-based maybe ? Or at the tube output maybe a transformer-stubstituting gyrator (-- a circuit I know little to nothing about, but am curious --) ?? I understand it's definitely not the CMR shorting-strangler suggested by abbey above. Could not find 'Bluebirds' experiments mentioned above. So any hints are welcome.

P.S.: BTW, Heikki needs 'bad' tubes 'badly' ;)
 

abbey road d enfer

Well-known member
Moderator
Joined
Jan 22, 2008
Messages
12,618
Location
Marcelland
Script said:
In a solid-state vari-mu we have no transformers to potentially shave off erratic transient spikes. We don't necessarily want a comp to react to such spikes (1ms, but also longer ones depending on chosen attack time, as well as strong signal onsets where we had to back off on the attack). So provided such spikes and signals made it through the tube stage without slamming against a transformer that could smear or swallow them, they would appear at the output. 
The interstage xfmr in a vari-mu does not smear/swallow differential mode signals. It does block common mode though.

What would be a smart solution here to get closer to classic vari-mu WITH TX behaviour?
I would think Heikki has provided a brilliant answer to this challenge.

(A) I'd guess driving the tubes harder (spikes dissolve into 'pleasing' distortion)
It would be worse, since tubes would go out of linearity and the differential signals would be even more uncontrolled.

(B) some kind of more or less crude diode clipper (diode clipping can sound unobtrusive when used sparingly but can easily just as well get plain stinky).
Such a clipper should be dimensioned as to not clip the desired differential signal; so it would allow thumps to pass at a level similar to the maximum operating level.

Something inductor-based maybe ?
That's the role of the interstage xfmr.

Or at the tube output maybe a transformer-stubstituting gyrator (-- a circuit I know little to nothing about, but am curious --) ??
A gyrator would replace an inductor. Indeed some vari-mu limiters use inductors instead of xfmrsI. As a result, they shave off some of the LF content of thumps but don't achieve the same level of choking of CM signals as a xfmr.

it's definitely not the CMR shorting-strangler suggested by abbey above.
Why not?
 

Script

Well-known member
Joined
Nov 26, 2008
Messages
1,357
Location
Japan
Thanks for the hint to inductors in place of interstage transformer.

And my bad. I probably didn't make it clear enough that I was not referring to CMR, thump and compression artifacts within a comp, but very short, erratic and musically mostly useless spikes already in some program material before it is sent into a compressor (trouble ahead). Such spikes are probably clashing frequencies that build up in a mix (bad mixing, we could say, so go remix). Also often seen in mixes of electronic music that use digital synths, drum machines, samplers (bad way of using them, we could say). They sit on top of the 'desired' signal, but undesired.

Now some people seem to smooth out such spikes with transformers near saturation, if they are to be believed -- made me wonder (I'd think input or output, not interstage). Or use tubes for soft clipping, which would point to a tube preamp stage or output driver. I do see that a vari-mu tube is not the place for it (even more artifacts, fartz and knartz).
 

5v333

Well-known member
Joined
Jun 30, 2013
Messages
477
Location
Gothenburg
thanks. i am actually aware of this phenomenon in opamps for instance. i just have a hard time picturing what happends at the cathodes. i guess that in order to have rejection, Vgrid and Vcathode must be equal but how does that add up if you get Vgrid from two tubes.
I know this is basics so i just have to look it up really.

But am i right that if small imbalances exists, synthetic inductors with fixed voltage at the plates will decrease thumping alot because dc variations will only have vertical effects and not horisontal?
Im probobly wrong but now ive said it...

JohnRoberts said:
A differential stage is just math, in fact early analog computers did addition and subtraction using antique op amps. 

A differential audio stage with + and - inputs will add the + input and subtract the - input. CMR (common mode rejection) means that if you put 2V into the + input and 2 volts into the - input,  +2V -2V =0V.

CMRR (common mode rejection ratio) is a measure of how well a differential stage rejects common mode (same in both inputs) signal. Of course this is a simplification and cancellations can be degraded by polarity, phase errors, component precision, and more.

JR
 

Script

Well-known member
Joined
Nov 26, 2008
Messages
1,357
Location
Japan
abbey road d enfer said:
A gyrator would replace an inductor. Indeed some vari-mu limiters use inductors instead of xfmrsI. As a result, they shave off some of the LF content of thumps but don't achieve the same level of choking of CM signals as a xfmr.
Found this thread about a vari-mu (by AWA?) with "inductors wound on a common core" to address thump. I love this place. Still reading and trying to digest...

https://groupdiy.com/index.php?topic=52228.msg666564#msg666564
 

Heikki

Well-known member
Joined
Jul 31, 2008
Messages
230
Location
Finland
In case it was still unclear what I was going on about earlier this should clear things up. I made a simulation to show what happens with badly balanced tubes when compressing.

Here is what is being fed to the circuit.
thump-input-small.png

Picture 1.



Here is the output. On green are perfectly balanced tubes and on blue badly balanced tubes.
thump-small.png

Picture 2



Here is the output with 150Hz high pass filter on sidechain.
thump-hpf-small.png

Picture 3

From the pictures it can be seen that HPF does nothing to the thump, but it lessens the over-compression it causes. To me even more annoying than the thump is the over-compression. I don't expect the tubes to stay in perfect balance forever, so I like to have HPF in the sidechain in case the compressor thumps. Of course sidechain HPF will also help with pumping with low frequency material when using fast release times.
 

Script

Well-known member
Joined
Nov 26, 2008
Messages
1,357
Location
Japan
I gave up on the idea of diode clipping in the SC (for now). I thought about a clipper ganged on a pot in parallel to the threshold, but it feels a bit "like trying to hit a bullet with a smaller bullet whilst wearing a blindfold, riding a horse" -- well, wrong metaphor, but effectiveness somewhat limited it seems. It might addresses overcompression, maybe, but HPF sure is easier.

Briefly toyed in my mind with, but completely ditched, the ithought-experiment of running extra tubes (tube halves) in antiphase and then flip for noise-cancellation in SC and in output stage. Doable but a lot of real-estate and, more importantly, unless filtered for the output, it would also cancel much of the uncorrelated differential parts of the signal introduced by the tubes ('sheen' or whatever) -- and with it the reason we want to use tubes in the first place. Might work for SC-fed, but only for absolute plonkers, cos HPF can be had cheaply.

However, I (re-)found a post with a pdf describing a patent from 1952 on some 'thump compensation' (abbey explainng). Interesting what they did in the old days. https://groupdiy.com/index.php?topic=67651.msg859835#msg859835

Also, this thread finally made me try feedforward vari-mu -- and Imust say I quite like it ! Still test-listening for 'FF or FB' and/or '50:50'.

@ Heikki, thanks for sharing your schamatics. I appreciate it very highly.
 

Heikki

Well-known member
Joined
Jul 31, 2008
Messages
230
Location
Finland
Gates had the best solution in the 50's for getting rid of the large common-mode voltages in compressors without a transformer. Gates Level Devil used a long-tail pair with a large tail resistor to get a decent CMR. https://groupdiy.com/index.php?topic=44868.msg983323#msg983323 (take look at V3). It doesn't have huge CMRR like a transformer but with little effort it could be improved.

Script said:
Also, this thread finally made me try feedforward vari-mu -- and Imust say I quite like it !
I think a well designed feedforward vari-mu will do the least damage to signal when tubes are less than perfectly balanced. The problem with feedforward is that you can get into negative ratio where the output decreases when input increases.
 
Top