Vari-mu caps, ripple and oscillation.

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

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

DaveP

Well-known member
Joined
Nov 8, 2005
Messages
3,019
Location
France
I thought I’d explore the phenomena around the audio voltage charging of a capacitor used to provide a DC control voltage (CV) used in many designs of tube compressors.  I shall be using the formula used by Morgan Jones in his book Valve Amplifiers, chapter on Rectifiers Page >300.  He uses it for mains rectification but it also applies to audio voltages too.
The subject is important because it shows how it’s possible to reduce the effects of the feedback oscillation that sometimes happens in these designs.  So, why does feedback even occur in a compressor when it’s supposed to be just a DC signal?  The answer lies in the ripple.

Morgan Jones’s formula for ripple is :-
Vp-p = t*I/C
                                                                                          Where t is in seconds, I is in Amps and C is in Farads. V is peak to peak ripple voltage.
For 50Hz mains rectification, t is 0.01seconds, which is 1/100Hz as the negative cycle is inverted, so two peaks instead of one.
To cut down on the variables I have chosen 10VDC as an arbitrary CV.  I have also assumed a 1M grid resistor so we can arrive at a current of 10/1M which equals 10uA.  I am not going to argue over exact figures, you can work those out for yourselves.
If you look through a lot of schematics, you will notice that nearly all the designs use CV caps between 100nF and 1uF, but I will just use those two values to illustrate the effects.

Let’s see what we get with 1kHz, that will have two peaks when rectified,
So t = 1/2000 = 0.0005, putting that into our formula we get:
0.0005*0.00001/0.000001 = 5mVp-p ripple with a 1uF cap and 50mV with a 100nF cap.  If the phase is wrong then these ripple signals could be positive feedback.
If we look at the bass frequencies it gets worse, 40Hz has two peaks so
t = 0.0125 seconds and in our formula that gives 125mVp-p ripple with 1uF and 1.25Vp-p with 100nF.

Now I know these figures are wrong and we don’t compress sine waves, we compress music, but they serve to illustrate why we can sometimes have problems.  If you have a regular beat of say 120bpm, then that translates to a very low frequency of 2Hz and t = 0.5 seconds.  Maybe that's enough to trigger motor-boating.  Check the number of caps in the signal path.  Check their values, you can see what happens with 40Hz, maybe you don’t want 40Hz reaching your CV rectifier?  I have found that going to a 1uF cap reduces audio ripple on the CV so that it's no longer a problem in some designs.

Best
DaveP
 
> will have two peaks when rectified

So full-wave rectifier?

> If the phase is wrong then...

Then one or the other phase WILL be "wrong".
 
> will have two peaks when rectified

So full-wave rectifier?

Yes, I was referring to the usual case where the audio is rectified by a 6H6 or a 6AL5


Image courtesy of Morgan Jones  (yes I bought the book)

Most compressors involve feedback and though it should all be DC, it often has a high ripple content depending on frequency.  If the gain of the amp is sufficient this ripple can become oscillation.

DaveP
 
I have seen a handful of comps with nasty dirty sound even below threshold, eventually traced to a bad side chain timing cap.....which measured fine with a meter.
 
DaveP said:
> will have two peaks when rectified

So full-wave rectifier?

Yes, I was referring to the usual case where the audio is rectified by a 6H6 or a 6AL5


Image courtesy of Morgan Jones  (yes I bought the book)

Most compressors involve feedback and though it should all be DC, it often has a high ripple content depending on frequency.  If the gain of the amp is sufficient this ripple can become oscillation.

DaveP
I generally try to avoid discussions about tube designs as that is not an area of much experience for me, but this seems like pretty basic dynamics side chain stuff. I do know a little about that.

I/C  describes terminal voltage relationship for a capacitor charged or discharged by a constant current source. Most dynamics side chains involve caps charged/discharged through a resistance limited current. As the capacitor voltage changes the forcing current changes giving the typical curved shape instead of straight line slopes. In fact most side chains use smaller resistance for faster charging, and larger resistors for slower decay.

I'm not a big math guy but I memorized this equation because it's useful to describe capacitor terminal voltage when driven through a resistor. V=1/(e^(-RC) )    For general approximations the RC becomes a time constant and you can estimate how many time constants it takes to attack or decay  x%... For some round numbers A step input feeding an RC will charge up to roughly 63% of the full voltage after 1 time constant. It will get to 95% of final voltage after 3 time constants. 

Even simple dynamic side chains involve at least a R in the discharge path, better different attack and release RCs. I wasted way too many hours designing tricky side chain circuits with adaptive time constants...  the holy grail design is fast when it needs to be, then slow when it doesn't need to be fast.

The typical cause of bad sounding side chains is from too fast release (like perhaps an open cap). You can see this distortion (actually a gain modulation) on low frequency sine waves.

I hope this is germane to your inquiry.

JR 

PS: Your reservoir cap waveform shows the effect of an RC time constant on the charging fraction determined by the transformer winding resistance, while the discharge fraction looks more like a current source.
 
I hope this is germane to your inquiry.

Yes it is JR.

I used the illustration of a power supply capacitor because it was handy.  The source resistance would be very low as you say and the current drain reasonably heavy making almost straight lines as shown, but in the side-chain with much weaker currents and much higher source resistances there would most likely be some sort of curved decay between pulses.

What I was hoping to understand better, was whether some manufacturers deliberately cut the bass going to the side-chain, and relied instead on the mid-range to trigger the compression threshold.  The compression would include the bass even though it would be the mid-range that generated the CV.

DaveP
 
There are too many variations in side chain design for me to make more useful generalizations.

I presume "vari-mu" is a specific dynamics product that I am not familiar with, so I really don't know.... :eek:

Perhaps somebody else can answer specifically about it's side chain design philosophy.

JR
 
With respect to timing, there is nothing really specific about the side-chain in vari-mu compressors, since the control law of the gain cell is hyperbolic, as it is for FET's, optocouplers, SS transconductance cells and diode strings. The only case that differs is logarithmic VCA's, with their dB/V response.
The problem of motor-boating resides in the stability of the feedback control loop. The basic stability criterions apply, except the oscillation risk is at VLF, but the difficulty is to evaluate the transfer function of the loop - which is also non-linear.
Most of the times motor-boating appears for a restricted range of signals; for levels under and close to the threshold, the loop gain is obviously small, and for high levels the compression ratio tends to reach a plateau. In normal conditions, the variable signal will trigger the motor-boating condition erratically and very briefly, so it will be inaudible, but there are cases (bass, synths, where it could be a nuisance.
It is quite difficult to solve this issue, since the VLF phase response of the complete loop is sometimes unpredictable; the most common remedy is to decrease the gain of the side-chain, at the detriment of ratio.
 
[quote author=abbey road d enfer
It is quite difficult to solve this issue, since the VLF phase response of the complete loop is sometimes unpredictable; the most common remedy is to decrease the gain of the side-chain, at the detriment of ratio.
[/quote]

Would dc gain of resulting cv null the ratio compromise?
 
DaveP said:
the transfer function of the loop - which is also non-linear

Abbey,
what does this mean in simple words for simple people like me ???

DaveP
That's unfortunately quite complicated.
The gain I'm referring to is that that defines how tthe overall gain varies vs. the control voltage. Due to the hyperbolic response of the gain cell (in Vari-mu gain varies inversely to the control voltage), the control loop gain is voltage dependant, hence the "non-linear" attribute.
Hope it's a tad clearer...? :-\
 
> I presume "vari-mu" is a specific dynamics product

Manley tried to defend that word/phrase as their own.

The irony is that Mu (amplification factor) is not really what is happening.

Their product is basically ancient (though super-nicely made-up). Vacuum tube push-pull pair. Give it a fixed load impedance, vary the current, Gm changes (because Gm always goes-to zero as current goes-to zero), voltage gain changes. You can vary the current by grounding the cathodes and dropping the grid voltage negative. This is a very high impedance, so can be direct-coupled to the timing capacitor (avoiding DC buffers, which were awkward in tube days).

As said, these are nothing-special for sidechain design. Perhaps more tolerant than some single-ended schemes.

Barry Blesser had some papers about limiter stability. He invents a novel notation, it appears to simplify the analysis, but I have never digested it.
 
DaveP said:
What I was hoping to understand better, was whether some manufacturers deliberately cut the bass going to the side-chain, and relied instead on the mid-range to trigger the compression threshold.  The compression would include the bass even though it would be the mid-range that generated the CV.

DaveP

People filter the side chain to de-emphasize the bass but not for stability or similar reasons. The main reason is more aesthetic than technical. LF information is generally less audible than MF or HF, so if it is allowed to cause gain reduction with the same threshold as MF or HF, it will appear to over process or 'punch holes' in the signal.  By simply EQ-ing a bit of LF out of the side chain, any LF dynamic information won't seem to have excessive control over the gain reduction of the entire signal.

Of course, if you're using a limiter to put a 'box' around a signal, as in the days of FM broadcast, this sort of filtering of the side chain signal will prevent the limiter from doing its job. However, that happens less now, and especially less with hardware limiters, now that high performance digital limiters can do such a good job with relatively few traditional audible problems.
 
Quick related question: Why are first-order (6dB/oct) filters always used in sidechain timing?

I haven't come around to experimenting much yet, but it seems that higher-order timing filters will do their job MUCH faster without distortion-generating-ripple.

Knowing how much speed-of-gain-change impacts the perceived sonic fingerprint of the compression, there surely must be some interesting possibilities here?

Jakob E.
 
gyraf said:
Quick related question: Why are first-order (6dB/oct) filters always used in sidechain timing?
Second-order is sometimes used (in feedforward topologies). But whichever parameter you improve, you end up sacrificing something else instead, e.g. you buy better ripple in exchange for longer release time or longer attack time. You're constrained by the physics of filters. So rather than increase the filter order, people tend to use non-linear techniques like 'adaptable' attack/release times, or whatever.
 
gyraf said:
Quick related question: Why are first-order (6dB/oct) filters always used in sidechain timing?

I haven't come around to experimenting much yet, but it seems that higher-order timing filters will do their job MUCH faster without distortion-generating-ripple.

Knowing how much speed-of-gain-change impacts the perceived sonic fingerprint of the compression, there surely must be some interesting possibilities here?

Jakob E.
Remember it's a feedback structure. Inserting a 2nd-order response in the loop may produce seriously ill effects; when 2nd-order is used, their response must be critically damped.
 
You can vary the current by grounding the cathodes and dropping the grid voltage negative. This is a very high impedance, so can be direct-coupled to the timing capacitor (avoiding DC buffers, which were awkward in tube days).

PRR,

The grounded cathode scheme was used in old radio circuits, and the original German tape-recorder circuits, which I believe you are familiar with.  Are there any special precautions to bear in mind?  If the cap is direct coupled to the grid, do the release caps serve as the grid leaks?



Merlin,

Thanks for contributing to the discussion, I use your web articles for reference all the time, they are very well written.

Best
DaveP




 
DaveP said:
If the cap is direct coupled to the grid, do the release caps serve as the grid leaks?
I think this is what PRR was describing.
 

Attachments

  • Vari_GM.jpg
    Vari_GM.jpg
    22.9 KB · Views: 31
Back
Top