Super Simple Opto comp design - ELOP-inspired

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

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

leigh

Well-known member
Joined
Jun 4, 2004
Messages
394
Location
Portland, OR
SSOPTO.gif


This is a design I've breadboarded and been happy with so far, on the bench. In testing, I'm getting between 12 and 16 dB of reduction at max, and it's sounding good. Would like to get feedback on it before building it into a box.

The inspiration for the design comes from discussion about the Manley ELOP. I have not seen the Manley schematic, however, and I imagine that the similarity ends with "driving a pair of vactrols with an LM386". Regardless, that basic idea stayed stuck in my head for months, and I finally had to give it a go.

As you can see, there is no makeup stage yet, or metering. I've got an idea sketched out for VU GR metering, and I'll post that in a bit. Also, for the power supply, I've just been using a 12 VDC bench supply.

Here is the parts list:

R1 = 5k
R2 = 100k audio taper pot
R3 = 680 ohms
R4 = 332 ohms
C1 = 10 µF
C2 = 2 µF
C3 = 1 µF
S1 = SP3T
D1, D2 = 1N34a
U1 = LM386
U2, U3 = VTL5C2 vactrols

D1 and D2 are in series with the vactrol LEDs, because the vactrol LEDs have a low reverse breakdown voltage (3 volts). I used 1N34a's that have a low forward drop (0.25 V), but other small signal diodes could be used without much change.

The value of R4 (current-limiting resistor) was determined so that the vactrol LEDs would see a max of 20mA, even if the 386 chip was swinging rail-to-rail on 12 volts (single-ended supply, so 6 volts per side). This might be a bit conservative, since it won't actually swing rail-to-rail, and 40mA is the listed "absolute maximum" on the datasheet - but I was getting enough gain reduction out of it, and hitting the LEDs with less current will help their longevity. NOTE: See my next post, below, for more about the actual value of R4.

Thanks for your feedback.

Leigh
 
About the R4 current-limiting resistor: the actual initial calculated value was 220 ohms.

I used this LED calculator:
http://www.ngineering.com/LED_Calculators.htm

And plugged in these numbers:
  • 6 volt supply (again, that would be an unobtainable maximum, since the LM386 is running on 12 volts single-sided, and does not swing rail-to-rail at max)
    2 volt LED drop (the vactrol LED + the protector diode in series)
    desired current of 20mA.
For those values, the calculator gives 200 ohms.

So I started with the closest value resistor I had (220 ohms), but then tried the next-biggest value I had (320) and was still getting enough compression for my ears. Erring on the side of caution, I went with the bigger value. This would also allow the circuit to safely run on a range of supply voltages, from +12 to about +20, and still keep the maximum LED current at around 25ma.

(Although, if you're going to use a supply voltage higher than 12V, make sure you have a flavor of LM386 that can handle it!)
 
You don't need the protection diodes---the other LED conducts and limits the reverse voltage across the nonconducting ones just fine.

Unless you are relying on the diodes to produce some extra thresholding...
 
You could put the vactrols into a schottky rectifier bridge and use each one for one channel, with a simple mod to sum the two into mono before the 386 (for stereo operation).

:?:
 
yeah but.... yeah but.... yeah but.... hmmmm, every "yeah but" I can offer has a "because!"

Very elegant simplicity.

If I must object: TWO costly Vactrols and TWO obsolete diodes.

Yeah but if I do it my way with one Vactrol and a Silicon Bridge, the threshold voltage rises from 2V to 3V, and as you say the 12V amp can barely swing 5V peaks. That gets down to transient action; do we need sidechain headroom? If you have it on breadboard, you might try it with real program material.

BTW: you don't need the series diodes. Small reverse voltage won't hurt the LEDs, and large reverse voltage can't happen because the other LED will clamp the signal. Saves 2 bucks.

An elaboration is to add a panel LED in series with the Vactrol inside the bridge. The relative blink brightness of this LED is a hint if limiting is slight or heavy. But that raises the threshold to 4.7V, so you need a higher-voltage amplifier.

It is possible to design this scheme so no limiting resistor is really needed. Say a safe 20mA in LED forces opto-R down to 500 ohms. And say this happens near 2V output. IF the sidechain amp were unity-gain, the 5K:500R divider means you need 22V input, and few studio sources can do that. (However with LM386 Gv=20, you'd only need 2V in to smoke the LEDs.

The other way to skip a protective resistor is to use a lamer amp. A TL071 or '741 won't put out more current than the LED can stand. Two 9V batts will give some sidechain headroom above even 4.7V threshold.

> there is no makeup stage yet

Sure there is. Take output from the '386. It is the same as the output, plus gain of 20, plus '386 distortions. For this type plan with all control time-constant in unselected opto-R, you can't sneer too much at the '386. But I've done the same plan with a Crown D-150 as sidechain/booster. (Done that way, you attenuate the speaker level signal down to LED level, so the speakers are not-quite smoking when the LED is shining bright and limiting hard.)
 
[quote author="bcarso"]You don't need the protection diodes---the other LED conducts and limits the reverse voltage across the nonconducting ones just fine.

Unless you are relying on the diodes to produce some extra thresholding...[/quote]

Ah, the benefit of fresh eyes. You are correct of course. I had made a note on the vactrol's datasheet to protect against low reverse voltage with a second diode, but didn't realize that it's taken care of by putting the LEDs in this kind of "anti-parallel" configuration.

Don't need them for extra thresholding, no. And removing them also enforces the case for keeping the higher R4 value in there.

thanks,
Leigh
 
[quote author="PRR"]yeah but.... yeah but.... yeah but.... hmmmm, every "yeah but" I can offer has a "because!"

Very elegant simplicity.[/quote]

Thank you!

[quote author="PRR"]If I must object: TWO costly Vactrols and TWO obsolete diodes.

Yeah but if I do it my way with one Vactrol and a Silicon Bridge, the threshold voltage rises from 2V to 3V, and as you say the 12V amp can barely swing 5V peaks. That gets down to transient action...[/quote]

Well, as you say, there's the aspect where a bridge would raise the threshold. Also, using two Vactrols gets you greater maximum gain reduction, all other things being equal, because then you have two LDRs to ground in parallel, rather than just one. (Maybe that's a dummy argument, since it's easy enough to change the value of R1, which is working as a voltage divider with the Vactrols' LDRs.)

Also also, as for using unselected Vactrols, using two averages their response, yes? So I "select" them as far as making sure they have an effectively infinite resistance in their dark state, and let the curves average out from there.

[quote author="PRR"]
> there is no makeup stage yet

Sure there is. Take output from the '386. It is the same as the output, plus gain of 20, plus '386 distortions. For this type plan with all control time-constant in unselected opto-R, you can't sneer too much at the '386.[/quote]

Again, the benefit of fresh eyes! I didn't see it that way, but you're right.

Unless... the diode limiting of the Vactrol's would be audible, when using the sidechain amp as your output. The VTL5C2 lists 1.65V as the typical LED forward voltage drop. So, I believe, if you were running the sidechain hard enough, you'd start to hear those LEDs clipping. ???

cheers,
Leigh
 
[quote author="leigh"]
Again, the benefit of fresh eyes! I didn't see it that way, but you're right.

Unless... the diode limiting of the Vactrol's would be audible, when using the sidechain amp as your output. The VTL5C2 lists 1.65V as the typical LED forward voltage drop. So, I believe, if you were running the sidechain hard enough, you'd start to hear those LEDs clipping. ???

cheers,
Leigh[/quote]

Depends on the output Z of the LM386. I've driven some pretty low impedances with that ancient part, quite a bit lower than the LED series R, which will dominate when voltages near the LED turnon threshold are reached.
 
> if you were running the sidechain hard enough, you'd start to hear those LEDs clipping. ???

Depends on the amp.

'386 with a few dozen ohms in series to the LEDs, the effect is mild, pretty inaudible. D-150 power amp with a 1K divider down to the LEDs, unmeasurable. TL071 with small resistor, you'll have outright peak-clipping. But hard-clipping peaks is an old tradition in limiters.

> using two averages their response, yes?

Perhaps. I was thinking more of time-response. The classic opto-limiters had very select opto-Rs to get a rapid attack and a long release tail, for most euphonious action. The Vactrols tend to be fast and more symmetrical attack/release. As I say, I done it, I can't be snobbish. (Actually they were surplus photo-Rs not Vactrols but I didn't have much choice of response.)

> Maybe that's a dummy argument, since it's easy enough to change the value of R1

That has other effects. At idle, R1's self-noise matters. But on the other hand, lay several values of R1 against the photo-R time response curves. A large value gives fast attack, deep max-GR, and slow release. A small value won't do much until light really soaks into the photo-R, won't GR deep, and will release pretty quick. So you have a compromise between idle hiss and max-GR, also time-constants to optimize (with different optimums for no-blattt speech communication versus classical piano solo clip-safety).

Rather than plot times or compute noise, just switch R1 maybe 3K, 33K, 300K, and let your ears tell you.

Another over-frill concern is distortion. It is low at 100mV and can be audible above several Volts. My monster limiter had a 5V peak threshold at the photo-R, and after some time I realized it was "going sour" before it actually did the limiting thing. My recording limiter is 0.5V at peak level and 0.1V at nominal zero VU, and I never felt it was un-clean. Your ~~2V peak at LED and Gv=20 in front means the photo-R feels about 0.1V max, which is good, but higher might be fine too.

Noise of 5K over the audio band is 1uV (plus '386 noise which is higher). 1uV to 100mV is 100dB hiss to limit level, excellent. Change 5K to 500K for different time-response, S/N is more like 80dB, which was fine for cassette but dubious for CDR or for live PA around mostly-acoustic performance. (Which is about why many opto-limiters and FET limiters use R1 in the neighborhood of 33K.)

> VU GR metering

If you are going to spend all your lunch-money on Vactrols, get the dual-element ones. Derive a 1.23V 1mA DC power point, put another R1 from there to the second opto-Rs, to ground. Bridge the meter (maybe buffered) across the opto-Rs. Dark, the VU feels the full 1.23V DC and sits at zero VU. And when the audio opto-Rs are pulling 10dB GR, the DC opto-Rs also pull 10dB reduction of 1.23V DC, and read down-scale. +/- the matching in the dual-elements, but clearly you are not aiming at great precision. Artistic limiting is "by ear", and the GR meter is just to let you know when you are getting some action or when you are asking too much of the poor limiter.
 
Awesome, this is great analysis guys. Thank you. I feel like I got lucky in picking some of the values in this circuit, since I certainly didn't understand all the effects they would have.

I will try listening to the output of the 386, and see what my ears think.

As for changing the value of R1 around, on the Hamptone opto comp, he was calling it a ratio control, to add a 10K pot in series with a fixed minimum resistance there. In reality, it sounds like it would be effecting several things at once, ratio yes but also the time constants... but I suppose when labeling knobs, "ratio" is as good as anything.

[quote author="PRR"]> VU GR metering

If you are going to spend all your lunch-money on Vactrols, get the dual-element ones. Derive a 1.23V 1mA DC power point, put another R1 from there to the second opto-Rs, to ground. Bridge the meter (maybe buffered) across the opto-Rs. Dark, the VU feels the full 1.23V DC and sits at zero VU. And when the audio opto-Rs are pulling 10dB GR, the DC opto-Rs also pull 10dB reduction of 1.23V DC, and read down-scale. +/- the matching in the dual-elements, but clearly you are not aiming at great precision. Artistic limiting is "by ear", and the GR meter is just to let you know when you are getting some action or when you are asking too much of the poor limiter.[/quote]

The metering solution I came up with was one to avoid the problems of matching between vactrols: Send the input and output nodes to the differential inputs of a JFET-input opamp, and thereby measure their difference. The opamp is set up basically as unity, with 100K resistors in series on each +/- leg, in the feedback loop, and from + to ground. For calibrating, there is also a trimpot in series in the feedback loop, and also of course a trimpot in series with the VU meter.

I've breadboarded this metering scheme, but I haven't gotten to the point of trying to calibrate the meter, so I'm not sure yet if it's got all the necessary pieces for an accurate response.

Currently, the way it's set up, the VU needle sits all the way left at idle, and bumps up during GR. By reversing the differential inputs and changing the feedback loop resistance, it could instead idle at 0dB on the meter, and bump towards the left during GR. Haven't gotten that far with it yet, though.

I'll post my schematic for the metering in a bit here...

cheers,
Leigh
 
Here's what I was talking about for a metering idea:

differential_meter.gif


This is just the basic idea. The feedback resistor would actually need to be smaller, with a trimpot put in series with it, to adjust the meter's response. Also, assume that the caps are all big enough to pass AC down to whatever's important, say 10 Hz.

Note that the JFET-input opamp I was using needs a bipolar supply, hence the biasing/virtual-ground business.

This circuit was quickly breadboarded once, but I'm re-doing it to test it out properly.

Thanks again for your feedback,
Leigh


*****
EDIT: downsized schematic graphic, and added notations...
 
[quote author="PRR"]> there is no makeup stage yet

Sure there is. Take output from the '386. It is the same as the output, plus gain of 20, plus '386 distortions.[/quote]

Whoops, for this circuit, the pot varies the output of the LM386. So it doesn't quite work as-is, since the threshold and output is controlled by the same pot.


In other news, my breadboarding of the meter circuit didn't go very well today. There's a fair amount of interaction between the meter circuit and the main audio output. In other words, when I adjust resistors in the meter circuit, I'm getting quite noticable level changes in the audio. My quick conclusion is that this metering scheme doesn't provide as much isolation from the audio path as I first assumed.

Perhaps the input and output nodes need to be buffered first, before being fed to the differential amp stage? The JFET-input opamp I'm using is a quad chip, so I've got other stages ready to go.

cheers,
Leigh
 
Your metering just won't work, not at that level of complexity. It is reading the instant difference, not the averaged ratio.

However it should not interact with the audio, not for R1 up to 20K or so. Therfore something else is going on with your breadboard.

> the threshold and output is controlled by the same pot.

True. Need to re-arrange the furniture.

Don't worry about dual-opto matching. I think 40% errors are allowed, and that's not many dB. Much of the static error trims-out when you set the meter to read on a mark at no-GR.
 
Am I the only one who has trouble with 900-pixel-wide images embedded in the thread?

Seems like half the size could be ample, and not push text off the edge of my monitor.

29akv7.gif
 
[quote author="PRR"]Am I the only one who has trouble with 900-pixel-wide images embedded in the thread?
[/quote]

In my world (1280x960 on a Macbook Air) Leigh's original schemo is perfect and the one in your last post is too small, or at best just about readable.

Is there a better compromise? Perhaps it's time for a sticky to remind everyone of an image size that works for all.

Cheers,
Ruairi
 
[quote author="ruairioflaherty"][quote author="PRR"]Am I the only one who has trouble with 900-pixel-wide images embedded in the thread?[/quote]

In my world (1280x960 on a Macbook Air) Leigh's original schemo is perfect and the one in your last post is too small, or at best just about readable.

Is there a better compromise? Perhaps it's time for a sticky to remind everyone of an image size that works for all.[/quote]

How about 750 wide?
 
[quote author="PRR"]Your metering just won't work, not at that level of complexity. It is reading the instant difference, not the averaged ratio.[/quote]

Instant difference should be OK, since the VU meter will smooth it over 300ms or whatever, and that's all this needs, a little peak-to-peak memory. The input and output for this circuit should remain perfectly in phase, so an instant read of their amplitude difference should work fine.

However, I forgot to convert the amplitudes to absolute value, so it's not working due to at least that. (When input/output swings positive, the difference is negative, and when I/O swings negative, the difference is positive, and those are averaging out to zero over the brief course of a full wave cycle.) So I'll start with a diode for half-wave rectification on the opamp's output, and see if I'm at least getting some readings that make sense then.

Or am I overlooking some other phenomenon?

thanks,
Leigh
 
I added a diode to the metering opamp's output (between the opamp and the VU meter), and the meter is finally responding in a way that makes sense.

It is, however, resting on the left side of the scale, and bumping towards the right when GR is happening. I want to go with the other style of metering (resting at 0 VU and bumping left), so that's next...

cheers,
Leigh
 
[quote author="leigh"]It is, however, resting on the left side of the scale, and bumping towards the right when GR is happening. I want to go with the other style of metering (resting at 0 VU and bumping left), so that's next.[/quote]

Worked on this for a bit, but feeling that it's a dead end. I came up with a slightly more complicated circuit than the last "differential meter" circuit I posted, but still with the same basic concept of measurement.

I believe the problem is that taking the difference of the two voltages is not what's wanted on a VU meter displaying gain reduction. As PRR hinted, one wants to see the ratio of the two voltages. Decibels are a measurement of ratio, and voltages in their "raw" form are a measurement of amplitude.

So, with the "subtraction" circuit I came up with, 6dB of actual gain reduction shows up differently on the GR meter depending on the amplitude of the input signal.

For instance, say the compressor circuit is causing of 6dB of gain reduction. For this example, I will take 0dB = 1 volt, to make the math simple. Remember, every 6dB drop is a halving of voltage:

1. If the input is at -12dB, the output would be -18dB. In voltages, that would be a difference of 0.125 V. (0.25 V - 0.125 V)

2. If the input is at -6dB, the output would be -12dB. In voltages, that would be a difference of 0.25 V. (0.5 V - 0.25 V)

So, when the GR meter should be reading the same amount, it in fact would be giving different readings.

Ah, hem... so my idea was crap. Now, off to wonder if there's a simple circuit to give a ratio of two voltages. Probably, for the Super Simple Opto circuit, will just drive an LED or VU meter off the LM386 output.

cheers,
Leigh
 
> so my idea was crap

Ideas are cheap.

The insight you gained in the process of discarding this idea is very valuable.

What you want is not subtraction, but division. Divide-by a fixed-ratio is trivial, voltage divider. Divide by a user-set ratio is a buck more: potentiometer. Divide one voltage by another, that's quite hard. The most straight-forward plan is a Multiplier and a feedback loop. But multipliers are nasty and tricky, even before we get to the feedback loop to invert their natural action. And any divider blows-up when you divide by zero, but zero is a common thing in audio. Perhaps never Dead zero, but so small the division product goes through the roof.

> If the input is at -12dB

Say the input was at -50dB. Not interesting for a peak limiter, but audio goes all the way down, and sometimes we wish to pull-up the soft voice from the back of the room, say at a press conference, and do it more consistently than a mere human.

Now we are dealing with few-milliVolt levels. That can be done, but takes more care than Volt signals. 3mV in and 10mV out (10dB boost) +/-5mV error in lowest-cost IC amps (50mV errors in quickie discrete design) is ambiguous.

We would like to work in dB, which is just LOG with arbitary meter numbers.

But log conversion is another nasty analog circuit.

> wonder if there's a simple circuit to give a ratio of two voltages.

Oddly, a Log Ratio circuit works fairly well even with simple parts. Two temp-sensitive log converter's errors buck each other, the residue tempco is small over the shirt-sleeve temperature zone. The output is log, the way we like to see any large (more than 2:1) audio change.

You do have to rectify and smooth first; otherwise every zero-crossing wants to divide-by-zero. If both inputs are zero the log-ratio may stay quiet, but there's always phase-shift.

There are ways around the rectification and smoothing... there's some way to do anything. I think Bob Pease had an elegant log-ratio for measuring utility power.

But note that dividers in general ARE "limiter gain cells". Here you have a 2-part $6.13 divider (opto and resistor), and you are working your way around to a 30+-part measurement plan. You will have more solder, and probably more money, in the metering than the audio itself. Is that wrong? No! More meters! But when you go way overboard on one side, it is worth thinking about the Total Box and if money/effort may be getting thin someplace it matters.

The usual path has been to use a divider with a known Law. A vacuum tube gain-cell will reduce (say) 1.6dB-2.1dB for every Volt of grid control. JFET is similar but lower voltage increments. And, for chopping over-shouts, nobody really cares if the Law is 1.6 or 2.1 (or you can re-mark the meter). You can instead watch cathode current, which may not be so loggy, but has a lot more power in it, enuff to swing a passive needle. Or a transistor VCA is 60mV/20dB +/-1dB all the way from -50dB to +50dB... watching this control voltage gives a high-precision measure of gain reduction.

The other path (rare partly for patent reason) has been the photo-resistor. Somewhere inside the LDR there is a molecule with resistance strictly proportional to light between some limits. But there are a zillion molecules, each a slightly different scale and limits. And few light sources are linear. Not to mention the x/(1+x) factor in the gain equation of a resistive divider. So the sum of light source, divider, and opto-R functions is fairly uncertain. You can't print a meter-scale and build a million and have them all accurate. (You can build one, hand-calibrate, and it will probably stay stable for decades.)

The real dumb yet very elegant meter for opto-compresser is: use two LDRs from the same batch. Shine the same light on both. In the mid-range of light, their absolute values will be different but their change of value will track pretty-much. So you feed audio to one, check that it limits. Measuring its audio output won't tell you the compression because the input is unknown. So you get a 1V DC source (the opto-R does DC same as AC/audio), and a 1V meter, with the second opto between. At zero limiting the second opto does zero reduction, meter shows 1V. When the audio opto is limiting, the DC opto will reduce the 1V DC in a similar proportion. You bench-test for say -3dB audio reduction, and change the resistor in the DC side so it passes 0.707V DC. You get the linear depth of reduction. You actually use a VU meter, which is linear with cramp/spread "dB" marks. An imperfect tool but very familiar (usta be).

This same trick works for JFET limiters. It is possible to pre-compute a JFET's law, but it is easy to take another JFET from the box, feed DC, and trim-out the mismatch between two JFETs.

And any tempco error (small in most opto-Rs but the LED may drift, small but annoying in JFETs) mostly cancels. That's the kind of elegance you need to build a million with cheap labor. Or to build one and then another without major re-discovery. Close and cheap often trumps precise and fussy.
 

Latest posts

Back
Top