Hi all, are there any DC servo experts in the crowd?
I'm trying a 2-stage non-inverting preamp design, each stage has a DC servo connected to the inverting input of its respective op amp.
The basic circuit of each stage looks like this (except there's no second servo, I'm not using the one connected to the microphone transformer, just the one in the feedback loop).
http://www.jensen-transformers.com/as/as019.pdf
It's what's "in between" the two stages that's of concern. At the moment I have a 330 ohm resistor. When this resistor is removed, each stage works fine by itself.
However, with the 330 ohm resistor in place, there is some kind of weird interaction between the servos. As the first stage gain is increased, the DC offset goes way up (far higher than it normally does with just the one stage).
Here's the problem - there are two symptoms:
1. It appears that the DC offset at the 2nd stage output no longer stabilizes at 0, it stabilizes at "some other value", and that value changes as the 1st stage gain is changed.
2. At some point the DC offset becomes excessive, causing either or both servos to slam to -10 volts, and they stay there until the gain is reduced.
I've tried inserting a large non-polar capacitor in series with the 390 ohm resistor. Done this way, it appears that there is an oscillation when the first stage gain is increased beyond about one third. I suspect that the phase shift through the capacitor is what's causing the oscillation. The oscillation doesn't appear to happen without the capacitor. There is also a "phase reversal", in the sense that the servo starts working the opposite way - it starts driving the output away from zero instead of towards it.
Right now both stages are set up with slow servos, 1 meg and .47 uF puts the time constant way below the audio range. However when looking at each stage alone, you can clearly see the servo action on a meter, the output offset returns to 0 within about 10 seconds. But when both stages are in series, I get either inability to zero (at low gain) or oscillation (at high gain), and if there's a capacitor in series between the stages (presumably DC-isolating both servos) I get the "phase-reversal" behavior where the output is driven away from 0.
What's up here? What should I be looking for/at?
I'm trying a 2-stage non-inverting preamp design, each stage has a DC servo connected to the inverting input of its respective op amp.
The basic circuit of each stage looks like this (except there's no second servo, I'm not using the one connected to the microphone transformer, just the one in the feedback loop).
http://www.jensen-transformers.com/as/as019.pdf
It's what's "in between" the two stages that's of concern. At the moment I have a 330 ohm resistor. When this resistor is removed, each stage works fine by itself.
However, with the 330 ohm resistor in place, there is some kind of weird interaction between the servos. As the first stage gain is increased, the DC offset goes way up (far higher than it normally does with just the one stage).
Here's the problem - there are two symptoms:
1. It appears that the DC offset at the 2nd stage output no longer stabilizes at 0, it stabilizes at "some other value", and that value changes as the 1st stage gain is changed.
2. At some point the DC offset becomes excessive, causing either or both servos to slam to -10 volts, and they stay there until the gain is reduced.
I've tried inserting a large non-polar capacitor in series with the 390 ohm resistor. Done this way, it appears that there is an oscillation when the first stage gain is increased beyond about one third. I suspect that the phase shift through the capacitor is what's causing the oscillation. The oscillation doesn't appear to happen without the capacitor. There is also a "phase reversal", in the sense that the servo starts working the opposite way - it starts driving the output away from zero instead of towards it.
Right now both stages are set up with slow servos, 1 meg and .47 uF puts the time constant way below the audio range. However when looking at each stage alone, you can clearly see the servo action on a meter, the output offset returns to 0 within about 10 seconds. But when both stages are in series, I get either inability to zero (at low gain) or oscillation (at high gain), and if there's a capacitor in series between the stages (presumably DC-isolating both servos) I get the "phase-reversal" behavior where the output is driven away from 0.
What's up here? What should I be looking for/at?
As there is only one gain stage your description makes no sense to me. Secondly reduce the 1M output resistor on the servo output to 470K. Third having built this circuit, it just ain't impressive. There are better sounding simpler circuits. Finally oscillation is easy to get if you have less than the best circuit layout and haven't read AD's design notes on how to bypass a 797.
Let's keep things simple. More servos mean more chances for things to go wrong.
As it is, I have the simplest possible design: one gain stage, and one servo.
It WORKS, as a single stage.
The question is, why doesn't it work when you put two stages in series?
Hi Mr. Gootee - okay, let's back up. First, thank you for your help, and for jumping into this thread. I found your other thread from a few years ago, it sounds like you've done considerable work with servos, so I appreciate your input and your willingness to help.
Next - yes, there appears to be some issue with putting the two stages in series. As you say, I tried disconnecting the servos and just measuring offsets. This is a two-stage preamp, the first stage is an AD-797 and the second stage is a 990c. All very vanilla, non-inverting throughout. For the second stage (990c), I have a 1.5k feedback resistor and 500 ohms in the gain leg from - input to ground, and I have a 390 ohm resistor from the + input to the previous stage. Compensated with 390 pf across the feedback resistor, as per Jensen documentation. So, very little gain, kind of roughly in the same ballpark you were investigating previously (gain of 3, or 4-ish).
Here's what I see: WITHOUT the second stage servo, looking at just the DC offset coming out of the AD797 (so, on the first-stage side of the 390 ohm resistor): if that AD797 output is connected to a 10k load, I see only 20 mV change in output offset as I rotate the gain pot through its range, however if that AD797 output is connected to the 390 ohm resistor I see a MASSIVE swing in the DC offset, like.... volts. Several of them.
Now, please pardon my extreme ignorance, I kinda "know enough to be dangerous", so please explain it to me nice and simple if you would please. (Pretend I'm a novice, which I mostly am).
What am I dealing with here, just the change in load resistance or is there something more going on?
This would explain the second-stage servo's inability to "catch", if the offsets are suddenly volts instead of milli-volts.
This is why I tried putting the capacitor in there in the first place - I figured, with the large-value cap in series with the 390 ohm resistor only the AC would get through, and the DC offset would become "not a factor" for the second servo - so I tried that, and that's when I noticed this strange "phase inversion" thing, where the servo was actually driving the offset AWAY from 0 instead of towards it.
Okay, further info - here's what I've discovered so far:
1. When the AD is driving a 10k load, everything is peachy. When it's driving the 390 ohm resistor followed by the 990c, I suddenly start seeing very large DC offsets at the AD output (not the 990 - the AD in the first stage).
2. In the absence of these large DC offsets at the 2nd stage input, everything seems to work fine, the second stage servo works correctly in that case - however, to accomplish that, I had to do the following:
a. Using a 1k pot connected to the + input of the 990c, measure the resistance at which the native DC offset is exactly zero. Turns out this is 416 ohms (makes ballpark sense, based on a 1.5k/500 ohm resistor ratio).
b. Install that resistor from + input of the 990c to ground. Now I can rotate the 990c throughout its gain range with impunity, with just the resistor in place and no signal, and I can see the servo having to adjust maybe 20-40 mV in either direction.
c. To get clean audio, I had to use a 100 uF NON-POLAR capacitor between the 390 ohm resistor (which is still connected to the AD output, so now I basically have a voltage divider consisting of 390 on top and 416 on the bottom, between the stages - I lose gain, but suddenly my servos are stable).
With this arrangement I can see the servo adjusting in the usual way, and I can see an additional "kick" which is much faster and which I ascribe to the capacitor.
So, okay...... not done yet. There's more. Here's the 990c circuit I'm using, it's basically a John Hardy M-1. http://www.johnhardyco.com/pdf/M1M2M1p.pdf (schematic at bottom)
Note the "bias adjust" circuitry. The presence of this circuitry is what seems to be causing my servo reversal.
At some point, the servo comes in between the zero and the bias voltage, but at another point the zero and the bias go in opposite directions, and I see the servo pushing the output away from zero instead of towards it.
So, I'm trying to quantify exactly what I'm dealing with here. Not the world's best math whiz, so it's taking some time. Apparently I need the maximum DC offset to stay "within" certain limits, otherwise the servo will reverse and drive the 990c to the rail.
1. When the AD is driving a 10k load, everything is peachy. When it's driving the 390 ohm resistor followed by the 990c, I suddenly start seeing very large DC offsets at the AD output (not the 990 - the AD in the first stage).
2. In the absence of these large DC offsets at the 2nd stage input, everything seems to work fine, the second stage servo works correctly in that case - however, to accomplish that, I had to do the following:
a. Using a 1k pot connected to the + input of the 990c, measure the resistance at which the native DC offset is exactly zero. Turns out this is 416 ohms (makes ballpark sense, based on a 1.5k/500 ohm resistor ratio).
b. Install that resistor from + input of the 990c to ground. Now I can rotate the 990c throughout its gain range with impunity, with just the resistor in place and no signal, and I can see the servo having to adjust maybe 20-40 mV in either direction.
c. To get clean audio, I had to use a 100 uF NON-POLAR capacitor between the 390 ohm resistor (which is still connected to the AD output, so now I basically have a voltage divider consisting of 390 on top and 416 on the bottom, between the stages - I lose gain, but suddenly my servos are stable).
With this arrangement I can see the servo adjusting in the usual way, and I can see an additional "kick" which is much faster and which I ascribe to the capacitor.
So, okay...... not done yet. There's more. Here's the 990c circuit I'm using, it's basically a John Hardy M-1. http://www.johnhardyco.com/pdf/M1M2M1p.pdf (schematic at bottom)
Note the "bias adjust" circuitry. The presence of this circuitry is what seems to be causing my servo reversal.
At some point, the servo comes in between the zero and the bias voltage, but at another point the zero and the bias go in opposite directions, and I see the servo pushing the output away from zero instead of towards it.
So, I'm trying to quantify exactly what I'm dealing with here. Not the world's best math whiz, so it's taking some time. Apparently I need the maximum DC offset to stay "within" certain limits, otherwise the servo will reverse and drive the 990c to the rail.
So... this is what I would like to have explained.
Why would a change in the "load conditions" cause a delta in the INPUT offset?
In other words, it makes sense that this is the purpose of the second servo (the one connected to the input transformer) - is to remove this additional DC offset in the case of series-connected stages (like... the Jensen Twin Servo design).
What I would like to understand though, is the mechanism by which this occurs. I thought the + input in follower mode was supposed to be more or less "completely isolated" from the load - is this not the case?
I really am not an expert at this stuff and it would take me a very long time to analyze it to figure out what's going on. And you have not provided a complete schematic of YOUR circuit, and keep mentioning more differences between the schematics that you did provide and what you're actually using. I am not smart enough to offer much help under those conditions. Sorry. But I'll hazard some guesses and comments, below, anyway, and maybe even launch into a diatribe or two (all in the spirit of trying to help you, of course, but I might have a bit of Asperger's Syndrome so please excuse any coarseness, or at least don't be too offended by it.).
The servos in the original Jensen schematic are being used in what to me are ways that seem peculiar to that particular circuit, and seem extremely circuit-dependent. As the servo descriptions there say, they are using tiny currents though certain resistors to do their jobs.
You messed with the circuit, in many significant ways, and then expected the servoing to still work as if you hadn't changed anything. For example, I would have instantly assumed that arbitrarily removing one of the two servos would make the entire circuit crash and burn fiercely. To think otherwise would be like arbitrarily "deciding" that the second servo had no reason for being there in the first place, which would make no sense.
I probably would have started, instead, with a simple classic summing amplifier architecture (and a plain differential integrator servo circuit with a voltage output), so that I had somewhere to sum a servo's voltage output with the signal, and could easily adjust its relative weighting. See AN-20 and AN-31, at national.com (NOT just a suggestion).
Also, once you stick two stages together, it's a whole different ballgame (again). The original circuit probably assumed some resistive-equivalent-to-ground load and source. You probably had no path to ground, or a much different one, after your first stage.
That alone could be a huge problem. Op amps are not ideal: You have to have proper DC circuit paths for sources and sinks for the tiny bias currents for each input pin, and have to balance the impedances, to keep the offsets within reason. Most people use an input resistor, to ground, at each stage. You can add, to that, a bias/offset adjustment network. Again see AN-20 and AN-31. Are you even sure that you are using op amps with the same type of input circuits, i.e. FET or bipolar? It would be hugely important, in Jensen's circuit.
Also, you changed the gain range used, and significantly changed the signal levels involved, at least in the second stage. I would automatically assume that stability and dynamic range could both instantly be problems and would want to carefully think everything through. (Actually, I might have been looking to make more of a composite amp, possibly, or would at least think about a possible need for or benefit from wrapping a feedback loop from the output of the second stage to the input of the first stage. Or maybe that would be crazy. I have no idea. But I would have investigated it, even if only briefly)
Here is a pretty good source for some ideas and information that are way better than mine:
ADI - Analog Dialogue | Op Amp Applications Handbook
There are some very good op amp circuits, and op amp circuit ideas and information, in that book.
I think that you can get where you want to go. But one problem, so far, is combining too many unknowns at the same time. Try a simple op amp amplifier circuit. Investigate it thoroughly. Then add a servo to it. Investigate thoroughly. Try a two-stage amplifier. Ditto. Add a servo to it. Ditto.
I have a gut feeling that one conceptual problem, so far, is the input offsets and bias currents and the DC paths needed, and their design. It is a non-obvious area of concern, if you've never had to worry about it in detail, before.
Also, I would probably start with an easier-to-use op amp, at least at first. For example, I always thought that the OP275 was very versatile, and forgiving. But at the least, realize that every op amp might be different, and have certain difficulties that might be unique to it. So VERY carefully study every op amp's official manufacturer's datasheet, even if you're an experienced expert engineer and designer (in which case you would already automatically do that).
The servos in the original Jensen schematic are being used in what to me are ways that seem peculiar to that particular circuit, and seem extremely circuit-dependent. As the servo descriptions there say, they are using tiny currents though certain resistors to do their jobs.
You messed with the circuit, in many significant ways, and then expected the servoing to still work as if you hadn't changed anything. For example, I would have instantly assumed that arbitrarily removing one of the two servos would make the entire circuit crash and burn fiercely. To think otherwise would be like arbitrarily "deciding" that the second servo had no reason for being there in the first place, which would make no sense.
I probably would have started, instead, with a simple classic summing amplifier architecture (and a plain differential integrator servo circuit with a voltage output), so that I had somewhere to sum a servo's voltage output with the signal, and could easily adjust its relative weighting. See AN-20 and AN-31, at national.com (NOT just a suggestion).
Also, once you stick two stages together, it's a whole different ballgame (again). The original circuit probably assumed some resistive-equivalent-to-ground load and source. You probably had no path to ground, or a much different one, after your first stage.
That alone could be a huge problem. Op amps are not ideal: You have to have proper DC circuit paths for sources and sinks for the tiny bias currents for each input pin, and have to balance the impedances, to keep the offsets within reason. Most people use an input resistor, to ground, at each stage. You can add, to that, a bias/offset adjustment network. Again see AN-20 and AN-31. Are you even sure that you are using op amps with the same type of input circuits, i.e. FET or bipolar? It would be hugely important, in Jensen's circuit.
Also, you changed the gain range used, and significantly changed the signal levels involved, at least in the second stage. I would automatically assume that stability and dynamic range could both instantly be problems and would want to carefully think everything through. (Actually, I might have been looking to make more of a composite amp, possibly, or would at least think about a possible need for or benefit from wrapping a feedback loop from the output of the second stage to the input of the first stage. Or maybe that would be crazy. I have no idea. But I would have investigated it, even if only briefly)
Here is a pretty good source for some ideas and information that are way better than mine:
ADI - Analog Dialogue | Op Amp Applications Handbook
There are some very good op amp circuits, and op amp circuit ideas and information, in that book.
I think that you can get where you want to go. But one problem, so far, is combining too many unknowns at the same time. Try a simple op amp amplifier circuit. Investigate it thoroughly. Then add a servo to it. Investigate thoroughly. Try a two-stage amplifier. Ditto. Add a servo to it. Ditto.
I have a gut feeling that one conceptual problem, so far, is the input offsets and bias currents and the DC paths needed, and their design. It is a non-obvious area of concern, if you've never had to worry about it in detail, before.
Also, I would probably start with an easier-to-use op amp, at least at first. For example, I always thought that the OP275 was very versatile, and forgiving. But at the least, realize that every op amp might be different, and have certain difficulties that might be unique to it. So VERY carefully study every op amp's official manufacturer's datasheet, even if you're an experienced expert engineer and designer (in which case you would already automatically do that).
Okay, thanks. Your hunches are good, the resistance to ground is an issue and I'll be playing with that tomorrow. I'm also thinking about currents going backwards up the bias adjust circuitry and through the power supply, and I'm going to look at that too. (I used John Hardy's bias adjust circuitry for a reason, I didn't use the rest of his design because I see some problems with it, like for instance using a 15-volt servo with a 24-volt gain block rated for +8 dBu input).
As far as making stuff more complicated, none of this is rocket science, right? The only major difference between my circuit and this one:
http://www.jensen-transformers.com/as/as083.pdf
is that mine has the servos connected to the - input instead of the +. It shouldn't matter which way you apply the corrective voltage, right? "Shouldn't", anyway.
Here's a specific question: what is the function of the resistor that's in series with the servo output? Notice in the schematic above, the series resistor on the left is smallish while the one on the right is rather large. Your previous thread seemed to indicated it might be related to the ability of the servo to capture extreme offsets.
Of course it matters! get it the wrong way round makes positive DC feedback
and the output hits a power rail.
R9 and R3 attenuate the A3a servo output by 1500, R15 and R10 attenuate the A3b servo output by 1000. This allows the servos to work at reasonable output levels into high enough impedance and also attenuates any VLF servo noise.
Please post your exact schematic, a photo of a hand drawn one is fine as long as it's legible - JUST POST THE SCHEMATIC
We're not remote mind readers!
So, upon further experimentation, what appears to be happening is the bias-adjust currents from the second stage are flowing backwards into the first stage, thereby disrupting the servo and causing the "settling at non-zero values" I was seeing, and also the variable and greatly increased offsets (from current flowing through a 10k feedback resistor and a few more ohms in the gain leg). These currents are ostensibly set at 2.2 uA, meaning 22 mV-ish across the 10k, which is definitely enough to cause problems. When the bias adjust circuitry is disabled, the offsets change "only slightly", I no longer see the 2-volt swings when I rotate the AD's gain control through its range. Still working...
Well, so, I'm still seeing something I can't explain. Maybe someone can figure it out.
What was happening here, is the "bias adjust" currents were running backwards into the previous stage, thereby disrupting both the stage itself (causing offsets) and the previous stage's servo (causing it to settle somewhere "other than zero").
So, the simple fix was to put a capacitor in series between the stages.
Now, the servo issue is fixed, things settle as they should - HOWEVER - I'm still seeing some very large offsets in the first stage, which are not present when the stage is used alone with an ordinary (10k) load.
When the stage is used alone, I can rotate the gain knob through it's range and I see a total change in DC output offset of maybe 10 mV. That, I believe, is "working as it should".
However when I connect the second stage via the blocking capacitor, rotating the first stage gain knob generates offsets OVER A VOLT at the output of the first stage.
What could be happening here?
What was happening here, is the "bias adjust" currents were running backwards into the previous stage, thereby disrupting both the stage itself (causing offsets) and the previous stage's servo (causing it to settle somewhere "other than zero").
So, the simple fix was to put a capacitor in series between the stages.
Now, the servo issue is fixed, things settle as they should - HOWEVER - I'm still seeing some very large offsets in the first stage, which are not present when the stage is used alone with an ordinary (10k) load.
When the stage is used alone, I can rotate the gain knob through it's range and I see a total change in DC output offset of maybe 10 mV. That, I believe, is "working as it should".
However when I connect the second stage via the blocking capacitor, rotating the first stage gain knob generates offsets OVER A VOLT at the output of the first stage.
What could be happening here?
Did you try inserting, say, 10k to ground at the first stage's output, before the blocking cap? I assume that you also have some sort of R to ground (a path for DC) just after the blocking cap? Usually that last one should be adjusted to equalize the DC impedances seen by the input pins, to minimize the output offset with no servo (or something like that).
Yes, I have the 10k direct load, followed by the cap, followed by 1k to ground. I selected the 1k by looking at the second stage output offset without any AC input, that's the value that seemed to bring the normal offset closest to 0 thereby keeping the servo happy.
At this point, the first stage servo does correct, but it seems like it's working too hard. I don't think there should be 1-volt DC offsets, and I can't imagine where they might be coming from. Same behavior with half a dozen chips, btw.
I have no way of posting a schematic. I already linked to the originals, the second stage is exactly a John Hardy M-1 except that instead of an input transformer it has a 330 ohm resistor (and now a 33 uF cap in series). The first stage is exactly the same thing, except that there's no "bias adjust circuitry" and no output transformer - the output feeds the 330 ohm resistor.
Really, it's the world's simplest circuit. It's just two non-inverting amps in series, coupled by a 330 ohm resistor (and now a 33 uF cap in series). Each amp has a servo, it's very vanilla, exactly like the John Hardy version. The stages work fine and "exactly as expected" by themselves - it's only when they're (AC) coupled that I see the DC offsets in the FIRST stage. They're very dramatic too, instead of 10 mV I see a whole volt.
What this would NORMALLY tell me is there's some DC creeping into the first stage somewhere. I can't figure out where that might be happening. There's no change in the impedance of either input leg, no change in the feedback path, and no DC link to the output. Doesn't make sense.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.