DC servo questions

Help Support GroupDIY:

chrissugar

Well-known member
Joined
Jun 4, 2004
Messages
1,315
Location
EU
OK, now that I have some real world info about the effects of DC servos in sound, I have some questions about this subject.

In my tests I used two different schematics. One circuit used for testing was based on the JLM hybrid schematic: http://www.jlmaudio.com/JLM%20Hybrid%20Opamp%20circuit.pdf and the other was based on Fred Forssell's circuits: http://www.forsselltech.com/schematics/Dual%20Amp%20PCB%20Schematic.pdf
The two circuits are very similar but there is one small difference. After looking to some other schematics that include DC servo, I concluded that most of the circuits take the DC offset voltage directly from the output of the opamp (gain block), and only Fred take the DC offset after the isolation resistor (the one used to isolate the capacitive load of the cable at the opamp output.
Question. What is the reasonig to do it one way or the other way, from a theoretical and practical point of view.

I was a bit intrigued by the fact that only in Fred's circuits I saw this way to extract the DC for the servo, but based on the fact that his tools are considered some of the cleanest sounding tools, I think he has some good reasons to do it this way.

chrissugar
 

Samuel Groner

Well-known member
Joined
Aug 19, 2004
Messages
2,935
Location
Zürich, Switzerland
I don't think that there is much difference between the two. Picking the voltage after the isolation resistor has the advantage of excluding any stray capacity of the track going to the large integration resistor from the audio opamp, but this should not cause any problems with carefull layout otherwise.

Samuel
 

chrissugar

Well-known member
Joined
Jun 4, 2004
Messages
1,315
Location
EU
Question number two.

I started to play with Joe's circuit: http://www.jlmaudio.com/JLM%20Hybrid%20Opamp%20circuit.pdf and in one particular situation I discovered that the circuit without the DC servo activated, at the discrete opamp's output the offset was around 1mV and with the DC servo activated the output offset became almost 2Volts :!: :?: .

This made me think that I did something wrong. Checked the circuit. Nothing wrong. Then came the idea to remove from the socket the OPA2604 and replace it with another. I had the surprise that the offset after very short time stabilised itself at around 0.5mV and remained around that value. WOW. I started to swap the the 2604 and measure the offset. I had 12 OPA 2604 so I measured each one and the results were surprising. Eight of them had around 0.4-0.8mV offset with the servo engaged, two of them around 2mV, one around 20mV, and one had 2Volts offset.

How is this possible? Even 20mV offset is a large value for DC coupling but 2Volts. That specific chip works without any problem if the DC servo side is not engaged, but with the servo you have 2Volts at the output. :?:
For those not familiar with the circuit, I would like to mention that one half of the 2604 is used as part of the hybrid opamp and the other as DC servo. Is it possible that this is because they thermaly interact, beeing in the same capsule, and the servo can't correct the output? Or this is just a bad chip?

chrissugar
 

mikep

Well-known member
Joined
Feb 18, 2006
Messages
450
Location
Philadelphia
[quote author="chrissugar"]
Question. What is the reasonig to do it one way or the other way, from a theoretical and practical point of view.[/quote]

This sort of impacts what I mentioned in your other thread. imagine this circuit is feeding an input transformer with a low DC resistance that happens to be at a different potential relative to your AGND (the value the servo is trying to force the output to). if you pick off the signal after the output isolator resistor like Fred shows, I would think this would lead to the servo driving the signal amp into a rail as it tries to zero the "offset". if you pick it off before the resistor then the situation *should* be happier. as long as the signal amp has enough drive capability it will be able to null, and the voltage will be dropped across that output resistor. I too wonder why Fred did it that way. note that I am really talking about a fault condition here, but this is what worries me about using a servoed output stage.

mike
 

chrissugar

Well-known member
Joined
Jun 4, 2004
Messages
1,315
Location
EU
In the last few hours since I posted the questions I was away and had some time to think about it. I think I have an answer for why Fred takes the DC after the resistor.

I was just thinking about the input receiver circuit of my AV0CET, which is a simple discrete opamp, with the differential inputs beeing a MAT02. This input receiver has the classic opamp with four identical resistors topology. The interesting thing is, that the input of this receiver is DC coupled. In that moment I realised that any significant DC at the input of this receiver would change the polarisation of the input transistors, so they would not work anymore optimally.

This lead to another thought. Why do we want to use a DC servo in an audio circuit? The main reason is to DC couple it to the next stage. So the important aspect is not to have zero DC at the output of the opamp, but at the input of the next device :!: .
If you sample at the opamp out you will have zero DC at the output but not on the cable. But we want to have zero DC on the cable.
Also the cable is in fact a capacitor that has to be charged or discharged to have zero DC, and you will do it efficiently if you "sample" the voltage across this capacitor.
What do you think?

chrissugar
 

mikep

Well-known member
Joined
Feb 18, 2006
Messages
450
Location
Philadelphia
[quote author="chrissugar"]So the important aspect is not to have zero DC at the output of the opamp, but at the input of the next device :!: [/quote]

which is exactly how I see it as well. unless the offset it so large that the amp clips assymetrically and you loose headroom, that is. I think inputs are friendly places to put coupling caps, you pick the load impedance and it never changes. use a FET input w/ high-value bias return resistor and a 1uf (or even less) film cap will perfom superbly.
 

Samuel Groner

Well-known member
Joined
Aug 19, 2004
Messages
2,935
Location
Zürich, Switzerland
In that moment I realised that any significant DC at the input of this receiver would change the polarisation of the input transistors, so they would not work anymore optimally.
What do you mean by "they would not work anymore optimally"? I don't see any reason why the pick-up place from the servo should have any reasonable effect in this cenario, especially as the resistors in the receiving differential amplifier are much larger than the build out resistor of the source and the input bias current is likely to be relatively benign. Please elaborate...

Samuel
 

chrissugar

Well-known member
Joined
Jun 4, 2004
Messages
1,315
Location
EU
[quote author="Samuel Groner"]I don't see any reason why the pick-up place from the servo should have any reasonable effect in this cenario, especially as the resistors in the receiving differential amplifier are much larger than the build out resistor of the source and the input bias current is likely to be relatively benign.
[/quote]

Yes, you are right, the effects at the receiver would be some order of magnitudes lower than at the 50 ohm output resistor. I posted in hurry and without thinking too much about the proportions.

chrissugar
 

bcarso

Well-known member
Joined
Feb 20, 2005
Messages
4,055
Location
San Fernando Valley, Los Angeles
[quote author="chrissugar"]Question number two.


For those not familiar with the circuit, I would like to mention that one half of the 2604 is used as part of the hybrid opamp and the other as DC servo. Is it possible that this is because they thermaly interact, beeing in the same capsule, and the servo can't correct the output?
chrissugar[/quote]

Naaah. I don't know what's happening to produce the huge offset, but thermal interaction on the chip doesn't make sense.
 

mikeg

Member
Joined
Jan 21, 2006
Messages
5
The large offset is coming from the input bias. Each chip has a different bias and the bias can be positive or negative. The servo has to sink/source this along with the current to servo the output.

The optimim solution is to trim the circuit if possible before adding the servo. This way the servo only has to handle thermal changes, not attempt to control the entire loop.

How long after turning on power did you wait before making your measurements? You will need to wait at minimum 10 seconds, if not more.
 

Samuel Groner

Well-known member
Joined
Aug 19, 2004
Messages
2,935
Location
Zürich, Switzerland
The large offset is coming from the input bias. Each chip has a different bias and the bias can be positive or negative. The servo has to sink/source this along with the current to servo the output.
Your statement is somewhat confusing, as "bias" is not a term that makes much sense in this context. Do you mean "input bias current" or "offset"..?

Samuel
 

chrissugar

Well-known member
Joined
Jun 4, 2004
Messages
1,315
Location
EU
[quote author="mikeg"]How long after turning on power did you wait before making your measurements? You will need to wait at minimum 10 seconds, if not more.[/quote]

I left each chip to stabilise for at least three minutes till they were hot. Of course I watched the DC at the output all the time from power up to the last minute, but considered the final measurement the one that was stable for the last minute.

[quote author="Samuel Groner"]
The large offset is coming from the input bias. Each chip has a different bias and the bias can be positive or negative. The servo has to sink/source this along with the current to servo the output.
Your statement is somewhat confusing, as "bias" is not a term that makes much sense in this context. Do you mean "input bias current" or "offset"..?

Samuel[/quote]

I think I understand what Mikeg is talking about. Probably the same thing like the one presented in John Hardy's 990 paper at page 6 called I bias (current bias): http://www.johnhardyco.com/pdf/990.pdf

But this is still not a reason for a servo to not correct a 2V voltage at the output. :?

chrissugar
 

mikeg

Member
Joined
Jan 21, 2006
Messages
5
Sorry. Input bias current and input offset voltage

Check the voltages at the output of the servo amp. For the amps your using that have a high offset even with the servo I'm going to bet (not much money) that the servo amp is railing or close to it.

This can be due to a number of sources; input bias current, the offset of the amp being used as a servo which is added.

If the schematic is correct, I'm not familiar with it. I don't see where the feedback loop is closed around amp. This makes the problem worse since the amp is trying to stabilize open loop.
 

bcarso

Well-known member
Joined
Feb 20, 2005
Messages
4,055
Location
San Fernando Valley, Los Angeles
Chris, what do you have surrounding the "JLM Hybrid" circuit? In particular, what is the d.c. reference for the (noninverting) input? Is the circuit just the four-resistor diff amp configuration? If so what are the resistances?

EDIT: looking at the TI spec on the 2604, I see a typical bias current of 100pA and offset current of +/- 4 pA. Although no maxima are given, to have one amp produce an error of 2V at any conceivable circuit output is a stretch. The loop is closed when the servo is applied even in the "naked" JLM schematic, although there has to be something to reference the n.i. input to common*. Clearly you have it closed by additional components since it behaves with no servo at all.

The 47k however, and the method of implementing the servo action, could well be inappropriate for a given overall circuit. Note for example that, in a four-resistor diff amp, at midband frequencies the 47k acts like a resistor to common, and unless accounted for somewhere else alters the circuit differential and common-mode gains.


*at low frequencies though, only, hence will have huge gains above without additional feedback to limit.
 

chrissugar

Well-known member
Joined
Jun 4, 2004
Messages
1,315
Location
EU
Hey Brad

Before doing the balanced receiver test with or without servo, I played with my existing API topology mic preamps.
Lundahl input transformer, connected to GND and to the noniverting input of the Hybrid opamp, feedback network is 10Kohm from out to inverting input parallel with a 47pico for stability, inverting in to attenuator (20ohm to 10Kohm) in series with 4700micro going to GND. Opamp output conected to a 50 ohm isolation resistor.
Circuit powered from differential +/-24V.
This was the circuit where I first detected this anomaly.
Later built the ballanced receiver test circuit.

chrissugar
 

bcarso

Well-known member
Joined
Feb 20, 2005
Messages
4,055
Location
San Fernando Valley, Los Angeles
[quote author="RogerFoote"]Thought I'd jump in here, since I have been playing with JLM's "Hybrid Op Amp".

After building it without servo, I measured a 630mV offset at the otput of the BD139/BD140 pair after the emitter Rs.[/quote]

If the feedback loop were only closed around the chip opamp, without the servo, and not therefore the whole circuit, then you would see that sort of offset from the output stage.
 

mikep

Well-known member
Joined
Feb 18, 2006
Messages
450
Location
Philadelphia
[quote author="RogerFoote"]Thought I'd jump in here, since I have been playing with JLM's "Hybrid Op Amp".

After building it without servo, I measured a 630mV offset at the otput of the BD139/BD140 pair after the emitter Rs.[/quote]

Like brad said, it sounds like you are taking the feedback from the wrong place. the output pair should be in the loop. it will sound better and the offset, even without a servo, should be just a few mV.

mike
 

bcarso

Well-known member
Joined
Feb 20, 2005
Messages
4,055
Location
San Fernando Valley, Los Angeles
All in all, it sounds like the 2V offset that Chris reported is indeed a bad 2604, probably one with way-excessive bias current at one pin. It would take about a microamp to account for the results. Even though there is no max spec for the part, I'm sure the manufacturer would agree that such a sample was defective.

In the midst of thnking about servos again I did a bit of work on the idea of replacing the differential integrator with a noninverting integrator (apparently first suggested by Deboo, it's basically a Howland current generator with a cap to common). I've done this before but never analyzed it in detail. It turns out that with suitable choice of resistors it has almost as low of noise gain as the differential integrator, and if the input bias current of the opamp used is negligible can have lower noise overall when part of the negative immitance converter at the core of the Howland circuit has lower R's in the inverting portion. It requires accurate resistor values but has the advantage of needing only one capacitor.

In the diff integrator the effect of cap mismatch is a longer settling time with a nonmonotonic step response, so there is something to be said for a one capacitor circuit.

If your overall configuration applies the servo feedback to the main amp's noninverting input OTOH, then a simple inverting integrator can be used for the servo.
 

Latest posts

Top