I don't recall ever seeing a circuit which had no compromises. DC-servo is itself a compromise: you swap the problems of capacitors in one part of the circuit, for the problems of a servo and capacitors in another part of the circuit. You are swapping a simple passive component for an active solution - in other circumstances people try hard to go the other way!
Neither! In addition to all the prior comments you want to choose an op-amp with very low DC errors for the DC servo. Choosing an op-amp with spectacular AC performance doesn't make a lot of sense for this application as it is used in an integrator with an HF cut-off well below the lowest audio frequency. (High open loop gain, and low input referred DC errors are what you want.)
I experimented extensively with DC servos in the distant and not too distant past and frankly preferred a really good film coupling cap to the servo.
Sonic...
A DC-Servo, after all, is a filter and every analog filter it's not a brick-wall so also some audible low frequency could be affected.... isn't it?
Too me both seems far superior to TL071 (see attachment)
Which parameters should I consider?
Sadly the cap the DC-Servo should avoid is 220uF...
Attachments
Hi,
On paper they appear broadly similar.
The real concern is however is if the Servo is implemented to minimise it's pull range to what is needed.
I recently came across a design which had a 100K feedback resistor and a 22k resistor from the servo into the feedback node. The Servo was a 072 running on 15V rails, the offset it actually corrected caused less than 0.1V to appear on the Servo Op-Amp output.
So the servo was injecting whatever it did into the feedback node with five times as much gain as the actual signal and it was doing so unnecessarily.
Using a 1M resistor to feed in the servo output caused much more voltage on the servo, but no more actual DC output from the circuit, meanwhile the noise, distortion etc. from the servo entering the signal path was reduced by 34dB!
I have yet to apply an OPA627 (I keep a substantial stock, but need to solder them to an SMD Adapter), but I doubt the effect from this would be as large as optimising the servo's pull-range, at least in this case.
Actually, my next step will be to seek a different point in the circuit to feed the Servo signal into which is inside the feedback loop and with a low amplification factor for AC, so any potential impact is further minimised.
Then we can apply a premium OPA, in which case the OPA627/827 is a very good choice, both due to the DC and AC precision.
Ciao T
On paper they appear broadly similar.
The real concern is however is if the Servo is implemented to minimise it's pull range to what is needed.
I recently came across a design which had a 100K feedback resistor and a 22k resistor from the servo into the feedback node. The Servo was a 072 running on 15V rails, the offset it actually corrected caused less than 0.1V to appear on the Servo Op-Amp output.
So the servo was injecting whatever it did into the feedback node with five times as much gain as the actual signal and it was doing so unnecessarily.
Using a 1M resistor to feed in the servo output caused much more voltage on the servo, but no more actual DC output from the circuit, meanwhile the noise, distortion etc. from the servo entering the signal path was reduced by 34dB!
I have yet to apply an OPA627 (I keep a substantial stock, but need to solder them to an SMD Adapter), but I doubt the effect from this would be as large as optimising the servo's pull-range, at least in this case.
Actually, my next step will be to seek a different point in the circuit to feed the Servo signal into which is inside the feedback loop and with a low amplification factor for AC, so any potential impact is further minimised.
Then we can apply a premium OPA, in which case the OPA627/827 is a very good choice, both due to the DC and AC precision.
Ciao T
Hi,
This can work okay, however two things to watch, if the injection node is very low impedance the pull range may not suffice and for some reason sims show often severe infrasonic peaks...
Ciao T
What about trying a one point DCservo measuring and compensating in the same point at the output using AD712/711. It works well with linestage.
This can work okay, however two things to watch, if the injection node is very low impedance the pull range may not suffice and for some reason sims show often severe infrasonic peaks...
Ciao T
Hi,
I have no idea what "MyRef Evolution" is, however, just because a Servo "works" in practice does not mean it works optimally.
You may wish to ask your friend to measure the actual voltage on the Servo Op-Amp's output. if it is not at least around 3V or so DC (for the common 12...15V rails for the servo Op-Amp) the servo is not optimised enough...
Ciao T
I have no idea what "MyRef Evolution" is, however, just because a Servo "works" in practice does not mean it works optimally.
You may wish to ask your friend to measure the actual voltage on the Servo Op-Amp's output. if it is not at least around 3V or so DC (for the common 12...15V rails for the servo Op-Amp) the servo is not optimised enough...
Ciao T
If you mind about that MyRef Evolution is an evolution of the MyRef, an interesting 'open-source' design by Mauro Penasa presented some years ago here on DIYAudio.
It's a chip implementation (LM318+LM3886) of MF A370 circutiry evolved into a voltage amp followed by a Howland Current Pump.
Mauro Penasa is a competent designer so I would trust that his servo works well...
Sadly MyRef Evolution schematic it's not public so I can't publish or link it...
Anyway when my friend completes the build I'll suggest him to check servo's output voltage as you suggested.
I'm not following your reasoning on this...
The servo output is dependant on the voltage that the main amp requires as a correction voltage. If the output of the servo feeds into a feedback node (it doesn't have too) then the impedance of the feedback node to opamp output must have been figured into the complete design from the start.
The output of the servo should be just a DC voltage with all audio removed by the integrator action. What is of prime importance is that the opamp uses clean reference points as any voltage appearing at the servo inputs (the ground reference) is translated into an unwanted component that appears on the correction voltage.
Hi,
Really?
My point is simple.
The output of the Op-Amp has a certain range it may swing, for the TL071 maybe +/-10V on +/-12V rails.
If there is too much "pull range", that is if the full output swing of the Op-Amp can correct much more than the worst case offset of the circuit then the circuit is not optimised.
An indirect indication of the offset in relation to the maximum "pull range" is the voltage on the output of the Op-Amp. If the circuit has no offset the Op-Amp output will also be zero. If the circuit has the maximum positive or negative offset the Op-Amp output will at the respective opposite maximum.
And you really believe that? I'm surprised anyone would hold believes that are so radically in opposition to reality. Would you mind to show the circuit of that magic integrator that outputs pure DC?
Well, so you seem to allow that there may be something that produces more than DC on the output of an integrator... You just do not seem to allow for any of the major sources of this. So I guess you really do have a magic integrator circuit that behaves in opposition to the laws of physics, do please, tell...
Ciao T
Really?
My point is simple.
The output of the Op-Amp has a certain range it may swing, for the TL071 maybe +/-10V on +/-12V rails.
If there is too much "pull range", that is if the full output swing of the Op-Amp can correct much more than the worst case offset of the circuit then the circuit is not optimised.
An indirect indication of the offset in relation to the maximum "pull range" is the voltage on the output of the Op-Amp. If the circuit has no offset the Op-Amp output will also be zero. If the circuit has the maximum positive or negative offset the Op-Amp output will at the respective opposite maximum.
And you really believe that? I'm surprised anyone would hold believes that are so radically in opposition to reality. Would you mind to show the circuit of that magic integrator that outputs pure DC?
Well, so you seem to allow that there may be something that produces more than DC on the output of an integrator... You just do not seem to allow for any of the major sources of this. So I guess you really do have a magic integrator circuit that behaves in opposition to the laws of physics, do please, tell...
Ciao T
Hi,
Ah, the multiloop amp. The choice of LM318 still seems very odd though...
Ahh, this version makes some more sense. I see Mauro lowered the 3886 gain and applied lead/lag compensation. The howland current pump is an interesting circuit, I am unsure I would use to drive a speaker.
I suspect that much better results would come from using a better (faster than LM318) op-amp and using it in a parallel feed forward error correction circuit together with a simple inverting LM3886.
Looking at the circuits I have at hand I am unsure, Mauro seems to pursue certain technical objectives that seem to have limited reference to the requirements of "good sound". So I would suggest to check.
BTW, as long as the Amp is multi-loop the way it is shown and uses the LM318 I doubt there is much point to try exotic Servo Op-Amp's. Even an NE5534 would likely perform better.
Seeing the circuit with the Howland Current Pump and the fact that it really likes a differential input, I'd be tempted to use something like the OPA1632, much lower impedance feedback networks around the LM3886 and so on...
There is much optimisation, both conceptual and actual that I see remaining in this circuit. Then again, a fully optimised the LM3886 in it's own right is already rather good (objectively and subjectively) so I personally would likely not invest the time to try improving it with loads of extra circuitry.
Ciao T
Ah, the multiloop amp. The choice of LM318 still seems very odd though...
Ahh, this version makes some more sense. I see Mauro lowered the 3886 gain and applied lead/lag compensation. The howland current pump is an interesting circuit, I am unsure I would use to drive a speaker.
I suspect that much better results would come from using a better (faster than LM318) op-amp and using it in a parallel feed forward error correction circuit together with a simple inverting LM3886.
Mauro Penasa is a competent designer so I would trust that his servo works well...
Looking at the circuits I have at hand I am unsure, Mauro seems to pursue certain technical objectives that seem to have limited reference to the requirements of "good sound". So I would suggest to check.
BTW, as long as the Amp is multi-loop the way it is shown and uses the LM318 I doubt there is much point to try exotic Servo Op-Amp's. Even an NE5534 would likely perform better.
Seeing the circuit with the Howland Current Pump and the fact that it really likes a differential input, I'd be tempted to use something like the OPA1632, much lower impedance feedback networks around the LM3886 and so on...
There is much optimisation, both conceptual and actual that I see remaining in this circuit. Then again, a fully optimised the LM3886 in it's own right is already rather good (objectively and subjectively) so I personally would likely not invest the time to try improving it with loads of extra circuitry.
Ciao T
Designing an optimum Servo...
Folks,
It may be worth demonstrating an optimum Servo...
Given is an inverting LM3886, buffered by a single LSK389, DC coupled input, maximum expected DC offset on the Input +/-50mV.
Feedback Network we set to 10K/220K for 27dB Gain (basically a power amp). We use the inverting input to ground.
This means we will inject the servo output into the inverting input feedback node. This means we use an inverter after the inverting servo, to get the polarity right. The commonly (in DIY) seen use of a differential servo is a really poor idea, as this only works with precisely matched resistors and capacitors, very hard to do in DIY with hand selected components, impossible using normal production methodes.
So we use a OPA2604, due to it's acceptable DC precision, FET Inputs and high voltage rails. Running at +/-24V the DC output range from this servo is around +/-22V if not loaded heavily.
If we use 10K for the inverter and 1M/1uF for the integrator we would have something akin to "how things are usually done"...
The LM3886 has a worst case input offset of 10mV. The chosen Impedances in the feedback network together with the input current add another 10mV.
The actual gain for this is 22, so the worst case is +440mV offset (clearly a problem). The +/-50mV from the source we allowed give us another worst case +/- 1.1V extra offset, so -1.1 to +1.54V is the maximum offset we need to deal with.
This means the gain for the 22V maximum from the OPA2604 Servo output must produce 1.54V output from the gain of the LM3886. For simplicity of calculation we take 2.2V (+/-) Pull Range, meaning the actual gain needs to be -20dB. So with a 220K feedback resistor we need a 2.2M resistor from the OPA2604 servo output to the feedback Node.
Now let's look at the AC side of things.
With 1uF/1M we get a turnover of 0.16Hz. So, at 16Hz we have 40dB Attenuation of the AC from Amplifier, at 160Hz we have 60dB, at 1.6KHz we have 80dB and at 16KHz the attenuation is 100dB. All this assumes perfect Op-Amp's, perfect Capacitors and perfect resistors.
So, from the 22V output from the LM3886 the AC output from the servo is 220mV for 16Hz, 22mV at 160Hz and 2.2mV at 1.6KHz. First of all, even if the servo Op-Amp has a significant DC voltage offset on the output we are unlikely to clip the servo Op-Amp with the AC signal, at least not before the main amplifier clips.
Thankfully, as suggested to be implemented the Servo's output is attenuated another 20dB to the output of the main Amp, so the output from the servo will produce at the LM3886 a voltage opposing the actual Signal of respectively 60dB below the signal for 16Hz, 80dB below the signal at 160Hz and 100dB below the signal, so the non-ideal characteristics of Op-Amp and passive parts should be low enough to not constitute a problem.
Had we instead (for arguments sake) used 0.1uF as servo capacitor for a 1.6Hz turnover and a 220K resistor from the servo to the feedback node the inverse signal at 16Hz would be only 20dB below the actual signal, and only 60dB at 1.6KHz, clearly allowing the limited performance of the servo to impact on the performance of the Main Amplifier.
So, correct choice of the turnover (or pre-clipping and pre-filtering the AC part of the signal into the servo) and correct servo gain matters greatly.
Ciao T
Folks,
It may be worth demonstrating an optimum Servo...
Given is an inverting LM3886, buffered by a single LSK389, DC coupled input, maximum expected DC offset on the Input +/-50mV.
Feedback Network we set to 10K/220K for 27dB Gain (basically a power amp). We use the inverting input to ground.
This means we will inject the servo output into the inverting input feedback node. This means we use an inverter after the inverting servo, to get the polarity right. The commonly (in DIY) seen use of a differential servo is a really poor idea, as this only works with precisely matched resistors and capacitors, very hard to do in DIY with hand selected components, impossible using normal production methodes.
So we use a OPA2604, due to it's acceptable DC precision, FET Inputs and high voltage rails. Running at +/-24V the DC output range from this servo is around +/-22V if not loaded heavily.
If we use 10K for the inverter and 1M/1uF for the integrator we would have something akin to "how things are usually done"...
The LM3886 has a worst case input offset of 10mV. The chosen Impedances in the feedback network together with the input current add another 10mV.
The actual gain for this is 22, so the worst case is +440mV offset (clearly a problem). The +/-50mV from the source we allowed give us another worst case +/- 1.1V extra offset, so -1.1 to +1.54V is the maximum offset we need to deal with.
This means the gain for the 22V maximum from the OPA2604 Servo output must produce 1.54V output from the gain of the LM3886. For simplicity of calculation we take 2.2V (+/-) Pull Range, meaning the actual gain needs to be -20dB. So with a 220K feedback resistor we need a 2.2M resistor from the OPA2604 servo output to the feedback Node.
Now let's look at the AC side of things.
With 1uF/1M we get a turnover of 0.16Hz. So, at 16Hz we have 40dB Attenuation of the AC from Amplifier, at 160Hz we have 60dB, at 1.6KHz we have 80dB and at 16KHz the attenuation is 100dB. All this assumes perfect Op-Amp's, perfect Capacitors and perfect resistors.
So, from the 22V output from the LM3886 the AC output from the servo is 220mV for 16Hz, 22mV at 160Hz and 2.2mV at 1.6KHz. First of all, even if the servo Op-Amp has a significant DC voltage offset on the output we are unlikely to clip the servo Op-Amp with the AC signal, at least not before the main amplifier clips.
Thankfully, as suggested to be implemented the Servo's output is attenuated another 20dB to the output of the main Amp, so the output from the servo will produce at the LM3886 a voltage opposing the actual Signal of respectively 60dB below the signal for 16Hz, 80dB below the signal at 160Hz and 100dB below the signal, so the non-ideal characteristics of Op-Amp and passive parts should be low enough to not constitute a problem.
Had we instead (for arguments sake) used 0.1uF as servo capacitor for a 1.6Hz turnover and a 220K resistor from the servo to the feedback node the inverse signal at 16Hz would be only 20dB below the actual signal, and only 60dB at 1.6KHz, clearly allowing the limited performance of the servo to impact on the performance of the Main Amplifier.
So, correct choice of the turnover (or pre-clipping and pre-filtering the AC part of the signal into the servo) and correct servo gain matters greatly.
Ciao T
Thanks... I understand what you are getting at now
and thanks for the detailed write up, I follow now what you are saying.---------------------------------------------------------------------------------------------
As for me having a magic servo... I'm afraid I haven't. The one I designed into my amp (-/+45 volt rail DC coupled) uses a single non inverting servo (TL071) operating on a single neg 12 volt rail. The servo output sits at -6 volts DC.
---------------------------------------------------------------------------------------------
Thanks...
Mooly: It's instructive to stick a spectrum analyzer on the output of the servo to see what AC residual you have. In a well implemented circuit, it will be quite negligible. If you do things right, the tiny residual AC returned to the amp input can also be greatly reduced without losing any of the DC gain.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.