Simulation of the above schematic. Close to zero audio signal at the servo input. It takes 90 seconds to reach zero offset. Could this be correct?
I just found a circuit that you may find of interest on page 10 of this document,
http://www.ti.com/lit/ds/symlink/lme49610.pdf
It is basically the same concept only it is using a LME49610 as an output buffer for only +/-250ma.
FWIW
jer 🙂
http://www.ti.com/lit/ds/symlink/lme49610.pdf
It is basically the same concept only it is using a LME49610 as an output buffer for only +/-250ma.
FWIW
jer 🙂
it is equivalent to the cap, thus not as much effective as a servo could be, but it's definitely a quite interesting idea! [emoji106]
Of course it's impossible to say in advance whether it would be better or worse than just a plain cap, especially sound wise.
A similar approach could be to remove C9 and use a gyrator (a virtual inductor, of large value) in parallel with the global feedback resistor (same effect: bringing down to unity the gain at DC).
Could be worth trying both.
Last edited:
Back to the servo, the more I think about it the more I believe that the best place to inject the integrator output is between the 318 and the 3886, on one of the two inputs of the Howland Current Pump.
That would require much more (DC) gain by the integrator, but OTOH it would avoid interfering with the input differential.
Moreover, being in the middle of the overall NFB loop, any spurious signal (noise, etc) added by the servo will be drastically reduced by the NFB.
Though I wonder why Mauro have chosen a more conventional approach, acting on the input, for his own "Evolution" design...
That would require much more (DC) gain by the integrator, but OTOH it would avoid interfering with the input differential.
Moreover, being in the middle of the overall NFB loop, any spurious signal (noise, etc) added by the servo will be drastically reduced by the NFB.
Though I wonder why Mauro have chosen a more conventional approach, acting on the input, for his own "Evolution" design...
Last edited:
Is the DC servo positive feedback?
The inverting input sees the injected error. this is inverted to become negative at the amp output.
Then inverted again in the servo and injected into -IN.
equals +ve feedback ??
Have I got too many invertings?
The inverting input sees the injected error. this is inverted to become negative at the amp output.
Then inverted again in the servo and injected into -IN.
equals +ve feedback ??
Have I got too many invertings?
Yes. If the global NFB loop is active till DC (as it is here), the main NFB loop will try to restore any DC offset voltage present at input (times gain) at the output.
But the servo may be set up to have more gain (at DC) than the main loop, so that (at DC) it will "win".
In fact, by subtracting input village from the 3886, the servo will kill the "open loop" amp gain at DC (and subsonic frequencies).
No, of course it must be NFB.
The servo integrator must be inverting or non-inverting depending on the injection point (in this case, which "side" of the HCP).
(After all the servo is nothing else than another loop of negative feedback, acting only at near-zero frequencies, and having 0V instead of the input signal as its reference or "input" so to speak).
Interesting. I'm not familiar with a gyrator. Could you provide an example of point us to information on the gyrator?
Thanks in advance.
Trying to offset inside the loop, i.e. at the current pump, does *not* work. You're fighting against some 100dB of open-loop gain of the 318.
The *only* place to offset adjust/servo any high open-loop gain "opamp" (no matter what topology, nested or not) is at the input transistors. Either at their bases/gates (inputs) or at the collectors/drains (outputs), the latter being used by the standard offset trims offered by many chips. Why at the input transistors? Because that is the only place you can manipulate the input diff-amp's Vbe or Vgs, with their mismatch -- no matter what is causing it -- being the origin and only source of offset, hence the only place it can be adjusted/servoed.
------:------
A floating gyrator accross the fb resistor will be *very* hard to implement, plus it will suffer the same problem what a synthesized cap at the bottom of the feedback divider would have: added noise and distortion.
The *only* place to offset adjust/servo any high open-loop gain "opamp" (no matter what topology, nested or not) is at the input transistors. Either at their bases/gates (inputs) or at the collectors/drains (outputs), the latter being used by the standard offset trims offered by many chips. Why at the input transistors? Because that is the only place you can manipulate the input diff-amp's Vbe or Vgs, with their mismatch -- no matter what is causing it -- being the origin and only source of offset, hence the only place it can be adjusted/servoed.
------:------
A floating gyrator accross the fb resistor will be *very* hard to implement, plus it will suffer the same problem what a synthesized cap at the bottom of the feedback divider would have: added noise and distortion.
Cool. More to read and understand.
Bob,
I commend your research abilities. I guess I was just lazy. I search diyAudio threads and didn't find anything appropriate right away, so I posted the question. Meanwhile, your links are good and I have a lot to learn.
My guess, based on early reading, is that the gyrator would definitely have implications on the sound, but possible a good trade-off. Back to reading.....
Jac
Bob,
I commend your research abilities. I guess I was just lazy. I search diyAudio threads and didn't find anything appropriate right away, so I posted the question. Meanwhile, your links are good and I have a lot to learn.
My guess, based on early reading, is that the gyrator would definitely have implications on the sound, but possible a good trade-off. Back to reading.....
Jac
So far, I like Mr. Elliot's description of the variable capacitor (figure 20 of his Gyrator article)<see Bob's link above>. It would take some experimentation/simulation to understand the effects of the source resistance, but I could imagine replacing C9 with an opamp realized capacitor multiplier based on that circuit. For example, R1 were 47R and R2 were 100k, C1 could be 0.1 uF to have the circuit represent 220 uF. Assuming my uncerstanding and calculations are correct, that is.
I'm sure there is a catch here, beyond any coloring that the active circuit might add.
Any thoughts?
Jac
I'm sure there is a catch here, beyond any coloring that the active circuit might add.
Any thoughts?
Jac
You're right, it can not be done that way. Problem is, by forcing 0Vdc at the output acting "in the middle" (which could be done), the 318 would see the whole input offset voltage across its inputs. Given its high gain (and no local NFB loop reducing it), it will almost surely end up sticking to one supply rail or the other, screwing up everything... 😱
Indeed. In our case the latter are available: what would be the advantages and disadvantages of using one solution or the other?
well, sorry, not quite. That's definitely not the only source of offset in the circuit. For instance, in our case we have at least also the extra offset added by the 3886 (yet, by acting at the input, of course we can compensate for any offset).
Indeed (I was not suggesting to go that way, I was only "thinking out loud" about the proposed idea).
OTOH, unfortunately a servo would not be free of the latter problems, either. No free lunch... 🙁
Last edited:
Mathematically, the offset of the 3886 does contribute to the total offset but in practice it's irrelevant because, again, it is divided by the open-loop gain of the 318. Assume you'd need to pull the input of the current pump to 1V to get zero output offset (that would be an enormous, completely unrealistic offset, but let's use it for sake of easy numbers), this will reflect back an additional voltage offset at the amp input of 0.01mV if we assume 100dB (100,000x) open-loop gain.
of course... but that's taking into account the action of the overall NFB loop. As you state it at first, I had (possibly mis-)understood that you were talking in a much general way. And of course, in general, the (main) NFB loop need not to be necessarily active all the way down to DC, the first stage(s) may have little or no gain (or even attenuation), there may be other "internal" (local) NFB loops around them, etc. Anyway, don't get me wrong: that was just a pointless aside. No pun intended. Sorry. 😱
Now... let's go back in topic. 🙂
I was asking about the relative merits (and drawbacks) of the two different options that you have pointed out. That is, having the servo acting on the inputs or the "offset null" pins of the 318, which is to say on the bases or the collectors of its input LTP.
If I'm not wrong, acting on the collectors (pin 1 and 5) should still keep the servo "inside" the NFB loop (the collectors are the outputs of the input LTP) thus still allowing the NFB action to attenuate (at least to some extent) any unwanted signal introduced by the servo.
Am I wrong?
I wonder: which are then the drawbacks of such an approach, with respect to the other (acting on either one of the inputs)?
And, BTW: should the servo act on just one of the collectors or should it rather act differentially on both of them?
(I mean: acting on either one should be enough to do the servo job... but are there any reason to go differential instead? would that be better, worse or just the same - thus pointless?)
Not directly related to the servo, but interesting nonetheless...
from here: "Musical Fidelity A1 › Technical", talking about the MF B200 (similar in design to the one which have inspired the my_ref idea).
Last edited:
Here is a method that would outperform a servo or a capacitor.
A micro-controller system measuring the amp's DC output with a hardware or software LP filter, The micro-controller uses a digital-potentiometer to regulate the amps offset.
The cost would not be high, needing only one system for multiple channels and can also operate the speaker protection relays.
A micro-controller system measuring the amp's DC output with a hardware or software LP filter, The micro-controller uses a digital-potentiometer to regulate the amps offset.
The cost would not be high, needing only one system for multiple channels and can also operate the speaker protection relays.
Basically yes and I did something like this -- auto-calibration -- for industrial measurement systems, but you'd need as much care and attention to details as with the simple analog integrator, you're just shifting problems into a different domain (need AD, DA and digital filtering).
I dare repeat my personal opinion on the whole servo thing for the MyRef. You don't need it and in case you're picky about the last mV then use the 318's offset pins with a trim pot as per datasheet, plus use good 'lytics at the critical positions (bipolars, i.e.). We have not heard from any offset issues from any MyRef user after all, no?
- Status
- Not open for further replies.