I observed the "transient DC offsets" in actual music, and posted the data. You were the one who was merely asserting, contrarily, with no evidence or theory to back it up, except for your inability to imagine how it could happen. The burden of disproving the possibility of the existence of what was apparently observed in real world music signals is on you, in this case. I was just trying to help you imagine how it was not only possible but extremely likely, even just in theory.
But you did not define, at all, for anyone else (the ones you were wanting to try to communicate with), what is in your mind. When I mentioned that your "inaudible" was not well-defined, I was trying to use a delicate way of saying that what you believe to be inaudible might still be affecting sound quality. i.e. Without knowing more about your knowledge, skills, and abilities, there was no way to know whether or not what you said could be taken at face value.
Obviously, we DID want to talk about it, four years ago, and found it quite interesting. But thank you for giving us your permission to do so.
If you were not interested in this thread, then why did you post?
Personally, I am hugely interested in the inaudible aspects of music reproduction, since they are extremely important and are required in order for us to even be able to create a high-fidelity music reproduction system, i.e. by making noise, ripple, and distortion, and all other undesirable effects, inaudible.
Last edited:
I got pointed at this thread by someone who questioned the servo design in my moskido amp design. I did not mean to be rude - apologies if you got that idea.
I pointed out to him that the servo is "inaudible" in the sense that I cannot hear any difference when it's connected and active, and when it's disconnected. Normally I just use my ears and well known test tracks. But nowadays it's also possible to use tools like Diffmaker which can profile A/B differences.
I certainly agree with the idea that you want to make noise, etc, inaudible. But I was querying the statement:
Better in what sense? Maybe one might be cheaper as it uses less parts?
Surely it cannot be because one sounds better, as we already know that both are inaudible.
Anyway, I won't bother you again.
I pointed out to him that the servo is "inaudible" in the sense that I cannot hear any difference when it's connected and active, and when it's disconnected. Normally I just use my ears and well known test tracks. But nowadays it's also possible to use tools like Diffmaker which can profile A/B differences.
I certainly agree with the idea that you want to make noise, etc, inaudible. But I was querying the statement:
Better in what sense? Maybe one might be cheaper as it uses less parts?
Surely it cannot be because one sounds better, as we already know that both are inaudible.
Anyway, I won't bother you again.
Thanks for the clarification.
It's been a long time since I thought about this stuff. But there were other qualities besides inaudibility (but which are almost always trade-offs vs inaudibility); speed, for example, and correction range, and, of course, DC accuracy. And I guess there would have to also be "more inaudible" or "less inaudible", depending on the gain vs frequency plot through the servo. I believe that was our main criterion, i.e. minimizing what was injected back into the signal path, especially as frequency rose of course. Aguably, we probably went way below the threshold of inaudibility, just to see how low we could possibly make it, while also trying to meet other requirements, such as minimizing time-to-correction and maximizing correctible range. I think that we also toyed with the idea of incorporating a catastrophic-case speaker protection mode, triggered when the offset passed beyond a threshold.
I believe that we almost always wanted two opamps. One would be the differential integrator and one would be used for a DC-accurate active low-pass filter.
It's been a long time since I thought about this stuff. But there were other qualities besides inaudibility (but which are almost always trade-offs vs inaudibility); speed, for example, and correction range, and, of course, DC accuracy. And I guess there would have to also be "more inaudible" or "less inaudible", depending on the gain vs frequency plot through the servo. I believe that was our main criterion, i.e. minimizing what was injected back into the signal path, especially as frequency rose of course. Aguably, we probably went way below the threshold of inaudibility, just to see how low we could possibly make it, while also trying to meet other requirements, such as minimizing time-to-correction and maximizing correctible range. I think that we also toyed with the idea of incorporating a catastrophic-case speaker protection mode, triggered when the offset passed beyond a threshold.
I believe that we almost always wanted two opamps. One would be the differential integrator and one would be used for a DC-accurate active low-pass filter.
Hi all, I found this thread, which contains some servo information at the level of detail I probably need. (I explained my servo issue in the other thread, I'm trying to make a 2-stage preamp work, and I'm getting what appears to be some servo-related oscillation or "maximum range exceeded" or some such thing).
Let's see, I've never used SPICE but I'm good with electronics/schematics, should I be trying to simulate my circuit this way? I have 1meg/.47uF integrator, and 100k resistor in series with servo output. I noticed specifically Mr. Gootee's comments about "the resistor isn't big enough", maybe this is what's happening to me?
How can I find out? What's the best path to the goal (smooth servo operation)?
http://www.diyaudio.com/forums/analog-line-level/238825-some-questions-about-dc-servo-mic-pres.html
I'm probably seeing two separate symptoms: one is offsets outside of the catch range, the other is some kind of phase inversion that causes the output offset to grow larger instead of smaller.
The thing I'm most grappling with is the oddball behavior when I put a series capacitor between the two gain stages.
This definitely changes the servo behavior, and makes the "phase inversion" appear more prominently and more easily.
Both my gain stages are variable gain, the first one has 20-60 dB and the second one has -6 to + 14. The first one is an AD797 and the second one is a 990c. They're both very vanilla non-inverting amps, there's a 300 ohm resistor between the amps (which is where I put the series capacitance to get the oddball "phase shift" or whatever's happening there).
So, I have an OPA-602 as the first servo (with the AD797), and typically as I rotate the gain control I see maybe 100 mV maximum change in DC offset. This is when the gain stage is operated by itself, without the second gain stage following it. The servo works just fine this way. The servo here is "slow", it's set to 1meg/.47 uF. But it corrects the 100 mV offset in a few seconds, everything's peachy when the stage is operated by itself.
In the second stage I have an OPA-604, running at +/- 24 like the 990c, and that uses the same component values as the first stage, 1meg/.47uF for the integrator and a 100k resistor in series with the servo output.
This definitely changes the servo behavior, and makes the "phase inversion" appear more prominently and more easily.
Both my gain stages are variable gain, the first one has 20-60 dB and the second one has -6 to + 14. The first one is an AD797 and the second one is a 990c. They're both very vanilla non-inverting amps, there's a 300 ohm resistor between the amps (which is where I put the series capacitance to get the oddball "phase shift" or whatever's happening there).
So, I have an OPA-602 as the first servo (with the AD797), and typically as I rotate the gain control I see maybe 100 mV maximum change in DC offset. This is when the gain stage is operated by itself, without the second gain stage following it. The servo works just fine this way. The servo here is "slow", it's set to 1meg/.47 uF. But it corrects the 100 mV offset in a few seconds, everything's peachy when the stage is operated by itself.
In the second stage I have an OPA-604, running at +/- 24 like the 990c, and that uses the same component values as the first stage, 1meg/.47uF for the integrator and a 100k resistor in series with the servo output.
Right. I have non-inverting servos feeding into the - input of the amplifiers. My schem looks just like a Jensen Twin Servo except the servos are connected to the - input(s) instead of the + input(s).
It appears that this is specifically the source of my problem: something is increasing the input bias current in the first stage AD-797 (thereby increasing the input offset and therefore the DC at the output).
I'm theorizing that it's the "bias adjust circuitry" in the second stage that's doing it. That circuit looks exactly like it does in a John Hardy M-1, it's just a small positive voltage that gets applied to both inputs simultaneously.
The AD, for some reason, doesn't like this "small positive voltage" at the output (ie the load returns to non-zero/non-ground). Not sure, but it almost seems like it's working its way backwards into the first stage bias current.
a non inverting servo feeding the +IN input will drive the output to rail.
An inverting servo feeding the -IN input will drive the output to rail.
A proper selection uses a Non inverting servo to feed the -IN input.
or an inverting servo to feed the +IN input.
Think it through.
Output offset goes +ve. Non inverting servo goes +ve at both it's input and output.
That +ve is sent to the -IN and forces the amp output towards -ve, i.e corrects the +ve offset.
An inverting servo feeding the -IN input will drive the output to rail.
A proper selection uses a Non inverting servo to feed the -IN input.
or an inverting servo to feed the +IN input.
Think it through.
Output offset goes +ve. Non inverting servo goes +ve at both it's input and output.
That +ve is sent to the -IN and forces the amp output towards -ve, i.e corrects the +ve offset.
Adding such dc servo loop to a otherwise flat-frequency-response amplifier is effectivly changing amplifier to a high pass filter, the corner frequency is K/(2*Pi*R5*C4) as in #4 schematic, where K is amplification from servo opamp output to main Amp output. In this case, K is roughly 0.5. so the corner frequency is 0.15Hz, change R5 to 100k, this corner is moved to 1.5Hz, to a 20Hz music frequency lower limit, it is going to be hard to tell the corner difference between 0.15Hz to 1.5Hz, unless other conponent has different value so your corner is not 0.15Hz to start with.
To a music signal, I do not believe 0.15Hz or lower corner frequency can make an audable difference, but lower corner frequency will make the whole amplifier slower to settle at power up.
To a music signal, I do not believe 0.15Hz or lower corner frequency can make an audable difference, but lower corner frequency will make the whole amplifier slower to settle at power up.
Jeez.... yeah, I'm not THAT much of a beginner.
Inverting servo should feed the non-inverting input, or a non-inverting servo should feed the inverting input. Easy.
And, take your pick, there should be zero difference in terms of the result, right? Either way, your output gets driven to 0 DC and stays there.
This is a much more sophisticated problem. Something is actually changing the behavior of the first stage, not just the servo but the amplifier itself. I'm thinking it's the second stage "bias adjust circuitry", and I'm about to start playing with that, so I should have some details for you by the end of the day. The thing is, when the AD is plugged into an ordinary 10k load (resistor) there's only 20 mV of output offset change when the gain pot is rotated through its range. When the AD is connected to the second stage instead, suddenly there are VOLTS of offset when the first-stage gain pot is rotated through its range. That's the first problem I need to understand, before I can select the proper component values for the servo. Yes?
Yes. In my case it turns out to be a little more complicated than that - I'll explain it in the other thread. (As best I can, at this point).
I believe I already linked to the John Hardy schematic, yes? If not, here it is again. http://www.johnhardyco.com/pdf/M1_M2_M1p_20031025.pdf
Note the "bias adjust circuitry". That's the culprit, in my case.
And if you look at the 990c schematic here http://www.johnhardyco.com/pdf/990.pdf you'll see the pair of diodes across the inputs...
Anyway, so, if you're DC-coupled to the previous stage the currents can run backwards into the previous stage's servo.
I am not an expert, but the type of ceramic cap that is available in large values (i.e not NP0 or C0G) might have issues with being microphonic, and with temperature or voltage coefficient of capacitance (capacitance might vary when temp or voltage varies). Or maybe not. But you should check. And if any of those ARE significant, then you should also check how they might affect servo operation.
Other than issues like those, polypropylene and ceramic usually both have very low dielectric absorption ("voltage memory" effect); better than most other types except teflon, I think.
Edit: WIMA had some little red polypropylene box caps that came in 1uF, that I think were roughly 1 cm x 1 cm footprint, or less.
Other than issues like those, polypropylene and ceramic usually both have very low dielectric absorption ("voltage memory" effect); better than most other types except teflon, I think.
Edit: WIMA had some little red polypropylene box caps that came in 1uF, that I think were roughly 1 cm x 1 cm footprint, or less.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.