Capacitors in negative feedback network

I'm thinking about something for some days and I'm curious about your opinion and experience.

In all the classic noninverting opamp topology mic preamps like the A*P*I and co. there is always a high value capacitor in the negative feedback network. This capacitor will make sure that there is no DC component amplified by the opamp, but at the same time it is necessary a large value to have the lowest possible cutoff freq and low phase distortion.
The negative aspect is that beeing a large value it has to be some electrolyt condenser (usualy bypassed with a small value polyprop).
Look at the schem at Fred's site:
http://www.forsselltech.com/schematics/JE16%20Mike%20Preamp.PDF
this is a tipical A*P*I topology mic pre. There is the 4700micro capacitor I'm talking about.
But if you look at John Hardy's M1 at page 7:
http://www.johnhardyco.com/pdf/M1_M2_M1p_20031025.pdf
you will see there is no capacitor in the feedback network. It seems logical because the transformer will not let DC into the opamp, there is the BIAS circuit that will minimise any DC offset and there is also the DC servo that will kill any kind of DC to the lowest value.
Question. I read tons of posts at different places about the pros and cons for DC servo. Anyone tried to do some serious listening tests with a similar preamp with the servo circuit in and out, capacitor in and out?
What is the best compromise? To servo and not include the capacitors (feedback and output) or not. How do you feel it affects the sound?
Anyone has the possibility to do a listening test of the M1 with the servo and without ( JP3 jumper permits disabling the servo)

My intention is to start to do some tests in the near future but before I do anything I'm curious about your opinions.

chrissugar

My fully biased opinion is that servos are not all that good. I have tried them in three different amplifiers and the capacitor version was always preferred in listening tests (API type pre, Etude2 pre, etc.) Reportedly, the low frequncies got messed with by the servo. Spent a lot of time in the past agonizing over the same issues and tried to eliminate caps from circuits. Initially, my search was not based on listening results, but driven by the notions of others that capacitors are bad. These days I just use decent capacitors. By the way, until you try it for yourself you will not know and will have to rely on other's opinions. :wink:

I haven't done listening tests, and suspect that either can be satisfactory if done with great care, but I do take issue with people that say the servo eliminates the "effects of capacitors in the signal path." IMO they are in the signal path (along with a lot of other supporting components), just in a less obvious way.

[quote author="tk@halmi"]My fully biased opinion is that servos are not all that good. I have tried them in three different amplifiers and the capacitor version was always preferred in listening tests (API type pre, Etude2 pre, etc.)[/quote]

Hey Tamas,

This is what I want to avoid, to draw conclusions based on my own built preamps. I also tried for example the SSL 9k preamp with servo and without. I dissabled the servo circuits and tried different NE553X opamps till I found some that had very low DC at the output (a couple of milivolts, good enough for the test).
I'm more curious about auditive tests with industry standard equipment made by the gurus like the M1 by John or others, and if in that case there is a consensus about the pros and cons of servo then we can draw some conclusion.
I have the feeling that the problem with the servo apears when it has to work too hard. I supose if you have a DC BIAS circuit like the M1 to minimise offset then the servo will be more relaxed and probably will not affect audio. Although this is just speculation.

Don't missunderstand me, I'm not against capacitors and I'm not a maniac purist, I like transformers and anything that can colour the sound in a good way (for modern music productions), but for classical music and mastering applications I would like the cleanest audio possible. They are two different things.

chrissugar

[quote author="chrissugar"]for classical music and mastering applications I would like the cleanest audio possible.[/quote]

What are you looking at servos for, then? :?

Peace,
Al.

If you trim your op amp carefully, or just luck out and wind up with an op amp with low offset from the git go, and you want to hear what no cap sounds like, you can just-straight wire the darn thing into your next stage, servo be damned. A few millivolts ani't gonna kill anybody. I have tried this with an old Melcor and nobody had to call 911. I also could not hear the cap. I wouldn't leave it like this permanately, but for a quick test, you can A/B the cap vs no cap thing to see if you really want to hassle with the servos.

This two minute test (alligator clip across the cap) could save a lot of TMA. (time, money, aggrevation)
It could also eliminate a valuable learning experience.

[quote author="chrissugar"]if in that case there is a consensus about the pros and cons of servo then we can draw some conclusion. chrissugar[/quote]

Is that a royal "we"? I already have my conclusion. What is the last time we had a consensus here? :green:

[quote author="CJ"] I also could not hear the cap. [/quote]

That's the syndrome I am also suffering from, may be I am not listening hard enough????

two minute test (alligator clip across the cap)

Click to expand...

Hey, that's the method I used also. May be I should patent, copyright and register that........... :grin:

Well, servos can be good, but are not necessarily good. The problem is that there's alot more places to screw up with a servo than there is with a cap.

For a cap, you have general two mistakes you can make, too small a value, or a bad cap (polar where non polar should be, or high ESR, whatever). A servo, you can mess up by having the wrong attenuator network on the output (which lets AC take a different path around the main signal path op-amp), or you can choose to put the break point too low, or you can get noise put back into the audio op-amp from the servo itself.

My view of servos is that you want the servo op-amp running at around 1/2 - 2/3 its rail voltage for the expected DC offset of the op-amp, and the have a nice pad, and filter on the servo output to reduce its effect on the pass band. PM me your email addy, and I can send you a good article on servo design.

Cheers,

Kris

Or you could buy something from John Hardy and be done with it. Considering the R and D you go thru, you might come out ahead. His prices are pretty reasonable.

[quote author="DrFrankencopter"]Well, servos can be good, but are not necessarily good. The problem is that there's alot more places to screw up with a servo than there is with a cap.

For a cap, you have general two mistakes you can make, too small a value, or a bad cap (polar where non polar should be, or high ESR, whatever). A servo, you can mess up by having the wrong attenuator network on the output (which lets AC take a different path around the main signal path op-amp), or you can choose to put the break point too low, or you can get noise put back into the audio op-amp from the servo itself.

My view of servos is that you want the servo op-amp running at around 1/2 - 2/3 its rail voltage for the expected DC offset of the op-amp, and the have a nice pad, and filter on the servo output to reduce its effect on the pass band. PM me your email addy, and I can send you a good article on servo design.

Cheers,

Kris[/quote]

One tricky part about filtering the servo output is that it introduces another pole in the whole system. Unless a compensatory zero is also introduced you will have long tails on your transient response, or worse.


The idea of using just low d.c. offset low drift components is fine, but bear in mind that offsets sometimes originate upstream depending on the source component, so some kind of d.c. reduction may be needed even if your amplifiers are "perfect".

One tricky part about filtering the servo output is that it introduces another pole in the whole system. Unless a compensatory zero is also introduced you will have long tails on your transient response, or worse.

Click to expand...

Don't you want the servo to have a slow transient response?

I don't quite see how it can really be a problem, as long as you choose the break points appropriately, and keep the overall system stable. I guess in a certain sense it depends on what the servo is working on, if the DC value is expected to change often, like on a variable gain amplifier, then maybe you need better transient response from the servo circuit, but if its fixed gain, and just tracking a slowly moving DC value, then I don't really see the problem.

I'm interested in your response, as in simulation I get good performance with a filtered servo output.

Cheers,

Kris

[quote author="DrFrankencopter"]

One tricky part about filtering the servo output is that it introduces another pole in the whole system. Unless a compensatory zero is also introduced you will have long tails on your transient response, or worse.

Click to expand...

Don't you want the servo to have a slow transient response?

I don't quite see how it can really be a problem, as long as you choose the break points appropriately, and keep the overall system stable. I guess in a certain sense it depends on what the servo is working on, if the DC value is expected to change often, like on a variable gain amplifier, then maybe you need better transient response from the servo circuit, but if its fixed gain, and just tracking a slowly moving DC value, then I don't really see the problem.

I'm interested in your response, as in simulation I get good performance with a filtered servo output.

Cheers,

Kris[/quote]

I guess what I should say is that the response will not settle at the optimal single-tau-based rate, but will have a long persistent tail beyond the primary response. For many applications this may well be inaudible.

What about taking the feedback from a transformer winding? No more DC after the trip to magnetic flux and back. You also cancel some of the trafo distortion - a bad or a good thing.

What about taking the feedback from a transformer winding?

Click to expand...

Without further studies I believe that this is not as easy as it might look; I think you need a coupling cap to make a stable feedback connection. And you want to have low DC at the driving amp output for most transformers anyway, so you need a cap or servo here too.

Regarding step response of filtered servo output: I guess that if the main integrator has a -3dB point of < 1 Hz and the output filter sits somewhere at 10 Hz, the impulse response would be almost indistinguishable from a ideal first order response, as the response is more than 20 dB down when the second filter sets in.

Probably related to this topic: If we have two cascaded and servoed stages, how do we choose the two servo frequencies for lowest step response overshoot (i.e. Bessel response)? Or are there other criterions? I believe that if we set the two freq. to the same value, we get a Butterworth response, right?

That's not what they did in the TwinServo (as083.pdf); the second servo is about 1.5x lower in frequency. Any reason for this? Could be a Bessel...

Samuel

Two cascaded integrators will tend to give you a response with arbitrary amounts of peaking, depending on the overall gain---the system looks like a Kerwin-Huelsman-Newcomb state variable filter.

But my guess is that you are right: for many applications Bessel reponse would be optimal. And state-variable filters can be made to give you whatever you like. However, we are after reduction of integrator noise here, so it seems like a final passive network is the way to go.

Another possibility might be to get your final pole from a transconductance amp driving a capacitor, although the loading on the cap has to be light. The appeal here is that the noise gain keeps rolling off indefinitely, unlike a feedback integrator which has no less than unity voltage noise gain until its own open-loop gain starts to fall. I've looked into this for low-noise voltage references.

One could also explore nonlinear techniques, especially if you have to cope with really big offsets coming from upstream...

Two cascaded integrators will tend to give you a response with arbitrary amounts of peaking.

Click to expand...

Just to make sure: I was talking about two (noninverting) gain stages with a servo each (as the TwinServo) and not about a single gain stage with two cascaded servos (which would not make much sense to me).

If we were talking about the same thing, this would be an indication why (in the TwinServo) they always use the same gain in the two stages and not higher gain in the first (which would be advantageous regarding distortion and noise). If the two gain stages would have different gain, this could result in a wild low-freq. response, right? Once I'll have to throw that into SPICE...

Another Q regarding integrators for servo usage: How do we optimize the RC-network for lowest nois? It looks to me as if R and C were in parallel, which means the R-value would contribute to the source impedance at very low freq.s only. In turn this means that for low audio band noise, we basically only want low e-noise from the opamp. Does this sound right?

Samuel

[quote author="Samuel Groner"]

Two cascaded integrators will tend to give you a response with arbitrary amounts of peaking.

Click to expand...

Just to make sure: I was talking about two (noninverting) gain stages with a servo each (as the TwinServo) and not about a single gain stage with two cascaded servos (which would not make much sense to me).

If we were talking about the same thing, this would be an indication why (in the TwinServo) they always use the same gain in the two stages and not higher gain in the first (which would be advantageous regarding distortion and noise). If the two gain stages would have different gain, this could result in a wild low-freq. response, right? Once I'll have to throw that into SPICE...

Another Q regarding integrators for servo usage: How do we optimize the RC-network for lowest nois? It looks to me as if R and C were in parallel, which means the R-value would contribute to the source impedance at very low freq.s only. In turn this means that for low audio band noise, we basically only want low e-noise from the opamp. Does this sound right?

Samuel[/quote]

Yes, I was speaking about a single servo system around a single gain block. Cascaded servos don't interact the same way, clearly. Simulations should be done---hit the candidate circuit with a voltage offset step and look at the evolution---it should be most instructive.

I'm not sure what R-C you are talking about in "Another Q" above.

But if you look at John Hardy's M1 at page 7:
http://www.johnhardyco.com/pdf/M1_M2_M1p_20031025.pdf
you will see there is no capacitor in the feedback network.

Click to expand...

I'm sure you meant that there is no capacitor in series with the output of the 990. There IS a 91pF capacitor in the feedback loop of the 990.

The DC servo circuit works as a high-pass filter at such a low -3dB point that the capacitors are, for all practical purposes, out of the circuit. It is also very predictable, and there is no strange behavior.

John Hardy
The John Hardy Co.
www.johnhardyco.com

Hi John

I'm talking about the output cap and mostly about the cap used in series with the gain control device to the ground (just look at the schematic from the other link). That is a large value cap, usually thousands of microfarads so it is an electrolyt. In your preamp that cap do not exist because the problem is solved with the DC servo, and the DC offset circuit.

chrissugar