Let's Discuss Servos (Previously OPA188 Thread)

[stoeffle]

@ Wayne: hope the pic upload worked...

The amp's output runs through a Sallen&Key 3rd order Bessel LPF with a 1 Hz corner freq. Its output passes a lead-lag network before entering the servo integrator itself.
The overall group delay performance including the power amp itself results in a completely flat group delay curve, within less than +/- 1µs from 10Hz upwards.
All caps are WIMA (the red ones) MKS and FKS foil type, offering low DA values. Using the OP97, offset matters are dealt well with. Trimmer P301 proved to be superfluous. Below 10 Hz the group delay curve is a real roller-coaster one!


@ JR: of course, one could go to extremes and place the servo's corner freq at some tens of mHz, to get a better group delay performance but the reaction times would be unbearable...

On looking at the Jensen schematics posted somewhere on this forum (Jensen_Twin_Servo_Micriphone_Preamp_1984.JPG) one is faced with two DC servos in cascade with a 4:1 ratio in time constants. Albeit the RC time constants are rather short (16.9k & 0.1µF and 66.5k & 0.1µF), indicating that the designer wanted a HPF function. The newer design, dated of 1997 uses 1M & 470nF / 2M & 470nF, which is way of a difference in lower -3dB freq. This latter used AD706 as integrators.

stoeffle.+++++

[JR.]

The concept of a servo is to use an opamp so you can get away from large caps, or are your 1UF caps film? If they are they will probably be physically large.

What kind of reaction time does a DC servo need?


JR

[stoeffle]

@JR:

The concept of a servo is to use an opamp so you can get away from large caps, or are your 1UF caps film? If they are they will probably be physically large.

All caps are film caps (metallized foil), WIMA MKS2 series. The footprint of the latgest one, 1µF/63Vdc & 40VAC is WxLxH in mm: 5 x 7.2 x 10; see http://www.wima.de/DE/mks2.htm . In my eyes this is a rather compact film cap...

Reaction times: In my application, where it needs to just correct long term time drifts, this servo settles within < 1mV (amp output) in 2 sec, for a large transient input signal.

In the case of mic input amps, the response of the loop sometimes is designed to act as a HPF to cut off unwanted spectral signal components. As such it has to have a quite fast response.

My point in this discussion was to draw your attention to the large induced group-delay behaviour impressed by the DC-servo in the loop and perhaps spend a thought or two on the possibly resulting sonic implications. Our hearing perception capabilities at the lower end frequencies of the audio range is rather low (Fletcher Munson diagrams), the curves showing that we are handicapped regarding amplitude discrimination. But are we that deaf that we can't discern phase differences between the fundamental and its related overtones (=harmonics)? I could imagine that we get the hearing impression of something not-natural in the reproduction of sound material.

stoeffle.++++

[ricardo]

Its output passes a lead-lag network before entering the ...

Are these non-minimum phase circuits? eg All-pass networks?

[stoeffle]

@ Ricardo

Are these non-minimum phase circuits? eg All-pass networks?

Hi Ricardo, no, not really. Allpass networks produce a constant phase delay keeping the amplitude rather untouched.
Here we produce a kind of phase equalization in context with the amplifier being corrected, so as to get a truly flat group delay performance (input to output) in the 10..20k Hz audio range.

stoeffle.++++

[JR.]

I am a little busy to dig into this too deeply right now. I am unclear whether you are talking about the group delay of a servo per se, or a 3 pole Bessel filter.

Once again, good practice in my judgement is to place the servo poles way down in frequency, then use a high quality film cap to generate a one pole HPF at an appropriate skirt frequency for audio. This one real pole should be pretty well behaved. The only place where I found a steeper than one pole HPF useful was for warped vinyl record playback.

Metallized polyester is a decent dielectric while I generally use smaller cap values and larger resistor values. Your servo topology does have a real pole before the active electronics so the servo opamp is not seeing the full passband edge rates (a good thing).

You tell us, do you hear a difference between the servo with and without the added phase shift tweak (R305,R306, C305)? I am also a fan of a real pole after the servo stage to protect the servo amp from high edge rate signals coming from the other direction, but that will be influenced by the actual circuitry it interfaces with.

Interesting questions I have no answer for.

JR

[ricardo]

Here we produce a kind of phase equalization in context with the amplifier being corrected, so as to get a truly flat group delay performance (input to output) in the 10..20k Hz audio range.

Are these special phase equalisers minimum phase?

[stoeffle]

Are these special phase equalisers minimum phase?

can't tell, as this correction network was a cut&try work done in simulation...

stoeffle.++++

[JR.]

I have never tried to make a proper filter from a servo but I suspect it will behave similarly to the class of derived crossover filters, where you subtract a LPF from unity to make a HPF response. The resulting response curve for the HPF is not remotely symmetrical.

I would suggest researching "derived" crossovers for some discussion but for that application the desired feature is a unity sum of the two outputs, not a very steep filter response for the derived passband.

I stand by my original design philosophy, make the servo very LF and out of the picture for audio so you can define the bandpass with more well behaved filters.

JR

[mediatechnology]

I have never tried to make a proper filter from a servo but I suspect it will behave similarly to the class of derived crossover filters, where you subtract a LPF from unity to make a HPF response. The resulting response curve for the HPF is not remotely symmetrical.

IIRC, the HPF derived is always 6 dB/oct regardless of the LPF order isn't it? (Or vice-versa if the derived is the LPF section.)