Low Noise DC Servo

[Samuel Groner]

For some particular application I need a DC servo with large authority but low noise contribution. At the same time I'd like to avoide the cost of a low-noise JFET opamp. Is there a possibility I'm so far missing to parallel opamps (or entire servo configurations) without having the offset voltages fighting each other? This would allow me to use a cheap TL074.

The particular configuration is an inverting opamp (current-input amplifier) with a noninverting servo feeding the error signal back to the inverting input. An inverting servo is not possible for this application.

Thanks,
Samuel

 

[jdbakker]

For some particular application I need a DC servo with large authority but low noise contribution.

What do you mean by "large authority"?

How low's the input noise of the amp you're servoing?

What kind of servo range are you looking for? What's the max acceptable post-servo offset / drift?

At the same time I'd like to avoide the cost of a low-noise JFET opamp. Is there a possibility I'm so far missing to parallel opamps (or entire servo configurations) without having the offset voltages fighting each other? This would allow me to use a cheap TL074.

I don't see a way to directly parallel op-amps here, and if you parallel servos they'd better be fairly overdampened or you'll get VLF ringing when their not-quite-the-same impulse responses start clashing.

When implementing servos for audio I usually find that the caps dominate board area and cost, unless you're going for something as exotic as an OPA627. What's wrong with a cheap oldskool OP-07 here?

(I'm probably not telling you anything new, but for limited servo output range I like putting an RC filter at the output (with time constant << that of the servo, of course), followed by a resistor divider. This rolls off servo amp noise fairly well).

JD "uC plus TrimDAC" B.

 

[Samuel Groner]

What do you mean by "large authority"?

What kind of servo range are you looking for?

The servo needs to inject about 2.5 mA.

What's the max acceptable post-servo offset / drift?

Uncritical, say 10 mV.

How low's the input noise of the amp you're servoing?

Servo voltage noise should ideally be below 6.5 nV/sqrt(Hz), although 10 nV/sqrt(Hz) would be acceptable.

What's wrong with a cheap oldskool OP-07 here?

Sure, although offset current is on the high side (10 M resistors). Just want to make sure I'm not missing an easy possibility.

I like putting an RC filter at the output.

Hopeless as noise needs to be low down to 10 Hz.

Samuel

 

[JohnRoberts]

I have paralleled 074s a few times, just because they were there and I needed more output current without using a different part. Simple play is to sum the outputs through low value resistors, where each opamp is regulating it's own output (NI follower), but buffer opamp's inputs connect to output of first opamp so it defines the overall transfer function. Open loop DC output is arithmetic average of the several opamps, but taking the feedback path for the first opamp from the sum of all the opamps makes the DC performance of first opamp dominant. This is not very fast due to lag of two opamps in series, but for a servo it should be fast enough to be stable (I've used this at audio frequency before). 

To make a low noise servo, sometimes you can connect the DC corrective output to a less noise sensitive node than the very input devices.  I did a phono preamp back in the day where I connected the DC servo to the current source feeding the input device. This changed the input DC operating point, and therefore the output DC, but was not injected as 1:1 audio signal.

Of course no matter where you connect it, it will modulate the output, so the obvious technique is to scale back the range so full scale servo output only translates to a modest change in path DC output. Also the noise of the servo device should already be LPF by the servo loop filtering. 

Sorry for restating anything obvious.

JR

 

[PRR]

100H 4K choke between the servo amp output and the inverter's summing node?

Taking sum-node impedance as 1 ohm, you have 4000:1 voltage loss at DC and more at 10Hz and above. 10V swing out of servo will force 2.5mA error-fix current.

How it looks in Bode, I'm not going to think about.

 

[ppa]

my idea is two inv servos connected in series so it would be a non inv servo. the 1st servo use a TL071 and the second one use a 5534 that filter the noise of the 1st. So you can use large resistors in the first servo.
At this point the second servo has not any resistor in the positive input to reduce as well as possible the noise at the 5nV/rt Hz at 1khz.

 

[abbey road d enfer]

I think if we had some hints at the structure the servo is to be grafted to would help a lot answering this subject.
So far JohnRoberts' suggestion seems to be the ticket, provided a suitable control port...

 

[mikep]

I think if we had some hints at the structure the servo is to be grafted to would help a lot answering this subject.

I agree.  so I have no idea if this is relevant, but in some cases a vactrol can be used in a servo loop which, among other things, tends to filter out the control amp noise.  I can think of several other general approaches but have no clue what your circuit looks like.

mikep

 

[Samuel Groner]

Thanks for your responses.

Simple play is to sum the outputs through low value resistors.

Unfortunately this is not possible for a servo as it is an open-loop connection at DC. Imagine two paralleled servos where one opamp has slightly positive, the other negative offset. With the input at 0 V one servo output will swing to the negative rail, the other to the positive due to the large DC gain.

Sometimes you can connect the DC corrective output to a less noise sensitive node than the very input devices.

In fact it doesn't really matter where you connect the servo; as long as its authority RTO stays the same, its noise contribution will also be the same. In some special circumstances a particular node might provide low-pass filtering which will reduce noise--however at the cost of amplitude flatness and other issues. As noted in my second post filtering is not an option here.

I think if we had some hints at the structure the servo is to be grafted to would help a lot answering this subject.

I'd say the description in my initial post should be informative enough (or can I elaborate on a particular point?), but the exact configuration doesn't matter anyway. The problem with fighting offset voltages (as described above) is an inherent one, related to the DC open-loop condition of servos and not to a particular configuration.

In the meantime I've come up with a possible solution where the paralleled servos have artificial local DC feedback (i.e. a resistor in parallel with the feedback cap). After output summing a second stage can be connected in series which correctes the frequency response and provides high DC gain again. I doubt it is worth the trouble though. I might just live with the noise contribution of the TL071.

BTW, this is part of a low noise (< 0.5 nV/rtHz) preamp. If the design ever materialises I'll post it here.

Samuel

 

[JohnRoberts]

Thanks for your responses.

Simple play is to sum the outputs through low value resistors.

Unfortunately this is not possible for a servo as it is an open-loop connection at DC. Imagine two paralleled servos where one opamp has slightly positive, the other negative offset. With the input at 0 V one servo output will swing to the negative rail, the other to the positive due to the large DC gain.

I guess for circuit design a picture is worth more than words but it's easier for me to type. I guess I glossed over the description too much.

Yes the added opamps use 100% local NFB (their outputs tied to - inputs). So the one or more added opamps are each simple non-inverting followers. Their + inputs connect to the first opamps actual output pin. Equal value low resistance summing resistors from each opamp output combine to form a new composite output. Overall feedback (like your feedback cap) connect to this new summed output node.

The DC precision is degraded by the average of the sundry buffer opamps. 

Sometimes you can connect the DC corrective output to a less noise sensitive node than the very input devices.

In fact it doesn't really matter where you connect the servo; as long as its authority RTO stays the same, its noise contribution will also be the same. In some special circumstances a particular node might provide low-pass filtering which will reduce noise--however at the cost of amplitude flatness and other issues. As noted in my second post filtering is not an option here.

Yes... indeed, I was able to modulate a current source for the input device that was already heavily filtered.

I think if we had some hints at the structure the servo is to be grafted to would help a lot answering this subject.

I'd say the description in my initial post should be informative enough (or can I elaborate on a particular point?), but the exact configuration doesn't matter anyway. The problem with fighting offset voltages (as described above) is an inherent one, related to the DC open-loop condition of servos and not to a particular configuration.

In the meantime I've come up with a possible solution where the paralleled servos have artificial local DC feedback (i.e. a resistor in parallel with the feedback cap). After output summing a second stage can be connected in series which correctes the frequency response and provides high DC gain again. I doubt it is worth the trouble though. I might just live with the noise contribution of the TL071.

BTW, this is part of a low noise (< 0.5 nV/rtHz) preamp. If the design ever materialises I'll post it here.

Samuel

Not sure I follow your parallel servo approach. I will be able to relate better to an actual schematic.

Good Luck.

JR