Stabilize "composite" amplifier at low gains

[RawCode]

Hello everybody.

I was simulating the circuit in The Art of Electronics at page 154 fig. 3.37 (you can find a schematic attached)

The book says that this circuit has low noise (2nV/sqrt(Hz)), gain of 50 and 20 MHz of bandwidth.
I like this circuit, but I would like to change its gain on demand from 50 to 1. Anyway, studying its stability with MicroCAP 12 loop-probe, I noticed that it has low phase margin (28 degrees) even at gain 20. At gain 1 it became an oscillator.
I would like to ask you how would you make this circuit stable, and in general, any circuit at low gain? Is it possible? Do I have to change JFETs(LSK389B) or opamp (LM6171) to achieve stability?
Thank you in advance

Off-topic questions:
Is it good circuit? Can I achieve better performance (input impedance, noise, bandwidth) with different components, circuit or just with a single opamp?

[moffy]

The major problem with the circuit is too much gain in the first stage, if you put resistors in the source of J1 and J2 you might be able to get the gain manageable but then you would probably degrade the noise performance. The greater the level of feed back the greater the potential for oscillation, you either reduce the gain, the bandwidth or the feedback. Or choose a different op amp like the LTC6227 with 1nv/(Hz)^0.5, 420MHz GBW and unity gain stable.

[RoGeorge]

Let the gain 50, then divide the output voltage by 1...50. 

[moffy]

Let the gain 50, then divide the output voltage by 1...50. 

I thought that too, but refrained.

[Kleinstein]

I see 2 ways to add compensation for lower gain:
1) add a R C series element across the OP-amps inputs. This reduces the input stage gain at the high frequencies and keeps the high gain at low frequencies.

2) Add capacitors and maybe series resistors around the OP amp to make it a differential integrator. So OP-amp output to inverting input and non inverting input to ground (or maybe the possible supply).

The circuit has some quirks with the current source that does not provide good PSRR.  The circuit makes some sense at high gain (quite good GBW), less so at a low gain, as the common mode range can be a bit limited.

[blackdog]

Hi,

Why do you want to convert this schematic to "low gain"?
It is not suitable for that and only with sufficient knowledge can it be converted somewhat for that function.
Kleinstein already indicates how you could start with this.

But as almost always, if a particular schematic has been figured out for a particular function and here that includes low noise and wide bandwidth, then converting it to another function (low gain) never delivers the same good characteristics.

My advice is find a configuration that fits what you need, read up on it,
the Art Of Electronics is a good start and also the application notes by Jim Williams is a good analog bible.

Kind regards,
Bram

[Terry Bites]

Its probably a good circuit, I mean it looks reasonable. Have you simulated it?
Addtional gain needs lag compensation- a reduction in BW to eliminante overshoot.
I'd simulate it and check the pulse response. Add capacitance to the fb path untill he repsonse is free of overshoot.
You can go the hard route and calculate the compensation from the phase response. I always take the shorter path.

 I think there may be more modern monlithic opamps with internal compensation that can compete with the stated performance. maybe an LTC6228.

[Kleinstein]

The LTC6227 is a completely different type of OP-amp:  limited suppl (11.7 V max) and BJT inputs with high bias.
With low gain a low nois is often less of an issue and one often does not need the high GBW.  There may well be modern OP-amps to fit the needs at low gain (e.g. OPA827 could be a candidate).
That type of circuit was popular in the early days when FET bases OP-amps were still quite limited.

For finding suitable values for the compensation it often is better to start from the slow side and than try increasingly less until one gets in the right BW range and keeps peaking (or overshoot in time domain) small.
If the circuit is unstable the simulation may not run at all.

[MathWizard]

In general, whats the advantage of using the external diff-amp, over using a dual op-amp ?

[Kleinstein]

The external JFET differential stage allow for getting very low noise (e.g. 1-2 nV/sqrt(Hz) range).
One has the option (but also the need to) choose the compensation that one want's. So one can make the amplifier intentionally slow or still relatively fast with a high gain.
The choice of OP-amps with external compensation is rather limited. Low noise OP-amps also tend to be rather fast, which sometimes is an issue.

The extra gain from the JFETs also helps with the loop gain, which can help with accuracy and linearity in a high gain (e.g. 100 fold) case.


A 2 op-amp composite (compound) amplifier can also provide extra GBW and loop gain for a high gain case, though often with 2 different OP-amp types, like a low drift/bias/noise one for the input and a fast one for the output. So 2 OP-amps can be an alternative in some cases.

[magic]

Off-topic questions:
Is it good circuit? Can I achieve better performance (input impedance, noise, bandwidth) with different components, circuit or just with a single opamp?

Maybe it would be possible to get similar results by placing JFET source followers in front of a fast and low noise bipolar opamp such as AD797. Some commercial JFET input opamps were actually made that way. I suppose it will be stable as long as the FETs are able to drive the inputs of the chip without excessive phase delay up to tens of MHz.

Another straightforward technique is to parallel and average the outputs of several JFET opamps of sufficient speed in order to decrease noise. IIRC, JFET opamps with noise near 3~4nV/rtHz are readily available so it would take just a few of them.

Note that whatever technique you use to make a unity gain stable amplifier with 20MHz unity gain bandwidth, its closed loop bandwidth will fall to (typically) 400kHz at gain 50x.
The only way out is to employ some current feedback technique, as is common in single chip instrumentation amplifiers for example.

I see 2 ways to add compensation for lower gain:
1) add a R C series element across the OP-amps inputs. This reduces the input stage gain at the high frequencies and keeps the high gain at low frequencies.

This results in a conditionally stable amplifier, i.e. unstable at certain closed loop gains. For stability up to 50x gain, JFET input stage gain must be flat at unity starting from the aforementioned 400kHz (likely somewhat lower in practice). So the resistor should be equal to input stage transconductance (maybe a little lower) and the capacitor must be sized to give appropriate time constant.

2) Add capacitors and maybe series resistors around the OP amp to make it a differential integrator. So OP-amp output to inverting input and non inverting input to ground (or maybe the possible supply).

This is the usual solution in vintage three stage opamps like NE5532 or OP07. It can achieve reasonably flat 20dB/decade open loop gain rolloff and stability at all gains from unity up. There are two singnal paths through X1: integration of input stage current appearing at X1:IN- and direct voltage follower action at X1:IN+. The latter must be suppressed at high frequencies by AC grounding IN+ as mentioned by Kleinstein, otherwise the composite amplifier will never be stable at gains lower than the gain of the input stage alone.