ChrioN said:
I'm toying arond with a precision dc circuit that requires something like a jfet op amp -
choppers and such are out of the question.
I claim that choppers and such are not out of the question. The Analog ADA4522-1 or -2 or -4 are quite wonderful amplifiers for DC servos, and I have been using them successfully as integrators in DC servo circuits for the past few years with a variety of amplifier circuits. They are low cost zero drift (chopper) amplifiers, they have very low residual offset (~7µV), practically unmeasurable output spurious tones, and great basic linearity to start with.
Because of the zero drift "chopper" system, they also have extremely low 1/f noise, far less than any of the typical JFET amplifiers that get pressed into integrator service for a DC servo. The integrator stage can inject its 1/f noise into the controlled amplifier stage because of the loop gain structure of the servo and the controlled amp. So, it makes sense to select an integrator amplifier that has low 1/f noise, and low noise in general. The ADA4522-2 (the dual that I'm using to control a balanced amplifier) works quite well in regards to this noise.
Additionally, the two amplifiers in the ADA4522-2 are run with the same switching clock, so there is no chance of 'birdies' caused by slight inaccuracies in the switching rate of one amplifier relative to the other. In short, there is no way to make these amplifiers present measurable switching tones into an amplifier when used as a DC servo as an integrator amplifier. I have experimented with passive RC post filters, and they do not have any effect at all - there are no output spurs to eliminate with post filtering.
The biggest drawback of these amplifiers is a synthetically high input "bias current" caused by charge switching effects from the rapidly switching input stage. This is a CMOS amplifier with an extremely low inherent input bias current, but leakage currents are generated by the switching action of the zero-drift switching mechanism as it switches the input stage at the chopping frequency. These charge glitches, when impressed upon the input passives of an integrator, get lowpassed and act more like an input bias current than a switching noise.
These currents therefore require some caution when scaling the impedance of the components around the input stage, to prevent the total offset voltage from being increased excessively as this bias is impressed upon the input resistors. This means that larger capacitors and smaller resistors should be used around the input stage to minimize the effect of these "bias" currents. I'm using 100nF caps and 66K5Ω resistors in a noninverting integrator circuit, and this seems to increase the offset by only a handful of microvolts - an acceptable range to me. This behavior seems to be modeled realistically by the SPICE model for the amplifier, so you can use SPICE to see how your overall design will perform relative to these "bias currents".
If you've designed filters before, you're probably thinking that the time constant provided by an integrator using 100nF and 66K5Ω is not far enough below the audio band (-3dB is around 24Hz). So, how does this work out in a DC servo, where the -3dB point ought to be only a few Hz at greatest? This trick is done by attenuating the "force" output of the DC servo, so that the integrator amplifier is attenuated into the input of the controlled amplifier. This attenuation factor increases the servo's time constant by an amount proportional to the attenuation factor. With this 24Hz -3dB integrator, I typically use a 20:1 output attenuation and then end up with a ~1.2Hz corner frequency of the controlled amplifier, and a generally acceptable time response of the controlled amplifier.
The DC servo drives an inverting amplifier with a single resistor into the controlled amplifier's inverting input node. The inverter uses a 10kΩ feedback resistor, so I feed the DC servo integrator output signal into the amplifier's inverting node with a 200kΩ resistor to get the 20:1 attenuation.
The tradeoff of this output attenuation is that the correction range of the servo is diminished by the servo output attenuation factor. With ±15V supplies, the 20:1 attenuation means a correction range around ±749mV. Still, for these amplifiers in this application, this is completely sufficient.
On the positive side, the noise of the servo integrator stage is also reduced by the same factor. With a 20:1 attenuation, this results in no appreciable added noise from the servo, even with extremely low noise stages.
A final and significant benefit of all of this is that the integrator requires only a 100nF integrator capacitor. These capacitors can then be a small 3216 size C0G SMD capacitor, which is extremely tiny, extremely linear, non-microphonic, and extremely reliable compared to a film integrator cap.
Overall, using a modern zero drift amplifier as an integrator with carefully sized passive components can result in a very high performance integrator at low cost and small PCB area. IMHO you should indeed consider using them.