Hi Frank,

It would be good if you could describe how your circuit degrades when you have the chopper active with the actual OTA circuit.
The switches connection seem to be ok concerning the chopper phases. If you say the chopper works well with an ideal OTA I would imagine you might be running into slew rate or bandwidth problems when you use the actual OTA: I imagine you might have some BW limitation at the output of your OTA due to the large output impedance. If this BW is not much larger than the chopping freq then you are in trouble. Also it is usually recommended for SC amplifiers to run the chopper at exactly half of the switched cap clock frequency.

Regards
Tosei

Have you run some sims with a large input signal?  DC is fine, but you really want to verify that the amp can handle the slewing from a large input.  I am assuming that your ideal OTA is some finite gain element with a possible pole to look at finite bandwidth, but I am pretty sure that you have not put in any slew limiter in your model.  When you added the additional gain boosting, did you break it in a way that it would give ideal slewing?

There is no slewing limiter in ideal OTA. What puzzles me is that the modulator works with real OTA and no-chopping. So the OTA slewing should meet the requirement. And supposedly chopping should not increase the slewing requirement.

First off, the major drawback of chopping is a greatly increased slewing requirement.  Let's assume that your capacitor gain in your integrator without chopping is 1/4.  For every full scale (FS) worth of input signal, you output is 1/4*FS.  So, without chopping, if your input is FS and your output is 3/4*FS, you output would still only need to step by 1/4*FS (with ideal rail-to-rail outputs) to bring the output to +FS.  With chopping enabled, you would need to drive the output from 3/4*FS to -FS, which is a huge difference in output step.  Since most OTA's enter a slewing condition during these large transients, chopping exacts a huge slewing burden.  As an aside, continuous time chopping exacts a huge bandwidth burden.

Quote:

If your OTA does not settle properly, either through bandwidth limitations or slewing limitations, that shows up as a gain reduction.  This is simply due to the fact that the charge on your caps are not what you expect them to be.  There would always be less charge, so therefore it would be indistinguishable from an integrator with lower gain.

Quote:

I don't know why you say that the error voltage on the summing junctions would be a noise-shaped signal, though.  The very-low frequency errors--i.e. offset, 1/f noise, cap mismatch, and some charge injection--would be up-converted to the chopping frequency.  However, this is not a noise signal.  It can be explicitly defined by the charge equations, and hence, I would not agree that it would move the quantization noise into the signal band.

I was thinking that you were chopping before the input caps, but I realize that is not the case.  As far as the slewing requirement, then, it depends on your amp architecture.  If you have a two stage with a Miller cap, you still have to slew that cap.

Quote:

Is this a bandpass Delta-Sigma?

Quote:

Remember that with Delta-Sigma's, there are two transfer functions, the signal TF and the noise TF.  It depends on your exact architecture, but there are some that treat the noise at the input to the first integrator as a signal.  The real question here is what is the transfer function for any input--noise, offset, charge injection, etc.-- at the input to the integrator.  At the output, all error sources and the signal are indistinguishable.

Quote:

This was the confusing sentence to me.  The quantization noise should be unaffected by any of this.  Other noise sources may be shifted into the signal band, though.

Quote:


no it is not.

Quote:


Well, if the input cap of the first integrator is not considered and chopping is not used, I agree.

Quote:

I do agree that there will be an error term on the summing junctions due to this.  Also, there will be shifting of noise in the frequency domain, but low freq noise would get shifted up while nyquist freq noise would get shifted down.  The noise would be filtered by the integrator transfer function, and the swap would take place again.  This is a better scenario since flicker noise is removed by the chopping and shaped by the integrator.  Do have some reason to believe that this would ultimately cause more noise in the signal band?