Hi,

I've spent the last couple of weeks trying to equate simulator output to the analytical derivation of kT/C noise in a switched cap setup, to very limited success. I realize that this is probably due to a failure in my reasoning to set up the simulation environment, or misinterpreting the results, so I'd really appreciate some pointers to where I'm doing things wrong.

I set up a simple circuit that consists of a switched noisy element (I have tried two different approaches to this; one, a MOSFET, and the other was an ideal switch referenced from the Cadence analogLib and a resistor in series) connected series to a capacitor. I set up a PSS simulation, choosing a variety of values for the number of harmonics, from 50 to 1000, and adjusted the maxacfreq in the options accordingly (usually choosing a value that was in the range of 'number of harmonics*beat frequency'). I then set up a PAC simulation to verify that the circuits showed the RC behaviour in their switched state, which always returned reasonable results.

Referencing several papers including Mr. Kundert's, I was expecting the whole noise power to be condensed to the frequency band between 0 and Fs/2. Thus, integrating the noise power from 0 to Fs/2 would have let me arrive to the analytically derived kT/C expression. While the noise bandwidth stayed approximately three decades below my clock frequency, I did get reasonable results but when the noise bandwidth started to increase (by using a smaller C) the result of the integration in the baseband started to deviate a lot from the analytical value. However, I did see some noise folding, since changing C should not have affected the DC level of the output noise, but it did steadily increase as expected when I decreased C.

Is there something wrong with this approach?

Thank you for your input.