Time Varying Noise - Whitepaper

Dear analogy,

> The problem back here is that integrated noise power is calculated for bandwidth from 0 Hz to 1 GHz

I believe the integration limits are a result of computing the impact of the time-vary noise on phase noise since this analysis utilizes the pnoise analysis.

Shawn

pnoise - tdnoise 

Yes, but the integration limit is somewhat vague (or maybe its my misunderstanding).Because having a periodic square wave , one would expect spectrum to have dual sideband and thus bandwidth would be between f1 Hz and f2 Hz centered around 1 GHz.Don't you think ?

Having quickly scanned through Joel's paper (not looked at this one for many years, since this is pretty much the original paper on tdnoise from 1998), I think there's a mistake. The pnoise frequency sweep should only go up to half the PSS fundamental, and similarly the integration limits should be up to a maximum of half the PSS fundamental.

The reason is that when you use tdnoise (or use the jitter noisetype for driven circuits, or jitter noisetype with "PM" jitter selected for oscillators - these both use tdnoise under the hood), it is actually performing a sampled or "strobed" noise analysis. When you do that, the simulator is inserting an ideal sampler at each time the noise is sampled at, and so the sample rate is the PSS fundamental (for each time point you give) - i.e. it samples once per period. Because of this ideal sampler, the noise from the entire bandwidth (at least all the sidebands that are computed, or better still using the "fullspectrum" option), gets folded into the 0->PSSfund/2 band - so integrating the noise in that band will give you the total noise across the whole spectrum (sampled at that time point).

If you sweep the noise beyond half the PSS fundamental, you end up double counting (or even more than double if you extend the sweep even further), so you don't want to do that.

With respect to Shawn's comments - I didn't quite understand what he meant. pnoise analysis isn't necessarily anything to do with phase noise - phase noise is something that can be computed given the result of a pnoise analysis. In this case you're not computing phase noise at all, but the output noise at a particular sample point and from that you can see how the noise varies throughout the period. To calculate jitter (for example), which is related to phase noise, this can be done with knowledge of the noise at the transition and the slope of the signal at that point - and that's exactly how the metrics on the direct plot form for jitter mode work. It's a bit more complicated, because some depend on the number of cycles (e.g. K-cycle jitter), and the integration limits have to be adjusted to make sure that what you're integrating makes sense if you're only looking at a small number of cycles - but there's more on this in the Jitter app note that is in the MMSIM installation hierarchy (under <MMSIMinstDir>/tools/spectre/examples/SpectreRFworkshop - forgive me if I typed it incorrectly as I'm doing this from memory).

Regards,

Andrew.