The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Analog Verification >> Analog Performance Verification >> dynamic comparator noise measurement.
https://designers-guide.org/forum/YaBB.pl?num=1489515640

Message started by yvkrishna on Mar 14^th, 2017, 11:20am

Title: dynamic comparator noise measurement.
Post by yvkrishna on Mar 14^th, 2017, 11:20am

Hi,

I am simulating dynamic comparator noise using pnoise setup.

In the interest of simulation time I am simulating with 10X clk freq than what I designed for, does it impact my final result in any manner?

Rgds,
yvkrishna

Title: Re: dynamic comparator noise measurement.
Post by DanielLam on Mar 15^th, 2017, 11:47am

Why don't you just simulate with different Fclks? The sims should take less than 5 mins. Pnoise sims are pretty fast. Transient noise sims are very slow.

In order to teach others, can you please post what you see?

I'll post what I've seen afterwards.

Regards,
Daniel Lam

Title: Re: dynamic comparator noise measurement.
Post by yvkrishna on Mar 15^th, 2017, 6:47pm

Hi,

My clk freq is quite low today so the pnoise sim takes 6 hrs atleast which is bothering me to try out few simple experiments also even during design optimisation.

and the recommendation was to increase sidebands , maxacfreq based on bandwidth of the circuit.

Sure I will getback on my observations.

Rgds,
yvkrishna

Title: Re: dynamic comparator noise measurement.
Post by DanielLam on Mar 15^th, 2017, 7:20pm

For the pnoise sim, I basically run a jitter sim, integrate the jitter, and divide by a peak gain. I was running a 2-4 GHz clock. Each sim (PSS+Pnoise) took about max a minute. 10 sidebands, moderate tolerances.

For my comparator with different fclk (which means sidebands and integration limit changes). Also, the clock was a square pulse with the same rise/falltime of 5 ps:

100M : 210 uVrms (jitter 70m)
1G : 373 uVrms (jitter 124m)
2G: 376 uVrms (jitter 123m)

I had a difference, and I think it was due to the integrated jitter range (along with the comparator characteristics). I didn't look any deeper after that.

Title: Re: dynamic comparator noise measurement.
Post by Ken Kundert on Mar 15^th, 2017, 7:35pm

Why would a low clock frequency result in a long simulation time?

-Ken

Title: Re: dynamic comparator noise measurement.
Post by yvkrishna on Mar 15^th, 2017, 11:00pm

Ken,

Since the bandwidth of my circuit is about 1000X than the clock freq, I think atleast few thousand sidebands are needed in pnoise setup to take of folding effects.
(some people even suggest 10X the ratio of bw to clk frq which is 10,000 sidebands )

This is the cause of slowness in my setup.

Let me know if I am missing something here.

Thanks,
Vamshi

Title: Re: dynamic comparator noise measurement.
Post by Andrew Beckett on Mar 16^th, 2017, 3:56pm

You should use the fullspectrum option for pnoise rather than specifying a huge number of sidebands. You need to specify the number of sidebands to be high enough to be above the flicker knee, but other than that, using the fullspectrum method will allow it to use an adaptive approach which is much more efficient.

Title: Re: dynamic comparator noise measurement.
Post by yvkrishna on Mar 16^th, 2017, 8:22pm

Thanks for the suggestion about fullspectrum Andrew.

So this would be in a 2 step, first give only few (~100) sidebands to get a crude estimate for flicker noise corner at the output of the comaparator and then use that to get the correct estimate for sidebands in the next iterations (??)

Thanks,
yvkrishna

Title: Re: dynamic comparator noise measurement.
Post by yvkrishna on Apr 18^th, 2017, 9:16am

DanielLam,

My noise result with 1X ,2X , 4X clk freq are identical, so this method is beneficial to speedup simtimes.

only thing to make sure is to have sufficiently large maxacfreq in pss setup w.r.t the bandwidth of the DUT.

Thanks,
Vamshi

Title: Re: dynamic comparator noise measurement.
Post by Andrew Beckett on Apr 21^st, 2017, 1:36pm

yvkrishna wrote on Mar 16^th, 2017, 8:22pm:

So this would be in a 2 step, first give only few (~100) sidebands to get a crude estimate for flicker noise corner at the output of the comaparator and then use that to get the correct estimate for sidebands in the next iterations (??)

No, it's not in two steps. I didn't say that... it doesn't work the way you are suggesting.

Did you try it?

Andrew.

Title: Re: dynamic comparator noise measurement.
Post by yvkrishna on May 20^th, 2017, 9:22pm

Andrew,

I have tried using 20 sidebands for flicker noise first, then tried with default.

both of the cases result is pretty close.

Thanks,
yvkrishna