The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> Mixed-Signal Design >> Number of Monte-Carlo runs
https://designers-guide.org/forum/YaBB.pl?num=1148993375

Message started by Visjnoe on May 30^th, 2006, 5:49am

Title: Number of Monte-Carlo runs
Post by Visjnoe on May 30^th, 2006, 5:49am

Hello,

I had a discussion with someone and since statistics is not my cup of tea, I'm posing the following question:

1) Is there any formula that relates the number of Monte-Carlo runs to the confidence level (e.g. 6σ)?
2) Can that number depend on the type of circuit? For example, in a flash ADC one could consider each preamp/comp out of 2^N
to be already an independent, statistical sample.

Kind Regards

Peter

Title: Re: Number of Monte-Carlo runs
Post by ACWWong on May 30^th, 2006, 5:59am

Statistics aren't my cup of tea either (although like most, I took a few sips from the stats cup many moons ago at university)

but i was once told (on good authority i was assured) that:

"The worst case of 10 MC runs gives an indication of 2sigma for a gaussian distribution"

Often I use it and find its quite good (and definately speeds up my design work). My head tends to hurt if I start to think too much about the maths... maybe someone with a clear head can educate us with a proof/disproof ??

Title: Re: Number of Monte-Carlo runs
Post by raf on Jun 27^th, 2006, 5:43am

Hmmm if you mean +/- 2 sigma around the mean then 10 MC runs seems too small. If you mean +/- 1 sigma around the mean, well is that really good enough anyways?

I don't know the answer either. I use brute force approach. I keep running MC sims until my answer doesn't change anymore. Then this is back annotated into simulink sims to run system level MC.

-Raf

Title: Re: Number of Monte-Carlo runs
Post by mikki33 on Jun 27^th, 2006, 7:46am

Very long time ago, I red an article proving that it is enough 95 runs to get Sigma with 5 % accuracy (after plotting hystogram, of course)...