Re: Add Noise

From: Richard Maine (nospam_at_see.signature)
Date: 03/18/04 

Date: 18 Mar 2004 12:39:07 -0800

Rich Townsend <rhdt@barVOIDtol.udel.edu> writes:

> Joost VandeVondele wrote:
> > tholen@antispam.ham wrote in message news:<dJg6c.7907$h85.3597@twister.socal.rr.com>...

> > noisy_vector=vector
> > DO i=1,N
> > call random_number(noise)
> > noisy_vector=noisy_vector+(noise -0.5)*scale
> > ENDDO
>
> Still ain't Gaussian...

But by the central limit theorem, it approaches Gaussian as N
approaches infinity (assuming you normalize scale appropriately).
And in this case, numbers on the order of 12 or so are a decent
approximation to infinity. I've seen code that uses this kind
of method to approximate Gaussians (and I'm thinking I recall
N=12 being used; might have even been N=6).

Perhaps not the world's most efficient method, but it is hard
to get simpler, and for some applications it is fine.

Hey, if you needed to be really picky about things, you'd be using a
3rd party generator instead of whatever the compiler vendor happened
to throw in for random_number. 

-- 
Richard Maine                       |  Good judgment comes from experience;
email: my first.last at org.domain  |  experience comes from bad judgment.
org: nasa, domain: gov              |        -- Mark Twain

Relevant Pages

  • Re: Add Noise
    ... >But by the central limit theorem, it approaches Gaussian as N ... >approaches infinity. ... The variance of the common uniform is 1/12 so the sum of 12 of them ... It is a workable approximation at 1:00 ...
    (comp.lang.fortran)
  • Re: Photographic Size of a Star
    ... pattern extends to infinity, exactly as a gaussian. ... matter of a specific PSF but rather that the PSF extends to infinity, ... your purely Bessel PSF becomes approximately Gaussian. ...
    (sci.astro.amateur)
  • Re: Add Noise
    ... > But by the central limit theorem, it approaches Gaussian as N ... > approaches infinity (assuming you normalize scale appropriately). ...
    (comp.lang.fortran)
  • Re: Mixture of Gaussian Integral
    ... Let N_adenote Gaussian ... Given a mixture of two gaussians distribution, ... an approximation, useful asymptotic results may be possible. ... small or large (-> infinity). ...
    (sci.math.research)