Bayesian estimation of structured correlation/covariance

Hi,

I have a problem that comes from VLSI that I've been wrestling with -
and I was hoping someone would tell me if I'm going in the right
direction.

Essentially, there are many delay elements that have are normally
distributed - and I have an assumed structured correlation matrix for
these delay elements - in the simplest case, suppose I have the
symmetric, positive-definite correlation matrix symbolically and the
only values allowable are {1,0,p}

I would like to apply bayesian parameter estimation for the unknown
variable given measurements.
The issue is, I cannot directly measure the delay elements, I can only
measure path delays - which are sums of various delay elements.

If I'm only allowed to measure a single path, I can proceed fairly
straightforwardly with estimating an unknown variance of a gaussian
distribution with a chi-square distribution - in which I can formulate
the variance as function of p.

Of course, I can measure several paths, so from what I've read, I
should formulate this as covariance matrix estimation with the Wishart
distribution.

I'd prefer correlation estimation which I understand from a bivariate
angle with the Fisher transform - but from what I've read from
discussion in this group, the multivariate case gets quite nasty.

Any comments or advice would be greatly appreciated,
thanks,
Ben Lee

.

Relevant Pages

  • Re: Help with BUGS model
    ... With the model where the Lambdas are drawn from only one gamma prior I ... sure why your MLE estimation procedure had problems. ... the expected value of the cell under the assumption of independence of ... The posterior distribution of lambda given N has a closed form. ...
    (sci.stat.math)
  • Least Squares solution for fitting beta distribution to empirical distrbution
    ... can use as a probability distribution using kernel density estimation. ... I'm wondering if there is an analytic solution to this. ...
    (sci.stat.math)
  • Re: significant test of the 2 approaches
    ... Let's say your estimation methods A and B were ways to estimate the mean of the distribution from which you are drawing data. ... You now have empirical estimates of the sampling distribution of your methods under the assumption of Nfor sample size n. ... Now consider using different sample sizes or parameter values or distributions. ...
    (comp.soft-sys.matlab)
  • Re: percentiles estimation
    ... The estimation depends on the data. ... If the data set is quite nice, say has a shape of known distribution, ... using parametric method to estimate the distribution and calculating ... Subj: percentiles estimation ...
    (sci.stat.math)