POD analysis, variance criterion, monte carlo simulation

[laila17]

Hi matlab users,

I have experimental digital data that have been analysed by using the POD approach. A visual examination shows the cummulative variance of the first POD modes rapidly converge to unity. I'm looking for a more scrupulous method to decide on a threshold criterion for separating modes that contain coherent structure from modes that contain negligible or no coherent structure (noise),(statistical representative from non representative POD modes).
I have found some literature on MOnte Carlo simulation methods for detemination of a variance threshhold, but have some doubts on how to apply this tool to my data. 1) Is it correct to generate a set of random samples by looping the normrnd command of matlab? (toghether with looping the normrnd command that changes the matlab state) 2) Following the socalled N-RULE described in PCA analysis texts i should multiply this random data set by a diagonal sigma*Ip (Ip = p-dimensional identity matrix),but am not sure how i determine this sigma (variance)? In other words what is the right N(0,sigma) distribution to use in this rule?

Any help, is highly appreciated!!!!
thanks ahead, laila.

 

[GregLocock]

You might want to try posting this in the matlab forum. Red flag this thread if you do.

Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.