Hi all,

Does anybody know if there's a traditional method for organizing linear data, such as mono PCM data, for use with SVD?

i.e. If I have 1 million samples.... Do I split it into arbitrarily sized chunks (say, 10x10 squares) and perform SVD on it?

Any help would be greatly appreciated!

(Also: if you know of a good site which talks _conceptually_ about SVD, that'd be great! I want to know all the interesting and useful properties it has. Most sites I can find give you an introduction into how to do it... or they say "it can be used for x and x"... or they say "Here's how I applied it to this"... or "here's a computationally efficient method of calculating it"... I want more of "These vectors tell you **this** about the original matrix)

This may be useful
http://www.math.ucdavis.edu/~saito/cours…
Stewart is the author of Afternotes on Numerical Analysis and Afternotes Goes To Graduate School.

Also, searching for singular value decomposition and Gene Golub may yield something. I found these which may be a little elementary for your purposes.
http://mathworld.wolfram.com/SingularVal…
http://mathworld.wolfram.com/SingularVal…

Cleve Moler's paper on Golub and SVD, titled "Professor SVD" may also be useful.
www.mathworks.com/mason/tag/proxy.html…

Golub and Van Loan's Matrix Computations probably should be in your personal library.

Regarding your original question, it is not clear what you are trying to do and why SVD would be the right tool. I would have expected someone with PCM data would have questions about the Fourier transform to go from the time domain to the frequency domain. Maybe I've done too little with such data to be aware of some obvious application for SVD, but I'm not seeing it.