Help With Matrices Whitening Problem in Random Vectors

46 Views Asked by At

Problem: Given Gaussian RV X with Cov Matrix $C_{xx}$.

Find Matrix A such that Y is an RV, Y=AX and Cov Matrix of Y $C_{YY}=I$.

Now it's clear that according to some rule, $C_{YY}=AC_{XX}A^T$.

BUT $C_{YY}=I$, so A is diagonalizing!!!! Hence $C_{XX}=I$!!!!

I'll show you. Please help me find when I'm wrong.

$C_{YY}=AC_{XX}A^T$

$A^{-1}=A^T$ (Since A is diagonalizing) And hence -

$A^TC_{YY}A=C_{XX}$.

But $C_{YY}=I$ and $A^{-1}=A^T$ hence

$A^{-1}IA=C_{XX}$ =>

$C_{XX}=I$

What is the mistake? It's a common matrices whitenning problem.

1

There are 1 best solutions below

0
On

I found the problem. It's the second time I make this mistake, hence I update for everyone:

$A^{-1}=A^T$ always holds true only if A is diagonolizing Mat X Orthogonally!!! If A is just a simple form of diaginolizing, it's not always true.