Hi Kiki,

First, read through a decent comm. textbook about how why/how a
wireless multipath channel can be represented as a time varying digital
filter. Try the book by Rappaport, it's easy to read.

Once you're through with that, you should be able to appreciate the
fact that if you pump a signal through a channel, it's the same as the
signal getting convolved with the channel's impulse response.

Convolution can be expressed as a matrix multiplication. Figure out how
:-)

Okay, so let's say you pump in a signal x' = {x_0, x_1, ..., x_(n-1)},
of finite duration, and you have a channel h' = {h_0, h_1, ...,
h_(m-1)}.
The convolution can now be represented as

y = H.x + w, where H is the channel convolution matrix, x is the signal
vector and w is the noise vector. The crudest way of recovering x is to
multiply the recd. signal y by the inverse of H. All that OFDM does is
transform the Toeplitz matrix H (why is H necessarily Toeplitz?) into a
circulant, which is easily invertible.

This is a hurried post, so I've missed out on a lot of things. I
suggest that you read through this paper for a better explanation:

"Wireless Multicarrier Communication: Where Fourier meets Shannon",
Zhengdao Wang, Georgios Giannakis, IEEE Signal Processing Magazine, May
2000. It's available off Wang's site.

HTH,
Karthik.

Hi all,

Can anybody enlighten me a little bit using concept of eigenvalues of the 
channel matrix?

I vaguely heard about that the eigenvalues correspond to the impulse 
response of the channel matrix and then it somehow did blah blah blah, so 
the OFDM idea is good... blah blah blah,

I am not sure I understood it... anybody provides some intuition and 
thoughts?

Thanks a lot!