math geeks: MATLAB/statistics help?

[klar]

so, seeing as maggot wisdom helped me install this thing, i thought i would appeal to the collective again with some questions about how it actually works. i have to do a "project" which is making me feel like a complete idiot.

basically i have a set of data that i am supposed to make a periodogram out of. i transformed the original numbers (subtract the mean, fft) and now have a set of wonderful complex numbers, but i don't get how i can plot these against frequencies? how the hell do i get the frequencies? i just don't understand the theory behind this, i guess

also, i am supposed to show that the parseval relation works for these values.

for y=[set;of;numbers;]

sum(y) should be 1/N*sum(y.^2) but it's not.

aaaaahhhh!!

sorry for nerdy nsr post.
any suggestions would be very much appreciated.

[BLOOD SWEAT STEEL]

Thanks for that. I was just thinking how badly I needed my daily reminder of how dumb I am.

[Ireallyliketoski2]

By the way, you're welcome.

[eirikainersharp]

By the way, you're welcome.

I have seen it in a slightly different expression but that looks right to me.
Regarding the proof... it might help if we say your data set. If its two big... you should be able to find some sort of algorithum in a text (I can't think of a good on of the top of my head) which you can base off.

[BeanDip4All]

Parseval's theorem relates the ENERGY of the signal in time with the ENERGY of the signal in freq. Thus, what you wrote (um(y) should be 1/N*sum (y.^2) ) is incorrect, that's why probably itdidnt match.

Second, the frequencies of the periodogram: For every point in the array that Matlab returns you associate a frequency. The maximum of that is 2*pi*(sampling_rate)... or something like that. Each frequency will be MAX_FREQ/i for every i=0:N...


Right now Im speaking from the top of my head. Im too lazy to pull out my undergrad books/Matlab Manual. But it's all pretty easy.... Check your books.

try Y = mx+c

where m = slope

If you want me to do it further, PM me for my e-mail and then send the problem in it's entirety, exactly as stated. Unchanged in wording. All data, all given info, all assumptions, the questions you've been asked to answer, and I can run it for you.

[davep]

i transformed the original numbers (subtract the mean, fft) and now have a set of wonderful complex numbers, but i don't get how i can plot these against frequencies? how the hell do i get the frequencies?

to be a bit more explicit than TLuv -
let's say your orignal data has N samples, and is sampled at times
t = 0:dT: (N-1)*dT

The frequencies of your fft will then be (in 1/s)
f = (0: (N-1))/N/dT

if your 'time' axis is something other than real time, then just substitute appropriate units for dT...

also, i am supposed to show that the parseval relation works for these values.

again like Tluv said, what you quoted is not parseval's relation. For matlab's fft, if
Y = fft(y);

then parseval's relation is
sum(abs(Y).^2)/N = sum(abs(y).^2)

don't forget the 'absolute value' bit, since Y will usually be complex.

Good luck.

[BeanDip4All]

Davep makes some good points and makes my rambles a little more clear.

Also, you might want to check out these pages:
http://www.mathworks.com/access/help.../helpdesk.html
Just look for periodogram in the search.

Mathworks is a really helpful reference

[klar]

thanks all!


still not sure i understand the frquency thing, don't have time now, will try again and come back here later

t.luv, thanks for the link, will check it out. as for the offer to do it for me, very very tempting, may get back to you if i really can't figure it out.


you guys rock.