From: Erwin on 8 Jun 2010 21:12 dbd <dbd(a)ieee.org> wrote in message <a3e6aaba-382d-46fb-8898-230d66fefba8(a)6g2000prg.googlegroups.com>... > Erwin > > I'm not going to waste time trying to keep up with a constantly moving > target when you fail to even clean up the problems people help you > with and persist in using function calls I don't have the toolbox for. > Here is an example of what I have said with vectors usage added and > Fortran-like indexing removed: > > %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% > clear all > beta=2; %slope of the log-log power law > a=500; %amplitude > N = 500; % half the number of samples > Fs = 1/5; > f = 2*(1:N)/Fs; % generated frequency space, includes no DC term > > S = (1./f).^(beta); %power function S(w)=(1/w)^beta > Y = randn(1,N).*S + i*randn(1,N).*S; > newY = [0, Y(1:N-1) 0 conj(Y(N-1:-1:1))]; % As I said > > figure > subplot(2,1,1) > loglog(abs(Y(1:N-1))) > grid on > title('Generated Spectrum Y') > > timecx = ifft(newY); > sum(abs(real(timecx))) > sum(abs(imag(timecx))) > timerl = real(timecx); > ftest = abs(fft(timerl)); > > subplot(2,1,2) > loglog(abs(ftest(2:N-1))) > grid on > title('Spectrum calculated from time domain real samples') > %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% > > That leaves you with the exercises of determining unit scaling and of > controlling the bandwidth you wish to generate. Then you can try to > figure what you want about Parseval. > > Dale B. Dalrymple Hello Dale, Thanks for all the support. I didn't mean to be persistent, I'm just new to Matlab and to signal processing... Sorry. I think that the algorithm finally gives me credible results! :) Thanks a lot.
From: dbd on 8 Jun 2010 23:30 On Jun 8, 6:12 pm, "Erwin " <car...(a)gmail.com> wrote: .... > > Hello Dale, > > Thanks for all the support. > I didn't mean to be persistent, I'm just new to Matlab and to signal processing... Sorry. > > I think that the algorithm finally gives me credible results! :) > > Thanks a lot. Erwin Persistent is fine. But try to get one code fixed or question answered before adding new features to broken code. It's easier to follow and fix/answer. For the advocates of phase randomization, Erwin's cited paper appropriately dismisses phase randomization for it's failure to generate data with a proper variance (non-zero) and distribution (chi- squared) for the periodogram. Dale B. Dalrymple
From: Greg Heath on 10 Jun 2010 04:03 On Jun 8, 11:30 pm, dbd <d...(a)ieee.org> wrote: > On Jun 8, 6:12 pm, "Erwin " <car...(a)gmail.com> wrote: > ... > > Hello Dale, > > > Thanks for all the support. > > I didn't mean to be persistent, I'm just new to Matlab and to signal >> processing... Sorry. > > > I think that the algorithm finally gives me credible results! :) > > > Thanks a lot. > > Erwin > > Persistent is fine. But try to get one code fixed or question answered > before adding new features to broken code. It's easier to follow and > fix/answer. > > For the advocates of phase randomization, Erwin's cited paper > appropriately dismisses phase randomization for it's failure to > generate data with a proper variance (non-zero) and distribution (chi- > squared) for the periodogram. A phase randomization model had been used in an attempt to model radar signal perturbations caused by ionospheric scintillation. However, predictions were incompatible with measurements. As a result, I was tasked to constuct a model with the following characteristics specified: Amplitude; Rayleigh distributed Decorrelation time, taua Asymptotic power law spectrum 1/f^1.5 Phase Gaussian distributed Decorrelation time, taup Correlation Crosscorrelation coefficient Cap I posted the reults but haven't been able to Google them up. My memory ain't that good these days. However, I remember the key to the solution was to use copulas and rank correlation coefficients. If I find the threads I will post the links. Hope this helps. Greg
First
|
Prev
|
Pages: 1 2 3 4 5 Prev: Sample time - sequence stair Next: Help test tomorrow in the MATLAB |