Spectral-Correlation
所属分类:matlab编程
开发工具:matlab
文件大小:6KB
下载次数:29
上传日期:2011-09-12 09:45:40
上 传 者:
holomm
说明: Cyclic Power Spectral Density \tab\tab\tab of signals X and Y using Welch s average
文件列表:
Pack Cyclic Spectral Correlation\SCoh_W.m (5633, 2007-09-21)
Pack Cyclic Spectral Correlation\CPS_W.m (4373, 2007-09-21)
Pack Cyclic Spectral Correlation (0, 2007-09-21)
{\rtf1\ansi\ansicpg1252\deff0\deflang1036{\fonttbl{\f0\fnil\fcharset0 Courier New;}{\f1\fswiss\fcharset0 Arial;}}
{\*\generator Msftedit 5.41.15.1507;}\viewkind4\uc1\pard\f0\fs20 CPS_W.m\tab\tab estimates the (cross) Cyclic Power Spectral Density \tab\tab\tab of signals X and Y using Welch's averaged \tab\tab\tab\tab periodogram method.\par
\par
SCOH_W.m\tab\tab estimates the coherence of signals X and Y using \tab\tab\tab Welch's averaged periodogram method.\par
\par
\par
\ul\b Input and output arguments\ulnone\b0\par
Both functions are used in the same way. For example\par
\par
Spec = CPS_W(y,x,alpha,nfft,Noverlap,Window,opt,P)\par
\par
returns the structure:\par
Spec.S = Cyclic Power Spectrum vector\par
Spec.f = vector of frequencies\par
Spec.K = number of blocks\par
Spec.Varduc = Variance Reduction factor\par
Spec.CI = P*100% confidence interval for Spec.CPS.\par
\par
The input arguments nfft, Noverlap, Window and P are as in function PSD.m or PWELCH.m of Matlab. Denoting by Nwind the window length, it is recommended to use nfft = 2*NWind and Noverlap = 2/3*Nwind (hanning window) or Noverlap = 1/2*Nwind (half-sine window).\par
Note : use analytic signal to avoid correlation between + and - frequencies\par
(For more details, type "help CPS_W" and "help SCOH_W" at the Matlab prompt.)\par
\f1\par
\ul\b Usage\ulnone\b0\par
\f0 CPS_W and SCOH_W operate at one cyclic frequency alpha only. In order to scan a full domain of the \{f;alpha\} plane, it is necessary to loop over alpha. The largest allowable increment of alpha (i.e. the cyclic frequency resolution) is 1/L where L is the signal length (never bypass this limit, or the results will be undersampled -- if cyclic resolution 1/L is deemed too fine with regard to computational demand, then the good solution is simply to shorten the signal length L). The following example illustrates the methodology:\par
\par
load x\par
L = length(x);\tab\tab % signal length\par
Nw = 256;\tab\tab\tab % window length\par
Nv = fix(2/3*Nw);\tab\tab % block overlap\par
nfft = 2*Nw;\tab\tab % FFT length\par
da = 1/L;\tab\tab\tab % cyclic frequency resolution\par
a1 = 51;\tab\tab\tab % first cyclic frequency bin to scan\par
a2 = 200\tab\tab\tab % last cyclic frequency bin to scan\par
\par
C = zeros(nfft,a2-a1+1);\par
S = zeros(nfft,a2-a1+1);\par
for k = a1:a2;\par
\tab Coh = SCoh_W(x,x,k/L,nfft,Nv,Nw,'sym',.01);\par
\tab C(:,k-a1+1) = Coh.C;\par
\tab S(:,k-a1+1) = Coh.S;\par
\tab waitbar((k-a1+1)/(a2-a1+1))\par
end\par
\par
alpha = (a1:a2)*da;\par
figure\par
imagesc(alpha,Coh.f(1:nfft/2),abs(S(1:nfft/2,:))),\par
colormap(jet),colorbar,axis xy,title('Cyclic Spectral Density'),xlabel('\\alpha'),ylabel('f')\par
figure\par
imagesc(alpha,Coh.f(1:nfft/2),abs(C(1:nfft/2,:))),\par
colormap(jet),colorbar,axis xy,title('Cyclic Spectral Coherence'),xlabel('\\alpha'),ylabel('f')\par
figure\par
plot(Coh.f(1:nfft/2),abs(C(1:nfft/2,[74:76])).^2,Coh.f(1:nfft/2),Coh.thres*ones(1,nfft/2),'m:')\par
title('Squared magnitude Cyclic Spectral Coherence and its 1% level of significance'),ylabel('f'),legend(['\\alpha = ',num2str((74+a1)*da)],['\\alpha = ',num2str((75+a1)*da)],['\\alpha = ',num2str((76+a1)*da)])\par
Coh.Varduc\par
\par
\par
\ul\b Parameter setting\ulnone\b0\par
The setting of Nw should achieve a good compromise between a fine frequency resolution 1/Nw and a statistically stable result (bias-variance trade-off). Our recommendation is to set Nw small enough so that the variance reduction factor returned by Spec.Varduc is no larger than 1/30. Values on the order of 1/100 are ideal.\par
\par
\ul\b Reference\ulnone\b0\par
J. Antoni, "Cyclic Spectral Analysis in Practice", Mechanical Systems and Signal Processing, Volume 21, Issue 2 , February 2007, Pages 597-630.\par
\par
\f1\par
}
近期下载者:
相关文件:
收藏者: