Calculate residuals of an ELiS fit.
Syntax
[rx,ry] = rdueelis(pdat,cdat,Fdat,vdat,expi)
[rx,ry,ryx,vryx,xe,ye] = ...
rdueelis(pdat,cdat,Fdat,vdat,expi,inp,outp)
Description
rdueelis calculates the complex residuals of a fit of elis (see "Study of the Residuals" in the User's Guide). pdat is the parameter vector of the fit (see exppar), or the name of the parameter file. cdat is the covariance array, or the covariance vector, or the name of the covariance file of pdat. If cdat is empty, the variance of the parametric estimate of the transfer function will not be considered.
Fdat is the Fourier vector, or the array [freqvect,x,y], or the name of the Fourier file. expi contains the number(s) of the experiment(s) in Fdat, for which the residuals are to be calculated. vdat is a 1-by-2 vector if the variances are constant (or a 1-by-3 vector if the covariances are also given), or an N-by-2 array if the variances are given point by point (or an N-by-3 array if the covariances are also given), or a long variance vector if the variances are put together by expvar, or a string if the variance file is referred to.
inp and outp select the serial numbers of the input and of the output port in the Fourier file.
The residuals are given in rx (complex residuals of the input amplitude vector), ry (complex residuals of the output amplitude vector), ryx (complex residuals of ym./xm vs. the estimated transfer function).
vryx contains the variance vector of the real and of the imaginary parts of ryx. If the distributions of the complex ryx values are circularly symmetric, this equals the halves of the variances of ryx.
vryx is the difference of two quantities (see "Study of the Residuals" in the User's Guide). Theoretically, this is never negative if the proper data are given (the same variance vector as used in elis; pdat and cdat that belong to this run of elis). However, because of the approximations, small negative values may occur: these usually indicate very small variances.
xe and ye are the estimated complex input and output amplitude vectors. These estimates are not consistent for one experiment. However, processing of several experiments at the same time does decrease their estimation error.
Default Argument Values
expi = [], inp = 1, outp = 1.
Examples
[rx,ry,ryx,vryx] = rdueelis('inpchans.pbn','inpchans.cbn',...
'inpchan.fbn',[9.61e-12,9.61e-10]);
Diagnostics
The validity of the variance values and the common length of the variance and frequency vectors is checked. If negative variance values are obtained, a warning message is sent.
Algorithm
The complex input and output amplitudes are estimated via weighted LS fitting, having the estimated transfer function parameters fixed.
References
[1] I. Kollár, "On Frequency Domain Identification of Linear Systems," IEEE Trans. on Instrumentation and Measurement, Vol. 42, No. 1, pp. 2-6, Feb. 1993.
[ Previous | Help Desk | Next ]