Hi, I go back to AR/trig, because maybe someone on the list try to do this and so can share experience/advice.
We got, on the website AFL library, AR prediction tool (search for "ar burg"). >From this tool we can extract frequency (there is a basic dominant frequencies extract with the AR AFL) for short lenght data. But now we got only the frequency. Purpose is to use those frequency to make a trigfit. AR spectral estimator loose amplitude (power of the frequency) and phase. So it is needed to use a fitting in the time domain to find back those two parameters. A very nice program can show this clearly, free for 30 days : http://www.weisang.com/index.php?id=2&L=1 It can find frequency (by AR Burg, and even Music or EigenVector!), and this frequency estimator can be followed by TirgFit to find back true Amplitude and Phase of the frequency. Look at the help, there is even a spectral analysis tutorial. By the way, Music algo seems damn powerfull but you need to say how many frequency you want... but seems computer very intensive too : )) I think trigfit can be done with the Appendix from Hurst book. There was two step : - one to find the frequency (done now by modern parametric estimation AR, should be ok... single value decomposition better but we don't have héhé) - one to find ampl/phase (this one we need !) I will try to look to compute the algorithm, but if it is already done here, .... think to share it plz. Cheers, Mich _________________________________________________________________ Live.com : le moteur de recherche aux réponses bien carrossées! http://www.windowslive.fr/livecom/
