fxcor fourier mode and filtering parameters question
sduffau wrote on Jan 15, 2008
I am using fxcor to estimate radial velocities from absorption lines. I would like to use the filtering option of fxcor. I looked at the help pages and saw that in the fourier mode the keystroke i displays what is called the period trend for a given wavelenght/frequency. Looking at the results of doing this I see the systme quotes a pixel value next to the wavenumber and frequency displayed for the cursor position. Is this pixel value the corresponding in linear space to the wavenumber position I selected? I mean if it says 3 pixels for a 150 wavenumber position, does it mean that a feature of characteristic size 3 pixels in my original input spectra (where pixels and angstroms are related linearly) reflects on a wavenumber of value 150 in fourier space? Thanks in advance for any help!
Mike Fitzpatrick wrote on Jan 15, 2008
I think you understand it correctly: The 'i' keystroke prints the "period trend in the data" in pixel space, i.e. a value of 3 means that a feature repeats every 3 pixels in the data. The equivalent wavenumber of and frequency are also printed and all values refer to the linearized log-dispersion spectrum that includes any padding and apodization.
More generally, features on the right of the default power spectrum plot represent high-frequency patterns such as pixel-to-pixel noise, features on the left tend to be the spectral lines or continuum background. In an ideal spectrum, the power should fall off on a curve and then level off where the noise begins to take over. The filter then should be set to cutoff the noise (or attenuate it) to improve the correlation of the spectral features. Of course, in the real world determining the "flat spot" is never that simple.
-Mike
More generally, features on the right of the default power spectrum plot represent high-frequency patterns such as pixel-to-pixel noise, features on the left tend to be the spectral lines or continuum background. In an ideal spectrum, the power should fall off on a curve and then level off where the noise begins to take over. The filter then should be set to cutoff the noise (or attenuate it) to improve the correlation of the spectral features. Of course, in the real world determining the "flat spot" is never that simple.
-Mike
Last post on Jan 15, 2008