autoidentify with non-linear dispersion
jacosta wrote on Sep 09, 2009
Hi there,
I'm trying to set up a semi-automatic wavelength calibration for near IR spectra. The spectral formats are very reproducible since the instrument is based on grisms which produce only few configurations. I mean the spectral range and dispersion are known with relatively good accuracy, expecting only small offsets due to flexures, etc. I have tried to do the calibration using autoidentify, I provide the
parameters crval and cdelt but the solution found is not very good.
There are two resolutions (700 and 2500), the number of features is around 20 and 5-8 respectively.
What could be the most reliable approach to have automatic feature identification??
I'm trying to set up a semi-automatic wavelength calibration for near IR spectra. The spectral formats are very reproducible since the instrument is based on grisms which produce only few configurations. I mean the spectral range and dispersion are known with relatively good accuracy, expecting only small offsets due to flexures, etc. I have tried to do the calibration using autoidentify, I provide the
parameters crval and cdelt but the solution found is not very good.
There are two resolutions (700 and 2500), the number of features is around 20 and 5-8 respectively.
What could be the most reliable approach to have automatic feature identification??
Francisco Valdes wrote on Sep 09, 2009
There are two approaches. One is based on having a previous solution that is very close. This would only be useful if your instrument really has only a few setups so that a particular wavelength will be within a few pixels of a standard point. Automatically ransferring a solution to a new spectrum (usually an arc) is done with REIDENTIFY.
The automatic approach is AUTOIDENTIFY. Automatic identification is a pattern recognition problem which is harder the less information can be supplied. But you probably know a reference wavelength and dispersion to within a fairly large interval in the spectrum. The main problem is that to find patterns of line spacings, which is how the task works, depends on their being a fairly linear relation between pixel and wavelength across the spectrum. Dispersers often don't satisfy this. The task provides a parameter, autoid.crquad which is a coefficient that allows an quick and approximate quadratic adjustment to roughly linearize the spectrum for the pattern finding. Read the help for aidpars as well as autoidentify. For what you say, you should be able to characterize this parameter for the limited modes. Once you have this and a good idea of a wavelenth and dispersion as a reference point the algorithm should work well. It also helps to tune the line list to include a modest number of strong lines rather than all the lines possible. The list can be made from your data itself because the strengths of arc lines vary a lot by the lamp and voltages.
I hope this helps,
Frank Valdes
The automatic approach is AUTOIDENTIFY. Automatic identification is a pattern recognition problem which is harder the less information can be supplied. But you probably know a reference wavelength and dispersion to within a fairly large interval in the spectrum. The main problem is that to find patterns of line spacings, which is how the task works, depends on their being a fairly linear relation between pixel and wavelength across the spectrum. Dispersers often don't satisfy this. The task provides a parameter, autoid.crquad which is a coefficient that allows an quick and approximate quadratic adjustment to roughly linearize the spectrum for the pattern finding. Read the help for aidpars as well as autoidentify. For what you say, you should be able to characterize this parameter for the limited modes. Once you have this and a good idea of a wavelenth and dispersion as a reference point the algorithm should work well. It also helps to tune the line list to include a modest number of strong lines rather than all the lines possible. The list can be made from your data itself because the strengths of arc lines vary a lot by the lamp and voltages.
I hope this helps,
Frank Valdes
Last post on Sep 09, 2009