How to best prepare spectra for CCF?

[Thomas Dickinson]

Happy Sunday!

Bottom line up front: what is the best way to prepare spectra for comparison with CCF in ISIS?

I am working with some other students on collecting data on the star Tau Bootis, which has a large exoplanet orbiting it that has been successfully studied by many people including Christian Buil. We are using a QSI 500 series camera, an eShel spectrograph, and Audela software for initial spectrum processing.

We are interested in using the Fourier Cross Correlation (CCF) functionality that is built into ISIS v5.4 under "Misc" --> "Radial Velocity" --> "Compute radial velocity difference" to analyze our Tau Bootis spectra and build a deltaRV curve to try and determine the exoplanet's orbit.

Two weeks ago we took two fifteen minute spectra of Tau Bootis one after the other and processed them in Audela to get the "B" plots that give intensity as a function of wavelength from about 5000 Angstroms to 6500 Angstroms. When we use the CCF functionality in ISIS to compare the two spectra it gives us a -0.715 km/s radial velocity difference from the CCF peak, when we expect is should be less than 0.100 km/s in magnitude given that the two spectra were taken one after another and the planet has an orbital period of 3.31 days. Cropping the two spectra down by about 5% on each side to remove some of the "noisy" looking data gives an improved Radial Velocity difference of -0.488 km/s, but if it could be lower that would be ideal.

I have attached screenshots of the CCF with ISIS and of the two spectra being compared. We are thinking that the discontinuities between orders caused by Audela stitching orders together (the periodic groups of spikes seen in both of the spectra) is throwing off the CCF, and we are wondering if there is a good way to remove these or lessen their negative effects. Any suggestions that are even remotely related to what we are doing will be most appreciated!

Thank you so much and I will be watching to answer any questions anyone willing to help may have!

[Martin Dubs]

Hi Tom,

there are a few things that show up at first sight.
First is the rather poor S/N. The precision of radial velocity measurement is directly related to S/N, so this would be the first point to start with. I do not know the telescope you are using, so it is difficult to say how much you can improve there.
Second, the flats and the corresponding instrument response are not ideal, I had to work about this a lot. In my case it was related to poor alignment of the fiber guide unit. Better results for instrument response may be obtained by the order by order method (see notes on echelle calibration).
In general I would study Christians notes about exoplanet observation here carefully: http://www.astrosurf.com/buil/extrasolar/obs.htm, especially the notes about frequent calibration and temperature stabilization. For high precision also make sure you use exactly the same mechanical setup and the same analysis parameters. The smallest changes have a dramatic effect if you look for such small changes. Without a temperature stabilized setup I normally achieved a precision of +/- several 100m/sec over several days with my mobile observatory.
For the analysis it is no use to calculate the CCF over such a large interval, 100km/sec is enough, with a finer spacing. Try using the part of the spectrum with best S/N (possibly calculating RV for diffferent orders separately and averaging afterwards). Exclude regions with telluric lines.

Hope you can take something useful from this.

Good luck, Martin

[Olivier Thizy]

Tom,


one thing that was done during acquisition is a themometer close to the spectrograph. We couldn't control temperature, but at least we were monitoring it and tried to close the telescope door to avoid temperature changes. From memory, temperature was stable within couple of °C.


A key thing was taking ThAr calibration all the time. Our basic acquisition was:
-take 3 ThAr
-activate guiding on the star
-acquire one spectrum (around 10minutes for tau Boo if I remember well)
-desactivate autoguiding
-take 3 ThAr
-activate guiding on the star
-acquire one spectrum (around 10minutes for tau Boo if I remember well)
-desactivate autoguiding
-take 3 ThAr
-activate guiding on the star
-acquire one spectrum (around 10minutes for tau Boo if I remember well)
-desactivate autoguiding
-take 3 ThAr

For each object spectrum, we took the ThAr before and after and average it; this was our ThAr to reduce the spectrum taking in the middle.

This gave us three reduced spectra to analyse; the three RVs obtained were average in a single value.


Each night, we were taking a reference star whose Radial Velocity was known to be stable. Using that reference RV measurement, the offset was introduced for each night/session.


I hope this helps,

Cordialement,
Olivier Thizy

[Thomas Dickinson]

Fantastic, thank you so much. We are waiting on good weather to collect more data.

Martin, our telescope was designed for observing the moon so the optics are very slow (f/16) so it is difficult to get a good SNR. These are 15 minute exposures so we are going to try and do 20 minute exposures once the weather clears up to try and improve our SNR. Excluding regions with telluric lines and averaging over the remaining regions was something we may try. Using Martin's instructions for the 1/2 interval and step size as well as reprocessing the data to order 48 instead of order 44 we were able to get down to about 50 m/s which is pretty good. We are looking forward to trying the same method as Olivier for collecting and averaging calibration files, because up until this point we have only been using a single ThAr exposure for each night of observing.

I have another question about Audela processing which I posted to the Yahoo group but is probably better to ask here:

When starting with raw data from the camera, what sort of interpolation is used in between data points to generate the A plots, since they show more detail than just one value for each pixel? What impact, if any, does this interpolation have on the Cross Correlation?

[Thomas Dickinson]

Also,
We are trying to figure out how to calculate the uncertainty and we are wondering how, from Christian Buil's deltaRV equation, to determine "w" (translated to English as the "observed linewidth?" Also, in that same equation where is FWHM (full width half max) value in the numerator coming from? Finally we are wondering if Audela normalizes the processed spectra with a built in quantum efficiency curve for our QSI 516 camera, or if that is something we need to do manually using the instrument response procedures?

Any insight anyone might provide would be much appreciated! Thank you!