CCD spectrographs

Current-day optical spectrographs almost exclusively use CCDs as their detector. The major reasons for this choice are the large free spectral range of modern CCDs (covering roughly 3000 to 11000 A), the linearity of the devices (better than 1% from zero counts to saturation) over a large dynamic range (allowing, for example, detection of absorption or emission lines as well as the continuum), and the large areal format of modern CCDs (2048 up to 4096 pixels or more in extent). This latter property is especially important for applications such as wide-field objective prism work with Schmidt Telescopes, multiobject fiber spectroscopy (Robinson, 1988c) in which the fiber fed spectra are placed in row order on the CCD, and for echelle spectroscopy for which many spectral orders are two-dimensionally imaged at once (Vogt & Penrod, 1988). The free spectral range obtained could, in principle, be as large as the detector's quantum response, but in practice limits in optical and grating design and CCD size restrict a single spectrograph coverage to somewhere near 4000 A or less in the optical band-pass. As we have mentioned, some observatories have solved this limited spectral coverage issue by designing and using double spectrographs having two separate spectrograph arms (one red and one blue) with a CCD for each (Oke, 1988). A double spectrograph almost always uses a different type of CCD detector in each arm; each CCD is customized for the best possible detection properties for its particular wavelength coverage. Both high resolution (R ~30 000-80 000 or more) and low resolution (R ~ a few thousand or less) spectroscopic applications are well suited to using CCDs as the detector; one simply has to match the CCD pixel size to the particular spectrograph and resolution being used.

Optimal sampling of a spectral line that is just unresolved occurs when the FWHM of the line is twice the physical pixel size (Nyquist sampling criteria). It can be assumed in this discussion that a spectral line has an approximately Gaussian shape for which a formal FWHM value can be determined. Note, however, that real spectral lines are not always this well behaved. In addition to matching the spectral line width to the pixel size, CCDs used for astronomical spectroscopy must also have very good charge transfer efficiency (CTE) in order to reduce smearing of spectral lines during readout, which would lead to a loss in spectral resolution. Also, small pixel sizes such as 15 or 9 microns are often desired to meet the Nyquist criteria discussed above.

Let us look at an example for a spectrograph that uses a CCD with 9-micron pixels as the detector. With this setup, the projected slit width size, must be near 18 microns to achieve optimal sampling. For an observing site with typical seeing of 1.5 arcsec, and using a 2- to 5-m telescope, we find (using the formulations given in Section 6.1) that the spectrograph camera must have a focal ratio near unity. This is a very fast focal ratio and requires excellent optical design and near perfect optical surfaces. For the CCD itself, this requirement means that its physical surface must be extremely flat throughout the entire extent of the chip (less than 0.5% rms for accurate spectrophotome-try), in order to allow all parts of the spectrum to be in focus simultaneously. As we have seen, this level of flatness can be a difficult requirement to fulfill for certain types of CCD (e.g., thinned devices).

The above example for a CCD spectrograph informs us that, for large-aperture telescopes (say 8-12 m), optimum spectral sampling can only occur if some combination of the following conditions are met. As the telescope diameter increases, the camera focal length must decrease, the seeing disk must decrease, and the detector resolution element (2 CCD pixels) must increase in size. Currently, the hardest requirement to meet in this list is the design and construction of very fast focal ratio cameras. Increasing the CCD pixel size while retaining the large range of total wavelength coverage is a major driving force behind producing even larger format CCDs.

Exceptional seeing, less than 1 arcsec for example, would seem to be the dream of any spectroscopist. However, let us look at an example when very good seeing can cause unexpected results in CCD spectroscopy. The problem is as follows: CCDs are mounted in dewars and attached to the end of a spectrograph in some manner. The dewars are then aligned in an attempt to have the observed spectrum fall onto the detector along either its rows or columns. Perfect alignment of the CCD across the entire spectrum is rarely achieved and thus the imaged spectrum centroid must cross pixel boundaries in the dispersion direction (i.e., as a function of wavelength).

If the object seeing disk becomes less than the projected pixel size, the position of the spectral centroid falls within the pixel itself, alternately occurring at the center of some pixel and then at the pixel boundaries themselves. Wavelength-dependent QE effects within the pixels, due to their gate structures and intra-pixel "dead" spots, will cause apparent flux variations that can be as large as ±10% in amplitude. Additional complexities, such as which type of CCD is used, telescope focus, and tracking changes, are harder to quantify and correct for but can have similar effects. The problem described here, that of undersampling, can also occur in CCD imaging applications as well, when the majority of the source PSFs fall within a single pixel. Optimum sampling in CCD imaging also occurs at the Nyquist sampling limit, that is, a point source FWHM should be imaged across about two CCD pixels (see Section 5.9). A several percent error in CCD photometry can occur for images that are undersampled (Howell et al., 1996; Holtzman, 1990).

Figure 6.5 shows an example of the ripple that can occur in a spectrum obtained under conditions of excellent seeing and for which intra-pixel QE effects are present (Rutten, Dhillon, & Horne, 1992). Possible methods of correction for spectral ripple include de-focus of the telescope or slightly trailing the spectrum up and down the slit during each integration,1 neither of which are desirable. Corrections to the spectrum can also be applied after the fact using some empirical model during the reduction phase (Dhillon, Rutten, & Jorden, 1993; Jorden, Deltorn, & Oates, 1993). Spectral ripple, as well as our discussion of pixel sampling in the chapter on photometry, indicates that while very poor sampling is not ideal (as the collected image is spread out over many pixels, each of which adds unwanted noise) undersampling is not necessarily better. Optimum sampling is the best but is not always possible with a given instrument and CCD combination. In addition, the conditions of optimum sampling can change with time owing to effects such as seeing, telescope focus, the wavelength of light imaged, and other more subtle effects.

1 This particular solution may remind us long-time observers of a Wobble Plate.

6000 7000 8000 9000 104

wavelength (A)

Fig. 6.5. Spectrum of the standard star Feige 92 taken during a time of excellent seeing. This 30-second exposure, obtained with the Faint Object Spectrograph on the Issac Newton telescope, has a pixel scale of 1.2 arcseconds/pixel and this spectrum was obtained when the seeing was sub-arcsecond. The ripple seen in the continuum near 6000 A and blueward is due to intra-pixel QE differences, which are especially prominent for short wavelength photons in the EEV CCD used. From Dhillon, Rutten, & Jorden (1993).

6000 7000 8000 9000 104

wavelength (A)

Fig. 6.5. Spectrum of the standard star Feige 92 taken during a time of excellent seeing. This 30-second exposure, obtained with the Faint Object Spectrograph on the Issac Newton telescope, has a pixel scale of 1.2 arcseconds/pixel and this spectrum was obtained when the seeing was sub-arcsecond. The ripple seen in the continuum near 6000 A and blueward is due to intra-pixel QE differences, which are especially prominent for short wavelength photons in the EEV CCD used. From Dhillon, Rutten, & Jorden (1993).

The two-dimensional nature of a CCD allows one to place the spectrum anywhere on the array, targeting good regions of the CCD that avoid manufacturing flaws such as areas of poor sensitivity or dead pixels. Their 2-D design also provides the ability to simultaneously image the (nearby) sky spectrum with the object spectrum. Accurate sky subtraction from a source spectrum increases the S/N of the final result and allows much fainter sources to be observed. It also increases the reliability of the flux measurements and is probably the greatest factor making CCDs superior to other detectors as spectroscopic imagers.

Although large-format CCDs for spectroscopy (2048 x 2048, 1024 x 3072, etc.) are useful in many ways, they have some drawbacks as well. Large CCD formats take longer to readout (if windowing in not available), they make areal flat fielding critical (important for accurate fluxes and line profiles), and they provide more stringent restrictions on mounting, dewar design, and overall cost. Physically nonflat CCDs under vacuum (such as thinned devices), shifting of the CCD within the dewar, or movement of the larger, heavier dewar itself can all cause slight movement of spectral features, leading to errors in the analysis. Loss of LN2 from the dewar throughout the night has been identified as one cause of spectral movement (Smith, 1990a).

Readout times for large-format CCDs (typically 1-2 minutes, but faster in newer instruments) do not seem like such a big deal compared with typical integrations of 30-60 minutes or more. However, the longer the readout time, the lower the overall observational duty cycle one experiences, and, after obtaining all the necessary calibration spectra, flat fields, comparison arcs, and standard stars, this extra time can become costly. CCDs that have a limited dynamic range (those with small pixel sizes) force the user to make multiple (shorter) exposures for each type of needed image, especially calibration data for which the high S/N desired can easily saturate the detector. In addition, to produce the final spectrum one often wishes to co-add shorter object exposures to avoid numerous cosmic ray events that can hinder accurate flux and line profile measurements.

The finite readout time of large-format CCDs has led to numerous attempts to obtain high-speed spectroscopy via some technique that takes advantage of the two-dimensional nature of the detector. For example, one could step the detector, or the spectrum, along in a direction perpendicular to the dispersion at some predetermined rate, reading out the CCD only after an entire frame of individual spectra is collected. A better idea might be to use the fact that the collected charge can be moved back and forth electronically on the CCD from row to row, without actually reading it out. This movement is quite fast, say 200 rows in a few milliseconds, and suffers no readout noise penalty but only a slightly increased dark current. Spectra placed on the top 50 rows of a device will allow faster readout times or one could use the bottom 50 rows with electronic charge movement used to shift the spectrum upward between each integration time. Both of these processes have been tried with success (Robinson, 1988b; Skidmore et al, 2004).

One could even slosh the charge back and forth periodically in order to buildup the signal in two spectra for say a given phase of a binary star. The spectrum not being exposed gets hidden under a mask and the final image produces two or more "phased" spectra. An idea such as this was developed for spectropolarimetry (Miller, Robinson, & Goodrich, 1988) with some success and led to the discovery of some previously unknown, yet interesting CCD effects (Janesick et al., 1987a; Blouke et al., 1988).

Was this article helpful?

0 0
Telescopes Mastery

Telescopes Mastery

Through this ebook, you are going to learn what you will need to know all about the telescopes that can provide a fun and rewarding hobby for you and your family!

Get My Free Ebook

Post a comment