System Design Study

The availablity of sensitive commercial CCD cameras aimed at the amateur astronomy community, as well as limited access to in-house workshop resources, made it desirable to seek an off-the-shelf camera system. Our requirement for high quantum efficiency excluded cameras on offer from the Santa Barbara Instrument Group (SBIG). Our budget cap of order $US10K excluded other manufacturers, such as PixelVision, Photometrics, and Princeton Instruments. Three possible CCD cameras manufactured by Apogee Instruments Inc. were considered; the AP7 camera selling for $\sim$ $US7500, the SPH3 camera selling for $\sim$ $US9000, and the KX260 camera selling for $\sim$ $US4000. Each uses a mechanical shutter.

The AP7 camera was considered for its high quantum efficiency (QE) and low dark current. It uses a SITe SIA502AB, 512 x 512, 24 $\mu$ m pixel back-side illuminated, full-frame CCD. It has a peak QE of $\sim$ 85%, 15 e read noise, 1 e/s/pix dark current at -40 $^\circ$ C, and a well depth of >350,000 e. The full-frame readout time was calculated to be 8.7 s.

The SPH3 camera was considered for its high QE and rectangular geometry which matches the DBS slit geometry. It uses a Hamamatsu S7030-0908, 512 x 250, 24 $\mu$ m pixel, back-side illuminated, full-frame CCD. It has a peak QE of $\sim$ 93%, 10-15 e read noise, 50-100 e/s/pix dark current at -10 $^\circ$ C, and >600,000 e well depth. Digitization is to 16 bits with a gain of 5 e/ADU, giving an effective full well of $\sim$ 320,000 e. The full-frame readout time was estimated to be 4.3 s.

The KX260 camera was considered for its fast, pseudo real-time, read speed. It uses a Kodak KAF-260 $512\times512$ , 20 $\mu$ m pixel, front-side illuminated CCD. It has only 38% peak QE, 15 e read noise, and 3 e/s/pix dark current at -10 $^\circ$ C. The controller uses a 14-bit, 1.3 MHz ADC so the estimated full-frame readout time is only 0.2 s. It has a gain of 8 e/ADU, so only just digitises read noise, but can sample up to 120,000 e of the CCD well depth.

Pixel Scales

Modification to the DBS slit viewing optics was required to produce an accessible focus for any of these cameras. The simplest proposed modification produced a pixel scale of $\sim$ 0.36''/pix. The AP7 and SPH3 cameras would then image the full width of the DBS slit and 184'' along the DBS slit (i.e., just under half of the 400'' slit length) with $\sim$ 50% vignetting in the corners. The ability to image this large fraction of the DBS slit length in one CCD exposure was seen as a significant advantage for identifying suitable offset guide stars, and for making accurate measurements of their offsets from the science object. The Fairchild camera had imaged 140'' along the DBS slit. With the same slit viewing optics, the KX260 camera would produce an image scale of 0.30''/pixel and would image a field-of-view of $153''\times153''$ .

Faint Object Performance

Approximate performance figures for each camera are listed in Table 1. The object brightness is that of a typical R=21.0 mag K giant star, which is likely to be the faintest object observed. We adopt a system optical throughput of 46% which is based on 5 reflections and 6 air-glass surfaces.

**Table 1:** Camera Performance Model
Parameter	B	V	R	I	Total
Zero magnitude flux (Jy)	4260	3640	3080	2550	...
Adopted wavelength (Å)	4360	5450	6380	7970	...
Adopted width (Å)	1130	800	1500	2500	...
Transmission of atmosphere	0.72	0.84	0.88	0.94	...
Transmission of optics	0.46	0.46	0.46	0.46	...
Object brightness (mag)	22.4	21.6	21.0	20.5	...
Object signal at CCD (photon/s)	25.1	29.6	73.4	136.6	...
Sky brightness (mag/arcsec²)	22.5	21.5	20.8	19.3	...
Sky signal at CCD (photon/s/arcsec²)	22.9	32.5	88.2	412.6	...
AP7 camera:
Q.E.	0.67	0.80	0.85	0.78	...
Object current (e/s)	16.8	23.7	62.4	106.6	209.5
Sky current (e/s/pix)	2.0	3.4	9.7	41.7	56.8
Dark current @ -40 C (e/s/pix)	1	1	1	1	1
Read noise (e)	15	15	15	15	15
Time for RN = (sky+dark) noise (s)	75.0	51.1	21.0	5.3	3.9
SPH3 camera:
Q.E.	0.70	0.90	0.93	0.70	...
Object current (e/s)	17.6	26.7	68.3	95.6	208.1
Sky current (e/s/pix)	2.1	3.8	10.6	37.4	53.9
Dark current @ -10 C (e/s/pix)	50	50	50	50	50
Read noise (e)	15	15	15	15	15
Time for RN = (sky+dark) noise (s)	4.3	4.2	3.7	2.6	2.2
KX260 camera:
Q.E.	0.10	0.30	0.38	0.35	...
Object current (e/s)	2.5	8.9	27.9	47.8	87.1
Sky current (e/s/pix)	0.2	0.9	3.0	13.0	17.1
Dark current @ -5 C (e/s/pix)	3	3	3	3	3
Read noise (e)	15	15	15	15	15
Time for RN = (sky+dark) noise (s)	70.3	57.7	37.5	14.1	11.2

When operated unfiltered on a dark sky, Table 1 shows that the AP7 camera would be largely sky-noise limited (read noise equals sky plus dark current shot noise in 3.9 s), the SPH3 camera would be largely dark current limited (dark current approximately equals sky signal), and the KX260 camera would be read noise limited for exposures shorter than 11 s unfiltered and for most filtered exposures on a dark sky. In 30 s unfiltered exposures in 1.5'' x 1.5'' seeing, a star with R = 21.0 mag has formal signal-to-noise ratios of 34, 26, and 18 with the AP7, SPH3, and KX260 cameras, respectively. These relative performances are shown visually in Fig. 1. This figure shows simulated 30 s exposures in 1.5'' seeing for the three cameras containing stars (bottom row), spiral galaxies (middle row), and elliptical galaxies (top row) with total R magnitudes from left to right of 18.0 to 22.0 in steps of 0.5 mag. On the basis of these calculations, the AP7 camera is to be preferred. It should outperform the SPH3 and KX260 cameras by 0.3 and 0.7 mag, respectively. However, all three cameras appear to be capable of achieving the required limiting magnitude of $R \sim 20$ on stars and galaxies in reasonable seeing.

**Figure 1:** Simulated 30 s exposure in 1.5'' seeing for the AP7 camera ( *top frame*), SPH3 camera ( *middle frame*), and KX260 camera ( *bottom frame*). Objects are (from top to bottom in each frame) elliptical galaxies, spiral galaxies, and stars with total R magnitudes (left to right) of 18.0 to 22.0 in steps of 0.5 mag. Greyscales range from -3 $\sigma$ to +5 $\sigma$ in each frame.
$\begin{figure} \centering\leavevmode \epsfysize=20cm \epsfbox{sim.ps}\end{figure}$

Bright Object Performance

Performance on bright objects was also compared. The minimum integration time is 0.03 s for each of the cameras. We again use our performance model for unfiltered exposures to crudely estimate the saturation magnitudes in 1.0'' seeing; these are 6.9, 7.0, and 6.7 mag for the AP7, SPH3, and KX260 cameras, respectively. Given the uncertainties in bias levels and point spread functions, these are essentially indistinguishable and a more conservative number of $\sim$ 7.5 mag is probably appropriate. It is clear, therefore, that a neutral density filter (at least ND2 = 5 mag) is needed with any of these cameras to reach the required bright limit of $R \sim 3$ mag.

Readout Speed

A major shortcoming of the AP7 and SPH3 cameras is their relatively slow readout times (8.7 and 4.3 s full-frame, respectively). The DBS slit face is $\sim 80''$ wide, so would occupy only 225 pixels in height on the AP7 or SPH3 CCDs with an imaging scale of 0.36"/pixel. Both cameras can be windowed (see below) so the effective readout times for the illuminated regions were estimated to be $\sim$ 3.8 s. Nevertheless, this would not appear at all like a real-time display to the observer, and could be a major inconvenience when working at high resolution on relatively bright objects.

Shuttering

Each of the three Apogee cameras is a full-frame device which must be shuttered. For a typical autoguiding exposure time of 5 s, continuous operation of the camera for 10 hr per night for 200 nights per year requires 1.44 x 10⁶ shutter openings per year. The AP7 camera uses a Vincent 25 mm blade shutter, with which RSAA has had no previous experience. RSAA has successfully operated several other Vincent mechanical shutters for extended periods, but never at this duty cycle. Vincent informed us that the mean time between failures for the VS25 is $\sim 1.3 \times 10^6$ cycles. We therefore expect a shutter failure rate of $\sim 1$ per year and consequently have instigated a preventative maintenance plan.

Control Software

Diffraction Ltd. offer MaxIm CCD camera control and image processing software for Apogee Instruments Inc. hardware running under Windows NT. In limited cross-evaluation this software appeared superior to the only other strong contender, CCDSoft by Software Bisque Inc., in its overall integration and ease of use. Furthermore, Diffraction Ltd. agreed to modify MaxIm CCD to output the control sequences necessary to interface the autoguiding function of MaxIm CCD to the 2.3 m telescope control system (§4.2).

MaxIm CCD acquires data in one of several acquisition modes, as well as monitoring and controlling the CCD temperature. The user can easily switch between these modes. In "Expose" mode, a windowed region of the CCD is read out and the result can be written to disk in FITS format (Wells, Greisen, & Harten 1981). On-chip binning is supported, as well as on-line bias subtraction, dark subtraction, and flat fielding. In "Focus" mode, a windowed subframe of the CCD is continuously read out. This mode can be used for focussing as well as for manual guiding. "Inspect" mode allows users to continuously monitor the centroid coordinates, peak intensity, full width half maximum, and profile of the images obtained in "Focus" mode. "Guide" mode is the most appropriate when operating as an acquisition camera. Both wide-field acquisition exposures and narrow-field guiding exposures can be initiated from "Guide" mode. Wide-field exposures are used for selecting offset guide stars. Narrow-field exposures are used for autoguiding. The standard autoguiding sub-frame is $32\times32$ pixels.

MaxIm CCD also has excellent features for combining tricolor images. This has proved invaluable for difficult acquisition tasks, such as aligning aperture plate masks.