CFHT, Instruments, Detectors, IR, Redeye Manual: Chapter 2.


Redeye Performance Specifications

Below is a summary of some of the most important performance specifications for both Redeye cameras. Many of the questions users have about the cameras can be answered by simply looking at this table. Note that, even though the cameras are designed to support interchangeable parts, in order to simplify summit operations a particular set of optics and a cryostat was dedicated to each array, hence users can find array specifications based upon which camera they are using. All pertinent values apply to the f/8 focus with a 1V bias voltage across the detectors (i.e., nominal conditions).

Table 2.1 - Many of the most critical performance levels of the Redeye cameras are listed above. An asterisk indicates values that are only estimates based on preliminary tests and will be updated with final values in V1.1 of the manual.


Both cameras house NICMOS3 Hg:Cd:Te infrared array detectors. These devices are manufactured by Rockwell International Science Center. They are hybrid devices, consisting of a photosensitive Hg:Cd:Te layer that is bump bonded through Indium columns onto a low noise silicon multiplexer. Each array has 256 x 256 pixels that are on 40 m centers, yielding detectors that are 10.24 x 10.24 mm across. Unlike a CCD, these detectors are literally composed of discrete photodiodes, with charge switched out of the unit cells and into one of 4 amplifiers located at the corners of the array. As a result, NICMOS3 detectors do not exhibit charge bleeding across rows or columns, as most CCDs suffer from, making them easy to use when bright sources are in the field. Each array is physically composed of four 128 x 128 pixel subarrays that are isolated from each other except for a common ground plane. This permits a wide degree of flexibility in tuning the performance of the device, as well as special applications that demand, for example, only single quadrant read-outs when data throughput is at a premium.

Figure 2.1 shows bad pixel maps for the science grade arrays originally in the cameras. A bad pixel in this measurement is one that has a signal that differs from the average signal in a flat field by more than a factor of 2. For the narrow field science array, this corresponds to ~550 bad pixels, of which ~90% are dead (have 0 signal). Most of these pixels are single pixels scattered across the entire chip, with a few clumps. (Note that due to a failure of one quadrant of the narrow field camera's array, a new device has been installed. As of this writing engineering tests are underway in the lab and performance characteristics will be determined on the telescope in the immediate future. As a result, all references to the narrow field camera's detector performance in this manual should be disregarded until these data can be updated.) For the wide field camera's array, there are ~150 bad pixels, with essentially all of these pixels dead.

Figure 2.1 - Bad pixel maps of the science arrays are shown. To the left is that housed in the wide field camera, to the right is the array in the narrow field camera.

Figure 2.2 shows flat fields made with both science arrays. Note how these devices exhibit "peaks" and "valleys" of responsivity across there surface, presumably due to variations in the thickness of the Hg:Cd:Te layer after polishing. The wide field array has a 1 dispersion from the mean of ~8%, while the narrow field array has a 1 dispersion from the mean of ~10%. These variations are a function of array bias, which is normally left at 1.0 V in order to make the well depth ~400,000 e-. Though this bias voltage increases the read noise slightly, most Redeye applications will be broad band imaging and will therefore be background limited in a few seconds of exposure time with the read noise induced by a 1.0 V bias.

Figure 2.2 - Flat fields for the wide field (left) and narrow field (right) science arrays are shown.

It is important to realize that the Redeye detectors are sampled in a very different manner than that used with CCDs. The array is first reset to clear all charge from the wells and then allowed to "settle" for a couple of seconds. It is then read-out twice. The first read-out is discarded but the second read-out is stored in memory. Next, the shutter is opened, thereby starting an integration on the sky. At the end of the integration, the shutter is closed and all four quadrants are read-out in parallel twice again. As in the beginning of the exposure, the first read-out is discarded but the second one is retained, then subtracted from the read-out stored at the beginning of the exposure, a FITS header is attached to the image, and the file is stored on disk. As a result, there is no DC or bias offset left in the image stored on disk, hence a 0 ADU value truly corresponds to 0 flux and dark current in the image. Implied by this description of the read-out process is the fact that NICMOS3 arrays have non-destructive read-out architectures. While it is theoretically possible to reduce the effective read noise in images through repeated samples by a factor of 1/*n (until 1/f noise dominates), this introduces other deleterious effects, described below.

Unusual Detector Characteristics

NICMOS3 arrays have a number of peculiar characteristics that users should be aware of when making observations. These features stem from design flaws in the arrays and are therefore found in all such detectors. A detailed physical explanation for these problems with NICMOS3 arrays is not available, even from the manufacturer, Rockwell. Hence our strategy has been to search the operational parameter space for the Redeye arrays and find read-out techniques that minimize the effects of these device flaws. While we are making every effort to reduce the impact of these problems through sophisticated controller algorithms, their effect is never eliminated completely in Redeye images. Each of these features are described in detail below.

Residual Images

At a low level, NICMOS3 arrays retain charge in pixels even after the devices have been reset. This problem probably stems from charge trapping sites in the array's lattice structure that are not flushed by reset pulses. The net effect of this phenomenon is that arrays retain a "memory" of past exposures. For example, if a bright star is imaged and then several dark frames are recorded, faint residual images of the star will be seen in the dark frames. Residual images typically retain only a ~1% percent of the flux level in the original image, hence if a star saturates the array, the residual image would only have ~1% of the full well charge in it. Residual images can be reduced through repeated reads and resets and they have a bimodal decay nature. More specifically, the first residual image only has ~1% full well of charge, but then the next read-out shows a reduction to ~0.3% full well, and subsequent images show a slow decay of this remaining charge over dozens of frames. Figure 2.3 illustrates the fast and slow decay components in a typical residual image. It was made by first flooding the array with light, then recording a series of 30 second dark frames. In spectroscopic (i.e., low background) applications this problem can be severe, since residual images can last for hours at an extremely low level, regardless of how many resets are issued. Since Redeye is not used with filters that are narrower than ~1%, and is most often used in broad band high background situations, residual images are usually overwhelmed by background flux and are therefore not a major problem. Still, in an effort to minimize the problem, the Gen III controller is configured to send the NICMOS3 array in each camera a steady stream of reset pulses when the camera is idle. This stream of reset pulses also helps prevent the camera from saturating when thermal (> 2.0 m) filters are used, allowing the array to integrate on the back of the warm external shutter. Furthermore, observers are encouraged to make several reads before a set of darks is acquired in order to flush as much accumulated charge in the array as possible.

Figure 2.3 - A plot of residual charge as a function of frame number is plotted. Note the steep drop in charge during the first few reads, then the slow decay thereafter.

Observers may notice in some images a set of circular dark spots on the arrays. At first glance, because they are circular and have the appearance of being somewhat fuzzy or "defocused", one might assume they are caused by cold dust in the optics. Actually, these spots are intrinsic to the arrays and probably stem from slight nonuniformities in the doping of the devices. They are removed from images through normal image processing techniques. Interestingly, Hodapp et al. (1992) noticed that these spots correspond to regions where residual image problems are minimized. There was speculation early after this discovery that they were caused by gaps in the epoxy used to bond the photosensitive layer to the underlying silicon multiplexer, but recent tests on a device intentionally left only partially glued proved otherwise. Since they occupy such a small fraction of the array surface area, trying to beat the residual image problem by placing targets in them is not practical.

Finally, one might suspect that this residual image effect can lead to accumulated charge over the course of a night of observations, effectively reducing the dynamic range of the device and progressively corrupting images at a low level. We have tested this possibility in the lab and found that the residual charge does not accumulate beyond a fixed level in the arrays. In other words, if a set of images of a galaxy are acquired with the camera in a simple staring mode, the flux is constant (within the various noise sources) for the entire data set with a faint and constant residual image component superimposed on all images.

Amplifier Glow

A well known feature of NICMOS3 frames is the excess charge seen in the corners of images. This charge comes from the output amplifiers in each quadrant during the read-out process and is believed to be due to luminescence from the amplifiers. This flux source is only injected into images during read-outs, i.e., turning off the amplifiers during exposures has a negligible effect. While this problem is not a serious one for typical high background Redeye applications, it is a serious noise source in low background applications. For this reason multiple sampling schemes, though known to reduce read noise in frames down to the 1/f noise limit, ultimately fail because the amplifiers must be repeatedly activated throughout the sampling period, thereby greatly increasing the glow from the array corners and potentially overwhelming object flux. Figure 2.4 shows the morphology of this glow in a dark frame. The high frequency noise in this image is pick-up noise in the lab where this image was recorded.

Figure 2.4 - A short exposure showing glow from the output amplifiers located on the corners of the array is shown.

Reset Anomaly

If a NICMOS3 array is read-out immediately after resetting the device, instead of the expected 0 ADU signal level, a significant charge is seen across the entire array. This charge decays very rapidly during a "settling" process after the reset and is due to electrical coupling between the reset signal and each pixel. It is crucial that the array be allowed to settle after a reset and before the first stored read-out in a double correlated sample so this charge does not interfere with images. Users of Redeye do not have to be concerned with this facet of NICMOS3 arrays since the Gen III controller pauses a couple of seconds after sending a reset pulse to the array. It is only for time-critical applications, where the array must be read-out as rapidly as possible, that the reset anomaly becomes a significant contributor to noise in images. And ultimately determines how fast images can be recorded.

Nonlinear Dark Current

Unlike most silicon detectors, NICMOS3 arrays do not have strictly linear dark currents as a function of integration time. For short exposures the dark current is fairly constant until 10 seconds of integration time. From that point on dark current adds quasi-linearly over time. Also note that when the camera has been blanked off at the filter wheel, there are still extremely low level light leaks that add to the signal detected during dark exposures.

Since dark current does not add linearly with time, it is imperative that observers not take short dark frames and attempt to scale them by some simple factor before using them to process frames. For example, if a 10 second dark is multiplied by 10 before being subtracted from a 100 second object frame, the results will be erroneous for at least two reasons. First, the dark current will not be removed since the current is not a linear function of exposure time. Second, as explained above, glow from the amplifiers is a one-shot injection of charge during each read and this charge would be over compensated for if a scaled dark current frame was subtracted during processing.


Figure 2.6 shows a number of aspects of the Redeye reimaging optics. First, each camera system is shown in perspective with the field lens in front and array in back. Rays from three points in the field corresponding to the on-axis, corner pixel, and mid-way point in the field are shown. These rays are drawn from the front of the field lens back to the detector. Also included in Figure 2.6 is a large grey square, representing the 2562 NICMOS3 array, with three smaller white squares extending to the left. These smaller squares represent, to scale, the 40 m pixels of the detectors. Spot diagrams for the three representative points in the field of view are shown in these pixels. Each polychromatic simulation includes flux from the blue, central, and red ends of the H band. In the case of the wide field optics, a slight (~1%) amount of flux falls outside of the 40 m bounds of a corner pixel while the narrow field optics contain all of the reimaged flux in a single pixel. Corresponding encircled energy diagrams are shown in Figures 2.7 and 2.8. When telescope aberrations, tracking errors, and seeing are convolved into the image, it is clear that the Redeye optics should not be a limitation in image quality, at least with the current generation of 256 x 256 pixel arrays. When future arrays based on smaller pixels and active telescope optics are used with Redeye it may be possible to resolve a slight amount of structure in the Point Spread Functions (PSFs) created by the camera optics.

Figure 2.6 - Depicted above is a layout of the Redeye Narrow field optics, while below is the wide field optical layout. Representative polychromatic spots are shown for each model.

Figure 2.7 - An H-band encircled energy diagram for the narrow field optics is shown. Point sources on-axis, 50% from corner pixel, and at corner pixel are used.

Figure 2.8 - Same as Figure 2.7 except the wide field optical performance is shown.

AR Coatings

Though the lenses used in the Redeye optics are all made of high transmission materials, they were still coated in order to minimize internal reflections and secondary images or ghosts when bright objects are in the field of view. As part of the AR coating test procedure, LiF and BaF2 windows were coated and scanned with a lab spectrometer to check their optical performance. A piece of bare LiF was also scanned as a control sample. Figure 2.9 shows the results of these tests. Note that only one side of each window was coated. From these scans it is clear that the coatings work fairly well across the 1-2.5 m range (the lab spectrometer did not work below ~1.3 m), with a predicted roll-off at ~2.5 m. The reflection per surface, based upon these tests, is ~1%, averaged across the 1-2.5 m range. It should be noted that the CaF2 Redeye windows were coated with the same material, meaning the theoretical throughput of the 12 surfaces in each camera's optics is 0.9912 or 89%. Tests of these coatings with astronomical images indicate that they give exceptional rejection of internal reflections, with > 10 mag rejection measured at J (no ghosts were seen in any test images). Comparable performance is also given at H and K, though the demand for ghost rejection is in general reduced at these longer wavelengths because of increased sky brightness, which in turns leads to shorter exposures and lower contrast between objects and the sky in images.

Figure 2.9 - FTIR lab scans made of test LiF and BaF2 windows with the same AR...

Measured Image Quality

The performance of the Redeye optics was first measured through test masks cut with LAMA consisting of a grid of 20 m holes placed precisely in the reimaging plane of the optics. Images made of these test masks matched extremely well predictions based on computer models of the narrow and wide field optics. From there, imaging performance was evaluated via direct imaging through JHK filters at the f/8 focus of CFHT, using globular clusters to provide a large number of point sources across the field of view. Figure 2.10 is one such image made of NGC 6553 through the wide field optics in August 1992. This particular image is actually the median of several dithered J-band images, each 30 sec long. Table 2.2 lists measured FWHM values from similar images made of globular clusters through the narrow and wide field optics. To generate these results, the CFHT program IQE was used to measure the FWHM values of stellar images in the globular cluster data across the entire field of view. The systematic improvement towards longer wavelengths is probably due to slightly improved seeing that inherently comes from longer wavelengths. The fact that the wide field optics yielded a PSF around 1" is not surprising, since digitization of the focal plane with the large (0.5") pixels created by this mode will typically spread light over a couple of pixels, no matter how good the seeing is. Consequently, the ~0.6" FWHM found with the narrow field optics and ~1.0" FWHM with the wide field optics is representative of what observers will probably find during a typical night on Mauna Kea. On nights of superb seeing, PSFs may drop by as much as ~0.2" below these values, particularly at K.

Figure 2.10 - An image of NGC 6553 using a J filter and the wide field optics is shown. This and many other frames of globular clusters were used in the image quality evaluation of the Redeye optics.

Table 2.2 - Image quality results with the narrow field and wide field optics are listed.

Figure 2.11 shows, in more detail, the results of the image quality analysis completed before the cameras were commissioned. Each plot shows FWHM evaluations at J, H, and K across the entire fields of view of the narrow and wide field cameras. Based upon these data we concluded that there is no significant variation of image quality across the field of view for either the narrow or wide field optics.

The plate scale for both modes was measured using binary stars at various position angles. The wide field optics were found to have a plate scale of 0.50" and the narrow field optics have a scale of 0.20", as designed. The change in plate scale as a function of wavelength is < 1 pixel across the entire field for both modes, hence multicolor photometry of extended sources or globular clusters should be straightforward with the cameras.

Background and Throughput

Tables 2.3 and 2.4 list background and system throughput measurements made with the science array in August 1992. In each column a note regarding a "stop" is used. This refers to a special occulting cold stop that, when placed in the pupil image of the reimaging optics, blocks thermal flux from the f/8 baffles and hole in the primary mirror from reaching the detector. Overall, based upon these measurements, we are getting very similar sensitivity levels as those found by the other NICMOS3 cameras used on Mauna Kea. It is important to note that the use of the occulting stops has no significant impact on camera throughput and their only effect is beneficial in the form of a reduced background in images beyond ~2.0 m.

Table 2.3 - Relative sky background measurements are listed for 2 NICMOS3 cameras used on Mauna Kea. The term "stop" refers to a special occulting stop that blocks thermal flux from the hole in the primary mirror and f/8 baffles.

Table 2.4 - Relative system throughputs are listed for the same NICMOS3 cameras listed in Table 2.3. No significant reduction in throughput was found with the occulting stop in place.


Hodapp, K-W., Rayner, J., and Irwin, E, "The University of Hawaii NICMOS-3 Near-Infrared Camera," Pub. Astro. Soc. Pacific, 104 (441).