CIRCE - Technical Details
University of Florida

CIRCE Technical Details

Optical Layout

CIRCE has an all-reflective optical system, consisting of 8 diamond-turned aluminum mirrors. Light enters the CIRCE cryo/vacuum space via an Infrasil entrance window. A linear slide at the telescope focal plane positions the Half-Wave Plate (HWP) and polarimetry/spectroscopy mask in/out of the beam. Two fold mirrors direct light into a 2-mirror collimator system which produces an image of the GTC secondary at the pupil location. In this region, CIRCE has selectable filters, a selectable Lyot mask for reduced background, and a selectable grism/prism wheel. The light then passes to the 4-mirror camera system. These are 16th-order polynomial aspheres, optimized to minimize the aberrations over the CIRCE field of view on the GTC. The camera focuses the light on a HAWAII-2RG 2048x2048-pixel detector array from Teledyne Imaging Systems.
CIRCE Optical Layout Schematic (Figure credit: M. Edwards)

CIRCE IR Detector System

CIRCE uses a 2Kx2K HAWAII-2RG engineering grade detector, with SIDECAR ASIC cryogenic readout electronics inthe dewar. The SIDECAR interface is via an I-SDEC card mounted outside the dewar, with USB connection to the CIRCE Local Control Unit (LCU - also known as "ganesh").
The CIRCE system gain is 5.3 +- 0.5 e-/ADU .

On "clean" channels, CIRCE has ~18 e- RMS read noise for a single read. The detector dark current is very small. In-dewar background is dominated by light leaks, but these are typically <10 e-/s/pixel (<0.02 times the sky background in J-band, which is the lowest-background band).

Generally speaking, NIR detectors have MANY more cosmetic defects than their optical CCD counterparts. However, the typical dithering patterns required for sky-background subtraction in the NIR effectively eliminate the systematic aspects of this problem. CIRCE is no exception to this - it's engineering-grade H2RG detector has more cosmetic problems than typical science-grade devices, though it is much better than most other enineering-grade devices. Analysis of CIRCE commissioning data from December 2014 and March 2015 both show that typical dither patterns combined with standard NIR data analysis techniques effectively remove the impacts of the comsetic defects. If you would like more details, you can read this report developed for the GTC in February 2014 (prior to shipping CIRCE to GTC).

Back to top



CIRCE Observing Information

Imaging Mode

CIRCE's imaging mode provides a 3.4x3.4-arcminute FOV on the GTC.

Imaging Filters

Currently, CIRCE has J, H, and Ks filters installed. There is room for additional filters in the future. Due to the high sky background in the NIR, the maximum recommended exposure time in each band is as follows: Ks - 25s; H - 45s; J - 60s . Substantially shorter exposure times are generally needed to allow dithering on timescales sufficient to properly subtract the rapidly-varying NIR sky background.

The CIRCE filters were part of the Gemini NIR filter procurement (see Tokunaga, et al.; 2002, PASP, 114, 180). Transmission curve plots (provided from the Gemini NIRI instrument) are here: J, H, Ks .

Back to top



Readout Modes

CIRCE's I-SDEC and SIDECAR readout control electronics provide a broad range of readout modes. For imaging, the standard mode is Correlated Double Sampling (CDS) readout. The read noise varies over the detector, but is typically ~30-40 e- RMS for a CDS pair. For Ks-band, CIRCE reaches the background limit in ~5s exposures. For H-band and J-band, we use Fowler sampling with 2 reads for exposures of >=12s, which allows CIRCE to reach the background limit in those exposures. The CIRCE FITS file structure is closely tied to the functionality of the I-SDEC/SIDECAR readout electronics in CIRCE. The FITS files are Mult-Extension FITS (MEF) files. The exact structure of each file depends on the readout mode selected for the exposure. We briefly review the readout process here.

The CIRCE H2RG detector, like essentially all NIR arrays used for astronomy, uses a correlated sampling readout mode to correct for electronic offset drifts in the detector. The basic readout process for a single pixel is illustrated in Figure 1 below. Here, the pixel voltage is set to a "Reset" value first. The pixel voltage level is then digitized in a "first read" of the detector. Then, after the integration time, the pixel voltage level is again digitized in a "second read" of the detector. For the 2048x2048-pixel H2RG array in CIRCE, this process happens sequentially for all of the pixels.

In other words, first all pixels are set to their "Reset" level using "Fast Reset" mode (which takes about 0.04s for CIRCE). Next, the pixel values are "clocked out" to the 32 parallel amplifiers of the H2RG detector. Each amplifier channel handles (2048/32)x2048 = 131072 pixels. The standard clock speed for CIRCE readout is 100 kHz, so this process about 1.45s from start to finish (including some extra clock cycles required for line shifts, etc.). At the end of this process, the pixel values have been digitized into a 2048x2048 array of first-read values, which are handled as 16-bit unsigned integers. This is written as the first (1st) extension to the MEF file. The array controller then waits for 1 integration time (T_int) and then repeats the readout, resulting in a second 2048x2048 array of unsigned integers. This is written as the second (2nd) extension to the MEF file. Subtracting the first read from the second read (Rd2 - Rd1) gives the integrated signal during the integration time. This process is known as Correlated Double Sampling (CDS) readout. For simple CIRCE CDS readouts (and ONLY for CDS readouts!), the CIRCE software will also take the additional step of subtracting the 1st read from the second read, and write the difference as a 32-bit signed integer 2048x2048 array as the third (3rd) extension to the MEF file.

Note that this process results in a couple of important features. First of all, the minimum integration time is equal to the "frame time" (approximately 1.45s) - this is the minimum time it takes to cycle through from a given pixel in the first read, to the same pixel in the second read. Secondly, the minimum time to take an exposures is 1 reset time (which is very short for CIRCE) plus two frame times - in other words, about 3 seconds. Note, however, that in practice it takes about 13s or so from start to finish of a CIRCE image, due to overheads in setting up the file structure, configuring the array control electronics, gathering FITS header information, and writing the file to the GTC archive.

Back to top



The I-SDEC/SIDECAR system for CIRCE is a powerful and flexible tool for H2RG data acquisition, and offers many "modes" of readout. In practice, the GTC Phase 2 tool for CIRCE and the CIRCE sequence generator normally select the best mode for the user. However, the mode options have an impact on the data structure, so we describe two important options here. The first is "NREADS". This is the number of readout frames per "group" of readouts. In the example shown above, there are 2 "groups" (this is the default number for CDS and for Fowler Sampling with CIRCE - the standard modes for CIRCE), with one group being located after the reset and the other group being located at the end of the integration time. For the example above, NREADS = 1 readout for each of these groups (and thus 2 total reads). An alternate option would be NREADS = 2 for CIRCE, with 2 reads taken in each group ( a total of 4 reads) - this is shown in Figure 2. In this case the MEF file would have four (4) extensions, each with a single 2048x2048 array of 16-bit unsigned integers. The integrated signal would be (Rd4 + Rd3 - Rd2 - Rd1)/2 for NREADS=2. This is also known as "Fowler2 Sampling". Using NREADS>1 has the advantage of reducing the effective impact of detector read noise by a factor of 1/sqrt(NREADS). This can be important for relatively low-background exposures. However, it also comes at the price of taking a minimum of 4 frame times (almost 6s) to execute, even for a minimum 1.4s exposure time. For CIRCE, our experience shows that we can reach the background limit in Ks band even for short exposures. The default NREADS selection is shown in Table DDD below.

Filter T_int NREADS
J-band <12s 1
J-band 12s+ 2
H-band <12s 1
H-band 12s+ 2
Ks-band Any 1

Back to top



A final very useful option for CIRCE is "N exp" (also known as "NRAMPS" to the CIRCE team). This is the number of times to repeat the readout scheme selected. The Figure below shows "N exp"=3 for two different scenarios (NREADS=1 and NREADS=2, respectively). One can accomplish the same data acquisition as "N exp"=M simply by taking M exposures, resulting in M different FITS files, rather than a single FITS file with "N exp"=M. However, as noted above, each exposure and FITS file has a significant (approximately 10s) overhead associated to setup, etc. Thus, if you wish to take M exposures at the same position, you pay this 10s overhead penalty M times if you take M different FITS exposures. On the other hand, if you set "N exp"=M, you get one FITS file and only pay the overhead penalty once. This time savings can be quite substantial - for instance, a dither5 sequence with 3 exposures of 5s at each position in Ks band would take approximately 30s at each position using "N exp"=3. Doing the same thing with "N exp"=1 repeated 3 times at each position would take approximately 50s - nearly 70% longer in wall-clock time. Thus, the default in the CIRCE Phase 2 tool and sequence generator is to use the "N exp" parameter for repeated exposures at a given position in the dither pattern. The resulting FITS file will be a MEF with the number of extensions determined by NREADS as above, but repeated "N exp" times.

The Table below presents some example data structures for CIRCE FITS files, for reference.
NREADS "N exp" Extensions
1 1 Rd1,Rd2,(Rd2-Rd1)
2 1 Rd1, Rd2, Rd3, Rd4
1 2 Rd1, Rd2, Rd3, Rd4

Back to top



Imaging Sensitivity

The following table gives the 5-sigma limiting magnitude (Vega system) for a point source in 0.6-arcsec FWHM seeing (typical for the GTC in the NIR). This assumes a 3600s on-source exposure time, and an airmass of 1.2.
Filter
J-band 24.0 mag
H-band 23.2 mag
Ks-band 22.5 mag

Back to top



Image Quality

CIRCE image quality is essentially always seeing-limited at the GTC. During CIRCE commissioning, the measured image quality was as good as ~0.32-arcsec FWHM. Typical seeing seems to be ~0.6-0.7-arcsec FWHM.


[add some stuff about distortion and distortion correction here]

Required Calibrations

CIRCE requires two basic calibrations, which should be taken at least once per observing run: dark frames and flat fields.

The dark frames are used primarily to subtract the electronic offsets of the detector (since the dark current is essentially negligible). The electronic offsets have significant structure, and depend on the exposure time and number of reads used in the readout (which is typical of H2RG detectors). Thus, we recommend taking at least 9 "dark" images with identical exposure times and NREADS for EVERY science image exposure/nreads combination used during the run.

Flat fields are used to correct the pixel-to-pixel response of the instrument to light. For an engineering-grade array such as the on used in CIRCE, these variations are significant, and science-quality analysis requires the use of flat fields. The GTC dome lamps are too bright to be used with CIRCE broadband filters, as they cause saturation in the minimum detector exposure tim even on their lowest power settings. Thus, we recommend twilight flats be taken during the observing run.

For photometric calibration, the observer should also be sure to observe standard stars. The exact frequency and pattern of standards observation of course depends on the experimental design, as for any optical/IR observation. One important things to note for CIRCE is that CIRCE matches the 2MASS bandpass, and has almost identical filter sets. Thus, 2MASS stars often provide very good in-field calibration stars for many targets. However, it is important to remember that not all fields will have suitable 2MASS stars, and that many stars in the sky (including 2MASS) will be intrinsically variable. Furthermore, due to the high sensitivity of CIRCE/GTC, even relatively faint 2MASS stars can saturate the detector in resonably short (say 10s) exposures (!). We recommend that the observer make sure that their field contains suitable 2MASS stars for calibration, and also that they take at least one set of short dithers (say dither5) over that field with short (5s or less) exposures in each filter of scientific interest.

Back to top



CIRCE Data Reduction

CIRCE data reduction is easiest using the SuperFATBOY data reduction pipeline. SuperFATBOY is an XML-controlled GPU-enabled Python pipeline developed at the University of Florida. More details can be found at the SuperFATBOY Web Page

Back to top



CIRCE Dither Patterns

For proper subtraction of the large, rapidly varying NIR sky background, CIRCE operates using dither patterns to "nod" the telescope during data observations. The sky background is highest for CIRCE in the Ks-band (where thermal glow matters), followd by H-band, and then J-band. The longer one "dwells" at a given location, the lower the overhead time lost to dithering. However, longer dwell times also can and do lead to systematic errors in sky subtraction, due to sky variations. The optimal dwell time depends on observing conditions as well as bandpass.

Based on CIRCE commissioning observations in March 2015, we recommend dwell times of ~30s in Ks-band, and <45s in H-bnd and J-band. To avoid saturation (from sky and/or targets), these may be divided into multiple images taken at the same dwell location. Dither "nodding" amplitudes should be sufficient to cleanly separate target images. For low surface-density stellar fields, nod amplitudes of ~10-arcsec should be sufficient. For extended objects, the nod amplitude should be greater than the maximum size of the object. If the object size is sufficently large (i.e. >1-arcminute diameter), we recommend that the observer use off-source skies to obtain proper sky backgrounds.

See the GTC Phase II manual for CIRCE for details on available dither patterns.

Back to top