KPNO WHIRC INFORMATION
Introduction
The WIYN High-Resolution Infrared Camera (WHIRC) is a near-infrared (0.9 – 2.5 μm) imager which installs on the WIYN Tip/tilt module (WTTM) port. The 0.1 arcsec pixel scale and 3.3 arcmin field of view are designed to take advantage of the excellent native seeing at the WIYN site and the near diffraction-limited image quality (~ 0.2 arcsec) which WTTM is expected to deliver in the 2 micron band. A selection of wide- and narrow-band filters allow WHIRC to achieve a broad range of scientific goals in stellar physics, star-forming regions, and the ISM in galactic and extragalactic sources.
As noted in the current observing information, WHIRC is being offered for the 2008B semester in shared-risk mode in direct imaging mode only (WTTM correction not available). This will permit the community to gain experience with the instrument as the commissioning effort is completed. WHIRC has had several engineering runs on WIYN during 2007 and early 2008; the weather limited the opportunity to gain on-sky experience, except in January 2008 (when a malfunction of the filter wheel limited the operation to the J band), although significant progress was made in the integration of the observing platform software. This webpage is intended to give prospective investigators information which is available as a guide for writing proposals and will be updated as information becomes available.
Useful Facts
|
Wavelength Coverage |
900 – 2500 nm |
|
Filters |
J, H, Ks; 10 narrowband |
|
Pixel Scale |
0.1 arcsec |
|
Field of View |
200 × 200 arcsec |
|
Detector |
Raytheon Virgo HgCdTe, 2048 × 2048 |
|
Detector Gain |
~ 3.7 e/ADU (0.7v bias) |
|
Read Noise |
~ 12 ADU (47 e) |
|
Full Well |
~ 32000 ADU (118000 e) |
Instrument Description
WHIRC is a straight-through all-refractive imager with no moving parts except for the two filter wheels. This design was dictated in part by the stringent instrument envelope and weight requirements of the WTTM port. The optical system consists of a five-element collimator and a five-element camera. A fixed cold stop is located at the pupil image formed by the collimator. The two filter wheels are located on either side of this stop, placing the filters very close to the pupil image. A single LN2 reservoir provides cooling for the optics and the HgCdTe detector, whose temperature is regulated by a servo control loop. The detector controller is the Monsoon system developed at NOAO. Figures 1 and 2 show schematics of the instrument and a closeup of the optical assembly.
Figure 1: Assembly drawing of WHIRC. The shim is used to adjust the axial location
of WHIRC so that the WTTM focal plane is imaged onto the detector.
Figure 2: Close up of Figure 1, showing the optical
elements in more detail.
Filters
The two 8-position filter wheels allow a total of 13 filters. Each filter wheel must have one open position and one wheel has an opaque blocker for taking dark frames. Table 1 lists the filter characteristics, as well as the observed signal in ADU/s corresponding to a mag=10.0 star determined from observations of the IR standard FS 23 at a bias of 0.8 v. Signal levels at the bias level of 0.7 v, which we are recommending for 2008A operation, should be similar. The sky background levels in ADU/s-pixel were measured in March 2008 at an ambient temperature of ~ 10 C. The background in the K band filters is dominated by thermal emission and can be expected to vary at least a factor of two on either side of the values in the table, depending on the season. Links to tracings of the individual filters are given at the bottom of this web page.
Note: The actual WHIRC filters were scanned by the vendor only at ambient temperature. Instead, standard size (25 mm) witness samples which were coated during the filter run were scanned at both ambient and cryogenic temperatures, and the difference between them was used to correct the ambient WHIRC filter parameters to their calculated cryogenic values. Because broadband filters often have oscillatory behavior in their transmission curves and narrowband filters of 1% fractional bandwidth rarely have a truly flat region at their peak transmission and may have broad wings, the definition of “average” transmission can be a matter of judgment. For the purpose of this table, we calculated the integrated transmission under the ambient filter curves, then divided by the vendor-calculated cryogenic FWHM of the filter to derive the number listed as average transmission. Since the product of these is actually used in throughput calculations this is a somewhat artificial definition, but the FWHM is an important parameter, particularly for the study of high-redshift emission line targets.
The three broadband filters are standard J, H, and Ks filters. The narrowband filters include those for He I (H II regions, PNe), Br γ and Pa β (ionized gas), [Fe II] (photodissociation regions and PNe), H2 S(1) (shocked molecular gas), and CO (cool stellar atmospheres). In addition, Br γ, Pa β and [Fe II] filters redshifted by ~ 4500 km/s are used to provide continuum images for emission line imaging in those filters or for observing these emission lines in redshifted galaxies. Finally a filter near 1.06 μm is located in a region nearly devoid of telluric OH line emission for very low-background deep imaging.
Table 1. WHIRC Filter Characteristics
|
Filter |
λ(μm) |
Δλ (μm) |
tavg |
Signal 10.0 mag |
Background |
|
J |
1.250 |
0.162 |
0.913 |
177000 |
5.5 |
|
H |
1.651 |
0.310 |
0.867 |
188000 |
25.3 |
|
Ks |
2.168 |
0.343 |
0.877 |
108000 |
73.7 |
|
Low airglow |
1.060 |
0.0132 |
0.638 |
13200 |
0.45 |
|
He I |
1.082 |
0.0094 |
0.706 |
9350 |
0.53 |
|
Pa β |
1.280 |
0.0158 |
0.872 |
17000 |
~1.0 |
|
Pa β (4500 km/s) |
1.303 |
0.0133 |
0.863 |
14300 |
1.07 |
|
[Fe II] |
1.646 |
0.0164 |
0.791 |
10700 |
2.03 |
|
[Fe II] (4500 km/s) |
1.668 |
0.0162 |
0.917 |
11500 |
2.86 |
|
H2 S(1) |
2.117 |
0.0216 |
0.680 |
7150 |
2.90 |
|
Br γ |
2.162 |
0.0215 |
0.849 |
7730 |
3.76 |
|
Br γ (4500 km/s) |
2.188 |
0.0237 |
0.940 |
8630 |
5.70 |
|
CO |
2.293 |
0.0228 |
0.797 |
5940 |
7.80 |
Imaging Performance
Commissioning and characterization have been limited by three issues encountered during the T&E testing to date. The most annoying has been the presence of electrical pickup which completely dominates the noise performance. This not only results in a read noise well above that anticipated, but the spatial coherence leads to additional systematic errors in an aperture photometry measurement. The noise has been reduced significantly since the first tests, but is still well above that which we should expect. Turning off the instrument rotator power reduces the noise by at least a factor of two, so efforts are directed towards minimizing pickup in the detector controller power supply cabling. A second issue was intermittent operation of the filter wheel, which resulted in the January T&E run in excellent conditions being restricted to the J filter. This was traced to a cold joint in the filter wheel cable, which has now been repaired; no further problems have been seen in filter operation. The third issue was an unusual spatially dependent nonlinearity in the detector response which made it difficult to characterize the flatfielding and photometric quality possible with the instrument. Very recent work has demonstrated that lowering the detector bias from 1.0 v to 0.7 or 0.8 v results in a much more “normal” linearity performance which appears to be the same within 1 – 2 % over the array.
During the January T&E run in the J filter, images as good as 0.29 arcsec FWHM were obtained without WTTM correction on a night of excellent seeing. Optical testing using a pinhole array at the input of WTTM yields images with FWHM ~ 0.17 arcsec. This gives some confidence that WTTM/WHIRC is not likely to limit the image quality and that images in the 0.2 – 0.25 arcsec FWHM are a realistic goal once WTTM is operating. The signal levels shown in Table 1 were obtained with 0.5 arcsec FWHM images and utilized a 2 arcsec diameter extraction aperture.
Operation
Progress in integrating the operation of WHIRC with the telescope has been limited in part because of the concentration on the noise and detector issues noted above. In addition, full operation, including guided offsetting with WTTM, is proving to be a challenging task which lies somewhat outside of the original WTTM operations requirements. A number of “canned” scripts to carry out dithered offsets have been written for the Monsoon Observing Platform (MOP), which is the standard user interface. In addition, one can utilize the WHIRC Observation Manager and Planner (WHOMP) to set up specific dither/offset scripts for particular targets in advance and execute them at the telescope. Currently, these appear to operate without the ability to guide during the individual integrations, which is adequate for fairly short (< 60 s) integrations on fairly bright targets. In addition, one can guide on static integrations using the IAS guider. However, combining the ability to guide and offset using the guider is currently not working reliably and should not be considered to be available for observing programs. Faint target programs should consider utilizing long (300 s or so) individual integrations with manual offsets and guider reacquisition. This will also improve the performance given the currently high read noise.
The WHIRC detector operation uses the double-correlated sampling (DCS) technique employed with virtually all infrared arrays. This involves a reset (biasing) of the detector, followed by a non-destructive readout of the array and a second readout starting after a predetermined time interval (which is the integration time). These two readouts are then subtracted to yield a bias-subtracted image. The minimum integration time is therefore the detector readout time, which is approximately 4 s for WHIRC. This mode, also referred to as “Fowler 1” is identified as “DA4F1” on the MOP interface (the DA4 refers to the use of four digital averages, which will become a default mode for WHIRC). A second user mode, “DA4F4” reads the array out four times at the beginning and four times at the end of the integration. This effectively reduces the read noise by a factor of two and is therefore a preferred mode for low-background, faint target observations, where one would be read noise limited. For such programs, one would want to use long frame times, so the longer detector readout time (16 s) does not represent a significant duty cycle penalty. Note: Multiple-readout data are not renormalized, so values in DA4F4 will be four times those given in Table 1.
Performance
The performance of WHIRC is currently seriously limited by the high read noise resulting from the electrical pickup. In addition, the small pixel size results in a background signal, even in the broadband filters, which is about ten times smaller than that which users of typical wide-field imagers may be accustomed to. The combination of these factors means that relatively long integration times will be required to achieve background-limited performance. Until the read noise issue is solved, narrowband filter (except for CO) operation may remain read noise limited.
Current Performance
One may roughly estimate the performance using the signal and background values listed in Table 1, under the assumption that the only sources of noise are the read noise of the detector and photon statistics (shot noise) from the signal and background. For an integration time ‘t’ and a gain (e/ADU) of ‘g’, the signal from a source of magnitude ‘m’ is:
S = 10-0.4 (m -15) * Ns * t * g (e), where Ns is the signal from Table 1.
The noise is given by:
N = [ S + Nb * t * g * Npix + rn2 * Npix ] 0.5 ,
where Nb is the background level from Table 1 and Npix = 78.5 * A2 is the number of pixels within the extraction aperture of diameter A arcsec.
To do sky subtraction, at least two images of a field are required, and the subtraction of two images will effectively increase the noise by 1.414, which is the same as the signal gain from the increased target time resulting from two observations. For very faint targets, one will take multiple dithered images and generate a master sky frame which can be subtracted from each raw image. For a large number of frames, the increase in noise resulting from the sky subtraction can be quite small, and one can approach the theoretical signal-to-noise given by the ratio of S to N above. For multiple frames of a target, the S/N should increase as nf0.5, where nf is the number of frames (total integration time = t * nf).
Table 2 gives examples of this calculation for the J, H, and Ks filters, where the performance is given as the source brightness corresponding to a 10σ detection in a total of 1 hour of on-source time. We assume the use of the DA4F4 observing mode, for which the read noise is assumed to be 50 e, and an extraction aperture 1.6 arcsec diameter, which is reasonable for an image FWHM = 0.5 arcsec. WTTM correction of the image in the future should yield further improvements in point-source limits, as should reduction of the read noise. The use of the DA4F4 mode will increase the signal and noise values in ADU by a factor of four.
Note that the backgrounds in the J and H filters are close
to the “typical”
Table 2: WHIRC Imaging Signal
and Background Levels (ADU-s-1) and Estimated Performance
Nread = 50 e; aperture = 1.6 arcsec diameter
|
FILTER |
SIGNAL (10 mag) |
BACKGROUND (pixel) |
BACKGROUND (mag-arcsec-2) |
PERFORMANCE (10 σ; 1 hr) |
|
J |
1.77 × 105 |
5.5 |
16.3 |
21.8 |
|
H |
1.88 × 105 |
25.3 |
14.7 |
21.2 |
|
Ks |
1.08 × 105 |
73.7 |
12.9 |
20.1 |
Given the high read noise, these performance values are quite good, but possibly optimistic. The calculation assumes that the signal from the source was negligible in comparison to the sky background, which is reasonable for a limiting calculation. The individual integration time of 300s was chosen to achieve background-limited operation in the H and Ks bands while staying well below the saturation limit of ~ 130000 e. The J band observation is barely background limited with a 300s frame time. They also assume that sky-subtraction and flatfielding add negligible noise to the image, assumptions which have not been empirically verified. If the read noise can be reduced to ~ 20 e (which has been measured in the laboratory for DA4F4 operation), one can achieve background-limited operation in much shorter integration times in the broadband filters and in the narrowband filters (except perhaps for the low-airglow filter) in times which would be appropriate for observing faint extended targets.
Linearity
All infrared arrays utilizing a unit cell architecture are inherently nonlinear, since the potential well created by the application of the bias voltage has a capacitance which increases as the collected charge fills up the well (one may think of the two capacitor plates moving closer together). In parallel with the capacitance of the rest of the unit cell, this yields a gain which varies slowly as the well fills up. Under the condition of constant signal flux, the plot of signal vs. time would begin at a slope near unity and slowly roll off until the array saturates. Alternatively, one can define a “linearity” function, which is essentially the slope of the signal vs. time plot normalized to the value at small signal levels. A plot of the WHIRC linearity for a bias of 0.8 v is shown in Fig. 3a. Note that the linearity decreases smoothly by about 4% at a signal level of 42000 ADU, then decreases suddenly as the array saturates. One can obtain a linearity correction function by plotting the well-behaved data (below 42000 ADU) and fitting a second-order polynomial to it (Fig. 3b).
Figure 3: (left
panel): Plot of linearity vs. signal
level for a bias value of 0.8 v. The
high values at low signal levels result from the actual minimum integration
time being longer than the assumed value of 3.3 s and should be ignored. The scatter results from the sampling of six
subregions of the array. (right
panel): Second-order polynomial fit to
the linearity data below 42000 ADU.
The quadratic fit in Fig. 3b is of the form y = A + Bx + Cx2, where
A = 0.9997
B = 1.7563 × 10-7
C = -2.6111 × 10-11
By inverting this function, one can derive a linearity correction function so that the corrected signal S’ is related to the raw signal S by
S’ = S * (A + B*S + C*S2), where
A = 1.0006
B = -2.3114 × 10-7
C = 2.8093 × 10-11
for a bias value of 0.8 v. For a bias of 0.7 v, the values are
A = 1.0002
B = -6.2122 × 10-7
C = 4.3382 × 10-11
The IRAF task irlincor is specifically designed to carry out this correction. It is critical that linearity correction be performed on the raw data, prior to any sky or dark subtraction.
Saturation
Because of the fine pixel scale of WHIRC, it is tempting to think that saturation on brighter stars is not as critical as for a wide-field imager with larger pixels, but the combination of a 3.5-m aperture, a 4 s minimum integration time, and good image quality does mean that one must consider this issue. If WHIRC attains its goal performance in conjunction with WTTM, it will yield image cores with FWHM < 3 pixels, similar to what one obtains with FLAMINGOS on the 4-m under median seeing conditions.
The data presented in Table 1 were obtained under seeing conditions ~ 0.5 arcsec (5 pixels FWHM). Empirically, the peak pixel flux was typically 0.03 of the integrated flux within the 1.6 arcsec diameter aperture. For a H = 10.0 star in the minimum integration time of 4 s, this yields a peak pixel signal of ~ 23000 ADU, seemingly at a safe level of slightly over half full well. However, seeing fluctuations over these short times could result in “good” images which push the peak pixel close to saturation.
Finally note that we are recommending a bias value of 0.7 v for 2008A operation, since the lower bias appears to give fewer “maverick” pixels. However, the saturation level is reduced to ~ 32000 ADU.
Flatfielding
Unfortunately, the quantitative performance of image flatfielding has not yet been established because of the above-mentioned work on higher priority issues such as the noise, filter wheel operation, and linearity. However, the recent linearity experiments which support the use of bias voltages in the 0.7 – 0.8 v range, as opposed to the 1.0v bias used previously, also lead us to believe that flatfielding should also work much better with the lower bias values.
In Fig. 4 we show the results of ratioing two flatfield images taken with integration times of 10.0 and 4.2 s, respectively, for bias voltages of 1.0 v and 0.7 v. In a sense, this gives a feeling for the precision which one might expect from flatfielding astronomical images. The two images are dramatically different. The 1.0 v bias ratio shows significant spatial variation which demonstrates a linearity relationship between signal and integration time which varies across the array, a situation which would not give confidence in the ability to achieve accurate photometry. The 0.7 v bias image, on the other hand, is spatially flat (even over a much smaller lookup table range) and centered near a value of 2.38, the ratio of the integration times.
Figure 4: Ratio of flatfields at integration times of
10.0 and 4.2 s, for bias voltages of 1.0v (left panel) and 0.7v (right
panel). The lower bias voltage yields a
much more linear response, both temporally and spatially. The noisy-looking columns on the right side
of the array are reference pixels and not part of the actual image.
The flatfielding performance is illustrated in another way in Fig. 5, in which flatfield ratios at several integration times (10.0 s and 4.2 s for 1.0, 0.7. and 0.5 v bias; 40.0s, 20.0s, 12.0 s vs 5.0 s for 0.8 v bias) are sampled within 100 × 100 subregions (the same as used for the linearity plots in Fig. 3) and compared to the expected ratio of the integration times. As one might expect from Fig. 4, the values for 1.0 v bias vary wildly, by more than 25%. On the other hand, those for the lower bias values seem to be within 2% of each other. The data points for the 40/5s ratio for 0.8v bias are a couple of percent below the expected value of 8.0, reflecting the nonlinear behavior of the array. However, as shown in the right panel of Fig. 5, applying the linearity correction function above to the 0.8v bias data results in values very close to 1.00, with a p-p spread on the order of 1%. This gives some confidence that we should be able to achieve flatfield corrections to this accuracy after some additional experience.
Figure 5: (left panel):
Plot of the mean values of ratio images for various flatfield
integration times at different bias voltages compared to the expected value of
the ratio of the integration times. Six
subregions within each array are plotted.
(right panel): Same plot, except
for the 0.8v bias results only, after the application of the linearity
correction function in the IRAF task irlincor.
Pupil Ghosting
One feature which has not yet been quantitatively investigated is a pupil ghost in the flatfield images. This is a characteristic which is found in many refractive imaging instruments, due in part to reflections from the detector and the optical surfaces. The effect can be seen at all wavelengths (Fig. 6), although it is exacerbated at thermal wavelengths (K band) in the case of WHIRC by the large number of ambient temperature mirrors (9) in the optical train. The pupil ghost manifests as a broad peak centered on the array, with a peak value about 5% above average at J, but almost 25% at Ks. This does NOT represent a real peak in the response, so the use of a flatfield constructed from raw sky or lamp flats will give artificially low flux values for targets near the center of the array. We are working on establishing a low spatial frequency “pupil flat” which would be applied to observed flats to yield a corrected flat for use in the data reduction.
Figure 6: Flatfields through the J (left panel) and Ks
(right panel) filters at a bias of 0.8 v.
The decreased response at the top and bottom of the arrays appears to be
a property of the system, but appears to calibrate out (Fig. 5). The pupil ghost, visible as a bright spot in
the center of the array, is not a true peak in the response, and must be
calibrated out of the flats.
Additional Information
More detailed technical information, such as the filter characteristics and detector gain analysis, can be obtained from the following links.
Filter Curves
Because the WHIRC filters are not the “standard” size, the vendor did not obtain cryogenic transmission information on the actual filters in the instrument. Instead, the vendor measured both ambient (300 K) and cold (77 K) transmission of a 25 mm diameter witness sample (which fits into their variable-temperature filter holder) and used the shift in the transmission characteristics with temperature to calibrate the 300 K transmission curve of the actual filter. These numbers are center wavelength and bandpass presented in Table 1.
The tracings on the links are the 300 K transmission plots of the actual WHIRC filters.
![[Fe II] filter](FeII.JPG){kind=link}
![[Fe II] (4500 km/s) filter](FeII_cont.JPG){kind=link}
Mean-variance Plot
The mean-variance (photon transfer) measurement is used to determine both the read noise and gain (e/ADU) of the detector. This analysis assumes the only sources of noise to be the signal-independent read noise and the shot noise from the signal itself. When the variance of the signal is plotted against the mean value on a log-log plot, the result is a line of unity slope in the photon-noise-limited regime which intercepts the X-axis at a value equal to the gain. The Y-intercept of the actual data is the variance of the read noise.
WHIRC Mean-variance Plot1 (21 August 2007; Vbias = 1.0v; global reset--obsolete)
WHIRC Mean-variance Plot2 (18 March 2008; Vbias = 0.7v; row reset)
Manuals
The most current version of the User’s Manual can be downloaded from the following link. Additionally, one may download a short summary which includes lists of the WHIRC characteristics and performance, as well as a short summary of the startup and shutdown procedures.
Updated 21 April 2008
NOAO Contact: joyce@noao.edu