arX

iv:a

stro

-ph/

0102

219v

1 1

3 Fe

b 20

01

An Infrared Camera for Leuschner Observatory

and the Berkeley Undergraduate Astronomy Lab

James R. Graham and Richard R. Treffers

Astronomy Department, University of California, Berkeley, CA 94720-3411

jrg, rtreffers@astron.berkeley.edu

ABSTRACT

We describe the design, fabrication, and operation of an infrared camera which is in use atthe 30-inch telescope of the Leuschner Observatory. The camera is based on a Rockwell PICNIC256×256 pixel HgCdTe array, which is sensitive from 0.9-2.5 µm. The primary purpose of thistelescope is for undergraduate instruction. The cost of the camera has been minimized by usingcommercial parts whereever practical. The camera optics are based on a modified Offner relaywhich forms a cold pupil where stray thermal radiation from the telescope is baffled. A cold,six-position filter wheel is driven by a cryogenic stepper motor, thus avoiding any mechanical feedthroughs. The array control and readout electronics are based on standard PC cards; the onlycustom component is a simple interface card which buffers the clocks and amplifies the analogsignals from the array.

Submitted to Publications of the Astronomical Society of the Pacific: 2001 Jan 10, Accepted 2001

Jan 19

Subject headings: instrumentation: detectors, photometers

1. Introduction

Many undergraduates enter universities andcolleges with the ambition of pursing several sci-ence courses, or even a science major. Yet sig-nificant numbers take only the minimum require-ments because science is often perceived as labo-rious, demanding, and irrelevant. The result isgraduates whose lives and careers as citizens arenot enriched by an understanding of science ortechnology. Since the Astronomy Department isfrequently an undergraduate’s only contact withscience we have a tremendous responsibility asa provider of undergraduate science instruction.Unfortunately, the students who are captivatedby introductory astronomy often conclude thatthe astronomy major consists of narrow technicalcourses designed only for those intent on pursu-ing an academic career. These classes often donot stress problem solving, critical thinking noremphasize the connections between broad areas ofscientific knowledge. The Berkeley undergraduate

astronomy labs offer a potent antidote to conven-tional lectures, and stimulate students to pursuethe astronomy major. We describe a state-of-the-art infrared camera which is used at Berkeley toinspire and instruct.

Prior to the construction of this camera theundergraduate lab syllabus consisted of an opti-cal and a radio class. The labs use theodolites, aCCD-equipped robotic observatory (Treffers et al.1992), and radio telescopes to provide the astro-nomical context to discuss the statistical descrip-tion of experimental data, including the notions oferrors, error propagation, and signal-to-noise ra-tio. These resources also provide students witha practical forum in which to explore signal pro-cessing techniques including power spectrum es-timation and convolution. The optical labs aretraditionally offered in the Fall semester. TheSpring climate in Northern California is unsuit-able for optical observations, so a radio-astronomylab based on a roof-top 21-cm horn antenna and a

1

two-element 12 GHz satellite-dish interferometeris offered (Parthasarathy et al. 1998).

These experiments have significant computa-tional demands that involve least square fitting,precession, coordinate conversion, Doppler correc-tion, Fourier transforms, and image processing.Students use a cluster of Sun Ultra workstationsand the IDL programming environment for datareduction and analysis. These machines provide,for example, the data storage and computationalpower required for four to five groups of studentsworking simultaneously to turn the visibility datafrom the interferometer into radio maps.

1.1. Development of an infrared under-

graduate laboratory

To maintain the intellectual vitality of the un-dergraduate lab, broaden the range of observingexperiences we offer, and satisfy the demand forincreased enrollment we have developed a new labprogram based on the theme of infrared astron-omy. Although radio receivers are commonplace,the notion of using radio technology for astron-omy remains exotic. Radio astronomy continuesto exemplify astronomy of the “invisible universe”,and detecting astronomical radio signals continuesto intrigue and captivate students. Perhaps, be-cause our eyes give us an intuitive sense of whatit means to “see” at visible wavelengths, and be-cause CCDs have become commonplace in prod-ucts such as digital cameras, the high-tech aspectof the optical lab has lost some of its gloss. Mod-ern CCDs have excellent cosmetic quality, andstudents report that calibration of CCD imagesappears to consist of annoying minor corrections.Faint stars and galaxies are readily observable inraw CCD images. In contrast, at infrared wave-lengths the brightness of the sky (e.g., 13 mag.per square arcsec at K-band), compounded bydetector flaws and artifacts, means that it is im-possible to detect interesting sources without sky-subtraction and flat-fielding. Digging the infraredsignal of optically-invisible dust-enshrouded youngstars out of noise can be presented as a challengingand worthwhile goal (cf. §3.4).

Thus, infrared astronomy is a natural nextstep for an undergraduate observatory that isequipped with optical and radio facilities. More-over, infrared observations are becoming famil-iar to students from introductory undergraduate

text books which rely on observations from IRAS,ISO, and and the NICMOS camera on the Hub-

ble Space Telescope to describe astrophysical phe-nomena such as star formation. We can expectinterest to soar as SOFIA and SIRTF becomeoperational in the next few years. Infrared ob-servations provide a unique context to discuss abroad range of physical phenomena and astrophys-ical processes: black-body radiation, solid statephysics and the operation of photon detectors,telescope optics, and absorption and emission bydust and molecules.

Because of the complexities associated withcryogenic optics needed for operation at wave-lengths longer than about 1.6 µm, a prototypecamera with a fixed H-filter located immediatelyin front of the focal plane array was fielded forinstructional purposes in the Fall of 1999. Thisversion of the camera provided the opportunity toverify the operation of the array readout electron-ics. A Mark II camera with cold re-imaging opticsand a cryogenic filter wheel was constructed dur-ing Summer 2000, and used on a class conductedin the Fall 2000 semester.

2. Design of the Infrared Camera

Given the restricted budgets available for un-dergraduate instruction the single most impor-tant factor influencing the design of the camerais inevitably cost. Traditionally, infrared camerasrequire expensive custom components—cryogenicvacuum vessels, cryogenic mechanisms, optics, andarray control and readout electronics. We havecontrolled costs by adopting a design which reliesalmost entirely on commercial parts.

The heart of the camera is the infrared sensor.Due to the generous support of the National Sci-ence Foundation and provision of matching fundsby Rockwell International we were able to procurean engineering grade 256 × 256 HgCdTe PICNICinfrared focal plane array (Vural et al. 1999). ThePICNIC array is a hybrid device with four inde-pendent quadrant outputs. This hybrid device issimilar to Rockwell’s familiar NICMOS3 array; ithas identical unit cell size, number of outputs andgeneral architecture, and uses a modified multi-plexer design and better fabrication techniques toimprove the noise and amplifier glow. The PIC-NIC array detector material is HgCdTe with a

2

band-gap corresponding to 2.5 µm. The require-ment for moderate dark current (< 102 e− s−1)necessitates operation below 100 K. A cold envi-ronment also is required to minimize the thermalradiation from room-temperature surfaces reach-ing the detector. A simple calculation shows thatformation of a cold pupil to match the acceptanceF-cone of the camera to the telescope is necessaryif useful operation in K-band is expected from theinstrument. Cold baffles alone provide an inef-fective and impractical method of controlling thisthermal background.

2.1. Optical design

The camera is designed to operate at theCassegrain focus of the 30-inch telescope ofLeuschner Observatory, which is located 10 mileseast of the Berkeley Campus in Lafayette. The30-inch is a Ritchey-Chretien with a nominal F/8final focal ratio, yielding a scale of 33.8 arc sec-onds per mm. The seeing at Leuschner is typically2-3 arc seconds FWHM at visible wavelengths. Atthis scale the 40 µm PICNIC pixels project to 1.35arc seconds on the sky.

Direct imaging is a simple yet natural solutionwhich takes advantage of the large corrected fielddelivered by the Ritchey-Chretien design. Unfor-tunately, direct imaging is an unacceptable solu-tion because of the detector’s sensitivity to ther-mal radiation. Consider installing the PICNIC ar-ray in a vacuum vessel typical of CCD Dewars.Suppose, optimistically, that the cold, dark inte-rior of the Dewar restricts the solid angle of ambi-ent radiation illuminating the detector to an F/3cone. The black-body radiation from the instru-ment flange and primary mirror support structureat 293 K in this solid angle in an H-band filter(1.5-1.8 µm) is 2500 photons s−1 per pixel. Thisshould be compared with the sky signal of 6000 s−1

per pixel (for H = 14 mag. per square arc second)and the detector dark current of a few electronsper second per pixel. At Ks (2.0-2.3 µm) the un-wanted thermal radiation is two orders of magni-tude greater (4.5×105 photons s−1 per pixel) thanit is at H . In Fall 1999 we deployed a prototypecamera to test the array read-out electronics, vac-uum Dewar, and the closed cycle He refrigerator.To speed the development we eliminated all opti-cal components apart from a fixed filter that waslocated directly in front of the focal plane array.

Stray radiation was controlled only by a cold baf-fle tube. This configuration, as predicted, sufferedfrom high background to the extent that it wasunusable in the Ks band because the detector sat-urated in the minimum integration time (0.66 s).Both theory and practice show that formation ofa cold pupil within the Dewar is an essential tosuccessful IR operation.

M1

M2C

B

A

Fig. 1.— The off-axis Offner system relays the imageat A to B with unit magnification. The concave mirrorM1 and the convex mirror M2 are concentric aboutthe point C. M2 is placed half way between M1 andthe focal plane AB to give a zero Petzval sum, i.e.,R1 = −2R2.

An infrared camera, known as KCAM, whichwas built with similar instructional objectives, isin use at UCLA (Nelson et al. 1997). This camerauses a ZnSe singlet to relay the image with unit-magnification from a 24-inch F/16 telescope ontoa NICMOS-3 IR focal plane array. Simultane-ously this lens forms a real image of the telescopepupil where stray light can be controlled by a coldstop. A refractive design combines the advantagesof simple assembly and alignment. The high re-fractive index of ZnSe (n = 2.43 at 2 µm) per-mits effective control of spherical aberration, whilethe relative slowness of the UCLA system meansthat off-axis aberrations are negligible. ZnSe ismoderately dispersive, which means that refocus-ing between J , H , and K is necessary—this is notan issue for KCAM which has a fixed filter. Ourfaster (F/8) configuration together with our desireto eliminate the need for refocusing forced us toconsider doublet designs. However, we were un-able to achieve good correction for geometric andchromatic aberrations over a 6 arc minute fieldwithout resorting to custom lenses, exotic salts orglasses, and expensive anti-reflection coatings (thereflection loss for uncoated ZnSe is 32%.)

3

Fig. 2.— Spot diagrams for an F/8 Offner relay. Thenine panels correspond to the center, edges and cornersof the PICNIC array. Each box is the size of a singlePICNIC pixel (40 µm) and is labeled with its coordi-nates relative to the center of curvature of M1 and M2

(point C in Fig. 1.) Dimensions are in mm. The geo-metric image size is limited by fifth-order astigmatism.

We therefore decided to evaluate the potentialof a reflective design. The telecentric, unit magni-fication Offner relay (Offner 1975) has been usedwith success in several infrared instruments (e.g.,Murphy et al. 1995). The Offner relay consists oftwo concentric spherical mirrors and is free of allthird-order aberrations. This geometry has tworeal conjugates in the plane which passes throughthe common center of curvature and is perpen-dicular to the axis joining the vertices of the twomirrors (Fig. 1). The first mirror, M1, is concavewhile the secondary, M2, is convex, with a radiusof curvature, R2 = −R1/2. The Petzval sum ofthe system is zero, because the combined power ofthe two reflections at M1 is equal and opposite tothat ofM2. BecauseM1 andM2 are concentric, anobject placed at the common center of curvaturewill be imaged onto itself with no spherical aber-ration. Vignetting limits the practical applicationof the Offner configuration on-axis. However, vi-gnetting drops to zero for off-axis distances equalto or greater than the diameter of M2. Coma anddistortion, aberrations that depend on odd powersof the ray height, introduced by the first reflectionat M1, are cancelled on the second reflection atthis surface because of the symmetry about thestop formed by M2. Fifth-order astigmatism lim-its the numerical aperture at which good imagery

is obtained. M2 is located at the mid-point be-tween M1 and its center of curvature; it thereforeis located at an image of the telescope pupil. Thispupil image suffers spherical aberration. Nonethe-less, the image quality is good enough so that acold stop at M2 suppresses light that does not em-anate from the telescope pupil and unwanted ther-mal emission from the telescope structure. Thestop will also act as a baffle to reduce scatteredlight at all wavelengths and should make it easierto measure accurate flat fields.

Fig. 3.— Spot diagrams for an F/8 corrected Offnerrelay. Layout is as in Fig. 2. Increasing the radius ofM2, while keeping it concentric with M1, introducesenough third-order astigmatism to cancel fifth-orderastigmatism at the central field point.

Fig. 2 shows the imaging performance of a clas-sical F/8 Offner based on a 6-inch (152.4 mm)focal length spherical primary (PN K32-836, Ed-mund Scientific) and a 76.2 mm separation be-tween the image and object. Clearly the perfor-mance is very good, with a central field point ge-ometric size of 5 µm rms. The worst field point isthe bottom right corner of the focal plane (-5.12mm, -43.22 mm) with 8 µm rms spot size.

Our method for fabricating the secondary mir-ror is to select a commercial plano-convex BK7lens and aluminize its curved surface. Choos-ing M2 from a catalog makes it likely that exactachievement of R2 = −R1/2 is not possible. Se-lecting M2 with a slightly larger radius of curva-ture is advantageous. Increasing the radius of M2,while keeping it concentric with M1 introducesenough third-order astigmatism to cancel fifth-order astigmatism at any given field point. This

4

Fig. 4.— Spot diagrams for the modified F/8 Offnerbased on 6-inch (152.4 mm) focal length spherical pri-mary and a R2 = 155.04 mm secondary. Layout is asin Fig. 2. This design has a slight excess of third-orderastigmatism, but perfectly adequate and uniform per-formance of 5 µm rms spots over the entire field.

cancellation, which occurs with a R2 = 153.6 mm,is shown in Fig. 3. The closest readily avail-able match plano-convex lens (R2 = 155.04 mm;PN K45-284, Edmund Scientific) has slight excessthird-order astigmatism, but perfectly adequateand uniform performance of 5 µm rms spots (Fig.4) over the entire field.

The alignment of the Offner relay presents noserious challenges despite being an off-axis config-uration. The performance is robust against decen-ter of the secondary, which can be displaced by upto 4 mm before the geometric spot size exceedsone 40 µm pixel. Secondary tilt is more critical;an error of 1 degree in tilt of M2 about its vertexswells the size to an rms of 11 µm and the spotsfill one pixel (Fig. 5). However, a secondary tiltof this amplitude also introduces 5.2 mm of imagemotion. Optical alignment, for example, can beachieved using a reflective reticle placed at point Bin Fig. 1 and an alignment telescope which viewspoint A. In practice, two alignment jigs were con-structed which fit over the ends of the snouts onthe optical baffle can (see Fig. 6). Each jig has asmall central hole which coincides with the focalplane. One jig is illuminated while its image isinspected at the other jig with a jeweller’s loupe.The tip and tilt of the primary mirror are then ad-justed using three alignment screws in the base ofthe mirror cell until the two jigs appear coincident.

Fig. 5.— The Offner design is robust against mis-alignments. The spot diagrams are for the modifiedF/8 Offner in Fig. 4, but with a 1 degree tilt of thesecondary about its vertex. The rms spot size is 11µm. A tilt of this amplitude introduces 5.2 mm ofimage motion which is easily detected.

2.2. Mechanical design

An overview of the layout of the Dewar is shownin Fig. 8 and an picture of the camera on thetelescope is shown in Fig. 9. The Dewar bodyis a semi-custom 8-inch diameter × 15-inch longcylindrical 304 stainless steel vacuum vessel (MDCVacuum Products Corp., Hayward, CA) with two4-inch ports. Both ends of the cylinder and thetwo 4-inch ports are fitted with ISO LF flange fit-tings. The top plate accommodates the quartzDewar window and two hermetic MIL connectors(Deteronics, South El Monte, CA); one for arraycontrol and read-out and one for the stepper mo-tor and temperature sensors. The chip carrier andstepper motor/filter wheel assembly are anchoredto the top-end plate by G-10 fiberglass tabs (Fig6). G-10 provides an extremely stiff and low ther-mal conductivity (0.53 W m−1 K−1) support. Thebottom plate provides the support for the primarymirror of the Offner relay (M1). Mechanical rigid-ity for the primary mirror cell is provided by fourfiberglass tabs (Fig. 7). A cylindrical optical bafflecan (Fig 6) doubles as an radiation shield with in-put and output snouts that mate with correspond-ing tubes on the filter wheel assembly and the focalplane array housing (see Fig. 8). The surfaces ofthe optical baffle can and other components adja-cent to the optical beam, such as the filter wheeland array housing, are anodized black to suppress

5

Fig. 6.— The Dewar and internal components. TheDewar consists of a stainless steel cylindrical vacuumvessel and top and bottom plates. The Dewar topplate carries the PICNIC array holder, the warm elec-tronics interface box, and the stepper motor/filterwheel assembly. Both cold assemblies are attached tothe top plate using G-10 fiberglass tabs. The quartzwindow, and the hermetic MIL connectors for arraycontrol, stepper motor, and temperature sensors arealso on this plate. The bottom plate carries the pri-mary mirror cell (supported on G-10 tabs) and theoptical baffle can. The main baffle can also providesthe mechanical support for the secondary and doublesas a radiation shield. When assembled the exterior ofoptics baffle can is wrapped with aluminized mylar.

internal reflections. The outer surface of the op-tical baffle can is wrapped with a single layer ofaluminized mylar to enhance its performance as aradiation shield. The optical baffle can attaches tothe base of the mirror cell, and thereby providesthe alignment and mechanical support for the sec-ondary mirror. One of the two 4-inch flanges isused as an access port for the model 21 CTI Herefrigerator, while the other carries a vacuum valveand pumping manifold. During final assembly thisport also provides access to the interior of the cam-era.

2.2.1. Cryogenic filter wheel

The camera is equipped with a cold filter wheelthat can hold up to six one-inch filters. The mo-

Fig. 7.— The primary mirror of the Offner relay inits cell. This F/1 spherical mirror is formed on a 6-inch diameter Pyrex 7740 blank. The mirror cell isattached to the Dewar bottom plate by four G-10 tabswhich provide a rigid, low thermal conductivity linkto the Dewar bottom. Three screws at the base of themirror cell provide tip-tilt adjustment.

ment of inertia of the filter wheel is low and the re-quired precision is undemanding so the filter wheelis driven directly by a stepper motor with no gearreduction.

Cryogenic mechanisms are frequently the mostunreliable part of infrared instruments. Thesemechanisms are usually actuated by a warm exter-nal motor via ferro-fluidic feed-throughs. A cryo-genic motor simplifies design and assembly of theDewar by eliminating the feed-through. Commer-cial vacuum-cryogenic stepper motors are avail-able, but they are more than an order of mag-nitude more expensive than conventional motors.

Fig. 8.— A drawing of the infrared camera Dewar,optics, cold head, and mechanisms.

6

Fig. 9.— The infrared camera mounted at theCassegrain focus of the 30-inch telescope at LeuschnerObservatory. The mechanical interface to theCassegrain flange is via a mounting box which accom-modates a computer-controlled flip mirror. The flipmirror directs the light to an eyepiece or a CCD cam-era; when swung out of the way the infrared cameraviews the secondary directly.

We have followed the approach developed forthe GEMINI twin-channel IR camera at LickObservatory which uses regular stepper motorswith specially prepared dry-lubricated bearings(McLean et al. 1994). The selected motor (PNVexta PX244-03AA, Oriental Motor, Torrance,CA) is two-phase, 1.8 degree per step, and rated12VDC at 0.4A. However, we have found that dis-assembly and degreasing of the motor bearings inisopropyl alcohol, omitting the time consumingstep of burnishing with MoS2, is satisfactory. Themotor home position is defined optically using anLED and photodiode.

2.2.2. Cryogenic design

A typical H-band sky brightness of 14 mag persq arc second yields about 6000 photons per sec-ond per 40 µm pixel on the 30-inch telescope.Since we would like to retain the option of us-ing narrow-band (1%) filters for imaging emis-sion lines such as H2 1-0S(1) or Brγ, our goalis to ensure that the thermal emission from theOffner relay optics and the detector dark currentis kept well below this level. Thus, if feasible wewould like to reduce any non-astronomical signalto < 102 photons s−1 pix−1. Cooling the Offneroptics by 70 K from 290 K is sufficient to reducethe λ < 2.5 µm thermal emission by three ordersof magnitude, so that the background level is re-duced to a few hundred photons per second perpixel. Reducing the dark current to 100 e− s−1

requires cooling the detector below 100 K.

The radiation load from the 290 K inner walls ofthe vacuum vessel amounts to 5 W (for clean stain-less steel walls with an emissivity of 0.05). Mul-tistage thermoelectric coolers can maintain tem-perature differences between hot and cold sides ofas much as 150 K, but their efficiency is typically< 10−3. Although thermoelectric coolers are prac-tical for CCD cameras which operate at smallertemperature differences, they are not suitable forthis application. Liquid nitrogen is readily avail-able; its 77 K boiling point and large latent heatof vaporization (161 J ml−1) render it an almostideal solution. A cryogen vessel containing, say,5 liters would be a practical volume, given ourdesign. However, this would require filling onceevery two days. Installation of a low-emissivityfloating radiation shield could increase this inter-val by a factor of two. Even with this extended life-time the use of liquid cryogen is incompatible withour objective of operating the infrared camera re-motely from Berkeley. Our dilemma was solvedwhen a Model 21 closed cycle helium refrigerator(CTI Corp.) was salvaged from the Berkeley Ra-dio Astronomy Lab’s 85-ft telescope. These coldheads are used extensively by the semiconductorfabrication industry in cryopumps and are read-ily available as second hand or refurbished items.The high pressure He gas for the CTI cold head issupplied by a compressor manufactured by AustinScientific, Austin, TX. We use only the 77 K sta-tion for cooling; the 20 K station has a vessel con-taining activated charcoal embedded in epoxy for

7

cryopumping.

On cool down from room temperature the PIC-NIC array reaches its initial operating tempera-ture of 100 K after about three hours. At thistemperature most of the functions of the cameracan be checked out. After 12 hours the tempera-ture of the optics and detector are sufficiently sta-ble for reliable operation. In this state the 77 Kstage of the Model 21 cold head is at 66 K, the fo-cal plane array is at 82 K, the optics are at 155 K,and the filter wheel is at 187 K. Over the courseof a night the temperature stability of the array is80 mK rms. The Dewar temperatures, includingthe focal plane array, show slow diurnal variationswith an amplitude of 1-2 K which are driven byvariations in the ambient temperature. These fluc-tuations will cause significant dark current varia-tions from night to night. To guard against thisthe detector block is equipped with a heater re-sistor that can be used in a temperature controlloop. To date, nightly dark frames have provenadequate and the temperature control servo hasnot be exercised. The closed-cycle refrigeratorhas one additional advantage related to the factthat the refrigerator can run unattended for manyweeks; the delicate detector and optical have suf-fered fewer thermal cycles, and consequently lessthermally induced stress due to CTE mis-matches,than in a conventional liquid cryogen system.

2.3. Array control and readout

The PICNIC array is read out by clocking itsmultiplexer so that each pixel is sequentially con-nected to one of the four output amplifiers. Whenthe pixel is accessed the charge is measured butnot removed—a reset line must be clocked to resetthe pixel and clear the accumulated charge. Con-sequently, the charge can be read multiple timesto eliminate kTC noise and improve the precisionof the measurement. Since the camera does nothave a mechanical shutter, timed exposures aremade by resetting the chip, waiting for the de-sired exposure time, then reading the chip again.When no exposure is underway the chip is con-stantly clocked with a reset waveform to flush outthe accumulation of charge.

The camera is controlled by commercially avail-able cards that are installed in a Intel/Pentium-based computer running Linux 2.0.34 which ismounted on the side of the telescope about 1 me-

ter from the camera. This machine communicatesto the outside world strictly through ethernet. Alevel shifter and interconnect box is mounted onthe vacuum Dewar (Fig. 10).

Fig. 10.— The warm electronics interface box mountson the top plate of the Dewar. The array lines forclocks, biases, and quadrant outputs pass through aMIL-style connector which is mounted directly on thecircuit card. The clocks from the AWFG card (§2.3)are buffered through a CMOS line driver. Each quad-rant’s output channel has a low noise JFET pre-amp(LF 356) and a sample and hold (S&H) circuit (LF398).

The PICNIC array requires six CMOS-levelclocks and two 5 V power supplies (one analog andone digital). Each quadrant has two shift registersfor addressing pixels in the array—one horizontalregister and one vertical register. Each registerrequires two clocks. To obtain a raster scan out-put, the horizontal register is clocked in the fastdirection and the vertical register is clocked in theslow direction. PIXEL and LSYNC are the tworequired clocks for the horizontal register. ThePIXEL input clock is a dual edge triggered clockthat will increment the selected column on bothedges (odd columns selected on positive edges andeven columns selected on negative edges). TheLSYNC clock is an active low input clock whichwill set a “0” in the first latch and a “1” in theremaining latches of the shift register, thereby ini-tializing the shift register to select the first columnin the quadrant. Since this is asynchronous to thePIXEL clock, LSYNC should be pulsed low priorto initiation of the first PIXEL clock edge. Thehorizontal register selects which column bus willbe connected to the output amplifier.

These clock sequences are generated by an arbi-

8

trary waveform generator card (PN PCIP-AWFG,Keithley Instruments). This card clocks out an 8-bit wide TTL-compatible word with a maximumpattern length of 32K. These words are convertedto the 0-5 V levels suitable for driving the PIC-NIC readout multiplexer and reset lines by a highspeed CMOS buffer/line driver (74HCT244). Thefour detector outputs (one for each 128 × 128quadrant) pass through warm monolithic JFET-input operational amplifiers (PN LF356, NationalSemiconductor) with a gain of 5.1 and a man-ually adjustable offset, a monolithic sample andhold circuit (PN LF398, National Semiconductor)to a 16 bit analog-to-digital converter card (PNDAS1602/16, Computerboards). The electronicsinterface box is shown in Fig. 10. This ADC hasa 10 µs conversion time so the entire chip can beread in 660 ms. The data from this card passthrough a very large FIFO (Mega Fifo, Comput-erboards) so that the data do not have to be readin real time.

The array controller supports several arrayoperations: 1) read—reads chip with no reset;2) reset—resets the chip but does not read; 3)frame—reset, expose, read; 4) correlated doublesampling—resets and read, expose, read; 5) Fowlersampling—resets, read, expose, read. Here exposemeans wait the requested duration. When a se-quence involves two reads the difference of thesecond minus first read is returned. This usuallyleads to a positive value of the signal.

2.3.1. Array clocking sequence generation

Generating the clock waveforms for the PICNICarray is central to the entire operation. The keyproblem is generating the sequence of bytes that isto be down-loaded into the AWFG card. The sixPICNIC clocks as well as the analog to digital con-verter trigger clock have been assigned to the two4-bit outputs of the AWFG. The AWFG is capa-ble of storing a waveform of up to 32K steps long.When executing a waveform you must specify:

• length of waveform;

• number of times to execute the waveform;

• clock divider specifying number of microsec-onds per step.

Unfortunately, the maximum waveform lengthof 32K is not large enough to readout the entire

PICNIC array. Instead, we must specify a se-ries of waveforms and execute them sequentially.The AWFG board does allow a new sequence tobe loaded as the last sequence is finishing, if thelength of the sequence is the same and the clockdivider does not change. We have chosen a wave-form of length of 4128 clock steps.

The basic step time is created from a 5 MHzclock divided by 25, yielding a 5 µs base step time.The full readout sequence is created by stringingtogether multiple waveforms. Various read modesare supported. For example, the read sequencewhich just reads through the entire chip and dig-itizes the result is the combination of two wave-forms. The frame sequence is more complex andis composed of three parts: 1) the chip is reset;2) a variable length wait time (the exposure); 3)The read sequence. The correlated double sam-pling sequence (CDS) is similar, except that thesoftware subtracts the first values that are readjust after the reset from the second reads after theexposure is done and reports the difference.

The software to control the camera was writtenin C++ for a client-server architecture. Linux de-vice drivers for the three PC cards were writtenin house. The camera server communicates viaTCP/IP sockets and has commands to

• read camera temperatures and status;

• set readout waveforms;

• read the PICNIC array.

The client program which is usually run on a re-mote machine (e.g., on a workstation in the Berke-ley undergraduate lab) reads the array and currenttelescope information and creates a FITS file withdata and detailed header. Since the observatoryoperations are automated, scripts can be writtento combine image acquisition and telescope mo-tions. This is vital in infrared observations whereimages are often built up from multiple jogged ex-posures.

The minimum readout time of 660 ms is deter-mined by the speed of the analog to digital con-verter card. Newer cards with 1 µs conversion timeare now available that would speed up the readout time by a factor of ten with little change inthe software.

9

3. Performance

The camera operated successfully during theFall 2000 semester and sufficient observing timehas been accumulated by a class of 15 students todemonstrate satisfactory performance. The designhas proven reliable. For example, the Linux-basedcontrol software has never crashed. The only sig-nificant problem was caused by the commercialhelium compressor. During early operation themodel 21 CTI refrigerator delivered only marginalcooling capacity, with the PICNIC array reachingonly 100 K. Intially, we suspected that the coldhead was defective since it was a salvaged unit.However, the helium compressor failed catastroph-ically after only a few thousand hours of operation;when it was replaced with a new unit the expectedcryogenic performance was achieved.

The adoption of a specially prepared motor forcryogenic operation was a risk which allowed usto eliminate vacuum feedthroughs. The steppermotor has worked flawlessly, and this elegant so-lution for Dewar mechanisms has paid off. Whenthe motor is driven continuously, or holding cur-rent is applied, sufficient heat is deposited withinthe Dewar so that after several hours of opera-tion the temperature of the optics and the detec-tor rises by 10–15 K. In normal operation, wherethe filter wheel is active for a few seconds and thenturned off, this heat source is insignificant. Appli-cation of holding current is unnecessary becausethe magnetic forces due to the permanent polesin the motor are sufficient to keep the filter-wheelfrom moving.

Inspection of images delivered by the Offner re-lay at room temperature leads us to suspect thatit is essentially perfect, but it has not proven pos-sible to quantify the optical performance becausethe image blur has been dominated by seeing andtelescope aberrations. These aberrations are dueto telescope collimation errors which are apparentin direct CCD images. The best image sizes de-livered by the camera have size of less than 2 pix-els FWHM (2.6 arcseconds). However, this is onlyachieved after careful focusing of the telescope sec-ondary mirror.

In the next three subsections we describe sev-eral characteristics of the infrared array and thecamera. Exploration and quantification of thecamera performance forms the basis of the first

Fig. 11.— A grey scale representation of the Ks-bandflat field. Bad pixels appear as black. The cosmeticproperties of this array are very good. See Fig. 12 forthe pixel response histogram. The flat has has beennormalized so that the median value is unity.

lab performed by students. Once they have under-stood the operation of the camera and masteredthe calibration of images they are ready to makeastronomical observations such as the one outlinedin §3.4.

3.1. Array uniformity

Under high-background conditions flat field ac-curacy is most important systematic error whichlimits sensitivity of an array detector to faintsources. Modern arrays deliver very uniform pixel-to-pixel response which means that students caneasily construct a flat field that is adequate forall practical purposes. Even this array, which isnominally designated an engineering grade device,has excellent cosmetic characteristics (see Figs. 11& 12). This is illustrated qualitatively by Fig.11 which shows a grey-scale image of a Ks-bandflat field. This flat is a weighted mean of 11dark-subtracted twilight sky images. Taking twi-light sky images requires careful timing; if theyare taken too early the detector saturates, takentoo late then there is insufficient signal to forma high signal-to-noise flat. For our pixel etendue(1.6 × 10−7 cm2 str) the optimum time for tak-ing Ks-band images is between sunset or sunriseand 6-degree twilight. A field at high galactic lat-itude is suitable to avoid stars. The telescope isjogged between each exposure so that stars occupy

10

different pixels in subsequent pictures. Any pix-els containing obvious stars are given zero weightand the frames are combined into a weighted meanwhere outlying pixels (faint stars or cosmic rays)are rejected if their deviation from the mean ex-ceeds three sigma. In a flat normalized so that themedian pixel value is 1.0 the best fit Gaussian hasan rms of 0.04 (Fig. 12). Three percent of thepixels fall in a tail with low response outside ofthis distribution, and the array can be considered97% operable.

Fig. 12.— The histogram of the normalized pixelresponse of a Ks-band flat. The rms of the best fitGaussian distribution is 0.04 and 3% of the pixels are“cold”, forming a tail with poor response.

3.2. System noise

Under the assumption that the variance in thecamera signal is the sum of a constant read noiseand Poisson fluctuations it is straightforward tomeasure the read noise and gain. The data re-quired for this measurement are a sequence of ex-posures of increasing integration time while thecamera views a source of spatially uniform illumi-nation. A pair of exposures is acquired at eachintegration time. The mean signal level was mea-sured by subtracting a bias frame and the variancecomputed by differencing the two frames—sincethe uniformity of flat field is very high the frameswere not flat fielded. A straight-line, least squaresfit to the linear mean-variance relation yields thereadout noise as the intercept and the gain as theslope. The results of this analysis yield 30 e− perdata number (DN), and a read noise of 70 e− rms.The majority of this is detector noise, since if onegrounds the input to the data acquisition system

the variation is on the order of 30 e− rms. A sci-ence grade PICNIC array is expected to have aread noise of about 10 e− rms. Despite this ele-vated detector noise, the data are background lim-ited in all but the shortest exposures. The arraysaturates at 22,000 DN (660,000 e−) and shows <1% nonlinearity at up to 80% of this level.

Table 1: Camera Zero Points

Band mzpt

J 18.0H 18.2Ks 17.5

3.3. Throughput

The efficiency of the camera is expected to behigh. The detector quantum efficiency is 60-70%over most of the operating wavelength range. Theaverage filter transmissions are 86% (J), 83% (H),and 92% (Ks) between the half-power points. Theprotected Al coating for the primary, M1, has aninfrared reflectivity of 97%, and the reflectivity ofbare Al on the secondary, M2, is probably similar.The quartz Dewar window has a transmission of94%, giving a predicted camera efficiency of 50%.The system efficiency, including the telescope andatmosphere, determined from observation of stan-dard stars is 30% (J), 39% (H), and 44% (Ks).The camera zero points in magnitudes on a Vegabased scale, defined as

mzpt = mstar + 2.5 log10(DN s−1)

are listed in Table 1, where DN s−1 is the countrate in for a star of magnitude mstar.

3.4. An example observation

The region NGC 2024 (Orion B, W12) is a well-studied site of massive star formation located at adistance of about 415 pc (Anthony-Twarog 1982).It is also known as the Flame Nebula, becauseof the shadow of prominent lanes of extinctionformed against a background of Hα emission fromionized gas. The ionizing stars for this nebula arehidden behind this dust lane. Near-infrared radi-ation from these stars penetrates the dust and a

11

Fig. 13.— A single raw Ks-band frame of NGC 2024.This is a 10 second exposure. Only the brightest stars(K ≃ 8 mag) can be seen in this image. One pixelcorresponds to 1.35 arc seconds and the field of viewis 5.8 × 5.8 arc minutes.

dense stellar cluster is revealed in the optically-dark region separating the two halves of the neb-ula. Observations of NGC 2024 provide a dra-matic demonstration of the ability of near-infraredradiation to penetrate dust.

Fig. 14.— Flat-fielding and sky subtraction of theraw frame in Fig. 13 reveals significantly more detail,including the low-contrast nebulosity. The coma andastigmatism apparent in this image are due to tele-scope colimation errors.

NGC 2024 was observed on 2000 Nov 4. Eight10 s exposures were taken in J , H , and Ks. Thetelescope was rastered in a 3 × 3 (leaving out thecentral location) with offsets of 36 arc seconds inRA and DEC between each pointing. Exposures

were collected in J , H , and Ks for the target andthen the telescope was offset to blank sky 6 arcminutes to the east and the filter sequence wasrepeated.

Fig. 15.— The exposure mask for this observation.Seven good frames were acquired, each one at a dif-ferent telescope pointing. This image represents theprojection of good pixels onto the field of view. Badpixels can identified as local minima in the exposuremask. Dithering the telescope ensures that every pointin the field of view is covered by a functioning pixel.

Fig. 13 shows an individual Ks band image.Only the brightest stars (K ≃ 8 mag) can be seenin raw images. A sky-subtracted and flat-fieldedimage (Fig. 14) shows significantly more detail,including the nebulosity. The exposure mask forthis observation is shown in Fig. 15, and the finalmosaic composed using using a weighted sum ofthe registered images is shown in Fig. 16. The J ,H , and Ks images have been registered and com-bined into a three color composite. The opticallyinvisible, highly reddened, stellar cluster behindthe central dark lane stands out clearly (Fig. 17).

4. Conclusions

We have designed and built a high performancenear-infared array camera for the 30-inch telescopeat Leuschner Observatory. Our design eliminatescustom parts in favor of commercial components.The array control and readout electronics are com-posed entirely of standard PC cards. The onlycustom board is the interface which buffers theclocks from the waveform generator and amplifiesthe analog signal from the detector outputs. The

12

Fig. 16.— The final Ks-band mosaic for NGC 2024.The total exposure time in the center of the mosaic is70 seconds. Heavy foreground extinction due to inter-stellar dust reddens many of the stars in this cluster.Even at this wavelength, where the extinction is onetenth of that at visible wavelengths, prominent darkdust lanes are apparent.

Offner-based optics relay the telescope image atunit magnification via a cold stop which provideseffective control of thermal radiation from the tele-scope and its environment. The camera has beenused successfully in an undergraduate laboratoryclass in Fall 2000.

Construction of this camera was made possi-ble by the National Science Foundation throughInstructional Laboratory Initiative grant 9650176and the generous support of Kadri Vural at theRockwell Science Center, Thousand Oaks, CA.The Berkeley Astronomy Department providedmatching funds for this enterprise. We thank Prof.I. S. McLean (UCLA) for advice during the designand construction of the camera. W. T. Lum andDr. D. Williams of Berkshire Technologies helpedwith the Dewar and C. Chang (UCB, Radio As-tronomy Lab) did the PC layout. Prof. C. Heilesprovided invaluable guidance and encouragement.

REFERENCES

Anthony-Twarog, B. J. 1982, AJ, 87, 1213

McLean, I. S. et al. 1994, Proc. SPIE, 2198, 457

Murphy, D. C., Persson, S. E., Pahre, M. A.,Sivaramakrishnan, A. & Djorgovski, S. G. 1995,PASP, 107, 1234

Fig. 17.— A three-color infrared image of NGC 2024.Red corresponds to Ks, green is H , and blue is J . Thereddened stars at the center of the image are responsi-ble for ionizing the gas in this region. The mosaic hasbeen cropped to a field of view of 5.8 arc minutes ×

5.8 arc minutes. The exposure time is 80 seconds perfilter and the limiting magnitude is 14.2 at J and 13at Ks. North and east are indicated.

Nelson, B., McLean, I. S., Henriquez, F. &Magnone, N. 1997, PASP, 109, 600

Offner, A. 1975, Opt. Eng., 14, 130

Parthasarathy, R., Frank, C., Treffers, R., Cud-aback, D., Heiles, C., Hancox, C., Millan, R.1998 Am. J. Phys., 66, 768

Treffers, R. R., Richmond, M. W. & Filippenko, A.V. 1992, ASP Conf. Ser. 34: Robotic Telescopesin the 1990s, 115

Vural, K. et al. 1999, Proc. SPIE, 3698, 24

This 2-column preprint was prepared with the AAS LATEX

macros v5.0.

13