[from "The Near-Earth Asteroid Tracking (NEAT) Program: An Automated System for Telescope Control, Wide-Field Imaging, and Object Detection," by Steven H. Pravdo, David L. Rabinowitz, Eleanor F. Helin, Kenneth J. Lawrence, Raymond J. Bambery, Christopher C. Clark, Steven L. Groom, Stephen Levin, Jean Lorre, and Stuart B. Shaklan (JPL) and Paul Kervin, John A. Africano, Paul Sydney, and Vicki Soohoo (AFRL), Astronomical Journal, 117, 1666 (1999)]
2.1 Camera hardware
The NEAT camera was designed and fabricated at JPL in 1995, and its performance has been improved several times since. It consists of a 4096 x 4096 Charge-Coupled Device (CCD) with 15 Ám square pixels, associated control and digitization electronics, a thermoelectric cooler, and a mechanical shutter (See Figure 2). At the focus of a GEODSS telescope, the pixel scale is 1.4". In principle, stellar images are undersampled by this pixel size. In practice, the combination of tracking error and seeing (which we have not disentangled) creates images from 2.5" to 3" in diameter, and thus the sampling is adequate. The 4K device covers much of the useable field of view, about 1.6 on a side. As much as 2 could be used, but with further image degradation at the edges.
Digital commands to control the operations of the camera are transmitted via an optical fiber from the on-site workstation computer. A second fiber transmits the returned imaged data to the workstation. An overriding design consideration for the camera was that it fit at the Cassegrain focus, which is a small confined space internal to a GEODSS telescope. This space is usually occupied by an AFSPC video camera, replaced by the JPL camera during NEAT operations.
The NEAT CCD is a commercial-off-the-shelf part manufactured by Lockheed-Martin Fairchild Systems of Milpitas, CA. It features good cosmetic quality and low dark current. The imaging area is 4080 x 4080 contiguous pixels with less than 0.3% unusable area due to blemishes. There are 4 output nodes or amplifiers that can be sampled in parallel one for each 2048x2048 pixel quadrant. The read noise is 20 electrons at a readout speed of about 200 kpixels s-1. The bandpass is about 4 - 8000 determined solely by the CCD response (i.e. no filters).
The dewar is aluminum and accommodates the CCD and associated electronics without room to spare. It is filled with dry N gas and sealed before use. Thermal modeling showed no cooling improvement in evacuating this dewar. A two-stage thermoelectric cooler (TEC), with its cold side in thermal contact with the substrate of the CCD, actively transfers heat from the CCD to the back side of the dewar through an aluminum block acting as a conducting path. A cool air loop then removes the heat from the back side of the dewar. This arrangement maintains the CCD operating temperature within ▒ 3░ of 0░ C. The temperature is determined from the voltage across a diode in thermal contact with the CCD support. With the diode conducting a small fixed current, the temperature is proportional to the voltage drop. With the CCD kept at 0░ C, the dark current is about 90 e- s-1 pixel-1.
The mechanical shutter was built at JPL and has a very low (2 mm), narrow (10 cm) profile to fit into the available space inside the GEODSS telescope. It consists of a metallic blade that rotates into or out of the field of view under motor control in about 0.1 s. Shutter position is commanded by the computer through the camera electronics boards, which provide switching signals to an electronic circuit controlling the shutter motor .
San Diego State University (SDSU) built the camera electronics (Leach 1996). This control system allows software modification of the operating parameters and thus can drive a variety of CCDs with minor hardware changes. Since NEAT's inception, an earlier CCD with 2048 x 2048 pixel has been replaced with the present CCD and earlier versions of the control electronics have been upgraded to increase the readout speed by a factor of 4 from 50 to 200 kpixels s-1.
The electronics controlling the CCD consist of 4 circuit boards: a "timing" board to control the phase and duration of the signals that drive the parallel and serial transfer of charge across the CCD; a "utility" board to control the shutter position and to sample the voltage across the temperature-sensing diode; and two "clock/video" boards which drive the voltages for the parallel and serial clocks and also sample and digitize the video return signals at the 4 quadrants of the CCD.
Both the timing and utility boards have their own digital logic that are separately programmable and addressable via an optical fiber link on the timing board. Precompiled Motorola machine code is thereby downloaded from the workstation computer to control the clocking waveform, shutter timing, and readout timing of the CCD. Upon receiving a signal to expose and readout the CCD camera, the timing board returns the digitized signal to the workstation as a multiplexed, serial byte stream through a separate return fiber.
The main, on-site controlling computer is a Sun Sparc 20 computer with two central processing units (CPUs) clocked at 75 MHz. Appendix A gives details of the operating system software. Until May of 1998, this one computer not only controlled the telescope and CCD camera, but also ran the software to identify asteroids. With recent upgrades to increase the camera readout speed and to improve the rate of sky coverage, a Sun Enterprise 450 with 4 CPUs, each clocked at 300 MHz, was added to run the search software.
The Sparc 20 is equipped with electronics built by SDSU to allow the computer to communicate with the camera via the fiberoptic link. The components are mounted on a single circuit board that connects directly to Sun's proprietary data bus (SBUS), internal to the Sparc computer. Software to control this SBUS card under the UNIX operating system was cooperatively written by JPL and SDSU engineers, and recently modified at JPL to allow software handshakes between the camera and Sparc 20 during image readout. The same code has successfully operated on other Sun computers, including Sparc 5 and Ultra 2's. The Sparc 20 is also equipped with a commercially available SBUS card (a DR-11 W emulator built by Ikon Corporation) to allow 16-bit parallel communication with the electronics controlling the drive motors of the GEODSS telescope and dome. The telescope communication link runs from the DR-11 W card through 2 multi-pin cables to a Binary Interface Unit, part of the GEODSS control system, and from there to the telescope tower using the existing GEODSS connections. The camera communication link is 2 ~100-m long optical fibers from the SDSU SBUS card to the telescope tower. Additional computer peripherals consist of: a Datum Global Positioning System (GPS) receiver (provided by the USAF) to synchronize the internal clock of the Sparc 20 to Universal Time with an accuracy of a few milliseconds using signals from the GPS; approximately 60 Gbytes of hard disk storage; and a 28.8 Kbps modem for transferring data to computers at JPL and for remote monitoring and control of the telescope and camera from JPL. A standard Sun monitor is provided for on-site operators to monitor image quality and system status.
The recently added Enterprise 450 is equipped with 95 Gbytes of disk space. It runs software (described in Appendix B) to search 4 image triplets in parallel for moving objects, running an identical version of the software on each of its 4 CPUs and with each CPU assigned to analyze a different triplet. At the current rate of 45 s per image (20 s exposure plus 25 s overhead), this computer is able to keep pace with the acquisition of data. Within minutes of the acquisition of the third image in a triplet, the search of that triplet is completed.
2.3 Noise performance
The sources of noise in the images were determined from analysis of dark and sky frames with varying exposure times. The results show a dark current of 90 electrons pixel-1 s-1 and a sky brightness of 69 electrons pixel-1 s-1. The read noise is about 15 electrons. The cooling is most efficient near the center of the chip and the thermal gradient results in higher dark current by about 2/3 toward the edges. Dark frame subtraction and local flat fielding are needed to enhance object detection over the entire frame.