J. Carlstrom, M. Dragovan, S. Myers, M. Joy & H. Quintana
4th September 1996
IntroductionThe CBI is a sensitive planar array which will image the cosmic microwave background radiation (CMBR) and measure the power spectrum of CMBR fluctuations on angular scales from ~ 5' to ~ 1.5°. Anisotropies in the CMBR on these angular scales contain a wealth of information about the early universe and provide a sensitive test of cold dark matter cosmologies. The ~ 5' to ~1.5° scales are essentially unexplored, so the CBI will open up a new window with the potential for exciting discoveries.Anisotropies in the CMBR are only ~ 10-100 µK (Scott, Silk, & White 1995), so fast imaging demands an exceptionally sensitive array. This requires low-noise receivers, a wide bandwidth and a high, dry observing site where atmospheric brightness fluctuations will not limit the sensitivity of the instrument. The CBI operates in the 26-36 GHz band. This was chosen as a compromise between foreground contamination due to point sources and galactic synchrotron and free-free emission, which dominate at low frequencies, and galactic dust and atmospheric O2 and H2O emission, which dominate at high frequencies. The CBI has thirteen 1-m Cassegrain antennas mounted on a hexagonal platform about 6 m in diameter. There are 208 antenna sites on the platform so many different array configurations are possible. This flexibility allows a fairly close-packed array for very fast low-resolution imaging, or a more sparse array for higher resolution observations. The array platform sits on a 3-axis mount (see Fig. 1). This has conventional azimuth and elevation axes for pointing the platform and an additional rotation about the axis of the platform which is used for aperture synthesis. Rotating the platform also allows us to remove systematic errors due to coupling between the antennas. To ensure that the antenna sidelobes see only ``cold'' sky, the CBI is surrounded by a conical ground shield approximately 70 m in diameter and 15 m high. Each antenna has a low-noise 26-36 GHz receiver, cooled to 6 K. Signals from the receivers are split into ten 1-GHz wide bands and cross-correlated in an analog correlator. This yields spectral information which is used to distinguish the CMBR from foreground emission (Brandt et al. 1994). The following sections contain a description of the technology of the CBI and a discussion of the key design problems.
Array configurationChoosing the configuration of the 13 antennas on the CBI platform involves a compromise between brightness sensitivity, resolution and the sidelobes of the synthesized beam. Close packing improves the sensitivity for imaging, at the expense of resolution. With 15 K receivers and 1 GHz bandwidth, a hexagonal close-packed array of CBI antennas has a sensitivity of ~ 5 µK in a day on an angular scale of ~ 9'. This makes the close-packed array attractive for some initial observations of the microwave background and for engineering tests, but because the array is very regular there are large grating lobes.The CBI must measure the power spectrum of microwave background fluctuations to provide definite tests of theoretical predictions. These measurements require uniform sensitivity over a range of angular scales, so a non-redundant configuration with uniform (u,v) coverage is needed. The interlocking spiral in Fig. 2 has reasonably uniform instantaneous (u,v) coverage and this type of array will probably be used for most CBI observations. When the array of Fig. 2 is rotated the (u,v) coverage is very uniform for baselines in the range 0.8-5.5 m so the array is uniformly sensitive to microwave background fluctuations with multipoles in the range l ~ 500-3500. This covers standard cold dark matter predictions of the third Doppler peak and the rapid cutoff in the spectrum due to the finite thickness of the shell of last scattering. To investigate the predicted first and second Doppler peaks at l ~ 200 and l ~ 400 respectively, we must mosaic a region several times the size of the 40' antenna primary beam. A mosaic may require a fairly close-packed array to attain enough sensitivity on baselines shorter than the antenna diameter. For observations of the Sunyaev-Zeldovich effect we need a resolution of 5' or better. Here a ring configuration is attractive because it maximizes the number of long baselines in the array and gives the highest resolution for a fixed platform size. To support the various arrays, we have made the CBI platform rather like an optical bench with many possible antenna sites. The platform is an irregular hexagon made up of triangular cells as shown in Fig. 3. There are 208 antenna locations with spacings ranging from 0.289 m to 5.508 m.
Antenna designThe design of the CBI antennas is severely constrained by the issue of inter-antenna coupling. In addition to generating the familiar receiver noise (ingoing noise), a receiver input also emits noise (outgoing noise). Some of the outgoing noise emitted from receiver x can scatter into an adjacent antenna and be cross-correlated with the receiver noise from receiver x, resulting in a signal at the correlator output. If we have receivers with 15 K receiver noise and 15 K outgoing noise, the signal at the correlator output is s = 15 c½ K where c is the power coupling between antennas. (If we transmit Pt from one antenna and receive Pr at another antenna, then c = Pr/Pt.) To observe 10 µK anisotropies in the microwave background, s must be less than 1 µK, which implies c < 10-12. This is very difficult to achieve, but the constraint can be relaxed somewhat because the coupled signals arrive at the correlator with a delay of ~ 6 ns. This reduces the coupled signal at the correlator output by about a factor of 20 if the correlator channels are 1 GHz wide. Also, we can measure the signal due to inter-antenna coupling by rotating the array platform about its axis. Signals from the sky remain fixed, but those due to coupling rotate with the platform. This strategy of keeping the relative antenna positions fixed and using rotation to measure and correct the now constant coupling is a key advantage of the CBI design. Rotating the array should easily reduce the false signal at the correlator output by an order of magnitude, but we still require an antenna design with low coupling.The CBI antenna is a conventional Cassegrain design with a shield can to reduce coupling due to scattering from the edge of the secondary and from the edge of the hole in the primary. The antenna is fed by a conical corrugated horn which is cooled to 6 K. The secondary is supported by a thin expanded polystyrene cone attached to the rim of the primary as shown in Fig. 4. The cone eliminates scattering from the feedlegs used in the standard Cassegrain design and is essentially invisible at 30 GHz (it contributes less than 1 K to the system noise). The shield can rim is rolled with a radius of a few wavelengths to reduce scattering from the rim and the inside of the can is corrugated to reduce its loss. The antenna is rather like a giant conical corrugated horn but with magnifying mirrors in the center to reduce the overall length. The beamwidth for the antenna varies from 40-50' across the 26-36 GHz band (see Fig. 5) and the inter-antenna coupling is in the range -100 to -130 dB for a 1 m spacing (see Fig. 6) which is the smallest in the CBI.
ReceiversThe CBI uses heterodyne receivers with a 26-36 GHz HEMT amplifier followed by a mixer and an IF amplifier in the 2-12 GHz band. All the components are cooled to 7 K by a closed-cycle 2-stage Gifford-McMahon refrigerator with an Er3Ni second stage regenerator (Model CSW-204SL-6.5, APD Cryogenics Inc., Sunnyvale CA 94086). The HEMT amplifiers are made by NRAO and use InP devices in the first two stages to give an exceptionally low amplifier noise temperature of ~ 10 K (Pospiszalksi 1993). The cooled mixer and first IF amplifier add ~ 0.5 K to the noise temperature and a coupler for a test signal and a polarizer ahead of the HEMT amplifier add ~ 2 K. The atmosphere contributes ~ 3 K and the microwave background another 3 K, so the system noise temperature is only ~ 20 K.A quarter-wave plate at the input of the HEMT amplifier provides circular polarization. This reduces the response of the CBI to linearly polarized galactic synchrotron emission, and reduces amplitude calibration errors due to a linearly polarized component in the calibration source (e.g., quasars often have 10% linear polarization). Between the polarizer and feed there is also a rotating half-wave plate ±360° phase shifter which can be used to discriminate between inter-antenna coupling and a real signal. If all the phase shifters in the array are rotated synchronously, the phase of the coupling between antennas varies by 720° while the sky signal is unchanged. The first local oscillator is at 38 GHz and this has a 180° Walsh function phase switch with a 12.8 µs period. This exceptionally fast phase switch is included to eliminate false signals due to offsets in the correlator and mains pick-up. The 2-12 GHz IF signal is split into 10 1-GHz wide bands in a filterbank and each of these bands is mixed down to 1-2 GHz. This scheme requires a comb of 2nd local oscillators at 3, 4, 5..13 GHz, all of which are above the 1-2 GHz correlator input band. Since the CBI is a planar array it is very easy to provide a stable correlated test signal that can be injected at each receiver input. The CBI has a wideband noise test signal so that all the correlator bands can be tested simultaneously. The test signal can also be used to measure the system gain between astronomical calibrations.
CorrelatorThe CBI correlator must process signals from 78 baselines with 10 1-GHz wide bands. This is quite a problem because in terms of multiplication rate, the CBI correlator is more than an order of magnitude bigger than the existing wideband correlators which are used in millimeter-wave arrays (Padin 1994). A digital correlator is far too expensive for the CBI so we are using an analog design with Gilbert Cell multipliers (Gilbert 1974). An analog correlator has the additional advantage that no sensitivity degradation occurs due to quantization of the signals (Thompson, Moran, & Swenson 1986).The key problem in a large analog correlator is signal distribution. In the CBI, the 1-2 GHz signal from each band in each receiver must be sent to 12 complex multipliers. Passband errors degrade the sensitivity of a correlator so the signal distribution scheme must introduce small errors. This requires wideband components and well-matched transmission lines or short connnections. In the CBI correlator, the signal distribution and all the multipliers for a single band are integrated onto one substrate. The multipliers are placed on a square grid as shown in Fig. 7. Signals from the receivers enter along the diagonal then travel along the grid boundaries. The multipliers above the diagonal compute the real part of the cross-correlation while those below the diagonal are fed with signals in quadrature to obtain the imaginary part. The auto-correlation is measured by total power detectors at the edge of the grid. There is just one quadrature hybrid for each receiver input so this configuration minimizes the number of components. The circuit is made entirely of chip components and fits on a 75×100 mm epsilonr = 10 substrate. Coupling between the multiplier inputs in this compact circuit can be as high as -40 dB and this causes a fairly large dc term at the output of each multiplier. The dc term is removed by the phase switch demodulator. At each multiplier and detector output there is a boxcar integrator synchronized to the 12.8 µs phase switch. The output of this integrator is digitized with a resolution of 16 bits and the signals are integrated for 839 ms in a digital accumulator which also handles the phase switch demodulation. Altera 10K50 field programmable gate arrays (Altera Corp., San Jose, CA) are used for the digital accumulators and their computer interface. A 10K50 chip has 50,000 equivalent gates and can provide 16 accumulator channels, so each 1 GHz band in the CBI correlator uses ten 10K50 chips. The electronics for a 1 GHz band fits on a 6Ux340 mm VME card so the complete system is quite small and can be attached to the back of the rotating platform on the CBI. This avoids cable wraps which can degrade the stability of the instrument.
MountAlthough the CBI operates at fairly low frequency, the pointing requirements are severe. A pointing error of 1 arcsec on a 5.5 m baseline at 31 GHz gives a phase error of 1° in the measured cross-correlation, so for mosaicing we can tolerate pointing errors of just one or two arcseconds. The SMA (SAO 1992) is one of the few radio instruments to achieve sub-arcsecond pointing accuracy, so we have based the CBI mount and drive on the SMA design. The azimuth axis has a 1.7-m pitch diameter ball bearing with an internal gear driven by a pair of 1000-Nm brushless dc motors. The elevation drive has a 100 mm ball screw driven by a similar motor. Since we cannot make microwave background observations through thick atmosphere, the elevation is limited to > 40°. The antenna platform is made up of many identical cast frames as shown in Fig. 8. A finite element analysis of this structure indicates a peak gravitational deformation of ±25 µm (Levy, 1996). This corresponds to a phase error of only 1.8° at the correlator output so we should be able to observe for long periods without phase calibration. The platform is mounted on a 3.7-m diameter bearing ring supported by two rows of cam followers. The platform is driven by three stepper motors through rubber wheels contacting the inside of the platform bearing. Inside the bearing is a cable wrap which carries 26 helium lines for the refrigerators, optical fibers for computer communications, power and cooling water for the electronics. The helium compressors are mounted on the azimuth platform so the gas lines must pass through the elevation and platform wraps. The elevation wrap is straightforward since the range is only zenith to 40°, but the platform wrap has a full turn so we have used a design with spools which take up excess line as the wrap rotates. To make the platform wrap compact, it is split into three separate wraps which lay lines on top of each other (see Fig. 9). The azimuth wrap is very simple since it carries only power, optical fibers for computer communications and cooling water for the helium compressors and electronics.
Ground shieldTo prevent ground emission entering through the antenna sidelobes, the CBI sits at the bottom of a conical reflecting ground shield, 70 m in diameter and 15 m deep. The shield reflects the antenna sidelobes so they see the ~ 6 K sky instead of the ~ 300 K ground. Since the CBI is a planar array, a source being tracked across the sky has zero fringe rate while a source on the Earth does not. This tends to wash out signals from the ground during the course of an observation. On the short (100 wavelength) baselines in the CBI the attenuation of ground signals will be on the order of 100. Longer baselines will have greater attenuation because the fringe rate is higher. If the antenna response is -50 dB at the closest approach of the ground shield to the beam, ground signals will be reduced by at least 70 dB in the image. So the ~ 6 K emission from the ground shield will cause < 1 µK false signals.The surface of the ground shield is at an elevation of 30°, 10° lower than the minimum elevation of the CBI. When the CBI is at its lowest elevation, the top of the shield is 15° from the highest point on the CBI platform. This is a few degrees above the mountain tops at the sites we have surveyed. With the shield, no part of the CBI can see ground features directly and the high local horizon increases the observing time with the Sun and Moon below the horizon. (The Sun is bright enough to cause huge false signals if it is anywhere above the horizon. The Moon may be a problem if it is near the beam.) The ground shield is such a large structure that wind loading is a serious problem. (The shield must survive a 45 m s-1 wind which causes a pressure of ~ 500 Pa at an altitude of 5000 m.) A free-standing structure is quite impractical, but an earth bank is robust and fairly inexpensive, so we have chosen this construction technique. The hole for the ground shield is 10 m deep and earth from the hole is piled up in a bank round the edge to a height of 5 m. The total volume of material moved is ~ 104 m3. The inside of the hole is dressed to remove large rocks and lined with sheets of copper foil with a plastic backing. Each sheet is pegged to the ground at ~ 2 m intervals using 0.5-m long rebar pegs with 10-cm diameter washers to hold the foil laminate. If parts of the shield are damaged by wind or intrusion it is very easy to patch the damaged area. At the top of the shield the foil laminate is rolled over the earth bank (so there is no sharp edge to cause scattering) and buried. Access to the CBI for both construction and maintenance is via a 6-m diameter culvert as shown in Fig. 10. During construction an underhung crane is erected inside the shield and the CBI parts are driven through the culvert on a truck. After construction the crane is dismantled and removed through the culvert. The culvert serves as a service tunnel and drain for the ground shield.
SiteThe Earth's atmosphere affects the response of an interferometer in several ways. Atmospheric emission increases the receiver noise, brightness fluctuations which are correlated between the antennas add noise at the correlator output, and path length fluctuations reduce the coherence. Path length fluctuations are caused by variations in the refractive index and are a problem for long baseline interferometers where the antennas look through different columns in the atmosphere. For short baseline interferometers, like the CBI, the antennas look through almost the same atmosphere so path length variations are small and the effects of atmospheric emission dominate.At Owens Valley Radio Observatory (altitude 1200m) the atmospheric opacity at 30 GHz is typically 0.04. 230 GHz opacity measurements at OVRO, Mauna Kea and the MMA Chajnantor site indicate a scale height for water of ~ 1800 m, so the 30 GHz opacity at a dry site at altitude h (meters) is roughly tau30 = 0.07 exp(-h/1800). Table 1 shows the opacity and atmospheric emission at various altitudes. At 4000 m the atmosphere contributes 2 K which is only ~ 10% of the system noise. Observations with the Owens Valley 5-m telescope suggest that atmospheric brightness fluctuations will not seriously degrade the sensitivity on the longest CBI baseline if the CBI is sited above ~ 4000 m. For shorter baselines the sensitivity degradation will be worse, so a higher site is favored. A site with high winds is also desirable because this increases the bandwidth of the fluctuations at the correlator output as clouds pass over the array, so the atmospheric noise integrates down faster. Siting the CBI involves a compromise between atmospheric noise, which favors a very high site, work efficiency, which can be severely reduced above ~ 4000 m (Cudaback 1984), accessibility and the cost of developing the site. Mines operate in the Andes at up to ~ 6000 m with local labor and up to ~ 4500 m with imported labor. There is also some experience at the MMA Chajnantor site at 5000 m and we have successfully camped in this area at ~ 4500 m. 5000 m is probably as high as we can expect to work with reasonable efficiency and this will require the use of supplemental oxygen for some tasks. There are only a few dry 5000 m sites on Earth and these are in the Atacama and Gobi Deserts. Another possibility for the CBI is the South Pole. This is only at ~ 3000 m but is probably the best observing site on Earth. Of these locations only the Atacama Desert is accessible and the MMA Chajnantor site near San Pedro is well characterized (http://www.tuc.nrao.edu/mma/sites/sites.html). The CBI will most likely be sited in this region.
Control systemThe CBI control system acquires data and controls the telescope hardware. It is run by two computers, a real-time computer on the correlator VME bus, and a Sun workstation for non-critical aspects such as scheduling and archiving the data. User interaction with the control system is through separate programs that connect to the Sun part of the control system via TCP/IP. This allows both control and access to diagnostics from the site, from the operations base, or from anywhere else in the world via the Internet.The real-time part of the control system runs on a Motorola MC68060 CPU under the VxWorks (Wind River Systems Inc., Alameda CA 94501) real-time operating system. Its activities are in turn scheduled by a control program on a Sun Ultra workstation. The real-time part of the control system continuously sends data to the Sun control program, where they are archived on high capacity optical disks. The recorded data include all the engineering diagnostics and the measured interferometer visibilities. These can be monitored in real-time via the Sun control program or at a later time by reading from the optical disk archives.
Power, communications, and domestic arrangementsSince the CBI is at a remote site we must generate our own electricity. This is quite expensive (~ $100k for fuel for a 70 kW year) so we have taken pains to minimize the site facilities. Fig. 11 and Fig. 12 show the telescope and site services respectively. Table 2 is a power budget for the site assuming that 2 people live there all the time. The power consumption is dominated by the helium compressors, telescope drives and cooling. We may be able to use some of the waste heat from the generators and telescope to heat buildings, though this does complicate the installation.The maximum load in Table 2 is just within the capabilities of an Onan 200DFAA generator (Onan Corp., Minneapolis MN 55432), derated to 5000 m. The power plant has of two of these units, one running as prime, the other as standby, with an automatic switch to change generators in the event of a failure. The generators cost ~ $30k each. Increasing the load requires a much larger engine and the cost of the plant and fuel increases substantially. The power plant is located ~ 400 m downwind of the CBI to ensure that water vapor produced by the generator does not affect observations and also to minimize the nuisance from noise and exhaust gas. The electronics, helium compressors and drives in the CBI are water cooled because air cooling at high altitude is difficult. The compressors and drives have water jackets and each electronics package has a water to air heat exchanger. The system requires ~ 100 l of 10 °C water per minute which is provided by heat exchangers to ambient and a water chiller. Communication with the CBI site is via a radio Ethernet link between the site and the operations base at lower altitude. The link provides telephone communication over Ethernet and allows remote control of the telescope. The link will probably be a Cylink 384 kbps unit operating at ~ 2.4 GHz with a range of 66 km (Cylink Corp., Sunnyvale CA 94086). A complete link costs ~ $10k. To avoid interference from harmonics of the link, it is located ~ 1 km from the CBI, below the ground shield. The operations base will have an Internet link, but this will likely be no more than a 9600 bps modem. Since we are working at high altitude the lab, control room and bedrooms have oxygen enriched atmospheres with ~ 24% oxygen concentration (compared with ~ 19% outside). This greatly improves sleep and efficiency at work. The oxygen is stored as liquid so the system requires electricity only for control. With two people on the site at all times we will use ~ 200 l of liquid oxygen per week. For working outside we will use small oxygen bottles with demand regulators. The goal is to avoid work in a critical or dangerous area without oxygen. The site buildings are ~ 12×3×2 m ocean cargo shipping containers. Four containers are comfortably fitted inside for living areas, a control room and a lab, but the containers for the power plant and water chillers are just bare shelters. Each fitted container has a water tank and sink. The bedroom and control room containers have chemical toilets and very basic kitchen facilities. We will probably need 2 or 3 other bare containers for storing equipment and vehicles. In addition to the site facilities we need an operations base at lower altitude. This is basically a place for the staff to live and the center for remote observing.
FundingThe CBI is funded by the National Science Foundation, Caltech, and a generous gift from Ronald and Maxine Linde.
BibliographyBrandt, W.N., Lawrence, C.R., Readhead, A.C.S., Pakianathan, J.N., and Fiola, T.M., ``Separation of foreground radiation from cosmic microwave background anisotropy using multifrequency measurements,'' Ap. J., 424, pp. 1-21, 1994.Cudaback, D., ``Four-km altitude effects on performance and health,'' PASP, vol. 96, pp. 463-477, 1984. Gilbert, B., ``A high-performance monolithic multiplier using active feedback,'' IEEE J. Solid State Circuits, vol. 9 pp. 364-373, 1974. Levy, R., ``Finite element model and performance analysis of the proposed Cosmic Background Imager deck truss,'' Pasadena CA:California Institute of Technology, July 1996. Padin, S., ``A wideband analog continuum correlator for radio astronomy,'' IEEE Trans. Instrum. Meas., vol.43, pp. 782-785, 1994. Pospiszalski, M.W., ``Ultra-low-noise Receivers for the 1 to 120 GHz frequency range,'' in Proc. 23rd European Microwave Conf., 1993, pp. 73-79. SAO, Design study for the Submillimeter Interferometer Array, Cambridge, MA:Smithsonian Astrophysical Observatory, February 1992. Scott, D., Silk, J., and White, M., ``From microwave anisotropies to cosmology,'' Science, vol. 268, pp. 829-835, 1995. Thompson, A.R., Moran, J.M. and Swenson, G.W., Interferometry and synthesis in radio astronomy, ch. 8, New York: Wiley, 1986. |