Canadian Hydrogen Intensity Mapping Experiment (CHIME) Pathfinder
Kevin Bandura, Graeme E. Addison, Mandana Amiri, J. Richard Bond, Duncan Campbell-Wilson, Liam Connor, Jean-Francois Cliche, Greg Davis, Meiling Deng, Nolan Denman, Matt Dobbs, Mateus Fandino, Kenneth Gibbs, Adam Gilbert, Mark Halpern, David Hanna, Adam D. Hincks, Gary Hinshaw, Carolin Hofer, Peter Klages, Tom L. Landecker, Kiyoshi Masui, Juan Mena, Laura B. Newburgh, Ue-Li Pen, Jeffrey B. Peterson, Andre Recnik, J. Richard Shaw, Kris Sigurdson, Michael Sitwell, Graeme Smecher, Rick Smegal, Keith Vanderlinde, Don Wiebe
CCanadian Hydrogen Intensity Mapping Experiment (CHIME)Pathfinder
Kevin Bandura a , Graeme E. Addison b , Mandana Amiri b , J. Richard Bond cd , DuncanCampbell-Wilson e , Liam Connor cf , Jean-Fran¸cois Cliche a , Greg Davis b , Meiling Deng b , NolanDenman f , Matt Dobbs a , Mateus Fandino b , Kenneth Gibbs b , Adam Gilbert a , Mark Halpern b ,David Hanna a , Adam D. Hincks b , Gary Hinshaw b , Carolin H¨ofer b , Peter Klages fg , Tom L.Landecker h , Kiyoshi Masui bd , Juan Mena a , Laura B. Newburgh i , Ue-Li Pen c , Jeffrey B.Peterson j , Andre Recnik f , J. Richard Shaw c , Kris Sigurdson b , Michael Sitwell b , GraemeSmecher a , Rick Smegal b , Keith Vanderlinde fi , and Don Wiebe ba Department of Physics, McGill University, 3600 University St, Montreal, Canada b Department of Physics and Astronomy, University of British Columbia, 6224 Agricultural Rd.Vancouver, V6T 1Z1, Canada c CITA, 60 St George St, Toronto, ON, M5S 3H8, Canada d Canadian Institute for Advanced Research, CIFAR Program in Cosmology and Gravity,Toronto, ON M5G 1Z8 e Sydney Institute for Astronomy, School of Physics, University of Sydney, NSW 2006,Australia f Department of Astronomy & Astrophysics, University of Toronto, 50 St George St, Toronto,ON, M5S 3H4, Canada g IBM Canada h National Research Council Canada, Dominion Radio Astrophysical Observatory, Box 248,Penticton BC V2A 6J9 Canada i Dunlap Institute for Astronomy & Astrophysics, University of Toronto, 50 St George St,Toronto, ON, M5S 3H4, Canada j McWilliams Center for Cosmology, Carnegie Mellon University, Department of Physics, 5000Forbes Ave, Pittsburgh PA 15213, USA
ABSTRACT
A pathfinder version of CHIME (the Canadian Hydrogen Intensity Mapping Experiment) is currently beingcommissioned at the Dominion Radio Astrophysical Observatory (DRAO) in Penticton, BC. The instrument is ahybrid cylindrical interferometer designed to measure the large scale neutral hydrogen power spectrum across theredshift range 0.8 to 2.5. The power spectrum will be used to measure the baryon acoustic oscillation (BAO) scaleacross this poorly probed redshift range where dark energy becomes a significant contributor to the evolution ofthe Universe. The instrument revives the cylinder design in radio astronomy with a wide field survey as a primarygoal. Modern low-noise amplifiers and digital processing remove the necessity for the analog beamforming thatcharacterized previous designs. The Pathfinder consists of two cylinders 37 m long by 20 m wide oriented north-south for a total collecting area of 1,500 square meters. The cylinders are stationary with no moving parts,and form a transit instrument with an instantaneous field of view of ∼
100 degrees by 1-2 degrees. Each CHIME
Send correspondence to K.Bandura: E-mail: [email protected] a r X i v : . [ a s t r o - ph . I M ] J un athfinder cylinder has a feedline with 64 dual polarization feeds placed every ∼
30 cm which Nyquist sample thenorth-south sky over much of the frequency band. The signals from each dual-polarization feed are independentlyamplified, filtered to 400-800 MHz, and directly sampled at 800 MSps using 8 bits. The correlator is an FX design,where the Fourier transform channelization is performed in FPGAs, which are interfaced to a set of GPUs thatcompute the correlation matrix. The CHIME Pathfinder is a 1/10th scale prototype version of CHIME and isdesigned to detect the BAO feature and constrain the distance-redshift relation.The lessons learned from its implementation will be used to inform and improve the final CHIME design.
Keywords:
CHIME, cosmology, SPIE Proceedings, Intensity Mapping, BAO
1. INTRODUCTION
The nature of the dark energy that drives the accelerated expansion of the Universe is one of the greatestmysteries in modern science. Of the observational techniques that probe dark energy, one of the most promisingis measuring the baryon acoustic oscillation (BAO) scale using fluctuations in the large-scale distribution ofneutral hydrogen. The BAO feature was imprinted in the matter correlation function at a scale of approximately 150 comovingMpc when baryons decoupled from radiation. By measuring the BAO standard ruler in the large-scale structureacross redshift, the expansion history of the universe is measured. In particular the period from redshift 1–2 hasthe most power to distinguish between dark energy models. The first clear detection of BAO came from analyzingthe Sloan Digital Sky Survey (SDSS) luminous red galaxies at redshift z ∼ The detection was verified in thetwo degree field galaxy redshift survey at redshift ∼ It has more recently been measured at three redshiftsof ∼ ∼ ∼ and Baryon Oscillation Spectroscopic Survey(BOSS) at z ∼ .
57. BAO were further detected in the Ly- α forest at redshift ∼
7, 8
The 21 cm neutral hydrogen emission is an accurate tracer of matter on cosmological scales.
The isolationof the 21 cm line eliminates the spectral confusion handicapping optical surveys operating at redshifts of 1 to 3.Since the BAO angular scale is on the order of a degree on the sky, the detection of individual galaxies as in anoptical BAO survey is not necessary. Instead, a 21 cm intensity map measures the aggregate 21 cm emission atlarger scales. These features make neutral hydrogen an excellent tracer of large scale structure and BAO at theredshifts where dark energy becomes dominant.The properties of the instrument are set to measure the BAO scale. The BAO first peak corresponds toan angular size of 1.35 ◦ at z = 2 . ◦ at z = 0 .
8. Nyquist sampling the BAO feature in a 21 cm maprequires baseline lengths of 15 m and 63 m at the corresponding redshifts. Along the line of sight, at redshift 0.8(800 MHz) the BAO scale corresponds to a correlation at 20 MHz separation, and at redshift 2.5 (400 MHz) itcorresponds to a 12 MHz separation correlation.In order to measure the BAO in neutral hydrogen, we are building a close-packed interferometer radiotelescope, the Canadian Hydrogen Intensity Mapping Experiment (CHIME) Pathfinder. It is a smaller scaleversion of the full CHIME instrument and is being built to inform the full CHIME design in addition to makingcompetitive measurements of the BAO scale. The pathfinder is comprised of two cylinders 20 m wide by 37 m longwhich are set 2 m apart. The cylinders are oriented along the north-south direction and fixed. The cylindricalnature of the structure gives the instrument a ∼ ◦ instantaneous field of view in the north-south direction,allowing a full survey of half the sky each sidereal day as the earth rotates. The close-packed configuration givesthe instrument a high intensity mapping speed.A summary of the Pathfinder specifications is in Table 1. All components have been designed with an eyetoward scaling the instrument by a factor of ten. The CHIME Pathfinder has 128 dual-polarization feeds andprocesses 400 MHz of bandwidth. Scaling to the 1280 feeds required for the CHIME telescope is made affordabletoday by advances in analog and digital processing. We have designed low-cost low-noise amplifiers based oncommercial components that achieve a noise figure of 35 K at room temperature. This drastically simplifies thecylinder feed line receiver design and number of components. For the channelizer we use field-programmablegate array (FPGA) processors which offer many digital signal processing blocks as well as many high-speed seriallinks, greatly simplifying the design of the digital network required to re-arrange and redistribute all the datarequency Range 400 MHz – 800 MHzRedshift Range 2.5 – 0.8Beam Size 1 ◦ – 0.5 ◦ Frequency Resolution 390 kHz, 1024 binsE-W FoV 2.5 ◦ – 1.3 ◦ N-S FoV ∼ ◦ Single Source Observing Time per Day Equator: 10 min – 5 min45 ◦ : 14 min – 7 minNCP: 24 hrStructure 2 cylinders – 37 m x 20 mNumber of Beams 128 dual polarizationReceiver Noise Temperature 50 K Table 1. Pathfinder design parameters for correlation. Our use of graphics processing units (GPU) for the correlation allows for a rapid developmentcycle and unique additional processing.The layout of the paper is as follows. Section 2 describes the science goals the CHIME Pathfinder hopes toattain. Section 3 describes the mechanical structure of the telescope, Section 4 describes the details of the feedand analog electronics and Section 5 describes the digital backend of the CHIME Pathfinder.
2. SCIENCE GOALS
The CHIME Pathfinder will measure the large-scale structure of the Universe from redshift 0.8–2.5 in 21 cmemission. Though the matter power spectrum contains a wealth of useful information, we are primarily interestedin measuring the BAO, relics of primordial sound waves that produce enhanced correlation at a characteristicscale. The physical observables corresponding to the BAO scale in the radial (redshift) and transverse (angular)directions are ∆ z = H ( z ) r s /c and ∆ θ = r s /D M ( z ), where H ( z ) is the Hubble parameter, r s is the soundhorizon at baryon-photon decoupling, and D M ( z ) is the transverse comoving distance. The sound horizon is well-constrained by Cosmic Microwave Background temperature anisotropy measurements,
12, 13 allowing measurementof the BAO scale as a function of redshift to directly probe the distance-redshift relation and expansion historyof the Universe.A major challenge faced by hydrogen intensity mapping experiments is the presence of bright astrophysicalradio emission at the same frequencies as the cosmological 21 cm emission we are searching for. These foregroundsare dominated by synchrotron emission from both our own galaxy and high redshift radio galaxies. They havebrightness temperatures up to 700 K, many orders of magnitude brighter than the 0.1 mK signal we are lookingfor.Fortunately the synchrotron emission is spectrally very smooth, and can be separated on this basis from thelarge-scale structure 21 cm signal. Though this is superficially straightforward, instrumental realities such asfrequency-dependent beams and polarization leakage present added challenges, which can be surmounted withprecise calibration and new analysis techniques. In Figure 1 we demonstrate the analysis of simulated timestream data, showing how we can effectively remove foregrounds using a Karhunen-Lo`eve transform to separatethe signals based on their statistics.The primary quantity we are interested in measuring is the spatial power spectrum of the large-scale struc-ture, which contains most of the useful cosmological information. Unfortunately the removal of foregroundcontamination inevitably reduces our sensitivity to the signal we are interested in. Using the m -mode formalism,the two-dimensional matter power spectrum can be reconstructed in the presence of foregrounds.
15, 16
As ex-pected we lose sensitivity to the largest scale fluctuations in the line-of-sight direction, but still retain significantsensitivity to the peaks of the BAO.
70 280 290 300 φ / degrees400420440460480500 F r e q u e n c y / M H z
330 340 350 360 φ / degrees400420440460480500 F r e q u e n c y / M H z
270 280 290 300 φ / degrees400420440460480500 F r e q u e n c y / M H z
270 280 290 300 φ / degrees400420440460480500 F r e q u e n c y / M H z
330 340 350 360 φ / degrees400420440460480500 F r e q u e n c y / M H z
270 280 290 300 φ / degrees400420440460480500 F r e q u e n c y / M H z − − µ K 140 µ K − µ K 30 µ K − µ K 0.5 µ K − µ K 120 µ K Unpolarised Foreground Polarised Foreground (Q) 21cm Signal S i m u l a t e d S kyF o r e g r o u n d F il t e r e d Figure 1. This plot illustrates the process of foreground removal on simulations of the radio sky. The top row of plotsshows sky maps of the individual components: unpolarized foregrounds, polarized foregrounds (showing Stokes Q only),and the 21 cm signal. On the bottom row we show the maps we would make after foreground cleaning visibilities from theCHIME Pathfinder. In each, the upper panel shows a frequency slice at 400 MHz, and the lower panel a slice through thegalactic plane from 400 to 500 MHz. Both the polarized and unpolarized foregrounds become substantially suppressed,whereas the 21 cm signal is largely unaffected. This leaves a clear correspondence between the original signal simulationand the foreground subtracted signal, while leaving the foreground residuals over 10 times smaller in amplitude than thesignal. Figure from Shaw et. al. 2014. In Figure 2 we illustrate the forecast constraints that the CHIME Pathfinder could place on the expansionhistory though measuring the BAO. The figure illustrates the forecast Pathfinder statistical limit of D V ( z ) = (cid:104) D M ( z ) czH ( z ) (cid:105) / with two years observation time in comparison to current measurements. The CHIME Pathfinder is an excellent platform to pursue ancillary science goals due to its design and operatingparameters. Some of those potential goals are listed below: D V r s , fi d / r s ( M p c h − ) BOSS6dFGS WiggleZSDSS-II BOSS Ly- α CHIME Pathfinder2-year Statistical Limit . . . . . . z . . . . . ( D V / r s ) / ( D V / r s ) fi d BOSS6dFGS WiggleZSDSS-II BOSS Ly- α CHIME Pathfinder2-year Statistical Limit
Figure 2. Forecast CHIME Pathfinder constraints on the expansion history parameterized using the ratio of D V (see text)to the sound horizon r s as a function of redshift, shown relative to a fiducial ΛCDM cosmology with h = 0 .
7, Ω Λ = 0 . m = 0 .
3. Also plotted are constraints from the 6dFGS, SDSS, BOSS, WiggleZ and the BOSS Ly- α results. Forthe BOSS Ly- α results the error bars are just a direct fractional error translation from the α iso constraints for illustrativepurposes, and should not be interpreted as their constraint on D V . For the CHIME Pathfinder, the forecast error barswere calculated using the methods from Shaw et. al. 2014 and represent the statistical limit with two years integrationtime. Pulsars: The CHIME Pathfinder’s wide field of view lends itself to the monitoring of pulsar timing. Thetelescope is able to see every pulsar in the northern hemisphere for at least ∼ The CHIME Pathfinder has the sensitivity and field of viewto discover many undetected transient sources. An exciting possible transient source is the radio afterglow fromcompact binary coalescence. Galactic polarization and Rotation Measure: In order to make a precise intensity map of the sky, the CHIMEPathfinder must be able to remove the foreground polarized emission from the galaxy. The CHIME Pathfinderwill produce an accurate map of foreground polarization, which can be used to further understand Galacticmagnetic fields.Cosmic rays: The close-packed nature of the CHIME Pathfinder will allow for the telescope to detect theradio emission from cosmic-ray air showers. Information about the air shower can be reconstructed from thearrival time, intensity, and polarization of the air-shower pulse.
3. TELESCOPE
The CHIME telescope structure consists of two adjacent parabolic cylinders, with a ground plane, feeds, andlow noise amplifiers held along the focal line of each cylinder beneath a walkway, see Figure 3a. The reflectivesurface is galvanized steel mesh bolted to the underlying structure. The cylinder axes run north-south.Each cylinder is 20 m across and 37 m long, see Figure 3c. The focal length of the parabola is 5 m andthe telescopes have a focal ratio of f/0.25. The telescope was designed to rely on standard steel constructionractices as much as possible. On-site welding is held to an absolute minimum to avoid RF disturbance of thehost observatory. The structure consists of parabolic steel trusses set every 5 m north-south supported by cementfeet which bear below the frost line. Purlins spaced 1 m apart run parallel to the cylinder axis, bolted to eachtruss. The structure is not very different from that of any warehouse roof, except that the east-west cross sectionis parabolic, the roof itself is instead mesh, and there is no floor. The structure is designed to have a peakdeformation of less than 2 cm under a wind and snow load of 1 kN/m .The mesh properties have been chosen as a compromise between reflectivity to radio wave and the ability toshed snow. CHIME has no moving parts and so, unlike a conventional radio telescope, it can not be tipped overto shed accumulated snowfall. A course mesh allows snow to fall through, which avoids loss of observation timedue to perturbed reflectivity. However, a course mesh also transmits more thermal radiation from the ground,raising the system temperature. The noise level from leakage of 300 K radiation through mesh is plotted inFigure 3d across the CHIME band for two different mesh options. The CHIME Pathfinder uses 19 mm (3/4in) spacing 14 Birmingham Wire Gauge (BWG) mesh, which is compared to a 25 mm (1 in) 10 BWG mesh.These are the heaviest wire sizes available at each spacing and thus the most reflective mesh at each size. Thenoise contribution of the 19 mm mesh is roughly 1 K lower across the upper half of the CHIME passband, whichcorresponds to a sensitivity difference per year equivalent to 15 days of extra observations. Over the 2013-14winter snow was on the CHIME mesh only immediately after the largest snow falls, for a total of four days. Themesh is available in rolls of 2 m width. The 1m purlin spacing is chosen so that the mesh sits on three purlinsand takes the shape of the reflector. The mesh is riveted every 75 cm along the purlin.Photogrammetry has been used to measure the shape of the steel support structure. The results are shownin Figure 3b as deformations with respect to the nominal design shape. The rms deformation here is 5.2 mm.Deformation of the mesh with respect to the steel structure has been measured by hand at hundreds of pointsacross the structure. The mesh surface error is 1.4 cm rms and dominates the telescope surface error budget.This rms gives a cylinder efficiency of 99% and 97% at 400 MHz and 800 MHz respectively.
4. ANALOG CHAIN
The analog system of the CHIME Pathfinder consists of the components shown in Figure 4. The feeds and LNAsare located along the focal line. The signals are then sent over coaxial cables to a shielded RF enclosure. Thesignals are then bandpass filtered and further amplified to achieve an input power of -21 dbm at the input ofthe ADCs. The overall receiver temperature design for CHIME is 50 K. This includes ground effects but doesnot contain any contribution from the sky. The analog components are described in more detail in the sectionsbelow.
The CHIME feed is a clover-leaf shaped compact dual-polarization feed. It is a modification of the four-squarefeeds developed for the Molonglo Observatory. The feed petals, balun stem, and support base are all madefrom printed circuits boards (PCB), as shown in Figure 5a. As shown in Figure 5b, the petals have curvedouter edges that broaden the frequency response by reducing the number of individual resonant dimensions. Thecurves are smooth and each petal is symmetric. The current pattern from a CST studio ∗ simulation is shownin Figure 5c for one linear polarization at 600 MHz. The currents near the gaps between petals run in opposingdirections so they cancel, and do not contribute to the radiation pattern. For this polarization, far-field radiationarises from the coherent currents running along the curved outer edges of the top and bottom pair of petals. Differential signals from pairs of adjacent petals are combined through tuned baluns to form one single-ended output. Thus each single polarization signal involves currents in all four petals. Full baluns, from bothpolarizations, consist of four identical microstrip transmission lines along four vertical support boards (the“stem”) and a horizontal base board. Both of the single-ended outputs are on the base board. Each transmissionline is varied in several abrupt steps, and the lengths and characteristic impedances of the transmission linesegments are carefully tuned to match impedance. Electrical losses in conventional circuit board materials ∗ to dB dB500 KFilter + Amp
BlockADC Coax
SMA(m)-N(f) Coax
SMA(m)-SMA(m)
SMA(m)-SMA(m)
2m RF
Room
C-CanFocus 60 m-5 to -8 dB-46 dBm50 K,0-3
GHz -57 dBm-21 dBm19 mV3.3 bits Bulkhead
SMA(f)-N(f)
Attenuator
SMA(m)-SMA(f)
Coax
N(m)-N(m)-6dB
Filter
Amp
SMA acts as bulkhead connector - Figure 4. The analog system signal chain overview. The clover-leaf feed receives the sky signal. The received signal is thenamplified by the low-noise amplifier block by 35 to 44 dB across the 400-800 MHz band while adding ∼
35 K noise. Thesignal is then transmitted over 60 m of LMR-400 coaxial cable to a central RF-shielded enclosure. The signal is attenuatedthere by approximately 6 dB, customized to each amplifier chain. The signal is then bandpass filtered to 400-800 MHzand further amplified by ∼
41 dB to achieve an input to the ADC of −
21 dBm power. generate unacceptable loss for astronomical instrumentation, so Teflon-based PCB is used for both the balun stemand support base. Between petals the PCB is completely removed to minimize loss along the slot transmission.The measured returned loss compared with simulation is shown in Figure 6a, and the measured beam patternis shown in Figures 6b and 6c.
The low-noise amplifier (LNA) is located at the focus and is directly attached to the feed. The LNA has two gainstages. The first is based on an Avago 54143 GaAs enhancement mode pseudomorphic high electron mobilitytransistor (E-pHEMT). It is followed by an Avago MGA-62563 E-pHEMT radio frequency integrated circuit(RFIC). The achieved noise figure with this design is 35 K across most of the band, as shown in Figure 7d.The input matching, output matching and feedback of the amplifier were all designed to achieve a low noisefigure. The resulting gain and matching S-parameters are shown in Figure 7b. The input has a relatively highreflection coefficient as the amplifier is noise matched instead of impedance matched. This does have the effectof reflecting more than ten percent of the incoming power back out of the feed.
A 60 m coaxial cable connects the ouput of each LNA to the sea container where each signal is received by afilter amplifier. This block comprises a custom-made Minicircuits band-pass 400-800 MHz filter followed by 3stages of gain. The S-parameters for this block are shown in Figure 8. The amplifier has a flat passband withless than 3 dB variation. It has greater than 20 dB rejection by 390 MHz on the low side and 815 MHz on thehigh side. It is a highly linear device, with a measured output power compression point of 29 dBm. a) (b) (c)Figure 5. (a) Photo of CHIME feed; (b) Geometry of petals with W = 138 . L = 131 . R = 20 mm. Thisgeometry is small enough that a feed element is compatible with any azimuth orientation within the array. (c) Thecurrent pattern from a CST simulation of the feed at 600 MHz for the horizontal polarization as indicated by the arrowlabeled E.
200 300 400 500 600 700 800 900 1000frequency/MHz4035302520151050 s / d B P1:396 P1:823P2:392 P2:832 return loss of designed clover antennameasurement of polarization P1measurement of polarization P2simulation of polarization P2 (a) (b) (c)Figure 6. (a) Measured return loss compared with simulation. Note the similarity between two polarizations. (b) MeasuredE plane of polarization P2. (c) Measured H plane of polarization P2.a) (b)(c) (a) In band (b) Broadband
Figure 10: LNA gain and noise measurements. (a) Return loss (b) Forward and reverse transmission
Figure 11: First stage LNA S parameter measurements15 (d)Figure 7. (a) Image of the CHIME LNA. The circuit board is soldered to the case to reduce internal resonances. (b)Measured S-parameters of the two-stage LNA. (c) Smith Chart of the LNA input matching relative to 50 ohms. (d)Measured gain and noise of the first stage of the LNA only.
5. DIGITAL BACKEND
The digital backend of the CHIME Pathfinder takes the overall structure of an FX correlator and is implementedinto the main components shown in Figure 9: • The analog-to-digital converters that sample the sky signal are located on daughter cards that attach tocustom FPGA motherboards. • The channelizer (F-engine) is implemented in each motherboard’s FPGA to split the 400 MHz-wide analogsignals into 1024 frequency bins 390 kHz wide. • The crossbar and shuffle modules re-organize the channelized data from all the motherboards in a crate inorder to concentrate the data for a subset of frequencies into a single FPGA. A 10 Gbps full-mesh networkconnecting every motherboard is implemented using a passive custom backplane driven by high speed serialtransceivers on the FPGA. • The offload link packetizes the shuffled data into 10 gigabit Ethernet packets and streams those packets tothe GPU correlator host computers. • The correlator (X-engine) is implemented in a dedicated computing cluster, where each node receives thedata with 10 gigabit Ethernet inputs, processes the packet header information and moves the data tosystem memory. • The GPUs are used to perform efficient, real-time full N correlation and averaging of the data. a) (b)Figure 8. (a) Image of the CHIME amplifier and band-defining filter. Input on the left. (b) Gain and passband of thefilter-amplifier block labeled filter amp plotted along with the full analog chain labeled cascade. The passband of the filteramplifier block is designed to be very flat with frequency. The entire analog chain has a slope from low to high frequencyprimarily due to the LNA gain and analog cabling. • Commodity gigabit ethernet switches are used to collect the data onto a server which stores the integrateddata. A server on this same network is used to configure and monitor the hardware in the array.Table 2 summarizes the key design parameters of the CHIME Pathfinder’s digital backend. In addition tothose, the system had to be designed with enough flexibility to allow testing of real-time gain corrections, RFIremoval, high-speed and triggered data tapping for ancillary science such as pulsar and radio transient signalanalysis, and beamforming along each cylinder. The design is also required to be scalable to 2560 inputs for thefull CHIME instrument.The hardware, firmware and software components of the digital backend are described in more detail in thesections below.
The analog-to-digital conversion of the feed signals is performed using custom double-wide FPGA mezzaninecard (FMC) compliant daughter boards equipped with two E2V EV8AQ160 analog-to-digital (ADC) chips (seeFigure 10). Each ADC chip has four inputs that can sample at up to 1.25 GSps at 8 bits.The sky signal in the absence of man-made signals is well encoded with only a few bits, with 4 bits having theeffect of increasing any properly amplified white noise system temperature by ∼ AQ N Point PFB/FFT S hu ff l e T X S hu ff l e R X G P U T x QSFP+QSFP+ C r o ss ba r C r o ss ba r CHIME PATHFINDER CRATE
Slot 1: ICEBoard motherboard
Kintex 7 FPGA Firmware
Slot 15Slot 16 G bp s F u ll m e s hba ck p l ane s hu ff l e ne t w o r k Channelizer
ADCBoardADCBoard
256 analog inputs from feeds (a)
Quad 10G EthernetCard
GPU Cluster
GPU Node 1GPU Node 15GPU Node 16 (b)Figure 9. Digital backend system overview. a) Data is sampled by ADCs and is acquired and framed by the FPGADAQ module, and is fed into the channelizer (F-engine) which performs a PFB/FFT and scales the data back into (4+4)bit values. The data is rearranged in the FPGA and through a high-speed backplane network and sent over 10 gigabitEthernet to a GPU farm to be correlated. b) Diagram of Pathfinder X-engine GPU system. Data is received by two quad10 gigabit Ethernet cards and passed into system memory. The data is then transferred in blocks to the GPU X-enginekernel for multiplication and accumulation. Finally the correlated signal is offloaded to a data server for storage andfurther processing. of each board, and are behind a shield that forms a Faraday cage. The analog traces are routed predominantlyon internal board layers to limit cross-talk. The neighboring inputs have been measured to have less than -50 dBcross-talk levels, limited by the ADC chip.Each daughterboard requires a 10 MHz clock input either through a SMA connector on the front panel, or asan LVDS or LVPECL signal from the host motherboard. The board has a software-programmable phase lockedloop (PLL) which creates the 1.6 GHz clock that drives both ADCs. A post-PLL ADC clock with less than 500 fsjitter is needed to ensure that the ADC performance is degraded by less than 0.1 bits.A synchronization input from the motherboard or an SMA connector on the front panel allows the ADC dataacquisition to be started in a precise way relative to the 10 MHz clock and ADC clock, which allows acquisitionof data frames with a deterministic phase across the array.An on-board I C EEPROM stores the digital serial number information along with individual board testinghistory, performance, and parameters. Also, an SPI bus allows the host FGPA to read out the temperatures ofumber of analog inputs 256Analog sampling 800 MSps @ 8 bitsChannelizer Type 2048 sample PFB/FFTFrequency channels 1024 bins, 390 kHz/binChannelizer data path Input: 8 bitsInternal: 18+18 bits complexOutput: 4+4 bits complexPower Consumption Channelizer: 1.2 kWCorrelator: 10 kWData rates Digitized analog inputs: 1.64 TbpsShuffle: 1.54 Tbps (Rx+Tx, plus overhead)Output to GPU correlators: 819.2 Gbps (plus overhead)Output from GPU correlator: ∼
100 Mbps (30 second integration)Baselines 32,896Computations Channelizer: ∼ Table 2. Key parameters for the Pathfinder digital backend.Figure 10. Two 8-channel, 8-bit, 1250 MSps, FMC-compatible analog-to-digital daughterboards (red boards) seated on aICE motherboard (blue board). the ADC cores and the board, as well as communicate with the ADC control registers.
The FPGA motherboard (also known as “ICE motherboard”, shown in Figure 10) accepts two FMC-complianthigh-pin-count CHIME ADC daughterboards that connect to a single Xilinx Kintex 7 XC7K420T FPGA. Thechoice of the FPGA was driven by requirements in terms of throughput needed for data shuffling and GPU links,Input/Output (I/O) pin count to the FMCs, backplane and other ICE motherboard subsystems, logic resources(gates, RAM, DSP blocks), cost, and power consumption.The FPGA board has 19 high-speed (12.5 Gbps GTX) serial links interfaced through the backplane connectorto support the Pathfinder and full CHIME data shuffling. It also has two QSFP+ connectors which connect to aGPU node and one SFP+ connector which connects to a control computer. The FPGA can also receive trigger,synchronization and GPS-based timestamp signals from the backplane.The board offers an identifying EEPROM, temperature sensors, FMC power control, and voltage/currentmonitoring for every power rail. This allows for self discovery and diagnostics of a large set of motherboards.Each ICE motherboard is equipped with a Texas Instruments AM3871 ARM processor with 1 gigabyte ofDDR3 DRAM, an SD card slot, two gigabit Ethernet ports, USB, UART, and SATA connections. The processoruns a Linux operating system, allowing for remote-programming of the FPGA as well as providing always-onmonitoring of the hardware and an arbitrated network-based control interface to the FPGA subsystems.The clocking for the entire board is derived from one 10 MHz clock which is received from the backplane.An on-board crystal oscillator and a front panel SMA connector are also available for single-board or bench topdevelopment work. The clocking is then distributed through a low-jitter network to the FMC mezzanine boardsand to the two onboard PLLs which create multiple clocks to drive the FPGA and ARM processor.The ICE motherboards are designed for a 9U standard VME physical crate design. They are 14 inches by 6.5inches, and can be spaced by 0.8 inch with the ADC mezzanines installed. The FPGAs and ADCs require activecooling provided by the host crate system. Each board requires approximately 75 W of power when runningthe Pathfinder firmware with two ADC mezzanines. The board operates from a single supply in the range of14-20 V, and the nominal 2 MHz buck converters can be synchronized by the FPGA to restrict the switchingnoise to known frequencies.
The firmware that operates on the FPGA is mostly custom code written in VHDL and simulated and compiledusing the latest Xilinx Vivado software suite. This approach enabled us to maximize the use of the Kintex 7FPGA and ICE motherboard potential as the new tools could easily place, route and meet timing closure in halfthe time required for older Xilinx tools on large FPGA designs.The firmware is subdivided into modules that interface with each other using the industry-standard AXI4-Streaming bus protocol to carry the data and control signals.All firmware configurations can be set talking directly to the FPGA using a simple UDP packet structure, orby communicating through the ARM processor.The FPGA has no embedded processor and any high-level functions and algorithms are performed by customcontrol software on either an external control computer or the ARM processor.In addition to the command system, the core firmware provides all the resources needed to operate theICE motherboard independently of the ARM processor. This includes access to the buck regulators, I C-baseddevices, the FMC hardware, etc. An internal frequency counter monitors internal and external clocks to confirmproper operation of the signal processing chain.The first signal processing module performs the data acquisition from the FMC ADC boards. The data isacquired at 800 MSps through 8 LVDS lines and a 400 MHz DDR clock. The module aligns the data acquisition ofeach line with a 78 ps resolution to compensate for the board and FPGA routing delays. The data is deserializedand combined into a 200 MHz, 32-bit wide AXI4 stream that is passed on to the channelizer module as framesof 2048 8-bit samples. The data acquisition module provides logic to deterministically start data framing on aknown edge of the ADC sample clock relative to the 10 MHz reference.The channelizer module starts its signal processing by selecting its source stream from the ADC or from anintegrated test pattern generator. The stream is fed to a customized poly-phase filter bank (PFB) and fast-Fourier transform (FFT) that has been generated by the CASPER † toolset and has been wrapped in an AXIinterface. The PFB includes a sinc-Hann window applied to 4 data frames, and outputs a frame of 1024 complexfrequency samples in a 18+18 bit format. The following scaler module applies a 16+16 bit complex gain to eachfrequency bin. The complex gain is stored in two tables that can be configured and switched in real time inorder to account for system gain and delay variations. The result is finally scaled to (4+4) bit complex valuesand saturates instead of folding. The output data is accompanied with ADC, FFT and scaling saturation flagswhich can be used to identify the strongest broadband and line radio frequency interference. A FPGA-basedstatistics subsystem also keeps track of ADC and scaling saturations for independent data monitoring.The data streams from the 16 channelizers are internally reordered by a crossbar module that aligns theincoming data streams and selects and routes specific frequency bins from every input to one of its 16 outputstreams. Each output stream is typically configured to contain a subset of 64 frequency bins from all channelizers. † https://casper.berkeley.edu/ he channel selection map is fully configurable to allow exclusion of unusable frequencies (due to RFI) and toadjust the downstream bandwidth. The crossbar repacks the data and saturation flags into larger blocks toincrease the efficiency of data transfers on the GPU host.The data shuffling module takes 15 of the 16 reordered streams and sends them to every other board in thecrate over the backplane using 10 Gbps links implemented using the FPGA’s high-speed serial GTX transceivers.The data is encoded in 64B/66B format, is scrambled to balance the DC content of the data, and is encapsulatedin simple packets with a cyclic redundancy check (CRC) code to detect transmission errors. The data comingfrom the corresponding boards is also received in the same way. After the full transaction, one FPGA nowpossesses a subset of 64 frequency bins from all 256 channelizers of the crate.The 16 data streams are then passed through another crossbar that rearranges the data into 8 output streams,each containing 8 frequency bins from all the channelizers of the array. An array of 8 custom 10 gigabit EthernetUDP packet transmitters is then used to send these 8 data streams to a GPU node through the two QSFP+connectors. Each packet is accompanied with a header identifying the data source, the format and size of thepayload data, and a timestamp. The CHIME Pathfinder channelizer consists of a single crate containing 16 ICE motherboards. The crate is 9Uhigh and uses standard VME mechanics but a custom high-speed backplane, shown in Figure 11. (a) (b) (c)Figure 11. (a) 9U crate partially populated with 8 ICE motherboards. (b) Front image of the custom 16-slot full-meshhigh-speed backplane. (c) Image of the rear of crate, showing the backplane’s power entry cables, 16 QSFP+ connectionsand SMA inputs for clock, trigger, timestamp and synchronization distribution to the motherboards. The image alsoshows the rear-accessible ports provided by each ICE motherboard: 2x Gigabit Ethernet, reset buttons, USB/UART,SFP+ and 2x QSFP+.
The backplane provides a low-jitter clock distribution system to the motherboards from a single SMA CMOSlevel input. It also distributes three additional digital signals: the ADC synchronization pulse, trigger pulse, anda timestamp signal. An I C interface allows any motherboard or an external controller to remotely power downor reset any board individually, and allows the backplane temperature and power to be monitored.A key feature of the backplane is its 10 Gbps full-mesh networking that connects every board with everyother one in both directions. This mesh connects directly to the motherboard’s FPGA multi-gigabit transceiversto implement the low-cost passive shuffling network. Extreme care was taken in maximizing the signal integrityof the 10 Gbps links. To achieve this, 25 Gbps Molex Impact connectors (Model 761657107) are used to matewith the motherboards, and transmission line discontinuities are reduced by using single-plane direct connectionsand back-drilled vias. The board is made with Panasonic Megtron 6, a low-loss material compatible with morestandard board assembly, and very low profile (VLP) copper foil is used to further reduce losses. The boardis laid out to reduce cross-talk between signals in the network. Strong and weak signaling levels are kept wellapart and are shielded by intermediate power planes. The FPGA’s internal dynamic equalizer compensates forhe low-pass frequency response and remaining distortions caused by the links. Preliminary tests between thetwo furthest boards (slots 1 and 16) show error-free transmission at 10 Gbps.The backplane offers 16 QSFP+ connectors connected directly to the motherboard FPGAs to provide anadditional 640 Gbps of off-crate data transfer through copper or optical cables, and will be used for the fullCHIME data shuffling between 5 crates. The QSFP+ can be interrogated to confirm cable connectivity andLEDs can be controlled to assist manual wiring and diagnosis.
Visibility calculation and time averaging of all baselines takes place in a dedicated GPU-based computing cluster.A diagram of the system is shown in Figure 9b.The operation is split across 16 fully independent and identical processing nodes, each responsible for process-ing the full set of baselines for 1/16th (25 MHz) of the CHIME bandwidth. A single control system is housed inthe same cluster, which serves the software and operating system used by the diskless nodes. This same systemaggregates and buffers the data prior to long-term archiving on a data server.Each node is housed in a 4U rackmount chassis and built primarily of high-end consumer-level components.The processing takes place in two AMD r9 280x GPUs and one r9 270x GPU. A pair of enterprise networkinterface cards (NICs) receives a total of eight 10 gigabit network connections, streaming a total of 51.2 Gbps ofradiometric data, along with associated headers and flags. This data rate sets the requirements for most systemcomponents.The incoming data is transferred over a third generation PCIe bus (8 lanes for each of the network boards) intoprimary system memory. Headers and flags are stripped from the data for further processing. Each packet headerhas a sequence number which allows the system to track and manage packet loss, and a stream ID identifies thefrequencies in the packet. The sequence number is identical across all links for a given ADC sampling period, soit is used to provide timing and synchronization between hosts. The flags are used to track and correct for ADCand scalar overflows in the data samples.The data is then DMA transferred from system memory into GPU buffers in large (256MB) blocks. EachGPU hosts 3GB of on-board RAM (2 GB for the 270x), used to buffer incoming data prior to processing. Formaximum computational flexibility, 3 connections are distributed to each r9 280x, 2 connections to the r9 270x –this results in all boards operating at roughly 2/3 utilization while performing the N correlation. The transferis controlled by an Intel i7-4820k CPU, allowing 40 total lanes of PCIe-3 communications. An EVGA x79 Darkmotherboard was chosen to allow 8 lanes to each of the 5 expansion boards (2xNIC, 2x r9 280x, 1x r9 270x).A primary bottleneck in the system was found to be the CPU-memory interconnect, and DDR3 2133 MHzoverclocked RAM is used to maximize performance.The correlation operation takes place in a custom processing kernel written in the OpenCL language ‡ . Detailsof this kernel will be presented in a future paper; we describe it briefly here. A single instance of the kernel(an OpenCL “Work Item,” WI) computes 4x4 correlations and accumulates them over 256 time steps, roughly0.6ms. These WIs are grouped into sets of 64 (OpenCL “Work Groups”) which share high-speed local memoryand compute a 32x32 correlation block. The 256x256 matrix of correlations is divided into these 32x32 sub-blocks,and the 36 upper-triangle blocks are computed and accumulated. (The remaining 28 blocks contain no additionalinformation, due to the symmetry of correlations.) Efficient operation requires that all calculations be pipelinedas multiply-accumulate operations (MACs), and the algorithm is able to operate efficiently by using integeroperations and packing two 4-bit values into each register. This packing sets the 256-timestep accumulationperiod, and requires a handful of book-keeping operations to take place at the end of each MAC loop, accountingfor the 8 least significant bit (LSB) offset on each sample, and accumulating the real and complex portions into32-bit buffers. Multiple kernel invocations result in longer accumulations, with correlation buffers read out forarchiving at 10-30s cadence.The control system collects the output correlations over a gigabit Ethernet network, merges the data andarchives it locally to an array of 3TB hard drives. This array can buffer at most several days of data, with the ‡ ong-term archive stored in another local building, on a much larger array of disks. Custom-written softwareregisters basic information about data products in a MySQL database for indexing purposes. The same softwarealso manages automatic transfer of data products between acquisition computers and long-term storage andanalysis nodes.Cooling of the GPUs is a significant concern. Consumer-level GPUs are assembled and sold by a varietyof vendors (using the same basic layout and identical processors), with various solutions for heat dissipation.A variety of brands were tested for thermal performance, and Sapphire “Dual-X” branded boards were chosenfor both the r9 280x and r9 270x, due to their remarkable cooling capabilities (keeping the processor die toroughly 70 C under load, 20-30 C cooler than all other models tested in our setup). We are presently exploringdirect-to-chip watercooling options, and anticipate retrofitting the system in the coming months.
6. OBSERVATION STRATEGIES AND CHALLENGES
The CHIME Pathfinder observing strategy is quite simple, since the instrument cannot move. Each day thetelescope observes the whole northern sky, with the observation time for every source set by the east-west field ofview of 2.5 ◦ -1.3 ◦ and the source declination: obstime / day = 4 min × fov × cos(dec). The full correlation matrixis saved at a 30 second cadence.One of the challenges of the system comes from the calibration needed in order to detect the BAO with theCHIME Pathfinder. From previous simulations of a Pathfinder-like instrument all telescope primary beamsmust be known to 0.1% and the system gain for each feed must be known to 1% in order to be able to properlyreconstruct the power spectrum. See the accompanying SPIE proceedings for full details about the calibrationof CHIME, which will be implemented on the Pathfinder. A brief summary of the calibration procedures plannedto achieve these levels of instrument stability is given below. • Measure the primary beam using bright point sources and pulsars. The bright points sources will give a firstmeasure of an overall complex gain calibration and beam-width as a function of frequency. Pulsars withtheir inherent on-off period allow one to remove all signals from that data that do not pulse at the frequencyof the pulsar. The beam will be measured further using pulsar holography by additionally correlating thesignal from the DRAO 26m Telescope tracking the pulsars. • Inject a broadband calibration signal. We will inject into every feed a broadband calibration signal, andmeasure and correlate that signal with all the CHIME feeds. This low-level injected signal is switched onand off on a ∼ second timescale. The switched signal is then used to measure and adjust for the complexrelative gain of every feed, removing complex gain changes on scales longer than the switching time. • Use the redundant baseline information. The CHIME Pathfinder is a very redundant interferometer bydesign. These redundant baselines can be used to calculate the complex gain of each feed without havingany previous knowledge of the sky.Another remaining challenge lies in interference management. Even though the CHIME Pathfinder is locatedin the radio astronomy reserve of the Dominion Radio Astrophysical Observatory, there are still man-made radiofrequency signals which will interfere with observing the sky. The CHIME Pathfinder plans to use real-time RFIflagging and excision techniques to mitigate those. The most extreme events which saturate the digital signalare flagged by the FPGA system and are passed to the GPU X-engine. Further processing will be performed onan intermediate integration scale of milliseconds to further process the data looking for excessive excursions ofthe signal. Post-processing will handle the final data clean-up.
7. CONCLUSION
The CHIME Pathfinder is currently being commissioned at the Dominion Radio Astrophysical Observatory.The cylinder structures have been completed and the analog and digital electronics are being phased in with 16feed channels currently installed. The first fringes observed with the telescope, from Cassiopia A, are shown inFigure 12. We are comparing measured sky maps to our galactic model to assess system temperature. We have igure 12. First fringes from an east-west CHIME Pathfinder baseline of . Shown are the real and imaginary parts of asingle visibility for a single frequency channel. performed holography with the DRAO 26m telescope to measure our beam shape. Finally, we have installed abroadband signal injection system to monitor complex gain in real-time.The CHIME Pathfinder is paving the way for all the aspects of the CHIME experiment. It is a 1/7th areaversion of the full system with 1/10th of the analog and digital electronics. All components being tested areto be scaled up to the full system. The Pathfinder will also be used to explore beyond the baseline design byinvestigating RF-over-fiber links, and real-time FFT beam-forming with the FPGAs and GPUs. Meanwhile,ground is currently being broken in preparation for the full CHIME construction. ACKNOWLEDGMENTS
We are very grateful for the warm reception and skillful help we have received from the staff of the DominionRadio Astrophysical Observatory, operated by the National Research Council Canada.We acknowledge support from the Canada Foundation for Innovation, the Natural Sciences and Engineer-ing Research Council of Canada, the B.C. Knowledge Development Fund, le Cofinancement gouvernement duQu´ebec-FCI, the Ontario Research Fund, the CIfAR Cosmology and Gravity program, the Canada ResearchChairs program, and the National Research Council of Canada. M. Deng acknowledges a MITACS fellowship.We thank Xilinx and the XUP for their generous donations.
REFERENCES
1. Weinberg, D. H., Mortonson, M. J., Eisenstein, D. J., Hirata, C., Riess, A. G., and Rozo, E., “Observationalprobes of cosmic acceleration,”
Physics Reports , 87–255 (Sept. 2013).2. Loeb, A. and Wyithe, J. S. B., “Possibility of Precise Measurement of the Cosmological Power Spectrumwith a Dedicated Survey of 21cm Emission after Reionization,”
Physical Review Letters , 161301 (Apr.2008).3. Eisenstein, D. J., Zehavi, I., Hogg, D. W., Scoccimarro, R., Blanton, M. R., Nichol, R. C., Scranton, R.,Seo, H.-J., Tegmark, M., Zheng, Z., Anderson, S. F., Annis, J., Bahcall, N., Brinkmann, J., Burles, S.,Castander, F. J., Connolly, A., Csabai, I., Doi, M., Fukugita, M., Frieman, J. A., Glazebrook, K., Gunn,J. E., Hendry, J. S., Hennessy, G. S., Ivezi´c, ˇZ., Kent, S., Knapp, G. R., Lin, H., Loh, Y.-S., Lupton,R. H., Margon, B., McKay, T. A., Meiksin, A., Munn, J. A., Pope, A., Richmond, M. W., Schlegel, D. J.,Schneider, D. P., Shimasaku, K., Stoughton, C., Strauss, M. A., SubbaRao, M., Szalay, A. S., Szapudi,I., Tucker, D. L., Yanny, B., and York, D. G., “Detection of the Baryon Acoustic Peak in the Large-ScaleCorrelation Function of SDSS Luminous Red Galaxies,”
The Astrophysical Journal , 560–574 (Nov.2005).. Cole, S., Percival, W. J., Peacock, J. A., Norberg, P., Baugh, C. M., Frenk, C. S., Baldry, I., Bland-Hawthorn,J., Bridges, T., Cannon, R., Colless, M., Collins, C., Couch, W., Cross, N. J. G., Dalton, G., Eke, V. R.,de Propris, R., Driver, S. P., Efstathiou, G., Ellis, R. S., Glazebrook, K., Jackson, C., Jenkins, A., Lahav, O.,Lewis, I., Lumsden, S., Maddox, S., Madgwick, D., Peterson, B. A., Sutherland, W., and Taylor, K., “The2dF Galaxy Redshift Survey: power-spectrum analysis of the final data set and cosmological implications,”
Monthly Notices of the Royal Astronomical Society , 505–534 (Sept. 2005).5. Blake, C., Kazin, E. A., Beutler, F., Davis, T. M., Parkinson, D., Brough, S., Colless, M., Contreras, C.,Couch, W., Croom, S., Croton, D., Drinkwater, M. J., Forster, K., Gilbank, D., Gladders, M., Glazebrook,K., Jelliffe, B., Jurek, R. J., Li, I.-H., Madore, B., Martin, D. C., Pimbblet, K., Poole, G. B., Pracy, M.,Sharp, R., Wisnioski, E., Woods, D., Wyder, T. K., and Yee, H. K. C., “The WiggleZ Dark Energy Survey:mapping the distance-redshift relation with baryon acoustic oscillations,”
Monthly Notices of the RoyalAstronomical Society , 1707 (2011).6. Anderson, L., Aubourg, ´E., Bailey, S., Beutler, F., Bhardwaj, V., Blanton, M., Bolton, A. S., Brinkmann,J., Brownstein, J. R., Burden, A., Chuang, C.-H., Cuesta, A. J., Dawson, K. S., Eisenstein, D. J., Escoffier,S., Gunn, J. E., Guo, H., Ho, S., Honscheid, K., Howlett, C., Kirkby, D., Lupton, R. H., Manera, M.,Maraston, C., McBride, C. K., Mena, O., Montesano, F., Nichol, R. C., Nuza, S. E., Olmstead, M. D.,Padmanabhan, N., Palanque-Delabrouille, N., Parejko, J., Percival, W. J., Petitjean, P., Prada, F., Price-Whelan, A. M., Reid, B., Roe, N. A., Ross, A. J., Ross, N. P., Sabiu, C. G., Saito, S., Samushia, L., S´anchez,A. G., Schlegel, D. J., Schneider, D. P., Scoccola, C. G., Seo, H.-J., Skibba, R. A., Strauss, M. A., Swanson,M. E. C., Thomas, D., Tinker, J. L., Tojeiro, R., Magana, M. V., Verde, L., Wake, D. A., Weaver, B. A.,Weinberg, D. H., White, M., Xu, X., Yeche, C., Zehavi, I., and Zhao, G.-B., “The clustering of galaxies inthe SDSS-III Baryon Oscillation Spectroscopic Survey: baryon acoustic oscillations in the Data Releases 10and 11 Galaxy samples,”
Monthly Notices of the Royal Astronomical Society , 24–62 (June 2014).7. Slosar, A., Irˇsiˇc, V., Kirkby, D., Bailey, S., Busca, N. G., Delubac, T., Rich, J., Aubourg, ´E., Bautista,J. E., Bhardwaj, V., Blomqvist, M., Bolton, A. S., Bovy, J., Brownstein, J., Carithers, B., Croft, R.A. C., Dawson, K. S., Font-Ribera, A., Le Goff, J. M., Ho, S., Honscheid, K., Lee, K.-G., Margala, D.,McDonald, P., Medolin, B., Miralda-Escude, J., Myers, A. D., Nichol, R. C., Noterdaeme, P., Palanque-Delabrouille, N., Pˆaris, I., Petitjean, P., Pieri, M. M., Piˇskur, Y., Roe, N. A., Ross, N. P., Rossi, G., Schlegel,D. J., Schneider, D. P., Suzuki, N., Sheldon, E. S., Seljak, U., Viel, M., Weinberg, D. H., and Yeche, C.,“Measurement of baryon acoustic oscillations in the Lyman- α forest fluctuations in BOSS data release 9,” Journal of Cosmology and Astroparticle Physics , 026 (Apr. 2013).8. Busca, N. G., Delubac, T., Rich, J., Bailey, S., Font-Ribera, A., Kirkby, D., Le Goff, J. M., Pieri, M. M.,Slosar, A., Aubourg, E., Bautista, J. E., Bizyaev, D., Blomqvist, M., Bolton, A. S., Bovy, J., Brewington,H., Borde, A., Brinkmann, J., Carithers, B., Croft, R. A. C., Dawson, K. S., Ebelke, G., Eisenstein,D. J., Hamilton, J. C., Ho, S., Hogg, D. W., Honscheid, K., Lee, K. G., Lundgren, B., Malanushenko, E.,Malanushenko, V., Margala, D., Maraston, C., Mehta, K., Miralda-Escude, J., Myers, A. D., Nichol, R. C.,Noterdaeme, P., Olmstead, M. D., Oravetz, D., Palanque-Delabrouille, N., Pan, K., Pˆaris, I., Percival, W. J.,Petitjean, P., Roe, N. A., Rollinde, E., Ross, N. P., Rossi, G., Schlegel, D. J., Schneider, D. P., Shelden,A., Sheldon, E. S., Simmons, A., Snedden, S., Tinker, J. L., Viel, M., Weaver, B. A., Weinberg, D. H.,White, M., Y`eche, C., and York, D. G., “Baryon acoustic oscillations in the Ly α forest of BOSS quasars,” Astronomy and Astrophysics , 96 (Apr. 2013).9. Chang, T.-C., Pen, U.-L., Bandura, K., and Peterson, J. B., “An intensity map of hydrogen 21-cm emissionat redshift z 0.8,”
Nature , 463–465 (July 2010).10. Switzer, E. R., Masui, K. W., Bandura, K., Calin, L. M., Chang, T. C., Chen, X. L., Li, Y. C., Liao, Y. W.,Natarajan, A., Pen, U.-L., Peterson, J. B., Shaw, J. R., and Voytek, T. C., “Determination of z 0.8 neutralhydrogen fluctuations using the 21 cm intensity mapping autocorrelation,”
Monthly Notices of the RoyalAstronomical Society: Letters , L125 (June 2013).11. Masui, K. W., Switzer, E. R., Banavar, N., Bandura, K., Blake, C., Calin, L. M., Chang, T. C., Chen,X., Li, Y. C., Liao, Y. W., Natarajan, A., Pen, U.-L., Peterson, J. B., Shaw, J. R., and Voytek, T. C.,“Measurement of 21 cm Brightness Fluctuations at z 0.8 in Cross-correlation,”
The Astrophysical JournalLetters , L20 (Jan. 2013).2. Hinshaw, G., Larson, D., Komatsu, E., Spergel, D. N., Bennett, C. L., Dunkley, J., Nolta, M. R., Halpern,M., Hill, R. S., Odegard, N., Page, L., Smith, K. M., Weiland, J. L., Gold, B., Jarosik, N., Kogut, A.,Limon, M., Meyer, S. S., Tucker, G. S., Wollack, E., and Wright, E. L., “Nine-year Wilkinson MicrowaveAnisotropy Probe (WMAP) Observations: Cosmological Parameter Results,”
Astrophysical Journal Sup-plement Series , 19 (Oct. 2013).13. Planck Collaboration, Ade, P. A. R., Aghanim, N., Armitage-Caplan, C., Arnaud, M., Ashdown, M.,Atrio-Barandela, F., Aumont, J., Baccigalupi, C., Banday, A. J., and et al., “Planck 2013 results. XVI.Cosmological parameters,”
ArXiv e-prints (Mar. 2013).14. Newburgh, L. et al., “Calibrating CHIME: A New Radio Interferometer to Probe Dark Energy,” in [
Societyof Photo-Optical Instrumentation Engineers (SPIE) Conference Series ], Society of Photo-Optical Instru-mentation Engineers (SPIE) Conference Series (2014). In preparation.15. Shaw, J. R., Sigurdson, K., Sitwell, M., Stebbins, A., and Pen, U.-L., “Coaxing Cosmic 21cm Fluctuationsfrom the Polarized Sky using m-mode Analysis,” arXiv.org (Jan. 2014).16. Shaw, J. R., Sigurdson, K., Pen, U.-L., Stebbins, A., and Sitwell, M., “All-sky Interferometry with SphericalHarmonic Transit Telescopes,”
The Astrophysical Journal , 57 (Feb. 2014).17. Lorimer, D. R., Bailes, M., McLaughlin, M. A., Narkevic, D. J., and Crawford, F., “A Bright MillisecondRadio Burst of Extragalactic Origin,”
Science , 777 (Nov. 2007).18. Thornton, D., Stappers, B., Bailes, M., Barsdell, B., Bates, S., Bhat, N. D. R., Burgay, M., Burke-Spolaor,S., Champion, D. J., Coster, P., D’Amico, N., Jameson, A., Johnston, S., Keith, M., Kramer, M., Levin, L.,Milia, S., Ng, C., Possenti, A., and van Straten, W., “A Population of Fast Radio Bursts at CosmologicalDistances,”
Science , 53–56 (July 2013).19. Spitler, L. G., Cordes, J. M., Hessels, J. W. T., Lorimer, D. R., McLaughlin, M. A., Chatterjee, S., Crawford,F., Deneva, J. S., Kaspi, V. M., Wharton, R. S., Allen, B., Bogdanov, S., Brazier, A., Camilo, F., Freire, P.C. C., Jenet, F. A., Karako-Argaman, C., Knispel, B., Lazarus, P., Lee, K. J., van Leeuwen, J., Lynch, R.,Lyne, A. G., Ransom, S. M., Scholz, P., Siemens, X., Stairs, I. H., Stovall, K., Swiggum, J. K., Venkataraman,A., Zhu, W. W., Aulbert, C., and Fehrmann, H., “Fast Radio Burst Discovered in the Arecibo Pulsar ALFASurvey,” arXiv.org , 2934 (Apr. 2014).20. Feng, L., Vaulin, R., and Hewitt, J. N., “Detectability of Late-Time Radio Afterglows from Compact BinaryCoalescence,” arXiv.org (May 2014).21. Leung, M.,
A Wideband Feed for a Cylindrical Radio Telescope , PhD thesis, University of Sydney (May2008).22. Deng, M. and Campbell-Wilson, D., “The Cloverleaf Antenna: A Compact Wide-bandwidth Dual-polarization Feed for CHIME,”
ANTEM 2014 (2014). In preparation.23. Mena, J., Bandura, K., Cliche, J. F., Dobbs, M., Gilbert, A., and Tang, Q. Y., “A Radio-Frequency-over-Fiber link for large-array radio astronomy applications,”
Journal of Instrumentation8