low-frequency, all-sky imaging with the paper array using aipy · movable (unburied tv cable) !...
TRANSCRIPT
Low-Frequency, All-SkyLow-Frequency, All-Sky
Imaging with the PAPERImaging with the PAPER
Array Using AIPYArray Using AIPY
A. Parsons1, D. Backer1, G. Foster1,
R. Bradley2,4, C. Parashare2, N. Gugliucci2,
E. Mastrantonio4, J. Aguirre3, D. Jacobs3,
C. Carilli5, A. Datta5,
M. Lynch6, D. Herne6, T. Colegate6
1 U. of California, Berkeley, 2 U. of Virginia, 3 U. of Pennsylvania,4 NRAO, Charlottesville, 5 NRAO, Socorro,6 Curtin U., Perth, Australia
The PAPER ArchitectureNon-tracking Crossed DipolesWide Bandwidth (125-205 MHz) !
Movable (unburied TV cable) !Smooth Beam
Flexible FPGA-based
Packetized CorrelatorFull-Stokes
Large # Ants (scalable)!Wide Band (up to 200 MHz)!
2048 Channel PolyphaseFilter Banks
4-bit Cross-Multipliers
AIPY: Model-based Imaging/CalibrationBootstrap Imaging/Calibration
W Projection, Multi-frequency Synthesis
Python-Based, Ties in MIRIAD, HEALPIXMeasurement Equation-based Parameter Fit
PAPER Antennas/Analog ElectronicsDeveloped by the Charlottesville (NRAO,UVA) Team
! Smooth beam (spatially and spectrally) ! Gain sensitive to ambient temperature
PAPER/CASPER Packetized Correlator
Two Dual 500 Msps ADCs + IBOB
Berkeley Wireless Research Center's BEE2
Two Calibration/Imaging Paths...
Bootstrap: Direct, fast, imperfect
Model-Fit: Clean, correct, expensive
“Sometimes you can't get started because you can't get started” – Don Backer
Parameter Space:
AIPY:AIPY: AAstronomicalstronomical
IInterferometry in nterferometry in PyPythonthon
an open-source toolkit targetingan open-source toolkit targeting
low-frequency interferometrylow-frequency interferometry
http://pypi.python.org/pypi/aipyhttp://pypi.python.org/pypi/aipy
What AIPY Is (and Isn't)!AIPY is:
AIPY isn't:
! A module adding tools for interferometry to Python (not visa versa)!! An amalgam of pure-Python and wrappers around C++ and Fortran! Object-Oriented! A collection of low- to mid-level operations (you write the program) !
! A toolkit (i.e. a grocery store, not a restaurant) !! In beta release (frequent updates/bug-fixes/changes) !
! A solution (it just helps you find it)!! A one-stop shop (it makes use of other open-source projects)!
! A replacement for other interferometry packages! Wedded to a file format (but only MIRIAD-UV, FITS currently supported) !
Philosophical Pedantry:
! Follow the (abbreviated) Zen of Python:! Explicit is better than implicit! Simple is better than complex! Complex is better than complicated
! Special cases aren't special enough to break the rules! Practicality beats purity! There should be one (and preferably only one) obvious way to do it, although that way may not be obvious at first unless you're Dutch! Now is better than never
! In the face of ambiguity, refuse the temptation to guess! Don't handicap the programmer; allow them all the rope they want! Don't hide data; return it (or at least grant access) at every turn
The ToolboxMODULES:
ant/sim/fit
const
coord
deconv
healpix
img
loc
map
miriadoptimize
src
----3rd party----
ephem
pyfits
numpy
pylab/matplotlib
geometry, phase / gain, amplitude / parameter fitting
physical constants
coordinate transforms (eq, ec, ga, precession, etc.)!
deconvolution (clean, lsq, mem, anneal) !
wrapper around healpix C++ (Max-Planck-Institut) !
imaging (gridding/weighting, W projection, beam calc) !
machinery for importing user-defined AntennaArrays
generation of all-sky maps, faceting
interface to MIRIAD filesparameter fitting, optimization code
lists of known sources and parameters
------------------------------------------------------------------------------
astrometry (relied upon heavily)!
reading/writing FITS files
numerical python (fundamental to everything)!
plotting, map projection
and example scripts for:
manipulating/visualizing MIRIAD files, flagging RFI, filtering out crosstalk,
fitting parameters, simulating arrays, making/viewing all-sky maps, phasing to/filtering out sources, imaging, etc.
Building a Simulator
AntennaArray
! set_jultime(...)!
! gen_phs(...) !
! gen_uvw(...) !
! sim(...) !
Antenna
! x,y,z
! dly,off
! bp_r,bp_i
Beam
! response(...) !
! <coeffs>
SrcCatalog
! compute(...) !! get_crd(...) !
RadioFixedBody
! compute(...) !
! ra,dec
! str,index
! angsize
RadioSpecial
! compute(...) !
! str,index
! angsize
Antenna
! x,y,z
! dly,off
! bp_r,bp_i
YourModule.py
! get_aa(...) !
! get_catalog(...) !
PAPER DeploymentsPGB: PAPER, Green Bank PWA: PAPER, Western Australia
38:25:59.24 N-79:51:02.1 W
Green Bank, WV
-26:44:12.74 S
116:39:59.33 EBoolardy, WA
! PGB-4: 2004-6, ants w/o flaps, 1 brd correlator! PGB-8: 2006-8, upgraded ants, packet correlator
! PGB-16: 2008-?, testing & characterization! Prototype facility for PWA deployments
! PWA-4: 2007, ants w/o flaps, 1 brd correlator! (PWA-32): 2009, ants w/ flaps, packet correlator
! (PWA-128): 2010?! Low-RFI, full-capability facility
Antenna Configuration
! Inner circle is PWA-4 configuration ! Outer circle is PGB-8 configuration! Shaded circles have radius = 10x elevation (dark is negative) !
Sample Data: 1 Day, 1 Baseline, PGB
TV/Aircraft TXFringes (Cas,Cyg,Sun)!
CrosstalkBeating sources
Intermittent TX
Satellite TX
Phase Amplitude (log)!
1-D “Delay” ImagingXRFI:! Remove model sky! Statistical thresholding! Manual flagging
DELAY TF:! iFFT of passband! Convolved by “delay beam”
1D CLEAN:! Compensate forholes in passband! Retain passband shape for self-cal
Delay/Delay-Rate Transform
! 1 hour of data with Cas A, Cyg A, Tau A! Phase to a source (here, Cas A)!! FFT of frequency axis = “Delay Image”! FFT of time axis = “Delay/Delay-Rate”! Cas A is confined to a region near origin! PSF determined by bandpass + time variabliity! Can recover bandpass + beam functions
Delay/Delay-Rate Transform! Two sources can show up at same delay for a given baseline! Over a time interval (say, 1 hr), can be differentiated by fringe rates (see below)!
! To prevent excessive blurring (changing fringe rate w/ time) best to phase to source before filtering! Clean delay axis before iFFT of time axis! Clean fringe rate axis to account for flagging of entire integrations
! Top to bottom: 3 baseline orientations! Left to right: delay bins, delay-rate bins, and combined
filter bins! DDR filters have 2 lobes of response: one that tracksyour source, one that sweeps across the sky
Recovering Bandpass and Beam
Kernels from DDR Transform
! Top plot: ! '+' = uncleaned bandpass recovered! heavy dots = cleaned bandpass! light dots = autocorr for comparison
! Bottom plot:! '+' = uncleaned delay image! line = cleaned delay image
! Cas A response versus time, Cyg A removed, 143MHz (top) and 165 MHz (bottom)!! '+' = beam response recovered using delay filter! dots = beam response recovered using DDR! lines = prediction of beam model
Modeling Beam of Dipole + Flaps
40dB zenith to horizon, 60 degree FWHM
Smooth spatially and vs. frequency
138 MHz 156 MHz 174 MHz
Model Beam vs. Source Strengths
Black: response toward Cas A
Blue: response toward Cyg A
Cyan: response toward Tau AGreen: response toward Vir A
Red: response toward Sun
Imaging with Non-Tracking Beams
Position-Dependent Weighting in Data Dirty Beam with Incorrect Weighting
! Only possible to weight for one track through primary beam! Dirty beam weighted for phase center imperfectly deconvolves away from center! Can evade problem with snapshot imaging
Using:8 Antennas at pos dec, 4 at negative3 days of data, 14s integrationsW projected, ~15 deg phase centers
138.8-174.0 MHz PWA-4/PGB-8
! Scale: log(Jy/beam) !! black->cyan = -1 to 1
! yellow >= 2, red >= 3
! Equatorial Mollweide Projection, RA=0 on right! Confusion at 80mJy/beam RMS at positive dec
! Sun at (1:10,+7:27), removed with DDR filter! Cyg,Cas,Vir,Tau peeled with ionospheric-dependent
solutions.
150 MHz PGB-8
! 1 day of 8-element single polarization data! Spans 138.8 to 174.0 MHz! Model subtraction of Cyg,Cas,Vir,Crab! Fit for ionospheric defraction
! Delay/F subtraction of Sun! Map has constant flux scale! Changing noise level due to beam response
RA,DEC = 12:00, 40:00
< "T > = 625 mK (10 mJy/bm)!
PGB-8 Thermal Noise Map
RMS Noise Map, log10(Jy/bm)!
Tsys = 250K
#FoV = 0.425 strradians
$0 = 162.5 MHz%$ = 35.2 MHz
#bm = 2.2e-5 strradians
Nant = 8 antennas
&int = 3 day observing = 3.5 hrs/pointing
Integrations added with alternating signs
PGB-8 System Temperature
Tsys (and predicted synchrotron) !
Tsky (confusion + synchrotron)!
Thermal noise (3 day integration)!
Using RMS pixel values from 4 patches nearRA=12:00,DEC=40:00 as a function of frequency, we
determine a perceived Tsky and thermal noise level.
Backing out integration, this gives us Tsys
Tsky
Thermal noise
“Beware the Jabberwock, my son!The jaws that bite, the claws that catch!”
-Lewis Carrol (1872)!
Power Spectra and Foregrounds
21 cm
EoR (z
=9.5) !
CMB Background
Confusion
Noise
Galactic Free-Free
Extragalac
tic Free-F
ree
Galactic Synchrotron
21 cm
EoR (z
=9.5) !
Power spectra at 146.6, 155.7, 164.5, and 173.3 MHz
(top to bottom w/ error bars), from 4 fields nearRA=12:00,DEC=40:00.
Point source dominated, starting to see synchrotronat lower frequencies, wavemodes
Summary!16 antennas deployed in PGB, 32 in PWA this fall!Analog system working well, going into production
!Packetized Correlator enabling build-out in staged deployments
! Working on QADC to lower digitizing cost, reducing to 100 MHz BW! Lowering integration time to ~8s
!AIPY-based processing pipeline maturing! Available at http://pypi.python.org/pypi/aipy
! Working on cluster/parallel computing
!Ionospheric refraction noticeable in data, compensated per source!Have a confusion/synchrotron-limited map (8.4K), ready for next tier of foreground removal
!Thermal noise in map (600 mK) continuing to integrate down
!Source removal currently limited by primary beam model! Using satellites to characterize per-antenna beam shapes
!Have not started investigating polarization or galactic synchrotron
Confusion NoiseNeed to balance array configuration between resolving and removing confusing sources (longbaselines), and optimizing for a power spectrum detection (concentrated core). Below, a map with nosources removed (left) and with the top 5 sources removed (right), with corresponding pixel histograms.
IonosphereTime scale: 10s Length scale: few km Refraction scale: few arcminutesPushes up data rates, need for instantaneous UV coverage (snapshot imaging) !Avoid dawn/dusk when ionosphere is highly variableAffects degree to which confusion noise can be suppressed
IonosphereTime scale: 10s Length scale: few km Refraction scale: few arcminutesPushes up data rates, need for instantaneous UV coverage (snapshot imaging) !Avoid dawn/dusk when ionosphere is highly variableAffects degree to which confusion noise can be suppressed
Black: Cas A, Magenta: Cyg A, 3 days of data, nighttime is shaded
38:25:59.24 N
-79:51:02.1 WGreen Bank, WV
38:25:59.24 N
-79:51:02.1 WGreen Bank, WV
-26:44:12.74 S
116:39:59.33 EBoolardy, WA
-26:44:12.74 S
116:39:59.33 EBoolardy, WA
Radio Frequency InterferenceSolution 1: Discretion is the better part of valor (run away) !Solution 2: Statistical thresholding of raw dataSolution 3: Manually flag contaminated channelsSolution 4: Delay/Delay-Rate FilteringSolution 5: Statistical thresholding after model subtraction
Boolardy, Australia Site Characterization
Light: PWA, 3 days of dataDark: PGB, 3 days of data
Blue:Peak, Green:Average, Red:Minimum
Interferometry FundamentalsOur Parameterized Measurement Equation:
Flat Field (Small Angle) Approximation:
! (u,v) coordinates represent the Fourier transform of the angular sky coordinates (l,m) !! Angular power spectrum directly measured in UV (visibility) domain! Inverse 2D FFT of gridded UV visibilities yields dirty image (convolved with synthesizedPSF) !! Synthesized PSF is inverse 2D FFT of visibility weights
Comparing with Other Experiments
MWA PAPER LOFAR 21CMA
Advantages/Philosophies:! Hands on: collect data early, address problems as they are encountered in data! Flexibility to adapt to new theory and new knowledge about foregrounds! Optimize smoothness of parameter space to allow faster, more precise fitting! Wide simultaneous bandwidth, full correlation of large numbers of antennas! Both fully functional prototype environment and low RFI location
Disadvantages:! Collecting area! Simulation
Unsure:! Spectral resolution vs. bandwidth! Group size
Sky |Vis| ø(Vis)!
Actu
al
Pri
ma
ry
Be
am
Re
str
icte
d
Vis
ibili
tie
s
For incoherent signalsummation, scaling of SNR with# samples depends on SNR ofsamples:
Sky-Vis Relationships
Implications formeasuring PS
Field of view (FoV)determines coherence regionin visibility domain.
Signal strength depends onFoV
Sample UV Optimized for 3 Scales
UV coverage Dirty Beam
Nested Triangular Configuration
with Outlying Calibrators
2008 May 13 Harvard-Smithsonian: 21cm Cosmology 38
Slices Through Image
Galactic Plane/Her AConfusion Sun Confusion
Virgo A
Power SpectraWhat an interferometer measures:
Related measures used in literature:
v
u
!̂
Sensitivity in Visibility DomainTsky ~ 160K#beam ~ 60 deg x 60 deg
%$ ~ 5 MHz#patch ~ 10 deg x 10 degNsamples ~ 81 baselines&s ~ 40 min (varies with patch size and baseline orientation) !
To hit SNR~1 in a UV pixel:' ~ 2 months for 0.5 mK
Minimum to constrain cosmic variance:- 3 sigma detection ~ 9 pixels around circle. - Using nested triangles with 3x3 antennas at each virtex,~ 64 antennas (being clever).
- Want to sample multiple rings to select for shape of power spectrum, so more like 128 antennas.- Need calibrator/imaging antennas, so more like 160 antennas.
BUT...
** UV pixel sampled is frequency dependent,undermining ability to subtract foregrounds.
What We Can Learn
What we can learn in first experiments:! Time of EoR degenerately constrains ! Dark Energy/Dark Matter (time for structure to collapse) + star formation rate! Confirm inside-out structure formation! X-ray heating by first quasars
What we can learn with SKA-class study:! Matter power spectrum through Dark Ages! Constrains dark energy, dark matter
(growth factor in collapse) !! Constrains time of radiation-matter
equality (transfer function) !! Geometry of Reionization! Star formation! Galaxy formation! Supermassive black hole formation! Feedback of reionization on
structure formation
“Richest of all datasets”
Inte
raction D
ista
nce S
cale
(lo
g)!
Time (log) !
Density P
ert
urb
ation (
log
)!
Inflationary Radiation Dominated Matter Dominated...
Recombination
Matter-Radiation Equality
Evolution of Perturbations
Simulated Power Spectra
Wouthuysen Field Effect
21 cm Spin Temp
Bounds on EoR (6<zEoR<12) !Gunn-Peterson Absorptionin high-z galaxies, (Becker, Fan)!
CMB Polarization from Thompson Scattering (Page, Spergel) !
Raw Sensitivity in Image DomainTsys = Tsky + Trx ~ 170K + 110K%$ ~ 1 MHzN ~ 256Aeff ~ 2#obs ~ 1 to 0.1 degrees
For necessary sensitivity:' ~ 1 month for 1 mK at 1 degree (l ~ 100)!
or more likely:' ~ 10 yrs for 1 mK at 0.1 degree (l ~ 1000)!
Have added flaps to! increase collecting area! reduce sky temp! suppress sidelobes of sources ! supress RF 21
cm EoR
(z=9.
5) !
CMB Background
Confusion
Noise
Galactic Synchrotron
Galactic Free-Free
Extragalac
tic Free-F
ree
Challenge: Starting from Scratch
Bootstrapping: ! Direct, imperfect, iterative! Do not rely (excessively) on priors
! Take advantage of wide bandwidth! Address degeneracies one at a time
“Sometimes you can't get started because you can't get started” – Don Backer
! Self-calibration needs single-source data! Source isolation requires calibration
! Parameter fitting needs a good starting guess! How do we get to first base?
The Problem:
Polarized Galactic Synchrotron
325 MHz Measured Polarization
150 MHz Power Spectrum Predicted from Komolgorov Turbulence
Faraday Rotation Screen:
! Smooth with frequency, but improperly calibrated linearpolarization can produce frequency-dependent structure.! Smooth power spectrum can allow further rejection.! Need a factor of 1000 suppression