Canadian Hydrogen Intensity
Mapping Experiment (CHIME)
title
Kevin Bandura
CHIME Collaboration
Overview
• CHIME Cosmology Science
– Measure Baryon Acoustic Oscillations (BAO) from
redshift 0.8-2.5
• Intensity Mapping Technique
• CHIME Design
• Pulsars/Radio Transients
log [energy density (GeV4)]
Annu. Rev. Astro. Astrophys. 2008.46:385-432. Downloaded from www.annualreviews.org
by Carnegie Mellon University on 04/20/11. For personal use only.
ΛCDM Cosmology
Frieman et al. 2008
–36
– 40
Radiation
Matter
– 44
Dark energy
– 48
4
3
2
1
0
–1
log (1 + z)
Figure 1
Baryons
evidencefromtheCMBthat theuniverseisnea
rly spatiaMatter
lly flat (seeSection 4.1), weassu
Dark
except whereotherwisenoted.
Energy
Figure 1 showstheevolution of theradiaDark
tion, matter,
and dark energy densitieswith
Evolution of radiation, matter, and dark energy densitieswith redshift. For dark energy, theband r
w = −1± 0.2.
The universe has gone through three distinct eras: radiation dominated, z ≥ 3000
dominated, 3000 ≥ z ≥ 0.5; and dark energy dominated, z ≤ 0.5. The evolution of
factor is controlled by the dominant energy form: a(t) ∝ t 2/ 3(1+w) (for constant w). D
radiation-dominated era, a(t) ∝ t 1/ 2; during the matter-dominated era, a(t) ∝ t 2/ 3; an
the dark energy–dominated era, assuming w = −1, asymptotically a(t) ∝ exp(Ht). F
universe with matter and vacuum energy, thegeneral solution—which approachesthel
√
2/ 3
aboveat early and latetimes—isa(t) = ( M / VAC)1/ 3(sinh[3
.
VAC H0t/ 2])
Thedeceleration parameter, q(z), isdefined as
4.5%
22.7%
q(z) ≡ −
ä
1
=
aH2
2
72.8%
i (z)[1+
3wi (z)] ,
i
where i (z) ≡ ρi (z)/ ρcrit (z) isthefraction of critical density in component i at redshift z
thematter- and radiation-dominated eraswi > 0, and gravity slowstheexpansion, so th
and ä < 0. Becauseof the(ρ+ 3p) terminthesecondFriedmann equation (Newtonian co
only have ρ), the gravity of a component that satisfies p < −(ρ/ 3), i.e., w < −
NASA / WMAP Sciencewould
Team
repulsiveandcancausetheexpansiontoaccelerate(ä > 0). Wetakethistobethedefining
of dark energy. Thesuccessful predictionsof theradiation-dominated eraof cosmology
Baryon Acoustic Oscillations as
Dark Energy Probe
Cosmological implications of the BOSS-CMASS ξ(s)
one, in which t he D D , D R, and RR count s are int egrat ed
over µ before t hey are combined as in equat ion (1) t o comput e ξ(s), ignoring t he fact t hat t he geomet ry of t he survey
int roduces a µ dependence on RR (Samushia et al. 2011;
K azin et al. 2012), alt hough t he differences between t he t wo
approaches are more significant for higher mult ipoles.
W hen comput ing t he pair count s in equat ion (1), a few
import ant correct ions must be t aken int o account . T his is
done by assigning a series of weight s t o each object in t he
real and random cat alogues. First , we apply a radial weight
given by
wr =
1
,
1 + Pw n̄(z)
WMAP Power spectrum along with data from
ACBAR and QUaD. Flat ΛCDM model to the
WMAP data alone
Komatsu et al. 2011
Wigglez - Blake et al, 2011
z-=0.6
(3)
where n̄(z) is t he expect ed number densit y of t he cat alogue at t he given redshift and Pw is a free paramet er.
Hamilt on (1993) showed t hat set t ing Pw = 4πJ 3 (s), where
s
J 3 (s) = 0 ξ(s )s 2 ds , minimizes t he variance on t he measured correlat ion funct ion for t he given scale s. Following st andard pract ice we use a scale-independent value of
Pw = 2 × 104 h − 3 M pc3 . Reid et al. (2012) show t hat t he full
scale-dependent weight provides only a marginal improvement over t he result s obt ained using t his const ant value.
We include addit ional weight s t o account for nonrandom cont ribut ions t o t he sample incomplet eness and t o
correct for syst emat ic effect s. T he incomplet eness in a given
sect or of t he mask has a random component due t o t he fact
t hat not all galaxies sat isfying t he CM A SS select ion crit eria
are observed spect roscopically. I n any clust ering measurement t his is t aken int o account by down-sampling t he random cat alogue in t hat region of t he sky by t he same fract ion.
However, t here are t wo ot her sources of missing redshift s
which require special t reat ment : redshift failures and fibre
collisions.
Even when t he spect rum of a galaxy is observed, it
might not be possible t o obt ain a reliable est imat ion of t he
redshift of t he object , leading t o what is called a redshift
failure. A s shown in Ross et al. (2012), t he probabilit y t hat
a spect roscopic observat ion leads t o a redshift failure is not
uniform across t he field since t hese t end t o happen for fibres
locat ed near t he edges of t he observed plat es. Hence, t hese
missing redshift s cannot be considered as an ext ra compo-
5
SDSS III Sanchez et al, 2012
Z=0.57
Intensity Mapping
NGC1232
Galaxy Surveys vs. Intensity Mapping
•
•
•
•
Galaxy Surveys
Small percentage of
the universe has been
mapped with optical
light from galaxies.
Galaxy surveys are
expensive
Rely on (non-linear)
galaxies as tracers for
BAO
BAO
Scale
Unresolved galaxies,
resolved BAO.
Blanton et al. (2003)
Neutral Hydrogen
21 cm Emission
N
S
N
N
S
S
N
S
21 cm
z
∞
CHIME will:
• survey BAO
• measure the
growth of space
• from 0.8<z<2.5
• over a volume of ~400 comoving Gpc3
0
Mapping
the
Observable
Universe
Age of
Universe
Big Bang
CMB
CHIME
9
5
3
2
1
13.7Gy
SDSS
WiggleZ
Reionization
Slide from K. Vanderlinde
Cylinder Telescopes
Illinois 400’ 1959
Ooty, India,
1970
Northern Cross,
Italy, 1980s
MOST, Australia,
1960
STE Lab,
Japan, 1980s
Cylinder Antenna
Cylinder Telescope
Cylinder Beamforming
CHIME
The Canadian Hydrogen Intensity Mapping Experiment
100m
100m
• Drift scan = 1/2 sky daily
•BAO scale to <10%
• 400-800MHz band
• w0 to ±0.05 (w0 ~ 1)
•0.8 < z < 2.5 (for 21cm)
• wa to ±0.2 (wa ~ 0)
• ~200 Gpc3 survey volume
•1MHz frequency resolution
•5-10Mpc
• 13’-26’ spatial resolution
•10-45Mpc
Z=1.5 2 years observing
Fractional Error (%)
CHIME Cosmology Forecast
Antenna
LNA
RF over Fiber
ADC
Two FMC Mezzanines
(analog signals come in the front)
FPGA
Custom Backplane
Power, clock, timestamp
Board inter-connect
Digital signals come in the back
Two Gbit
Ethernet
Correlator – FGPA Firmware
• ADC
board
• FPGA
GPU Correlator
The CHIME Pathfinder
“an end-to-end hardware, calibration, foreground suppression, and data analysis proof-ofconcept for CHIME”
40m
40m
• 64 dual-pol antennas per cylinder (256 total channels)
• 100’s Gpc3 Survey volume
• Construction beginning
• Due to be functional in 2013
•
•
•
•
Test CHIME hardware
Test Calibration Techniques
Test Foreground Removal
Preliminary BAO Measurement
Signal per 50MHz bin
1 month of pathfinder data
CHIME Pathfinder Construction
M-Mode Analysis
v=B a + n
6
Or iginal
Obser ved
270
275
275
280
280
280
285
290
285
290
295
295
300
400
300
400
420
440
460
480
500
φ / degrees
270
275
290
420
440
460
480
300
400
500
640.00
0.0000001
275
280
280
280
290
285
290
295
295
300
400
300
400
460
480
500
φ / degrees
270
275
φ / degrees
270
285
+/-1mK
500
0.000026
285
290
420
440
460
480
500
300
400
420
440
f / MHz
0.0011
480
295
f / MHz
− 0.0012
460
− 0.000023
430.0000000
275
440
440
f / MHz
270
420
420
f / MHz
0.64
φ / degrees
285
295
f / MHz
Signal
For egr ound Filt er ed
270
φ / degrees
φ / degrees
For egr ounds
0-600K
+/-0.02mK
− 0.00071
460
480
500
f / MHz
0.00071
− 0.00042
+/-0.5mK
0.00046
Shaw et al. 2013
F i gu r e 2. T his plot illust rat es t he foreground removal process in act ion on simulat ions of t he foregrounds-only (t op row) and signal-only
(bot t om row). Each plot has t wo element s, an image of t he 400 M Hz frequency slice on t op, and beneat h, a cut t hrough t he celest ial equat or
Recovered Power Spectrum
PP((kk))// PPss((kk))
0.95
0.95
N al l
0.90
0.90
l
†
! N al
s = P s (F + N ) P s
= I + P s N P †s .
(27)
(28)
t he t ransformed
inst rument
al noise
mat rix
will not re0.00
0.05
0.10
0.15
0.20
0.00
0.05
0.10
0.15
0.20
in diagonal t his gives a correlat
ed
component
between
−−11
Wavenumber
Wavenumber kk [[hhM
Mpc
pc ]]
our modes. However, as it is useful if our modes are
Foreground
—ions
log10 ( S/ F)
summat
of
basis
funct
summation
ionSignal/
of di↵erent
di↵erent
basisRatio
funct
ions
X
X
P
ppaaP
P(k)
(k) ==
Paa(k)
(k) ..
10000
KL mode number (sorted)
− 3 M pc33 ]
PP((kk)) [[hh− 3M
pc ]
e eigenvalues λ corresponding t o each eigenvect
or give
fg
fg++ thermal
thermal
thermal
e diagonal mat rix ⇤
thermal only
only
To isolat e10
t10he
21
cm
signal,
we
want
select
modes
wit h
44
envalue (signal-t o-foreground power) great er t han
me t hreshold (see Figure 1). To project int o t his basis
define t he mat rix P s which cont ains only t he rows
m P corresponding t o eigenvalues great er t han t he
3
103
eshold s. 10
For most analysis we can work direct ly in t he eigenbaHowever, for visualising our result s, we want t o be
e t o t ransform back t o t he sky (by way of t he mea22
ed visibilit10
To project back int o t he higher dimen10ies).
−2
−1
10
10
10− 2
10−−11
nal space we simply generat
e t he full inverse
P
and
−1
Wavenumber
Wavenumber kk [[hhM
Mpc
pc− 1]]
move columns corresponding
t o t he reject
ed modes (we
not e t his mat rix P̄ s ). T his is equivalent t o project ing
1.10
o t he full1.10
eigenbasis, zeroing t he foreground cont amied modes, and t hen using t he full-inverse P − 1 .
1.05
For furt her
1.05analysis, we must include all noise t erms,
h foregrounds and inst rument al. Writ ing t he t ot al
se cont ribut
ion as N al l = F + N , t he mat rix in t he
1.00
1.00
ncat ed basis is
aa
77
5
(33)
(33)
4.5
In
In Appendix
Appendix A
A we
we describe
describe how
how ttoo project
project tthis
his quant
quantity
ity3.0
int
o
t
he
angular
power
spect
rum
of
21
cm
fluct
uat
ions
int
o
t
he
angular
power
spect
rum
of
21
cm
fluct
uat
ions
8000
tthat
hat we
we use
use ttoo calculat
calculatee tthe
he visibility
visibility correlat
correlations.
ions. For
For1.5
simplicity,
each
of
our
bands
is
simply
equal
t
o
t
simplicity, each of our bands is simply equal t o the
he input
input
power
6000 spect
power
spectrum
rum wit
within
hin aa fixed
fixed k-band,
k-band, and
and zero
zero out
outside,
side,0.0
such
t
hat
t
he
fiducial
model
is
p
=
1.
such t hat t he fiducial model is paa = 1.
For
For tthe
heband-powers
band-powers ppaa tthat
hat we
weare
arettrying
rying ttoo est
estimat
imate,
e,− 1.5
4000mat rices C̃ are simply t he project ion of t he basis
tthe
he mat rices C̃aa are simply t he project ion of t he basis
funct
functions
ions P
Paa(k
(k)) int
intoo tthe
he eigenbasis.
eigenbasis. St
Start
arting
ing from
from tthe
he− 3.0
00
angular
power
spect
ra
C
(
⌫
,
⌫
)
corresponding
t
o
each
angular power spect ra Ca;l
, ⌫) corresponding t o each
a;l ( ⌫
2000 basis funct ions P (k ) (using Equat ion (A2))
of
of tthe
he basis funct ions Paa0(k ) (using Equat ion (A2))
− 4.5
−4
†† ††
4
C̃
(34)
C̃aa == R
RB
BC
C21,a
B R
R ..
(34)− 6.0
21,aB
0
0
50
100
150
200
250
300
350
400
pract
m ing
practice
ice explicit
explicitly
ly calculat
calculat
ing tthe
he C̃
C̃aa tthis
his way
way is
is
In
In
comput
at ionally expensive,
we
ead
use
Mont
ecomput
expensive,
we inst
inst
eadrum
useforaaall
Mont
eF i gu
r e 1. atTionally
he Signal-t
o-Foreground
spect
m -modes.
Carlo
t
echnique.
We
can
form
t
he
est
imat
or
q̂
=
a
t echnique.
We can form t he est imat or q̂a =
WeCarlo
have
−− 11plot t ed
−− 11log10 λ i m , where for each m t he eigenvalues have
Shaw
al.
been
sort
ed
by
in
ascending
order
hus
t heretet
no
x̃x̃††C̃
C̃
C̃
x̃
,
which
tthe
hat
ititphysical
ss2013
covariC̃ C̃aa C̃ x̃ , which has
has
he(tproperty
property
tis
hat
covari-int erpret
at
ion
t
o
t
he
vert
ical
direct
ion).
T
he
cont
ours
are
drawn
at − 4,
ance
ance hq̂
hq̂aaq̂q̂bbii −− hq̂
hq̂a ii hq̂
hq̂bii == FFab
(Padmanabhan et
et al.
al. 2003).
2003).
ab (Padmanabhan
− 2,T his
0, 2 means
and 4. we acan best imat
T his means we can est imatee tthe
he FFab
by averaging
averaging over
over
ab by
realisat
realisations
ions of
of x̃x̃.. For
For det
details
ails see
see Dillon
Dillon et
et al.
al. (2012).
(2012).
In
we
tthe
errors
for
two
(Wilson
et al.33 2009,
er 6).spect
To rum
ext end
t his
In Figure
Figure
we plot
plotchapt
he power
power
spect
rum
errors
foro↵twoaxis
cases:
in
t
he
presence
of
foregrounds
t
hat
have
been
in e
t he
of area
foregrounds
hat have giving
been
wecases:
modulat
bypresence
project ed
of t he t elescope
cleaned
hout
cleaned using
using our
our met
met
hod and
and wit
wit◆
hout foregrounds
foregroundsat
at all.
all.
✓hod
In
t
he
case
wit
hout
foregrounds,
F
=
0
and
we
only
perW
In t he
case wit hout
2 foregrounds, F = 0 and we only perA 2tthe
(he
n̂ )final
= sinc
⇡ n̂ · û ttransform
⇥( n̂ · ẑ)
· ẑ
(29)
form
K
ttoo n̂
diagonalise
form
final
Karhunen-Loève
arhunen-Loève
ransform
diagonalise
λ
tthe
he signal
signal and
and inst
instrument
rumental
al noise.
noise. For
For tthe
he foregrounds
foregrounds
CHIME Fact Sheet
Full CHIME Layout
Structure
5 cylinders, 100m x 20m each
Bandwidth
400-800 MHz
Number Feeds/cylinder
Frequency Channels
256 dual pol feeds per cylinder
(2560 digitizers total)
512 frequency channels, 781 kHz wide (1.28 μs)
Data Rate
2NFEEDS x 3.2 Gbit/s = 8 TeraBit/s
Observing Frequency
400 MHz
to
Wavelength
75 cm
37 cm
21cm Redshift
z=2.5 (11 Gyr ago)
z=0.8 (7 Gyr ago)
Beam Size
0.52o
0.26o
E-W FoV
2.5o
1.3o
-45o to +135o (max possible)
0o to +90o (more likely)
N-S FoV
Time/pixel/day
Receiver Noise Temperature
Flux Conversion
Daily Sensitivity
Final Survey
800 MHz
10min, 14min, 24hrs
5min, 7min, 24hrs
equator, 45deg, ncp
equator, 45deg, ncp
50k
~2K / Jy
~50 μJy / pixel
~1.5 μJy/pixel
(Approximate – for planning purposes only)
Digitize 8bits at 800 MSPS
~31cm spacing
(for cosmology, you can channelize
further!)
(assumes 4bit truncation)
CHIME Pathfinder output
• 256(256+1)/2 correlations
– Raw 3.2Gb/s per Antenna
– Integrated to ~1ms for RFI flagging, calibration, 2
Gb/s/frequency.
– Further integrated to ~30s snapshots
• Sent off site to Compute Canada
• 36Mb/s data
CHIME RFI
• Horizon pointed dedicated RFI monitors
DRAO RF Environment
Pulsar Calibration
• Pulsar Holography to map out Telescope
Beams
– Track with single dish and correlate
CalibraBon:&Pulsar&Holography&
• Pulsar Polarization Calibration
– Need a
• Track pulsars with large dish
• Correlate each feed against this signal
Noise
Injection/Unpolarized
• Map
primary beams & sidelobes known
DigiBzer&/&
Correlator&
source
10m&/&26m&
tracking&
dish&
Pulsar
Monitoring/Searching/Timing
• CHIME similar sensitivity to current instruments
• But can search ~half the sky every day
• Significant backend investment
• Watch know pulsars every day for ~10min
Radio Transients
• RRATs
– Transient radio bursts
• Lorimer Bursts
• Use ‘standard’ Pulsar
techniques on formed beam
• “easy” to send off ~1 formed
beam for processing at a time
Lorimer et
al. 2007
Summary
• Measure Baryon Acoustic Oscillations
(BAO) from redshift 0.8-2.5
• Measure Pulsars
• Measure Transients
The
Canadian Hydrogen Intensity
Mapping Experiment
UBC
•
•
•
•
•
•
•
Mandana Amiri
Greg Davis
Meiling Deng
Mark Halpern
Gary Hinshaw
Kris Sigurdson
Mike Sitwell
McGill U
•
•
•
•
•
•
Kevin Bandura
Jean-Francois Cliche
Matt Dobbs
Adam Gilbert
David Hanna
Juan Mena Parra
•
•
•
•
•
•
•
•
U Toronto / CITA
Dick Bond
Ue-li Pen
Richard Shaw
Keith Vanderlinde
Ivan Padilla
HIA, DRAO
Tom Landecker