BINGO simulations BAO from Integrated Neutral Gas Observations - M.-A. Bigot-Sazy
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
BINGO simulations
BAO from Integrated Neutral Gas Observations
!
M.-A. Bigot-Sazy
!
BINGO collaboration
!
!
!
!
!
!
!
Specialist intensity mapping meeting
Friday 9th May 2014Table of contents
• Instrument simulation & Foregrounds
component separation
• New possibility of the BINGO experiment
with the new design of the telescope &
Sensitivity of BINGO experiment
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 2Challenges with intensity mapping data
foreground component separation (intensity up to five orders of magnitude higher than the HI
signal)
Brightness temperature of the 21cm signal T = 20 mK
! Galactic Synchrotron emission T = 200 to 1000 K
-principal component analysis
-power law fitting 100
Beam profiles
GRASP simulations
noise
Bruno Maffei
-
Beams at 1400 MHz
uncorrelated noise white noise
-
10-2
correlated noise in time and frequency 1/f noise
- atmospheric 1/f noise
10-4
dB
systematics effects
- sidelobes: near, intermediate, far (mode mixing)
- bandpass calibration
10-6
- cross polarisation
-
Central pixel
beam ellipticity …. 10-8
Edge pixel
-200 -100 0 100 200
Angles (deg.)
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 3Horns in receiver plane
10
Simulated observations 5
Y (meters)
56 receivers
0
!
a resolution of 40 arcmin (at λ = 30 cm)
-5
21 frequency channels of 15 MHz BW (0.13 < z < 0.48)
-10
-10 -5 0
X (meters)
5 10
one year of integration time (about 2 years of data)
!
drift scan at -5 deg of declination
!
Sky - Synchrotron (Remazeilles et al. 2014 in prep.)
extrapolated to small scales
⇣ ⌫ ⌘
! Tsignal (⌫, n̂) = T408 (n̂)
! 408MHz with = 3
- 21cm signal using gaussian field with the approximation
of the flat field power spectrum ΔT ~ 1 mK
!
- Diffuse point sources (Santos et al. 2008)
Simulate of the contribution to each sky pixel independently (uncorrelated point
sources)
Coverage for 1 year of observation
Sources of Smax > 10 mJy removed
The brightness of each point source is randomly drawn from the source count
✓ ◆ 1.75
distribution (Di Matteo et al. 2012) dn 1 1 S
= (4.0 mJy sr )
dS 880 mJy
⌫ ↵
The spectral dependence of each point source is given S(⌫) = S(⌫⇤ ) ⌫⇤ = 150
⌫⇤ ↵ = 2.7
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 4Simulated observations
56 receivers
21cm signal at 1008 MHz
! (convolved with the beam)
a resolution of 40 arcmin (at λ = 30 cm)
Mask out the galactic latitude |b| > 20
21 frequency channels of 15 MHz BW (0.13 < z < 0.48)
one year of integration time (about 2 years of data)
!
drift scan at -5 deg of declination
!
Sky - Synchrotron (Remazeilles et al. 2014 in prep.)
extrapolated to small scales
Observed map at 1008 MHz
⇣ ⌫ ⌘ From map making
! Tsignal (⌫, n̂) = T408 (n̂) technics
! 408MHz with = 3
- 21cm signal using gaussian field with the approximation
of the flat field power spectrum ΔT ~ 1 mK
!
- Diffuse point sources (Santos et al. 2008)
Simulate of the contribution to each sky pixel independently (uncorrelated point
sources)
Sources of Smax > 10 mJy removed
The brightness of each point source is randomly drawn from
Point sources at 1008 MHz
(Di Matteo et al. 2012)
✓ ◆ 1.75
dn 1 1 S
= (4.0 mJy sr )
dS 880 mJy
⌫ ↵
The spectral dependence of each point source is given S(⌫) = S(⌫⇤ ) ⌫⇤ = 150
⌫⇤ ↵ = 2.7
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 5Instrumental noise
a gaussian process = sum of white noise component and correlated noise (1/f
noise)
!
Total noise described by a power spectral density [V2]
fknee ↵ 2 ↵ slope index
P (⌫) = [1 + ] fknee knee frequency
fs f
Arbitrary values
fs sampling frequency
1/f noise : knee frequency = 1.0e-3 Hz
need
slope index = 1
measurements
sampling period = 20 min
from the
correlation length = 10 min receivers
thermal noise : system temperature = 50 K
1/f noise at 975 MHz
Thermal noise at 975 MHz
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 6Time ordered data to map (Maximum likelihood mapmakers)
The receiver measures the temperature of the sky in a given direction through an instrumental beam
map
noise N = hnnT i
Model dt = Atp sp + nt
noise covariance matrix
Time Ordered Data
Pointing matrix
!Simplest solution given by the maximum likelihood estimate (AT A)ŝ = AT d
- average the data falling into each pixel without weighting them
- neglect the effects of the correlation of the noise (low frequency drifts can induce stripes in the maps along the scans)
1~
Optimal solution (noise model) (AT N 1
A)ŝ = AT N d
Hard to invert the pixel domain noise covariance matrix (AT N 1
A)
use of Preconditioned conjugate gradients method
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 7Foreground removal amplitude of the eigenvalues
eigenvalues are higher for the first
one of the primary technical challenge of 21cm values of the modes
mapping experiment
!
- foreground emissions smooth in frequency
!
- BAO signal uncorrelated in frequency
!
Two methods : Principal Component Analysis, Power
law fitting
Principal Component Analysis
Staking of the independent frequency maps into
amplitude of the eigenvectors
orthogonal modes according to the covariance in modes look like polynomials
!
frequency
!
each mode = a fraction of the total variance of the
data set
!
foreground emissions dominate the overall variance of
the sky = contain in the modes of the highest variance
!
foreground spectra do not contain any sharp
features
!
foregrounds well described by a small number of
eigenforegrounds
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 8Foreground removal
Principal Component Analysis Observed sky after
- subtract the mean of the simulated data
applying PCA with 1
mode removed
- compute the eigenvectors and eigenvalues of the covariance matrix
- find the principal components and remove these components in the
frequency space
Power law fitting
⌫
T (n̂, ⌫) = A(n̂) ⌫0
A, fitted values
⌫0 = 408 MHz
Observed sky after
With thermal noise and1/f noise applying PCA with 6
2 modes removed
Power spectra [mK ] from the maps at 975 MHz
Observed sky after
applying PCA with 8
modes removed
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 9Foreground removal
With Observed sky after
applying PCA with 1
thermal mode removed
noise and1/f
noise 21cm signal (convolved with the
beam)
Observed sky after
2 applying PCA with 8
Power spectra [mK ] from the maps at 975 MHz modes removed
Observed sky after
applying power law
fitting
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 10Foreground removal Residuals from PCA method with only thermal noise
Difference between the 21cm signal and the observed maps
Conclusion
!
!
- PCA method allows to subtract about 5
orders of magnitude of foreground signal
! - But simplest simulation !
With only thermal noise
-
6 modes is the optimal number of modes to
remove (more modes remove some signal)
-
l>100 reach the thermal noise level
!
With 1/f noise
-
8 modes is the optimal number of modes to
remove
-
PCA method removes some 1/f noise
!
To follow
!
- synchrotron emission with a curvature of the
spectral index
!
Try different methods for foregrounds removal :
parametric fitting, generalisation of ILC method,
wavelet based methods ...
!
- systematic effects
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 11Foreground removal Residuals from PCA method with 1/f noise
Conclusion
!
!
- PCA method allows to subtract about 5
orders of magnitude of foreground signal
! - But simplest simulation !
With only thermal noise
-
6 modes is the optimal number of modes to
remove (more modes remove some signal)
-
l>100 reach the thermal noise level
!
With 1/f noise
-
8 modes is the optimal number of modes to
remove
-
PCA method removes some 1/f noise
!
To follow
!
- synchrotron emission with a curvature of the
spectral index
!
Try different methods for foregrounds removal :
parametric fitting, generalisation of ILC method,
wavelet based methods ...
!
- systematic effects
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 12Table of contents
• Instrument simulation & Foregrounds
component separation
• New possibility of the BINGO experiment
with the new design of the telescope &
Sensitivity of BINGO experiment
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 13New possibility of the experiment
Two different designs
50 horns - 10 deg x 200 deg field of view Uncertainty on the 3D HI power spectrum ?
Projected errors on the power spectrum ?
70 horns - 15 deg x 200 deg field of view
3D HI Power spectrum
3
2 2 2 k Pcdm (k, z)
[ THI ] = T̄ (z) [b(k, z)]
2⇡ 2
b(k, z) = 1
Mean brightness temperature
(we assume Planck cosmology)
✓ ◆
⌦HI (z)h (1 + z)2
T̄obs (z) = 44µK
2.45 ⇥ 10 4 E(z)
We consider one central redshift z = 0.28
Projected error on the power spectrum measurement
s !
2
P (2⇡)3 1 pix Vpix
= 2 1+
P Vsur 4⇡k 2 k [T̄ (z)]2 W (k)2 P Seo et al. 2010
cosmic variance
- 0.13 < z < 0.28
- FWHM = 40 arcmin at 1000 MHz
- system temperature = 50 K
- integration time = 1 year
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 14New possibility of the experiment
!
!
!
increasing the number of receivers from
50 to 70 and the field of view from 10 to 15
deg allow to decrease the expected errors by a
factor of 16%
Uncertainty on the measurement of
the acoustic scale (Seo et al. 2010)
fit a decaying sinusoidal function to the data
in order to deduce a best fit of the oscillation
wave scale
P (k) k 1.4 2 k k = 0.016 Mpc 1
= 1 + Ak exp[ ( 1) ]sin( ).
Pref 0.1h Mpc kA
2.4 % for the considered designs
kA /kA ⇠ 1.93%
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 15Conclusion
We present a simulation for the BINGO experiment which consists of 21 frequency channels
from 960 to 1260 MHz, 56 pixels and one year of integration time.
Two different methods of foreground removal : PCA and power law fitting
PCA can remove foregrounds with a good accuracy.
The largest variance mode map is dominated by galactic foregrounds and point sources.
PCA allows to remove the 1/f noise.
Analyse of two different designs with nf = 50 or 70 and FOV = 10 deg x 200 deg or 15 deg x 200
deg.
Accuracy on the acoustic scale
kA /kA ⇠ 2.4 %
1.93%
With nf = 70 and FOV = 15 deg x 200 deg allows to decrease the projected errors on the 3D HI
power spectrum of a factor of 16% compared to the original design.
Thank you for your attention
Specialist intensity mapping meeting
Friday 9th May 2014
BINGO collaboration 16You can also read
