Foreground subtraction in radio surveys

Transcription

Foreground subtraction in radio surveys
Foreground subtraction in radio surveys:
intensity mapping and the EoR
1
Aeff S [ ]1 / 2
snr  2
for each polarization
kTsys
For SKA,  rms 
100 mJy
, two polarizations
[ ]1 / 2
Filipe B. Abdalla,
Emma Chapman,
Laura Wolz,
University College
London
+ collaborators inc,
Chris Blake,
Richard Shaw,
LOFAR-EoR team
Outline
• Brief background of why to study 21cm
• Foreground subtraction methods overview
• Application to intensity mapping simulations and
BAO
• Application to EoR data from LOFAR-EoR.
The History
of our Universe.
• We have a extremely
well measured CMB
sky.
• The EoR and the dark
ages remain a mystery
and unmeasured
territory.
• We are a long way
measuring the LSS of
the Universe near us.
• The Galaxy will pose a
problem into looking at
these fluctuations as
with the CMB
21 cm Observations:
Emission at small scales
z=18.3
z=16.1
z=14.5
z=13.2
z=12.1
z=11.2
z=10.4
z=9.8
z=9.2
z=8.7
z=8.3
z=7.9
z=7.5
z=7.2
10 Mpc comoving
credit: Furlanetto et al. (2003)
D=0.1 MHz
Predicted BAO constraints –
Would like a confirmation from the Radio,
with different systematic effects
Uses public code to estimate errors
from BAO measurements from Seo &
Eisenstein (2007: astro-ph/0701079)
See Chimes + Bingo talks
by Shaw and Dickinson
For full details...
Intensity mapping results to date:
• Cross correlations with optical
galaxies (Masui et al.)
• Auto correlations also detected
(Switzer et al.)
• Projects are planning to map this
better in the advent of the SKA:
Chimes, Tianlai, Bingo...
• The SKA could (depending on
configuration) do a huge amount of
such studies.
• These data have shown this is a
possible experiment but foreground
subtraction is important.
Current Results - EoR
Dillon et al 2013
MWA
Parsons et al 2013
Best results from PAPER give
A limit on the variance to be
~300mK
When the signal is expected to
be round ~5-10 mK
GMRT results
Foreground subtraction techniques
intensity mapping + EoR
FG simulations
(~29 %)
(~ 1 %,  ~ -2.15)
(~70 %, Tb ~ 240±2 K,  ~ -2.55)
Jelic et al., 2008, MNRAS
Simulations all sky by R. Shaw.
Problem Outline:
Spectral smoothness allows separation of 21cm. Options:
1 Fit power law to maps
2 Remove low order polynomials or some constraint fit (Harker et al.)
3 Measure components and model (Liu and Tegmark)
4 Measure modes of the foregrounds from a given FG model (R. Shaw 2013)
5 Model independent methods (Chapman et al 2012a,2012b)
Issues:
- Mode mixing of angular and frequency fluctuations by
frequency-dependent beams (esp. interferometers) [1, 2] method [2] does better in
fourrier space.
- Robustness Biasing introduced if foreground model poorly
understood (esp. non-gaussianities). [1, 3]
- Statistical Optimality Need to keep track of transformations
on statistics, for optimal PS estimation [1, 2]
- Model Dependent [4] although excellent results.
Slightly different issues than with CMB:
- UV coverage is crucial
- Side lobe noise can add to standard noise from the observations -> cannot be removed
- Ratio of foreground strength to signal strength is much larger
- Many other unused techniques available from the CMB community.
We put together 4 different pipelines for the
Results on simulations:
data presented later.
Power spectrum and maps
recovered...
With a foreground signal ~10^4
larger than the signal.
(left) simulation w/o FG subtraction
(right) simulations w FG subtraction
GMCA, Fast ICA, polynomial fits,
wp-smoothing fits.
• GMCA Find X where
Information maximisation:
Wavelet decomposition in multi-scales
Sparsity -----------------------------------------------------> solve:
Simulations on intensity mapping data
Intensity mapping pipeline:
Intensity mapping:
systematic plots from foreground subtraction
• Foreground subtraction if all sky
were available!
• Residuals are of the order of
0.1mK.
• If all sky available sims, with
proper masks, show biases at
scales of l~20, which means ~
10 degs scales. -> implications
on science
• Use a selected area for
analysis...
• Completely blind method,
assumes NOTHING about
foregrounds... Can be improved
Intensity mapping
BAOs
• Theoretical Model of the
power spectrum Cl
• There is a ~ couple of
sigma shift if we include
all shape informations
• BAO fit is not biased.
(Wolz et al.2014)
Foreground subtraction on LOFAR data
Cycle0 – observations
(Dec 2012 – Nov 2013)
Nighttime: 8-16h syntheses
Frequencies: 115-190 MHz
Resolution:
2s, 3.2 kHz
Raw data volume:
LOFAR spectral capabilities:
8-bit mode  488 subbands
1 subband = 0.195 MHz (64ch)
30 - 60 TB / night !!
25 x 8h
on 3C196
20 x (8-16h) on NCP
~ 200 h
~ 240 h
de Bruyn- LeidenMiley- symposium
96 MHz total bandwidth
1st stage processing completed (RFIflagging, averaging,
initial calibration)
Foreground separation on the data:
It works well...
3C61.1
Dec + 86o
8o x 8o
Foreground pipeline by Chapman, Abdalla, et al.
Confusion limited low resolution image
Current limits up to now: by Zaroubi et al.
On behalf of the EoR team at Gerfest
•
Some improvements can be done
from here:
– Add more data which we have
available
– Improve the foreground separation
in some small data related issues:
beam modelling
– Include on longer timescale
modelled sources outside the
beam to improve calibration
– One tricky bright source needs
international baseline data for
better calibration
•
Preliminary, LOFAR-EoR will
produce a final paper which should
be submitted shortly.
Parsons et al 2013
Conclusions…
• 21cm is a very rich area of research LSS + EoR
• Investigation of the effects of foreground
subtraction are crucial in these areas
• Intensity mapping data can be robust to
foreground subtraction for BAO measurements
• EoR data/sims can be robust to detection given
foreground subtraction.
• Competitive limits start emerging from
LOFAR-EoR with real data.
• More to come!
J. W. Award 2012
THE END