boloSource()

boloSource()

Techs

Motivation

• Extended emission analysis requires clean maps
• Compact objects contribute to the image power spectra with a significant power at a broad range of spatial frequencies:
• modify the image at frequencies comparable to the beam-size
• depending on the surface density and the clustering strength, lower spatial frequencies are contaminated with a smaller power density but typically at large bandwidth

Motivation

• Image analysis techniques are difficult to compare if sources are not subtracted, because their sensitivity to discrete sub-structures may be quite different
• techniques using sparsity information could be disturbed by even a few point sources
• techniques analyzing full intensity maps are more sensitive to clustering
• For extended emission analysis we need a technique to subtract sources that fall within a well defined range of spatial frequencies

• boloSource() input:
• source list (centroid coodinates)
• input L2 mask OR source cutoff frequency
• PACS OBSID pair

• boloSource() output:
• PACS L1 cubes with interpolated timeline

• compatible with any projection algorithm that can use PACS L1 frames cube structure

• Full code in Jython, will be available in build HIPE v10.0+

Motivation

• A major requirement: preserve noise properties of the image!

• Classical way: try modeling the source intensity I(x,y) in the position-position space and subtract from the image

• This is not easy, but one could reduce the problem to 1D in the detector timeline

• Subtract sources from the detector timeline and re-project the image

Original image

Extra motivation

getsources background

image

Motivation

Clean maps for analysis of diffuse background

A new source subtraction technique for extended emission analysis

Workflow

boloSource()

Source subtraction

in the timeline

• Reduced dimensionality in I(t) vs. I(x,y), but noise spectrum in timeline is more complex:

• Lower S/N in L1 timeline than in L2 reconstructed map
• Looks difficult but we mainly interested in to subtract high-frequency components. In the masked part of the timeline one could interpolate with simulated noise + sky background:

Interpolated intensity

in masked timeline

Simulated noise

Baseline estimate from data

Level 2 mask

aperture

source

• source centroid & size
• create L2 mask for circular aperture

• aperture radius defines the cutoff frequency
• backproject L2(x,y) mask to L1(t) mask

• L1 mask size depends on L2 mask size AND scan speed

Concept

Source substraction in the timeline

Perspectives

• Benchmarking by using simulated sources back-projection onto the L1 detector timeline
• Photometry on differential detection images
• Adaptive low-frequency cutoff (i.e. # of readouts) to limit maximum L1 mask size for any given source in any given pixel timeline
• Backproject flux in order to simplify noise spectrum on the L1 timeline
• Try it for SPIRE

Noise simulation

• The objective is to simulate intensity distribution of a single scan- leg with similar noise power spectrum as we experience in the observed data
• The simulation is non-parametric, i.e. one could not invert a noise power spectrum into an intensity distribution
• Monte-Carlo simulation in the wavelet space:
• Measure STDDEV of outlier filtered wavelet coefficients after last iteration
• Get random noise and scale each wavelet frequency coefficients to the measured STDDEV

Baseline estimation

• Take L2 mask size (i.e. effective diameter of input mask aperture) for a given source
• Decompose signal with SWT
• Flag high-frequency outliers above a given sigma level on the entire scan leg
• Interpolate flagged outliers with baseline estimate
• Solve inverse transform, reproduce intensities
• Create output products:
• Baseline
• Noise dataset
• Outlier Mask

Timeline iterpolation

Intermediate working mask

Optimized adaptive mask

Back projected input mask

L1 bolometer signal

Interpolated signal

Perspectives

getsources background image

• Optimize L1 mask size and location through the adaptive mask option:
• Extend the L1 mask to 2x larger working mask and find a maximum:
• defined by the baseline estimator outlier mask
• or if outlier mask is empty within working mask then try to find local peak of at least 3 readouts broad
• Find the 1st and last readouts below the baseline around the peak
• This section defines the final adaptive mask

boloSource() background image

Extra motivation

getsources background

image

Original image

Timeline iterpolation

boloSource subtracted image

• Take the previously identified baseline and add simulated noise components beyond the cutoff frequency within the section of L1 adaptive mask
• Do bit-rounding

Photometry

Original image

HI-GAL F297_0 "red" map

Photometry, SPIRE

getsources flux:

0.158 Jy

Implementation

boloSource() flux on differencial image:

0.220 Jy

difference: 39.84%

Photometry

Interpolation and noise simulation

getsources flux:

8.65 Jy

boloSource() flux on differencial image:

8.92 Jy

Photometry

difference: 3.19%

HI-GAL F297_0 "blue" map

getsources flux:

1.27 Jy

boloSource() flux on differencial image:

1.18 Jy

difference: -6.76%

HI-GAL F297_0 "red" map

Original

Interpolated

First results

HI-GAL Field 297_0

• Gábor Marton, Konkoly Observatory, HUNGARY
• Roland Vavrek, ESA HSC
• Csaba Kiss, Konkoly Observatory
• Davide Elia, IFSI