Download presentation
Presentation is loading. Please wait.
Published byRolf Carroll Modified over 7 years ago
1
A tunable weighting scheme for imaging extended sources
Ryan A. Loomis (Harvard-Smithsonian CfA) Collaborators: K. Öberg, E. Price, S. Andrews, and D. Wilner Harvard-Smithsonian Center for Astrophysics
2
Motivation ALMA Partnership + 2015
----- Meeting Notes (10/6/16 14:59) ----- maybe even explicitly mention brigg's thesis chapter (e.g.) ALMA Partnership Ryan A. Loomis CALIM 2016
3
Motivation Andrews + 2016 ----- Meeting Notes (10/6/16 14:59) -----
maybe even explicitly mention brigg's thesis chapter (e.g.) Andrews Ryan A. Loomis CALIM 2016
4
Motivation 50 AU Loomis + 2016 (in prep)
----- Meeting Notes (10/6/16 14:59) ----- maybe even explicitly mention brigg's thesis chapter (e.g.) Loomis (in prep) Ryan A. Loomis CALIM 2016
5
Motivation Many resolved sources show sub-structure
SNR limitations may prevent imaging this sub-structure Tapering increases SNR, but isn’t ideal, can wash out features ----- Meeting Notes (10/6/16 14:59) ----- maybe even explicitly mention brigg's thesis chapter (e.g.) Ansdell Ryan A. Loomis CALIM 2016
6
Motivation Uniform weighting SNR robust v2 (FOV x3) SNR = 49 SNR = 87
Ryan A. Loomis CALIM 2016
7
Definitions Weights: Briggs 1995 Ryan A. Loomis CALIM 2016
8
Definitions Weights: Taper weights Briggs 1995 Ryan A. Loomis
CALIM 2016
9
Definitions Weights: Taper weights Density weights Briggs 1995
Ryan A. Loomis CALIM 2016
10
Definitions Weights: Taper weights Density weights SNR weights = 1/σk2
Briggs 1995 Ryan A. Loomis CALIM 2016
11
Definitions Weights: Dirty image: Taper weights Density weights
SNR weights = 1/σk2 Dirty image: Briggs 1995 Ryan A. Loomis CALIM 2016
12
Definitions Weights: Dirty image: Variance of the dirty image:
Taper weights Density weights SNR weights = 1/σk2 Dirty image: Variance of the dirty image: Briggs 1995 Ryan A. Loomis CALIM 2016
13
Definitions Weights: Dirty image: Variance of the dirty image:
Taper weights Density weights SNR weights = 1/σk2 Dirty image: Variance of the dirty image: RMS thermal noise: Briggs 1995 Ryan A. Loomis CALIM 2016
14
Natural weighting Minimize (ΔID)2 using Lagrangian multiplier:
Thermal noise is lowest if Tk and Dk = 1, as wk = 1/σk2 Briggs 1995 Ryan A. Loomis CALIM 2016
15
Thermal noise and SNR Normalized RMS noise:
SNR at a given point in the image: ----- Meeting Notes (10/6/16 14:51) ----- the weights for natural weighting encode no information Briggs 1995 Ryan A. Loomis CALIM 2016
16
‘Natural’ weighting for a resolved source
Natural weighting is only ideal for a point source Maximize SNR for an arbitrary source: Vk TkDkwk /σk2 becomes: TkDkwk Re(Vk)/σk2 Gaussian tapering increases SNR of faint sources because the taper approximates a matched filter TkDkwk |Vk|/σk2 Briggs 1995 Ryan A. Loomis CALIM 2016
17
SNR natural weighting vs tapering
FFT Ryan A. Loomis CALIM 2016
18
SNR natural weighting vs tapering
FFT Ryan A. Loomis CALIM 2016
19
SNR natural weighting vs tapering
FFT Taper will damp out high frequencies Ryan A. Loomis CALIM 2016
20
SNR natural weighting vs tapering
FFT Ryan A. Loomis CALIM 2016
21
SNR natural weighting vs tapering
Gaussian taper Ryan A. Loomis CALIM 2016
22
SNR natural weighting vs tapering
Ryan A. Loomis CALIM 2016
23
SNR natural weighting vs tapering
FFT + noise FFT Ryan A. Loomis CALIM 2016
24
SNR natural weighting vs tapering
Matched taper Ryan A. Loomis CALIM 2016
25
Robust weighting Find a smooth knob between natural and uniform weighting through minimization Briggs 1995 Ryan A. Loomis CALIM 2016
26
Robust weighting Briggs 1995 Ryan A. Loomis CALIM 2016
27
Robust weighting Briggs 1995 Ryan A. Loomis CALIM 2016
28
Robust weighting Briggs 1995 Ryan A. Loomis CALIM 2016
29
A tunable SNR weighting
Can we find a way to connect SNR natural weighting with uniform weighting? Minimize: ||ID||2 + ΔID / ID Find that: Wpq /(S2 + 2σpq/|Vpq|) Analogous to robust weighting, but tunes between uniform and SNR natural weighting Ryan A. Loomis CALIM 2016
30
Test case - ring Robust Gaussian taper SNR robust Ryan A. Loomis
CALIM 2016
31
Test case - ring Robust Gaussian taper SNR robust Ryan A. Loomis
CALIM 2016
32
Test case - ring Robust weighting Gaussian taper SNR robust
Ryan A. Loomis CALIM 2016
33
Test case - ring Robust Gaussian taper SNR robust Ryan A. Loomis
CALIM 2016
34
Test case - disk Robust Gaussian taper SNR robust Ryan A. Loomis
CALIM 2016
35
Test case - disk Robust weighting Gaussian taper SNR robust
Ryan A. Loomis CALIM 2016
36
Test case - disk Robust Gaussian taper SNR robust Ryan A. Loomis
CALIM 2016
37
Test case – Gaussian ring
Robust Gaussian taper SNR robust Ryan A. Loomis CALIM 2016
38
Test case – Gaussian ring
Robust weighting Gaussian taper SNR robust Ryan A. Loomis CALIM 2016
39
Test case – Gaussian ring
Robust Gaussian taper SNR robust Ryan A. Loomis CALIM 2016
40
Investigating Dk - ring
Robust weighting Gaussian taper SNR robust Ryan A. Loomis CALIM 2016
41
Investigating Dk – Gauss. ring
Robust weighting Gaussian taper SNR robust Ryan A. Loomis CALIM 2016
42
New formulation Previously had: Wpq 1/(S2 + 2σpq/|Vpq|)
Can we prevent the fast damping of long baselines? Previously had: Wpq /(S2 + 2σpq/|Vpq|) What about: Wpq /(S2 + 2σpq/|Vpq|S /S ) This is just a hunch – I don’t know the proper form or minimization condition. Ideas? 2 2 max Ryan A. Loomis CALIM 2016
43
Comparison - ring Robust Gaussian taper SNR robust SNR robust v2
Ryan A. Loomis CALIM 2016
44
Comparison - ring SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
45
Comparison - ring Robust Gaussian taper SNR robust SNR robust v2
Ryan A. Loomis CALIM 2016
46
Comparison – Gaussian ring
Robust Gaussian taper SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
47
Comparison – Gaussian ring
SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
48
Comparison – Gaussian ring
Robust Gaussian taper SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
49
Synergy with super-uniform weighting
Robust Gaussian taper SNR robust SNR robust v2 SNR robust v2 (2x FOV) SNR robust v2 (3x FOV) Ryan A. Loomis CALIM 2016
50
Disk with multiple rings
Robust Gaussian taper SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
51
Disk with multiple rings
SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
52
Disk with multiple rings
Robust Gaussian taper SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
53
Disk with multiple rings (imaged)
Dirty image SNR = 4 Gaussian Taper SNR robust v2 SNR = 5 SNR = 5.5 Ryan A. Loomis CALIM 2016
54
Real data example – AA Tau
Robust Gaussian taper SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
55
Real data example – AA Tau
SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
56
Real data example – AA Tau
Robust Gaussian taper SNR robust SNR robust v2 Ryan A. Loomis CALIM 2016
57
Realistic example – disks with rings
Uniform weighting SNR robust v2 (FOV x3) SNR = 49 SNR = 87 Ryan A. Loomis CALIM 2016
58
SNR weighting – potential limitations
Beam pattern is more complicated, no longer close to a Gaussian; nasty shelves Multi-source and asymmetric sources? Deconvolution in high dynamic range cases likely more difficult due to shelves Potentially problematic for line imaging ----- Meeting Notes (10/6/16 14:51) ----- potentially problematic for line imaging Ryan A. Loomis CALIM 2016
59
Summary and future SNR weighting provides clear benefits over Gaussian tapering in the low SNR regime Our tunable weighting scheme can often outperform robust weighting alone What is the correct metric to minimize? Implement for CASA? ----- Meeting Notes (10/6/16 14:51) ----- remove rings Ryan A. Loomis CALIM 2016
Similar presentations
© 2025 SlidePlayer.com Inc.
All rights reserved.