ALMA Future Sc. Program Development Workshop Aug. 2016, - - PowerPoint PPT Presentation

alma future sc program development workshop
SMART_READER_LITE
LIVE PREVIEW

ALMA Future Sc. Program Development Workshop Aug. 2016, - - PowerPoint PPT Presentation

ALMA Future Sc. Program Development Workshop Aug. 2016, Charlottesville Wide-fjeld Wide-band Full-polarization Imaging: Impact of Direction Dependent PB efgects S. Bhatnagar, U. Rau, P . Jagannathan, K. Kundart S. Bhatnagar: ALMA Future


slide-1
SLIDE 1
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

1/20

ALMA Future Sc. Program Development Workshop

  • Aug. 2016, Charlottesville

Wide-fjeld Wide-band Full-polarization Imaging:

Impact of Direction Dependent PB efgects

  • S. Bhatnagar, U. Rau, P

. Jagannathan, K. Kundart

slide-2
SLIDE 2
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

2/20

What is wide-fjeld?

“True” sky

slide-3
SLIDE 3
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

3/20

What do we call wide-band?

  • When fractional signal bandwidth used for imaging > ~20%
  • Plus source spectral index >= -1.0
  • Plus target dynamic range > 1000
  • Spectral efgects for higher source spectral index will become

signifjcant at lower bandwidth ratios

  • Empirical Dynamic range :
  • Spectral line imaging, by defjnition, does not require wide-band imaging

algorithms Iα 100

S( ν)∝( ν/ νo)

−0.7

slide-4
SLIDE 4
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

4/20

Wide-band Wide-fjeld Imaging

  • Wide band data to image beyond the ~50% point of the PB at a

reference frequency

  • Bandwidth ratio > ~20%
  • FoV > ~HPBW @ reference frequency
  • Variable PB:
  • Long integration (rotation), Mosaicking (pointings at

difgerent PA), in-beam polarization is large (AA)

PB “Spectral Index”

PB Frequency dependence (blue curve)

slide-5
SLIDE 5
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

5/20

Wide-band Wide-fjeld Imaging

  • Characterization of the (WB) A-Projection + MT-MFS

MFS+SI MT-MFS+SI MT-MFS+ A-Projection MT-MFS+ WB A-Projection

2016, AJ (accepted) [arXiv:1605.07640]

slide-6
SLIDE 6
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

6/20

Wide-band Wide-fjeld Imaging

  • WB A-Projection + MT-MFS
  • WB A-Projection for PB
  • MT-MFS for sky

(2011, ApJ, 739, L20 [arXiv:1106.2796v2])

slide-7
SLIDE 7
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

7/20

Full-Mueller (Polarization) WF-WB Imaging

  • Direction Dependent (DD) Measurement Eq. (ME) in the

image domain:

  • DD ME in the visibility domain:

I

Obs = [ M ]⋅[ I

  • ]

[

I I

Obs

IQ

Obs

IU

Obs

IV

Obs]

= [ M 11 M 12 M 13 M 14 M 21 M 22 M 23 M 24 M 31 M 32 M 33 M 34 M 41 M 42 M 43 M 44] ⋅[ I I

  • I Q
  • I U
  • I V

0]

  • Diagonal: “pure” poln. products
  • Off-diagonal: Include poln. leakage

where Ai = [ A p A p→ q Aq→ p Aq ] V ij

Obs = [ Ai⊗A j T ] ∗[V ij

  • ] = [ Aij ] ∗[V ij
  • ]

Antenna Aperture Illumination

slide-8
SLIDE 8
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

8/20

DD Efgects in Full-pol. Imaging

  • DD “Mueller” matrix:

Stokes Basis [Jagannathan, Bhatnagar, Rau & Taylor]

  • Affects DR at the 103-4 level
  • PB Stokes-Q, -U is few% of Stokes-I

.

[

I I

  • I Q
  • IU
  • IV

0]

[

I I

Obs

I Q

Obs

I U

Obs

I V

Obs]

=

slide-9
SLIDE 9
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

9/20

DD Efgects in Full-pol. Imaging

  • DD “Mueller” matrix:

Stokes Basis [Jagannathan, Bhatnagar, Rau & Taylor]

  • Affects DR at the 103-4 level
  • PB Stokes-Q, -U is few% of Stokes-I

.

[

I I

  • I Q
  • IU
  • IV

0]

[

I I

Obs

I Q

Obs

I U

Obs

I V

Obs]

=

slide-10
SLIDE 10
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

10/20

Issues in Wide-fjeld Wide-band Full-Pol. Imaging

  • PB Efgects
  • In-beam efgects : DD Leakage
  • Parametric Aperture Illumination model (Holographic measurements

not suffjcient)

  • Pointing Errors
  • Mosaic patterns
  • Variations with frequency
  • Frequency dependence of intrinsic Q and U
  • Frequency dependence due to PB
  • Computing load
  • Fundamentally more expensive : Larger CF support
  • Larger memory footprint: Fundamentally required, any which way you

cut it

slide-11
SLIDE 11
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

11/20

Full-pol. Imaging: In-beam Leakages

  • Leakage (Ofg-diagonal elements of the Mueller matrix)
  • Vary with direction (position in the beam), Parallactic Angle (time) and

frequency

[Jagannathan, Bhatnagar, Rau & Taylor]

Radial Slice for Stokes-I and I->Q Leakage

slide-12
SLIDE 12
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

12/20

Full-pol. Imaging: In-beam Leakages

  • Leakage (Ofg-diagonal elements of the Mueller matrix)
  • Vary with direction (position in the beam), Parallactic Angle (time) and

frequency

[Jagannathan, Bhatnagar, Rau & Taylor]

Radial Slice for Stokes-I and I->Q Leakage

slide-13
SLIDE 13
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

13/20

Full-pol. Imaging: Mosaic Sensitivity Pattern

[Jagannathan, Bhatnagar, Rau & Taylor]

In-beam Stokes-Q pattern for a 11x11 point mosaick

  • Heterogeneous case;

rotation due to PA change also ignored

  • The resulting pattern is

combination of overlapping Clover-leaf pattern of each pointing

  • In-beam DD leakage spreads all

across the mosaicked region.

slide-14
SLIDE 14
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

14/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination
  • Difgerence between Ant6 and Ant10 in “homogeneous array”

R-Beam L-Beam Data: R. Perley Analysis: P.Jagannathan, S.Bhatnagar

In the graph below: Optical effects should be independent of frequency (e.g. Poln. Squint) Mechanical effects should show linear trends (e.g. Antenna pointing errors)

  • Poln. Squint

Antenna pointing error

slide-15
SLIDE 15
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

15/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
slide-16
SLIDE 16
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

16/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
slide-17
SLIDE 17
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

17/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
slide-18
SLIDE 18
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

18/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
slide-19
SLIDE 19
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

19/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
slide-20
SLIDE 20
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

20/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
slide-21
SLIDE 21
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

21/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
slide-22
SLIDE 22
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

22/20

Full-pol. Imaging: PB Efgects

  • Parametric model of antenna Aperture Illumination

[Kundert et al. IEEE Trans. (in review)]

Aperture Illumination

  • Strongest effects
  • Antenna size,Pointing errors
  • Quadrapods
  • Antenna-to-antenna variations
  • Cost equations
  • Cheaper to fix the hardware?
  • Cheaper to handle in post processing?
slide-23
SLIDE 23
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

23/20

Full-pol. Imaging: PB Efgects

  • EVLA Squint
  • Expected: Lateral translation between RR- and LL-beams
  • Measure: Translation + Rotation (~1 deg!)
slide-24
SLIDE 24
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

24/20

Full-pol. Imaging: PB Efgects

  • EVLA Squint
  • Expected: Lateral translation between RR- and LL-beams
  • Measure: Translation + Rotation (~1 deg!)
slide-25
SLIDE 25
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

25/20

Full-pol. Imaging: Q/U vs. Frequency

  • MFS approach to gain from WB sensitivity

[Jagannathan, Bhatnagar, Rau & Taylor]

  • PSF structure scales with frequency
  • Effective PSF amplitude also changes with frequency (Sp. Ndx., PB scaling)
slide-26
SLIDE 26
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

26/20

Full-pol. Imaging: Q/U vs. Frequency

  • MFS approach to gain from WB sensitivity

[Jagannathan, Bhatnagar, Rau & Taylor]

  • PSF structure scales with frequency
  • Effective PSF amplitude also changes with frequency (Sp. Ndx., PB scaling)

PSF(xo)Continuum=∑ν I (xo, ν) PSF(x−xo,ν)

slide-27
SLIDE 27
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

27/20

Full-pol. Imaging: Q/U vs. Frequency

  • Possible to use MT approach on Q and U for low RM cases
  • Polynomial model (a la MT-MFS) tested for RM = -25 to +25
  • Even easier for ALMA frequency range

[Jagannathan, Bhatnagar, Rau & Taylor]

slide-28
SLIDE 28
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

28/20

The Hot-spots

Gridding De-Gridding Image Reconstruction

(Convolutions Of large images)

Supply Convolution Functions FFT

2 1 3

Data

slide-29
SLIDE 29
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

29/20

The Hot-spots

Gridding De-Gridding Image Reconstruction

(Convolutions Of large images)

Supply Convolution Functions FFT

2 1 3

Data

Data scatter

+ Multi-process computing Vs Multi-thread CPU computing

70-80% of the compute load! 70-80% of the compute load! (The major cycle) (The major cycle)

slide-30
SLIDE 30
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

30/20

The Hot-spots

Gridding De-Gridding Image Reconstruction

(Convolutions Of large images)

Supply Convolution Functions FFT

2 1 3

Data

Data scatter

+ Multi-process computing Vs Multi-thread CPU computing

70-80% of the compute load! 70-80% of the compute load! (The major cycle) (The major cycle)

Compute and cache Vs Compute on demand (on the GPU/FPGA)

Can dominate the memory Can dominate the memory footprint footprint

slide-31
SLIDE 31
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

31/20

The Hot-spots

Gridding De-Gridding Image Reconstruction

(Convolutions Of large images)

Supply Convolution Functions FFT

2 1 3

Data

Data scatter

+ Multi-process computing Vs Multi-thread CPU computing

70-80% of the compute load! 70-80% of the compute load! (The major cycle) (The major cycle)

Compute and cache Vs Compute on demand (on the GPU/FPGA)

Can dominate the memory Can dominate the memory footprint footprint

Compute and cache Vs Compute on demand (on the GPU/FPGA)

Minor cycle compute load Minor cycle compute load (Can dominate the total run-time) (Can dominate the total run-time)

slide-32
SLIDE 32
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

32/20

Algorithm architecture

FLOPS I/O Memory FLOPS-to-Mem Ratio

More memory per FLOP Lesser memory per FLOPS GPUs?

Algorithm design

  • Move towards algorithms with higher compute-to-I/O ratio
  • Use cheap massively parallel h/w (low memory footprint at the cost of higher computing)
slide-33
SLIDE 33
  • S. Bhatnagar: ALMA Future Science Program Development Workshop, Aug. 24 – 25, 2016, CV

33/20

Issues in Wide-fjeld Wide-band Full-Pol. Imaging

  • PB Efgects
  • Develop parametric models for antenna aperture illumination
  • Extend full Mueller imaging tools for heterogeneous array case

Full-pol imaging for EVLA under test as part of P . Jagannathan’s thesis

  • Variations with frequency
  • Assess extension of MT-MFS for full pol.

Improved multi-scale algorithms in general

  • Assess the range of freq. dependencies of the PB to be included
  • Computing load
  • Computing: Parallel processing on a routine basis
  • Memory footprint: Use heterogeneous hardware (CPUs + GPUs, etc.)
  • Use Cloud Computing as an efgective development platform.