14
June 2009 ARPG Some Comments to the W-projection algorithm. L. Kogan National Radio AstronomyObservatory Socorro, NM USA

Some Comments to the W-projection algorithm

  • Upload
    osmond

  • View
    36

  • Download
    0

Embed Size (px)

DESCRIPTION

Some Comments to the W-projection algorithm. L. Kogan National Radio AstronomyObservatory Socorro, NM USA. - PowerPoint PPT Presentation

Citation preview

Page 1: Some Comments to the  W-projection algorithm

June 2009 ARPG

Some Comments to the W-projection algorithm.

L. KoganNational Radio AstronomyObservatory

Socorro, NM USA

Page 2: Some Comments to the  W-projection algorithm

June 2009 ARPG

1.W PROJECTION: A NEW ALGORITHM FOR NON-COMPLANAR BASELINEST.J. Cornwell, K. Golap, and S. Bhatnagar, EVLA memo 67, December 2003 “MEMO”

2. The non-complanar baselines effect in radio interferometry:W-projection algorithmT.J. Cornwell, K. Golap, and S. BhatnagarJuly 2008 “PAPER”

Page 3: Some Comments to the  W-projection algorithm

June 2009 ARPG

.

The visibilities with W 0 can be recalculated to visibilities with W=0 with the help of convolution.

Frater R.H. was the first one discussing this idea (“Image Formation from Coherence Functions in Astronomy”, edited by C. Van Schooneveld, 1978)He considered the unpractical case of full UV coverage at each W layer.

W-projection is implementation of this idea for the general case including the poor UV coverage in some W layers.

Page 4: Some Comments to the  W-projection algorithm

June 2009 ARPG

The projection of the measured visibilities at W 0 to the desired visibilities at W = 0

ondistributi brightness source theis )(

vectorbaseline theofcomponent U theis

plane picture source at the coordinate theis :where

)2exp()exp()exp()()( 220

lB

U

l

dlUljWljWljlBuVisw

)exp(function theof ansformFourier tr theis G

nconvolutiofor stands :where

)(G )( G )(

2

111000

Wlj

dUUUUVisVisuVis

w

wwwww

Page 5: Some Comments to the  W-projection algorithm

June 2009 ARPG

How wide should be the gridding function ?)(UGw

This question is very important because the wider it This question is very important because the wider it is the more computation time is required.is the more computation time is required.

The W projection memo says: The W projection memo says: ““The support the W-dependent gridding function The support the W-dependent gridding function grows with both W and the field of view typically up grows with both W and the field of view typically up to a largest value of about 70 by 70 pixels.”to a largest value of about 70 by 70 pixels.”

I evaluated the convolution (gridding) function I evaluated the convolution (gridding) function analytically to verify this statement in particular.analytically to verify this statement in particular.

Page 6: Some Comments to the  W-projection algorithm

June 2009 ARPG

Analytical evaluation of the convolution function

function complex a is )(

radians;in view,of field full theis 2

gth;in wavelen sy' visibilit theof valueis

gth;in wavelenlayer W at the sy' visibilit theof valueis where

)2exp()exp( )(L

L-

2

UG

L

UU

WW

dlUljWljUG

w

w

Page 7: Some Comments to the  W-projection algorithm

June 2009 ARPG

gthsin wavelen is W radians;in view,of field of half is L

82U

41 pixel 1 pixelsin

0.5 level at thefunction n convolutio theof width full The

integrals. Frenels are 2

sin (x)

;2

cos C(x)

2 ;2 where

arctan)(

)]()([)]()([)(

2pix0.5

2

0

2

0

21

)()()()(

2

212

2122

1

12

12

WL

L

dttS

dtt

WW

ULxW

W

ULx

W

UUGPHAS

xSxSxCxCUG

x

x

xCxCxSxS

w

ww

Page 8: Some Comments to the  W-projection algorithm

June 2009 ARPG

Page 9: Some Comments to the  W-projection algorithm

June 2009 ARPG

The full width of the convolution function for the 4 VLA configurations at wavelength 4m

VLA,config D C B A

250 750 2500 7500

21 62 210 625

/maxW

5.02 pixU

Page 10: Some Comments to the  W-projection algorithm

June 2009 ARPG

00

2

111000

ies visibilitdesired the to ies visibilitmeasured theof

n)(projectioion recalculatfor base almathematic theisequation This

)exp(function theof ansformFourier tr theis G

nconvolutiofor stands :where

(*) )(G )( G )(

ww

w

wwwww

VisVis

Wlj

dUUUUVisVisuVis

• But the relation (*) is correct only if the measured visibilities are available everywhere.

• In reality only few points of can be available at one W layer especially for large W!!!

)( 10 UVisw

Page 11: Some Comments to the  W-projection algorithm

June 2009 ARPG

narrow).ly indefinite is DM beamdirty (The

coverage. UVfull theof case at theonly

completely canceled is W term theofeffect that showsequation The

) )(exp()()(

)exp( )exp()()(

)exp())]exp()(( )([

)( )]( )([ )()(

function) ( ies visibilitmeasuredactually oflocation

at the functions delta ofset theand tsmeasuremen ofset indefinite

theofproduct a as ies visibilitmeasured ofset thengrepresenti

map),(Dirty (*)equation of )( ansformFourier tr down the Write

2211

21

21

21

2

001

dlllWjllDBlB

WljdlWljlBllDB

WljWljlBlDB

GVisVislDM

w

w

w

wwwww

Page 12: Some Comments to the  W-projection algorithm

June 2009 ARPG

)exp( of ansformFourier tr 3.

nconvolutio theof FFT 2.

)exp( of nsform tra

Fourier with plane UVat the pixelby pixeln Convolutio 1.

:includes n timecomputatio algorithm projection- WThe

)exp(by pixelby pixeltion Multiplica 2.

)(DM calculate toFFT 1.

: includes n timecomputatio The

)exp())]exp()(( )([)(DM

)exp(by )(DMMultiply 2.

))exp()(( )()(DM

:locations baseline

at the sampled theof ansformFourier tr Take 1.

:layereach W for sopperation following thefollowing

obtained becan )( for theequation identical The

21

21

21

1

21

22

211

21

0

1

Wlj

Wlj

Wlj

l

WljWljlBlDBl

Wljl

WljlBlDBl

Vis

lDM

w

w

w

w

Page 13: Some Comments to the  W-projection algorithm

June 2009 ARPG

)2(l

lat diminished is W term theofeffect thereforeand

)(2)( pixels) (several narrow is )(

))(exp()()( )(

DM mapdirty summed for the expression following get the llwe'

),( severemost islayer each W at W term theofeffect the

gconsiderin and togetherlayers W all summarized Having

1

122

11

221max11

sum

max

l

llllllDB

dlllWjllDBlBlDM

W

sum

sumsum

Page 14: Some Comments to the  W-projection algorithm

June 2009 ARPG

The computation time comparison

• The authors of the W-projection declare its advantage in computation times 10-50 times.• But I heard this number as several times only!• The W-projection paper compares the computation time

with uvw-space facets algorithm existed inside of CASA.• The big number can reflect the bad quality of uvw-space

facets algorithm but not the good quality of the W-projection algorithm !

• It’d be more fare to compare with image-plane facet algorithm existed at AIPS.