1

ME4: Extended Sources, Images, Image-Plane Effects

ME3: Calibration & Correction

2

So, Is There Always Such A Beast As Corrected uv-Data?

Objectives:  Learning to predict extended sources  Getting to grips with wide fields and n  Thinking about image-plane effects  Planting the seeds of some advanced topics and techniques

Say we now have some image-plane effects:

V pq =G p

F N



p





E p B E q N q G q

... and we know all of the G p , E p , and (of course) N p 's ; then is there a way to obtain "corrected" visibilities V ' such that V ' pq =

F B 

??? 

(or at the very least V ' p q =F N p B N q ) ???

svn up Workshop2007

ME3: Calibration & Correction

3

ME3: Calibration & Correction

Correcting At A Single Point

And The Answer Is...   

In general, NO! uv-plane effects (the Gs) can be taken out. Image-plane effects correspond to convolution in the uv-plane:  q

 q

 q

But we can correct for a single point l 0 , m 0 : 1



=

 q

...with time-variable kernels ...and with each baseline's uv-plane sampled along just a single track

(Note: Bhatnagar et al. (EVLA Memo 100) suggest a method for approximate correction during the imaging step. We'll return to this later.)

1



V '=E p l 0 , m 0  V E q l 0 , m 0  =

V p q =F N p E p B E N = F E p ° F N p B N ° F E  

4

F  E

p

l 0 , m 0 1 E p N p B N E E l 0 , m 0 1   q

=

F N

 q

 q



p

B ' N q

where B 'l 0 , m 0 =B l 0 , m 0  , but diverges further away. 



In general, uv-data can only be corrected for a single point on the sky. This is the motivation for facet imaging.

5

Divide And Conquer

6

Case 1. Point Sources 

Let's forget about the G s, as they're trivial to put in/remove. B (M.E.) V p q = K p E p E q K q d l d m =  K p E p B E q K q d  n sky lm

Point sources we have already done, their B is a delta function. V p q = K p E p B 0 l l 0, m m 0  E q K q d l d m = lm

= K p l 0, m 0 E p l 0, m 0  B 0 E q l 0, m 0 K q l 0, m 0  

Bs    V pq = V s E K dldm pq =   K p E p n q q s s lm

 

And we can examine the integral for each source separately.

Case 2. Extended Sources And Patches 

  

Any B that is not a delta-function gets interesting, especially if E is not trivial, and especially for wider fields (the not quite an F.T. problem.) ...but we need to solve this for two reasons. Reason 1: spatially extended sources... Reason 2: it is impractical to directly predict large numbers of fainter sources individually, so we need to organize them into patches, and find a faster way to predict a patch en masse.

=



= K p0 E p0 B0 E q0 K q0

then





This is linear over B , so if the sky is a sum of sources: B l , m =s B s l , m  , (or B 

=s B s 

 )



7

Product of five 2x2 matrices. We can predict point sources perfectly. Computationally practical for a limited number of sources.

8

First, Reduce To Center... Consider a source centered on direction 0

l 0, m 0  ,

with the phase center at

=0. E p 

B 

E q 

K q 

d = V p q=  K p 





sky

Let's integrate in the coordinate system

'=

0,



and define B B 

 , E 

B 

: 0

p

0

V p q=  K p 

0  E p 

0  B 

0  E q 

0 K q 

0 d  '= 



sky

=K p  0 

  K   E   B   E   K  d  K  = p

p

 q

 q

 q

sky

=K p l 0, m 0 

 lm

K p E p



  B  E q K q d l d m K q l 0, m 0  n

0

9

So All We Need To Figure Out Now...

...Where Life Gets Easier V p q =K p 0 





 K p E p nB E q K q d l d m lm





K q 0

In other words, to predict a source at l0,m0, it is sufficient to predict it at the phase center (shifting E accordingly...), then apply the phase terms Kp0, Kq0 to move the source to its true position. This is beautifully symmetric w.r.t. the ME for a point source: 

lm



lm







l v p q m w p q n 1 

dl dm

2  iu p q l v p q m  F.T. kernel

...and we only require the Weak Assumption that n1 over the extent of the source.

12

E p E p 0,0 over the source, then



 E q 0 K q 0

This is very good from a practical point of view, since:  F.T.'s are easier than strange integrals  phase shifts and (in the optimistic case) image-plane effects are applied in the same way to all kinds of sources  we only need to worry about predicting the source (or patch) at the phase center

"apparent sky"

If we also make the "Strong Assumption" that

0

pq

V p q =  E p B E q e d l d m 

Or Even More Optimistically...

F B

B  2  iu E q e n

When n  1 , this becomes a Fourier Transform:

11

V p q =K p 0 E p 0

...is how to predict extended sources at the phase center. We've simplified the problem considerably. If we now expand the K terms: V p q = E p

K p0 E p0 B0 Eq0 K q0

10





This is only an approximation. Error levels will depend on:  source flux (in a linear way, so we can talk about relative error);  patch/source size (because of the n term);  the variation in E over the extent of the patch. We need to develop some feeling for how small the patches/sources need to be to make the error suitably small. We can build some trees to simulate these errors (well this is actually good news, isn't it?)

13

Shapelets: The Ultimate In Analytic Sources

Case 2: Analytic Extended Sources 

For some kinds of sources (e.g., a 2D Gaussian), we can write out the F.T. analytically.



The shapelet transform decomposes an image (in l,m) into a sum of shapelet basis functions: b l , m =  c ns n l , m 

i.e. given a set of source parameters that determine its B

n

(for a Gaussian, this is its major/minor axis extent 1, 2 ,



position angle  , and integrated I , Q , U , V fluxes), we can immediately write out an expression for its

F B . 

(The F.T. of a Gaussian is a Gaussian.)  



Meow.GaussianSource implements the trees for this. Only correct under the Strong Assumption, obviously.

 

15



Finally, we can have an image  B (or perhaps just plain I) sampled on a RA/Dec (or other) grid. An FFT gives us the F.T. on a regular uv grid: {B ij =B l i , m j }

 

FFT



{X ij =

F B u , v } i

Implementing An Image Source

... u t  , v t  from one child, and interpolates  u t   v t  X t , = X within the {X ij } "brick" , c c







j

This is done by the Meq.FFTBrick node. The Meq.UVInterpol node then takes...



(which it gets from the other child.)

Complex source structure can (potentially) be represented with relatively few low-order {cn} coefficients. The basis functions sn have an analytic F.T. Therefore, we can quickly compute the F.T. of a shapelet source, given a set of coefficients. Strong Assumption not necessarily required... Just making its way into MeqTrees, so no demo...

Case 3: Image Sources 

14



Meow.FITSImageComponent implements the right combination of Meq.FFTBrick, Meq.UVInterpol, and Meq.FITSImage. Images can also have a frequency axis, this is seamlessly propagated through the tree. No demo  something is currently broken :(

16

17

Bonus Image: The True nJy Sky

18

Image Sources & Image-Plane Effects Given an image, {B ij }, we can easily apply E : B pq=E p B E q 



19

The Bhatnagar Approach: Coming Soon To a Tree Near You?

F E

B E q =F E p ° X ° F E q  

p

If E p is sufficiently smooth, then



F E

p

 can be approximated

n: Naughty, Nasty, Notorious, Nefarious, Nebulous, Noxious, or Just Negligible? 



with just a few low-order Fourier coefficients. 





Convolution with a small kernel can be computed directly, fairly cheaply. ...and in fact the interpolation function used inside UVInterpol to interpolate X to the uv track is a form of convolution. So we can apply E's with a suitable modification to the convolution function.

This can be done in a simple tree:  the FFTBrick node expects a brick child.  this doesn't have to be an actual FITSImage node, but can also be any subtree, i.e. one that does some operations on the image. If the Ep's are different per antenna, this is INSANELY expensive (an FFT per baseline!)







The Weak Assumption is that n=1 over the extent of the source. Let's make a differential tree to estimate the error that this assumption introduces. General idea: make a test grid of point sources, then:  predict them perfectly in one branch.  predict them with (l,m,1) in the other branch. The former are Meow.PointSources, how do we implement the latter? ...through OOP, of course!

20

21

Demo 1: So How Weak Is The Weak Assumption?

A Naughty Direction  







lmn's ultimately come from Meow.LMDirection. We're going to derive a subclass from Meow.LMDirection This subclass reimplements the n() function, which LMDirection uses to create its n node (which is subsequently used to make a K term). The end result: a direction that ignores the normal n term, and uses 1 instead. See ME4/NaughtyDirection.py.

Mega-Exercise 3: Replicating The Perley Movie The ionosphere introduces two effects: Faraday rotation RM F =Rot  2  , and phase delay: L=e i  .  Both RM and  are proportional to TECl , m . Let's simulate a TID (Travelling Ionospheric Disturbance), by making TEC  x , y  constant + a moving 1D sine wave. (And each antenna looks up through a different x , y point.) We'll make a grid of point sources, and apply F and L. With a bit of luck, we can even force the imager to make a series of time slices...

 





23

Load up ME4/demo1-naughty.py. This makes two sets of sources arranged in a cross:  Regular PointSources in one branch  PointSources with a NaughtyDirection in the other branch You can select the grid stepping size  compare results for 1' and 60'. Compare MFS and per-channel maps.

22

## GitHub

data can only be âcorrectedâ for a single point on the sky. ... sufficient to predict it at the phase center (shifting ... errors (well this is actually good news, isn't it?)

#### Recommend Documents

GitHub
domain = meq.domain(10,20,0,10); cells = meq.cells(domain,num_freq=200, num_time=100); ...... This is now contaminator-free. â Observe the ghosts. Optional ...

Torsten - GitHub
Metrum Research Group has developed a prototype Pharmacokinetic/Pharmacodynamic (PKPD) model library for use in Stan 2.12. ... Torsten uses a development version of Stan, that follows the 2.12 release, in order to implement the matrix exponential fun

Untitled - GitHub
The next section reviews some approaches adopted for this problem, in astronomy and in computer vision gener- ... cussed below), we would question the sensitivity of a. Delaunay triangulation alone for capturing the .... computation to be improved fr

ECf000172411 - GitHub
Robert. Spec Sr Trading Supt. ENA West Power Fundamental Analysis. Timothy A Heizenrader. 1400 Smith St, Houston, Tx. Yes. Yes. Arnold. John. VP Trading.

Untitled - GitHub
Iwip a man in the middle implementation. TOR. Andrea Marcelli prof. Fulvio Risso. 1859. Page 3. from packets. PEX. CethernetDipo topo data. Private. Execution. Environment to the awareness of a connection. FROG develpment. Cethernet DipD tcpD data. P

BOOM - GitHub
Dec 4, 2016 - 3.2.3 Managing the Global History Register . ..... Put another way, instructions don't need to spend N cycles moving their way through the fetch ...

Supervisor - GitHub
When given an integer, the supervisor terminates the child process using. Process.exit(child, :shutdown) and waits for an exist signal within the time.

robtarr - GitHub

MY9221 - GitHub
The MY9221, 12-channels (R/G/B x 4) c o n s t a n t current APDM (Adaptive Pulse Density. Modulation) LED driver, operates over a 3V ~ 5.5V input voltage ...

fpYlll - GitHub
Jul 6, 2017 - fpylll is a Python (2 and 3) library for performing lattice reduction on ... expressiveness and ease-of-use beat raw performance.1. 1Okay, to ... py.test for testing Python. .... GSO complete API for plain Gram-Schmidt objects, all.

article - GitHub
2 Universidad Nacional de Tres de Febrero, Caseros, Argentina. ..... www-nlpir.nist.gov/projects/duc/guidelines/2002.html. 6. .... http://singhal.info/ieee2001.pdf.

PyBioMed - GitHub
calculate ten types of molecular descriptors to represent small molecules, including constitutional descriptors ... charge descriptors, molecular properties, kappa shape indices, MOE-type descriptors, and molecular ... The molecular weight (MW) is th

MOC3063 - GitHub
IF lies between max IFT (15mA for MOC3061M, 10mA for MOC3062M ..... Dual Coolâ¢ ... Fairchild's Anti-Counterfeiting Policy is also stated on ourexternal website, ... Datasheet contains the design specifications for product development.

MLX90615 - GitHub
Nov 8, 2013 - of 0.02Â°C or via a 10-bit PWM (Pulse Width Modulated) signal from the device. ...... The chip supports a 2 wires serial protocol, build with pins SDA and SCL. ...... measure the temperature profile of the top of the can and keep the pe

Covarep - GitHub
Apr 23, 2014 - Gilles Degottex1, John Kane2, Thomas Drugman3, Tuomo Raitio4, Stefan .... Compile the Covarep.pdf document if Covarep.tex changed.

SeparableFilter11 - GitHub
1. SeparableFilter11. AMD Developer Relations. Overview ... Load the center sample(s) int2 i2KernelCenter ... Macro defines what happens at the kernel center.

Programming - GitHub
Jan 16, 2018 - The second you can only catch by thorough testing (see the HW). 5. Don't use magic numbers. 6. Use meaningful names. Don't do this: data("ChickWeight") out = lm(weight~Time+Chick+Diet, data=ChickWeight). 7. Comment things that aren't c

SoCsploitation - GitHub
Page 2 ... ( everything â {laptops, servers, etc.} ) â¢ Cheap and low power! WTF is a SoC ... %20Advice_for_Shellcode_on_Embedded_Syst ems.pdf. Tell me more! ... didn't destroy one to have pretty picturesâ¦ Teridian ..... [email protected].

Datasheet - GitHub
Dec 18, 2014 - Compliant with Android K and L ..... 9.49 SENSORHUB10_REG (37h) . .... DocID026899 Rev 7. 10. Embedded functions register mapping .

Action - GitHub
Task Scheduling for Mobile Robots Using Interval Algebra. MudrovÃ¡ and Hawes. .... W1. W2. W3. 0.9 action goto W2 from W1. 0.1. Why use an MDP? cost = 54 ...