[mmaimcal]Y+ Simulations

[John Conway]

Hi Mark et al,

 Some comments on Mars email of yesterday.


On Tue, 25 Jun 2002, Mark Holdaway wrote:

> 
> 
> I see three different sets of Y+ imaging simulations to do:
> 
> 1. Full Resolution Simulations
> 
> The main drawback of the Y+ configurations, which NOBODY
> seems to be picking up on, is it's probable inferior
> snapshot imaging.  I aim to investigate this in the full
> resolution array by looking at the imaging performance
> over 30 minute, 1 hour, 2 hour, 4 hour, and 8 hour integrations.
> I anticipate that the ring array's image quality will not 
> change greatly with increasing integration time, while the
> Y+ configuration will.
> 
> In addition, I'll add noise of varying amounts (just seems like
> the right thing to do) so I'll have, for a given configuration,
> a simulation result matrix that looks like:
> 
>                  integration
> noise     0.5    1.0    2.0     4.0   8.0
>  0.1
>  0.5
>  2.5
>  12
> 
> We'll have one such matrix for each of the array configurations
> in:  ring, loose Y+, tight Y+, strict Y.
> 
> This will take some time in doing the computing.
> 
> 

1) Well I was going to comment on this but I thought you 
were away etc. As I mentioned at the Socorro PDR 
snapshots for the largest array are important because
we will want to use the largest array for time variable 
sources, it also ensures good coverage for long tracks
at all dec.

Its important to distinguish here with 
aspects of the uv coverage which are fundamental 
properties of a Y topology versus rings and those 
which come from the particalar design/terrain constraint/
optimisation used. I worry more about the latter.


Specifc Properties of preset Y+ designs
----------------------------------------

Specifically in the present 
designs in your 'Y+2' proto-meo there are some pretty big
holes along the v-axis and othe rplaces for snapshots  - 
and they are still  pretty  big even after quite long synthesis. 
These holes are somewhat worse for the 'tight' version 
but are there also in the 'loose' one.

These type of holes are not fundamental to the Y philosphy 
-after all the VLA doesn't have them. The main question is 
are they fundamental to the combination of Y+ and terrain 
constraints we have?, if so then we have a problem.
Looking at the mask that has been adopted I suspect
that its not the terrain that dictates (although we 
must remember that a fairly liberal mask has been adopted 
with 15% peak gradients).  The 'arms'
of the allowed terrain are about 120 degrees apart and
only 20-30% of the area is blanked out. You could 
probably fit in a scaled version of the VLA with 
a few wiggles.  

Its more probably  that the holes along the 
v-axis instead come from the optimisation process. 
Specifically the array was built up optimising outward
a few pads at a time until the largest array was found, and the 
final array is at the end of this chain of optimisation. Even 
so I'm surpised that the optimisation has given more weight
to filling the large holes that exist than it has, but 
then again its easy for any optimisation algorithm to 
get stuck in a cul-de-asc.

In fact of course, as you indeed mention yourself below, 
the largest array is the most important to optimise. 
If we give the highest weight to optimising this, we know
we can probably achieve intermediate arrays between this and the 
3km one which are much better tha ring intermediates and use 
less resources, that  sort of comes with the toplogy. 
Also  the reason for having the  intermedates going 
from  3km to 14km is somewhat different than the ones smaller
than 3km, in the later case the case there is more astronomical
use of the intermediates, but in the former case we want 
useable  intermediates mainly  because its going to take 
a long time to reconfgure, not because of the large scientifc
demand for the intermediates - again arguing for more
weight to getting the best possible largest array. 


The present optimisation scheme stepping outward does
a good job of linking to the <3km array. But we ideally 
want some 'directed evolution' to the best possible 
largest array. I don't know how to exactly go about the
optimisation - and I don't minimise the difficulty 
of coming up with a robust optimisation menthod- but looking 
at the mask it seems that there should be  enough combinatorial 
space to get a very good largest array plus good intermediates.
Pehaps as part of the imaging simulation process you should
make a best possible largest Y optimised without regard to 
intermediates and see how tha performs (I think you mention this 
under 'Limiting cases' in your draft memo, but don't
see a uv plot).


General Peoperties
----------------

general properties of Y's versus rings are

- Partial Resonances between straight lines of uv points 
   gives large spikes in VLA snapshot- can be avoided by adding randomness
  to points, needed anyway to avoid terrain problems. Rings also have 
  similar problems broken up by terrain randomnenss

- star shaped outer boundary not circular.

   Fairly minor efefct I think will modulate 
   inner sidelobes but if these are small 
   anyway effect not large. 
     
- interarm baselines giving uv density spikes
 
   is a function of the 'tightness of the arms'
   - we have lived with these for a while now on the VLA
   - a lot more can be done with cleaver weighting I think to
     minise them - if you just don't use the interarm baselines 
     you can get fairlry uniform uv coverage and increase noise by
    sqrt(3/2), for cases when you are fidelity limited 
    and not sensitivity limited.

- Usually tapered uv density,

    Not really a fundanmental function of Y versus ring. 
    For instance a Y with a linear pad 
   distribution along arm is not tapered in uv radius. However for 
   VLA-like Y with maximum pad sharing we do generally 
   have decreaing pad density with radius. Here we are back
   to the old arguments for uv coverage; tapered versus untapered 
   which is best. If we can get two arrays with SAME 
    resolution then I believe that tapered is better. 
     
    If the terrain had a circular boundary - I would 
    always build a ring for my largest array -since 
    for the largest array one wants the max resolution
    and this overpower any arguments about degree of 
    uv taper. IF though as seems the case from Chajnantor the 
    geoegraphy gives a triangular star shaped terrain 
    mask then the highest resolution array can be a Y.

    As we learnt from the pre-Grenobel PDR discussion 
    more taper gives decreased inner sidelobes cf a ring, 
    but the peak far sidelobes are the same. People 
    often think that the sidelobes in a tapered 
    array come from the very low density of points 
    at the perimeter, but this is not so, if you 
    use UVCUT to get rid of them the sidelobes don't 
    change. Athough it seems significant to the eye 
    the very few points (about 2% or the total between 
     15db cutoff for a ring an 25db for a Y) are 
    not enough to influence much - its not worth getting 
     hot under the collar about these few points - 
     if they offend you just delete them and observe 
    2% longer to get the same edge density you would have 
    got with a ring array with natural 15dB cutoff - 
    but its wnot orth building a whole new toplogy of 
     rings which requires more pads just to  avoid this 
    low density of points at the edge.


- Different Self-cal toplogy

   This is one which nobody talks about, but for a ring 
   every anttena which participates  in a longe baseline 
   also is in a short baselines. This is not true for an
   open topology like a Y.  It would be nice to do 
   some simulations phase-calibration and phase-calibration
   with the different topologies.


-  Astrometric Accuracy

    I believe that there can only be minor differences
    between Y and rings of the same resulotion in astrometric accuray
    (see rough argument I sent a week or two ago). Still 
    would be nice to test it.


> 
> 
> 2. Sensitivity Loss with Resolution
> 
> No actual imaging needs to be done here, just simulation
> and reweighting to achieve the desired resolution.
> 
> We DO need to have a guess at the distribution of desired
> resolutions.  I would guess for a first pass that we want
> a large fraction of the Y+ observations at the full resolution
> (50% ?), and the rest are logrythmically distributed between
> the full resolution and the largest spiral configuration.
> 
> This will not take much time (which is why I've listed it 
> second priority and not third).
> 
> 
> 
> 3. Imaging quality of the incremental Y+ configurations
> 
> The incremental Y+ configurations will have generally superior
> imaging performance as compared with the ring array tapered to
> the proper resolution for comparison.  (The ring array will
> participate in observations in which the longest baselines
> resolve out the detectable structure, so tapering will happen.)
> We could quantify this.  Rather than perform an exhaustive
> comparison, I suggest that we take two of the incremental 
> configurations and compare them to the taperd ring array,
> but just for some simple cases.
> 
> We could pass on this if we run out of time.
>