S. Guilloteau

A. Wootten

D. Hoffstadt

Mark Rafal (unable to attend)

Richard Kurz

Jaap Baars

Darrel Emerson

Other affected DH/TLs.

I would like to have also some representative from the logistic (D.Hoffstadt if possible).

3) Documents

To be ready 2 weeks in advance for distribution, specially to the outside members. We already have a lot, but are weak on operation issues. Gross distribution

- Nested Ring concept (Yun / Butler)

- Spiral concept (Conway)

- Multi-configuration aspects (Viallefond)

- Compact array issues (Morita / Guilloteau)

- Site issues (Radford / Butler)

- Operation concepts (Wootten / Guilloteau) (oops, have to work here...)

- Cost issues (Hofstadt / ? )

Goal of the meeting: select a baseline design to continue with detailed implementation work ...

Updates

The CLEAN and difference results for B arrays for all three declinations and all five test images, with directly comparable grey scale difference images, have been posted on my website, available under Steven's Lair. This now shows CLEAN and difference images for all declinations and test images, with the grey scale difference plots directly comparable. Opinins?

Steven calculated fidelity images for the CYG, MPD and SDC examples for the B resolution at the same time as the C array ones posted earlier, so has posted them for perusal at his site.

Are the procedures used sensible? How do we improve on this? How do we make use of the outputs?

Koh-Ichiro Morita has made a summary of my recent imaging simulations for ACA. The summary may be seen at this URL. Any comments or questions are welcome.

Brief summary:

Please read Advance of Alma simulation at IRAM by F. Gueth, S. Guilloteau, R. Moreno and J. Pety considering the performance of ALMA with the ACA.

Simulation Strategy outline at the Tucson meeting is found here. What changes should we adopt based on our experiences so far?

Other Future Plans

The plans for the layout of the pads must be finalized by the beginning of construction. We hope that the first construction funds will arrive on 2001 Oct 1 when the FY2002 begins here. The project WBS has this date as 2002 Jan 1. Realistically, I think we need to inspect the pad sites before finalizing their location. As a start, Wootten compared pad locations for the spiral and nested ring arrays, and Butler and Radford did some studies in December 2001. There exists a list of possible problem pad locations for the 3km 1.6km and 0.8 km and Spiral array. Hopefully, these sites are, in fact, OK. Are there any others which folks have worried about?

Next Meeting?

Minutes of Meeting 4 Jan 2001

PDR

Since the teleconference, both Lee Mundy (lgm@astro.umd.edu) of the University of Maryland BIMA Group, and Eric Anterrieu (Eric.Anterrieu@cerfacs.fr) of (Signal & Image Processing Team), CERFACS, at Toulouse have agreed to serve as external referees. Two are required, so the quota is met, though more opinions may be sought.

Simulations

There was considerable discussion on the error images and statistics which Steven has placed at his WWW site. Best to use Steven's report here:

SH:"I'll just summarise my verbal report from the last meeting, and say what my immediate workplan is.

"At the time of the previous telephone conference I had just presented arithmetic difference images, and we discussed displaying image statistics, FFTs of difference images, and fractional difference images. The fractional difference measures primarily discussed were fidelity image results as presented by Stephane in his example, and the measures proposed by John which are a series of ratios of (arithmetic difference image)and (smoothed model magic blanked at a certain level) which gives rise to a corresponding series of rms values [This is my understanding of it].

"As these measures have very similar implications for scripting an AIPS implementation, I pressed on with implementing the fidelity image as it was slightly better specified. I originally intended to present just a few examples for comment, but progress was better than I expected so I was able to provide a complete set of C array examples and B array examples for MPD, CYG and SDC (please ignore bad links to the MAR and M51 cases). These examples are just a first step so feedback was and is welcomed.

"Mark asked about negative values in the Fidelity plots and histograms- these arise because some of the initial model images have negative values in them, viz. M51, SDC and CYG. M51 and CYG have negative values which are of the order of 0.1% of the peak values in absolute terms, but the SDC model has negative values which are up 5% of the peak value in magnitude. The use of CLEAN boxes was suggested, and also the use of a prior model for images such as MAR. I raised the question as to whether or not the denominator in the fidelity image ( the absolute values of the arithmetic difference image, truncated at 0.7 x rms value of the difference image) should use the same rms value for each of the corresponding ring and spiral cases... and the answer seems to be no, although some are uneasy with the fidelity image as a whole rather than the details of its implementation.

"I reported that FFTs of difference images can be churned out if necessary.

"So bearing all this and the discussions from last Thursday's telecon in mind, I have a number of things to do...

"1) Truncate M51, CYG, SDC model images so that they are greater than or equal to zero. Someone commented that this would take about an hour, which is true, but the implication in all this is that the imaging, CLEANing, calculation of smoothed models, differencing and presentation of results for these models should be redone as well, which takes considerably longer.

2) Import the 10K (dots) model into AIPS, and create down sampled versions to avoid registration problems. I may as well put the 10K model through the same process as the others, as (apart from disk usage and cpu time) the only overhead it imposes is adding a line to the setup file for each batch job.

3) Archive results from AIPS disks to CD, and also text and PostScript files used in generation of the webpages.

4) Redo imaging, CLEANing, differencing etc. for B and C resolutions using M51, CYG, SDC and 10K

5) Perform imaging of M51 and and CYG for D resolution for later combination with B array results, by way of trying multiconfigurations.

6) Try out DBCON using results of 5)

7) Implement John's metric from his specification.

8) Look at obtaining off-source rms values

"Use of CLEAN boxes is still an open issue as is the use of a prior model for Mars. If the consensus is to do either of these in the current round, I will, obviously redoing everything for all models consistently."

And, on 18 Jan, Steven reported:

I have posted new results (CLEAN, difference, fidelity) for the B and C resolutions. These results replace the 18th December results. The main changes are that the M51, CYG and SDC model images have had any negative pixels truncated to zero, so that they and their smoothed counterparts as used in the difference and fidelity image calculations are wholly positive (or zero). These results should now be complete, i.e. the links for the MAR and M51 cases should now work for the B resolution.

The results are of course at my site.

I have calculated results for M51 and CYG for the D resolution for the quick look at the hybrid array (although I imagine I only needed to go as far as getting the UV data), and may post these as well. Actually combing these snapshots with B array data will follow.

I hope to start looking at the 10K dots tomorrow, as well.

Metrics

John suggested that we arrive at the metric iteratively and collected the emails to alma-config on the subject. Summary of discussions at the teleconference:

Principle: The metric should not depend on the character of the array.

It was generally agreed that negative values should be suppressed in some way. There was general agreement that model image pixels could be magic value blanked at some percentage of the peak, then one could divide the model and produced images to obtain fractional error and fidelity indeces. The 'median fidelity' could be extracted at percentages of the peak--1 per cent, ten per cent, and so on. Then this and fractional error can be plotted as a function of which percentage was used in the model image The rms for an image could be defined as the mean from the whole difference image over a region where there is signal, e.g. the unblanked region. This is item 1 and 4

Discussion also centered on how to treat the fact that differing content of short spacings dominates the images. This could be addressed by adding in data from a smaller array. It was decided that it would be excellent if Heddle were to simulate some multiconfiguration images, including B and D array data, for e.g. M51 and Cygnus models images. This is item 5 on Heddle's list above.

Since last tcon, there has been some discussion on the dots image. There are ALMA simulations with FITS images at: http://www.cv.nrao.edu/~awootten/mmaimcal/newzlines.html or John has distributed his DOTS image. Steven is working on the latter.

Next Meeting

Another phone meeting was suggested, to discuss the metrics.