 News |
FAQ
|
| Search
|
Home
|
| Getting Started | Documentation | Glish | Learn More | Programming | Contact Us |
|
| Version 1.9 Build 1367 |
|
[This is a snapshot of the SDCI effort as of the AIPS++ review. It represents the state of affairs as of about December 2, 1994. There has beem some clean-up of the document that was available at the time of the review. In addition, the SDCI demonstration that was part of the review is now summarized as part of this document.]
This document describes the initial implementation of the basic components of single dish calibration and imaging in AIPS++ . This initial implementation was guided by the immediate goal of calibrating and imaging a very specific data set (on-the-fly spectral-line data from the NRAO 12-m). The lessons learned during this phase are being applied toward the design of more general single dish calibration and imaging classes and applications. They will also strongly influence the UV calibration and imaging design (especially with regard to the MeasurementSet that all forms of data in AIPS++ will share). Future development is outlined in the final section of this document.
The AIPS++ project has spent a long time struggling with the concepts of MeasurementSet, TelescopeModel, and MeasurementModel with little to show for our efforts. These abstract concepts were first described at a meeting in Green Bank, West Virginia in early 1992. They are described in more detail in the AIPS++ design overview (Glendenning).
This paragraph roughly summarizes the Green Bank model. The MeasurementSet is the class that is intended to contain astronomical data, it is little more than a simple container. The TelescopeModel consists of a number of TelescopeComponents. Each TelescopeComponent has an apply, corrupt and solve member functions. apply applies some calibration system to a MeasurementSet. corrupt is the opposite of apply. solve uses a model of the sky along with a MeasurementSet to solve for appropriate calibration parameters (which could then be ``applied'' to the data through apply). The job of the TelescopeModel is to coordinate the various TelescopeComponents that are present in the model. A MeasurementModel has an invert and a predict member function. invert ``inverts'' the data to produces a model of the sky. predict is the opposite of invert.
The lack of a concrete MeasurementSet has been a particularly large obstacle to forward progress. Without any place to put real astronomical data (the MeasurementSet) it has been difficult to make any real progress on a number of other fronts.
Faced with this observation, it was decided to adopt a well defined immediate goal in order to obtain concrete working examples of the MeasurementSet and TelescopeModel. The goal has been to calibrate and image a spectral-line on-the-fly data set from the NRAO 12-m telescope.
OTF observing at the 12-m consists of moving the telescope while recording data as fast as the hardware and telescope and instrument control software will allow (currently this is 1/10 second). The instantaneous telescope position is recorded for each data sample. The speed of the telescope along the sky is such that the telescope beam is over-sampled in the direction of telescope motion (typically more than 10 samples per telescope beam width). A data set consists of several horizontal OTF rows with the spacing between rows being close to the Nyquist size appropriate for the telescope beam. Although the pattern scanned on the sky is horizontal rows, various effects (wind, variations in drive speed during the scan, etc) mean that the sky has not been regularly sampled. The motivation for the OTF technique is to account for these variations in tracking during an observation as well as to map the region as fast as possible to minimize receiver drifts so that the spectral baselines are more consistent. OTF observing can be done in continuum and spectral-line mode. For the rest of this document, OTF refers to spectral-line data.
The OTF data samples are total-power data. Calibration is accomplished by observing appropriate OFF (emission/absorption free) regions. This is usually done at the end of each OTF row. It may alternatively be possible to use the emission free regions in the area being mapped during the calibration phase. However, this requires that the calibration occur after the data has been imaged onto a regular grid. The end result of a typical OTF spectral line observation is a data set with a number of rows consisting of several hundred individual spectra and the telescope position when each spectrum was taken. The rows may be interspersed with calibration (OFF) scans.
Calibration of the data within AIPS++ is done using a TelescopeModel consisting of any number of TelescopeComponents each having their own apply() method. The OTF observing mode at the 12-m consists of a number of samples from the region of interest (ON) and some samples from regions that the observer hopes are free of astronomical signal (OFF). One hopes that the non-astronomical signal does not vary between any of these regions and that the difference signal will contain the astronomical signal. Constructing this difference signal is a typical calibration at any single dish telescope and is not specific to OTF data. It is also necessary to correct the difference signal for the sensitivity or gain of the telescope. For position switched data this is generally done at the same time as the construction of the difference signal. At the 12-m, gains are recorded for each spectral channel while at the NRAO 140' telescope, a single value is assumed to apply to the entire spectra. This means that position switched calibration is a telescope dependent operation.
Although not yet implemented as an observing mode at the 12-m it should be possible to use the emission free regions in the OTF mapped area to construct appropriate OFFs when calibrating an OTF image (i.e.,it would not be necessary to make specific calibration observations, they would be constructed from the known emission-free regions of the OTF data). This calibration can occur after the data has been gridded onto a cube. This will reduce the observing time by eliminating the need for any specific off-source observations. It will also reduce the time required to produce a calibrated image by reducing the number of subtraction and multiplication operations. For large spectral-line data sets, that may be extremely important and may be required in order to keep up with the data rate.
The calibration and imaging of single dish data must keep up with the data rate. Observers at single dish telescopes generally plan observations over short terms based on the most recent observations. It is vital to be able to asses the quality of the data as soon as possible. A small 0.25x0.25 degree field mapped using the OTF observing mode consists of data from roughly 14,000 points on the sky for each feed. Currently there is only one feed but there will shortly be an 8 feed receiver. There are at most 1536 spectral channels that can be split in any number of ways between polarizations and feeds. Given the current practice of 1/10 second per data point and allowing for the calibration scans (10 seconds every other row is typical) and time to start and stop the telescope motion at the end of each row. This small field takes about 40 minutes to map.
The test data set used in this initial single dish calibration and imaging effort consists of OTF data from an 0.25x0.25 degree field. There are 4 spectra at each location, 2 polarizations at each of two spectral resolutions (128 channels and 256 channels). Execution times reported in this document refer to this test data set and were done on a single processor of a Sparc-10.
UniPOPS is the current NRAO single dish analysis package. It is primarily a vector calculator. For many single dish problems, this is quite adequate. It is especially suited to the needs of observers while they are at the telescope for most observing strategies. The command-line interpretor is a version of the POPS parser related to the AIPS POPS interpreter. Unlike AIPS, UniPOPS relies exclusively on verbs that manipulate several large internal arrays (there are 10 1-dimensional arrays and 4 2-dimensional arrays) as opposed to tasks that run independently of the interpreter. Users of UniPOPS, especially at the 12-m, rely heavily on POPS procedures to effectively create new verbs to do the specific tasks that their observations require.
UniPOPS is adept at analyzing 1-dimensional data. It is not adequate for serious analysis of multi-dimensional data. There is very little in the way of data-base management tools and no real ability to deal with data cubes. With the OTF observing mode described above as well as the multi-feed receivers that will be increasingly common at single dish telescopes, UniPOPS will shortly be inadequate for many observers.