Getting Started

Documentation

Glish

Learn More

Programming

Contact Us

News	FAQ
Search	Home

Strategies for processing

Much of the strategy for calibration and imaging is highly dependent on the context. Nevertheless, we give here some rules that are of general use.

imager can perform joint deconvolution of Stokes I,Q,U,V images, and self-calibration of both antenna gain (G) and polarization leakage terms (D). Either the Clean algorithm (in both Hogbom or Clark variants), or Dan Briggs' NNLS method or both together can be used. We recommend the following approach for making a deconvolved image.

• Calibrate the visibility data as well as possible in AIPS, write a single source fits file and convert it to a MeasurementSet,
• Plot the uv coverage and visibility amplitudes,
• Make a deconvolved image,
• Restore it, and display it
• Convert it to FITS format using tofits of the object image.

Both clean and NNLS use masks to control where flux is to be found (i.e. an equivalent to clean boxes). To construct these, use the imager derive_mask method.

If self-calibration is needed, the steps are:

• Apply image plane constraints to the existing model using using derive_model method.
• Predict the visibility using derive_ft method.
• Calculate the calibration for this model solve_cal method.
• Remake the deconvolved image as above,
• Repeat as necessary.

You might wish to restrict the calibration solution step to a limited uv range. The best way to do this is demonstrated above in the section on the sortandselect module.

The terms allowed in the calibration solution are:

P

Non-solvable feed position angle. This includes both feed position angle (especially for linear polarization arrays such as WSRT), and parallactic angle.

T

Solvable atmospheric gain and phase. Eventually this will become atmospheric gain and delay but for the moment, we recommend that you use this for phase changing atmospheric phases. It is the same for both polarizations.

G

Solvable electrical path gain and phase. These terms are different for each polarization. These usually vary more slowly than the T terms (but there is no assumption about time-variability).

B

Solvable bandpasses. Different for each polarization.

D

Solvable polarized radiation leakage. These describe the leakage of each hand of polarization into the other. The calibration matrices are therefore non-diagonal. These usually vary slowly, on the time scale of hours, if at all.

M

Solvable interferometer-based errors. These describe the lack of closure of a correlator. The timescale for this should be kept as long as possible, usually the entire observing time.

The order of terms is, reading from the left: M G B D P T. So the radiation for a given antenna goes through the T-matrix first, then P, then D, B, G and finally after correlation, through M. Since the solution for D-terms is more difficult than the solution for T or G-terms, one should perform a cycle or two of self-calibration of T and G before attempting a solution for D. It is usually also a good idea to solve for T (phase-only) before solving for G.

Finally, we should emphasize the power of the tools that AIPS++ provides for diagnosis of problems in calibration. In figures 10 and 10, we show the fit per antenna as a function of time, and of distance from the array center. The former is available as a method of gaintable: gaintable.plot, whereas the latter is a custom plot. Here is the Glish work that lead to these two plots:

- tt:=gaintable('tcal')
- tt.plot(,'ifit')
T 
- dp.setyscale(0,0.6)
ok 
- dp.psprinttofile('ifits.ps')
- ant:=table('JUPITER.MS/ANTENNA'))
- ant.getcol("NAME")
VLA:N32 VLA:W20 VLA:W12 VLA:E8 VLA:E28 VLA:E36 VLA:E16 VLA:W8 VLA:OUT
VLA:W28 VLA:N24 VLA:W4 VLA:N12 VLA:N4 VLA:N20 VLA:E12 VLA:E20 VLA:W32 
VLA:N28 VLA:W36 VLA:W24 VLA:E4 VLA:E32 VLA:E24 VLA:N16 VLA:N36 VLA:N8 
VLA:W16 
- antloc:=ant.getcol("POSITION")
- center:=antloc[,14]
- for (i in 1:28) antloc[,i]-:=center
- for (i in 1:28) r[i]:=sqrt(antloc[1,i]*antloc[1,i]+antloc[2,i]*antloc[2,i]+antloc[3,i]*antloc[3,i])
- ifit:=tt.ifit()
- dp.clear()
T 
- for (i in 1:117) dp.plotxy(r,ifit[,1,i],,"points")
- dp.psprinttofile('fits.ps')

In this example, the custom plot shows that the best fit improves with distance from the array center, presumably because the source is resolved on the longer spacings to which the antennas further out are biased. We therefore restricted the range of uv spacing from which the calibration was derived. In practice, of course, such a custom plot might be sufficiently useful that it should be incorporated in the system.

Figure 1: Fits per antenna plotted against time

Figure 2: Fits per antenna plotted against distance from the array center