Getting Started

Documentation

Glish

Learn More

Programming

Contact Us

Version 1.9 Build 803

News	FAQ
Search	Home

Controlling the Major Cycles

The key to making the incremental deconvolution in AIPS++ multi-field imaging successful lies in controlling just how deeply we deconvolve in the major cycles. The control parameters discussed here can be set with Imager.setmfcontrol. If we deconvolve too deeply with the approximate PSF, we will spin our wheels while adjacent fields with slightly different PSF sidelobes argue over where the low level model flux belongs; the answer will of course be in error, we will mis-subtract from the data visibilities, and we will have to correct our error in the early stages of the next major cycle of incremental deconvolution. If we don't deconvolve deeply enough in each major cycle, there will be more major cycles, which are dominated by the exact model subtraction.

One can see fairly easily if one is cleaning too deeply or not cleaning deeply enough in a given major cycle by looking at the plots given when the argument displayprogress=T (in tool function clean). When trying to clean too deeply, the peak residual level towards the end of a major cycle will flatten out (or may even increase and diverge) for many iterations. At the start of the next major cycle, the peak residual level will begin at a higher level than at the end of the previous major cycle. (This can be especially true for mosaicing with multi-scale clean, where the approximate PSF may actually be a pretty poor match for the true PSF's on the larger size-scales). If all is well, this major cycle will still get to a lower peak residual than the previous one. However, stopping the major cycles sooner will prevent us from spinning our wheels, resulting in a comparable deconvolved image in fewer iterations.

Eventually, there will probably be a smarter, more automatic way to determine if we need to stop the current major incremental deconvolution cycle. Right now, the tools are simple, but easy to use. If the progress display indicates that ending the major cycle sooner is appropriate, we can do that in one of two ways:

• by increasing the cyclefactor. The major cycle cleaning threshold is a fraction of the peak image residual. That residual fraction is determined by the cyclefactor multiplied by the peak negative sidelobe of the approximate PSF. Stopping the major cycle cleaning sooner can be accomplished by increasing the cyclefactor. Values ranging between 2 and 4 are common.

• by decreasing the cyclespeedup. What if we have set the cyclefactor too low? The cycle threshold as calculated above can be made to drift upwards by setting the cyclespeedup. The threshold will double every cyclespeedup iterations in the major cycle until the major cycle stops. If the cyclespeedup is less than or equal to 0.0, no adjustments to the calculated cycle threshold will be made.

In addition to these cycle control parameters, which are applicable to mosaicing with CLEAN, Multi-Scale CLEAN, MEM, or Maximum Emptiness, there are two more control arguments set by tool function imager.setmfcontrol which are applicable only to Multi-Scale CLEAN. These are stoplargenegatives and stoppointmode, which are discussed below.