Next: Formation of the CLEAN Up: CLEAN ideas Previous: Stopping criteria Contents Index
The simplest way to control the convergence is the CLEAN_STOPcharacter variable. It allows to express the stopping threshold in natural units, e.g. Jy, mJy, K, milliK, Noise or Sigma (the noise levels), or even % for peak percentage.
But Clean convergence can also be controlled by the usual CLEAN_ARES (#1, maximum Absolute RESidual value) , CLEAN_FRES (#2, maximum Fractional RESidual value) and CLEAN_NITER (#3, maximum Number of ITERations) criteria, plus CLEAN_NCYCLE for methods with major cycles. A fourth criterium (#4) is convergence, which is controlled by CLEAN_NKEEP, a number of components. Deconvolution of a given channel stops if the cumulative flux at iteration number N is smaller (resp. larger) than at iteration N-CLEAN_NKEEP for positive signals (resp. negative). In essence, CLEAN_NKEEP is the number of components when the signal is just above the noise. Experimentation with various types of images has shown that CLEAN_NKEEP = 70 is a good compromise.
However, criteria #1-3 can be set to 0, allowing IMAGER to automatically guess when to stop. In this case, IMAGER uses an absolute residual threshold equals to the noise level (available in dirty%gil%noise), and estimates a (very conservative) maximum number of Clean components.
Using CLEAN_STOP has precedence over the other ways of specifying the stopping criteria.