Regularized Calibration Solver

implementation of parallelelized code results are the same as for the serial code within numerical error limits of compute platform provide capability to perform calibration with smoothing constraints on several 1000 channels the quality of smoothing is shown for different weights description of empirical relationship between the smoothness constraint and the level of noise

Parallel smoothing constraints are implemented in askapsoft trunk, starting from svn commit #11105. Final calibration tables were compared to serial runs for different mpi partitioning, and results are exactly the same (with intermediate vector norms the same for 10 figures) Calibration with smoothing constraints can now be run on several 1000 channels, as each cpu only calculates its local matrix part (of the huge solver matrix where different channels are coupled via constraints). The channel chunk size can be defined in the input parameter file, with the minimum 1 channel per worker. The parallelization is done by model (i.e., by gain parameters being solved for), in the least squares framework. This framework allows adding various other types of constraints into the system in future, e.g. direction dependant calibration, ionospheric effects, uncertainty etc.