doamp, dofloat, solint Self calibrate for telescope phase (and amplitude) corrections. EXAMPLES ------- 1. Phase-only self-calibration: 0>selfcal Performing phase self-cal Adding 49 model components to the UV plane model. The established model now contains 49 components and 5.43211 Jy Correcting IF 1. A total of 2187 telescope corrections were flagged in sub-array 1. Fit before self-cal, rms=0.333969Jy sigma=1.037002 Fit after self-cal, rms=0.325501Jy sigma=1.012208 Reinstating the default IF. 0> 2. Amplitude and phase self-calibration with a solution interval. 0>selfcal true,true,1440 Performing amp+phase self-cal over 1440 minute time intervals Correcting IF 1. A total of 222 telescope corrections were flagged in sub-array 1. Fit before self-cal, rms=0.325501Jy sigma=1.012208 Fit after self-cal, rms=0.255853Jy sigma=0.747795 Reinstating the default IF. 0> All but the first line of each example are responses from 'selfcal'. PARAMETERS ---------- doamp - (Default=false) If this parameter is omitted or given as false, only phase corrections are calculated and applied. Alternatively, if doamp=true telescope amplitude corrections will also be calculated and applied. dofloat - (Default=false) If doamp=true, then dofloat specifies how the amplitude corrections are to be applied. Initially the corrections are applied verbatim to the visibilities. If dofloat is true then this is all that occurs. However if dofloat is false then the total amplitude self-cal corrections so far applied to the data, by this and previous invocations of selfcal and gscale, are summed over all integrations and telescopes to form a mean, and then all corrections are divided through by this number to normalize them. This provides an established but hard to justify means of anchoring the flux scale, in the event that feedback in the CLEAN+selfcal loop becomes slightly positive. It is also often abused as a means to get amplitude self-cal to work before the model has been completed at short UV spacings. If you are tempted to do this please look at the help given for the selftaper command. solint - The solution interval in minutes. Solutions are calculated within top-hat solution bins of this width and then are smoothed with a gaussian filter. The half power point of the filter in the Fourier plane is at 1/(2.solint). According to the Shannon sampling theorem this is the highest frequency that can be sampled with a sample grid spacing of solint. CONTEXT ------- Self-calibration is a means to finding self-consistent telescope based amplitude and phase corrections, through comparison of the observed data to an approximate model of the source structure. The algorithm implemented here performs least-squares minimization. It was based on sample code provided by Tim Cornwell. As explained in the 'clean' help topic, self-calibration with 'selfcal' is usually combined in a loop with 'clean' and a number of auxiliary data editing, display etc.. commands. The initial model to use with selfcal may either be provided by a few iterations of CLEAN, or by an external model read in with the rmodel command. Such an external model could come from model fitting, or be the model from a different map of the same source. It is strictly recommended that one initially avoid amplitude self-calibration, and only use phase self-calibration until the model flux from CLEAN reaches that of the data, as displayed in radplot. Only at this point should amplitude self-calibration be used to remove amplitude errors. If you have trouble getting all the flux at short UV spacings, try switching to natural weighting (see help uvweight) and/or to using a UV-taper (see help uvtaper) before continuing CLEANing. This may reveal the extended emission implied by the extra flux at short spacings. If this still doesn't reveal any more believable flux then, and only then attempt amplitude self-cal, but be sure to use the selftaper command to weight down the significance of the badly fitting short UV spacings. If you fail to use selftaper in this way then selfcal will happily attempt to drag the short UV spacing visibilities down to fit the erroneously low model. When interleaved applications of clean and selfcal start to provide diminishing returns, the residual map (as displayed by 'mapplot map') should appear to contain only noise and be devoid of beam-shaped features. Depending on the quality of the data, this may not always be achievable. If artifacts remain after many cycles of clean and self-cal, then this may indicate, either low-surface brightness extended emission that will take some time to clean out, or real problems with the data. Both cases can be determined by examining the fit between the model and observed visibilities with the 'vplot' command and the radplot command. Judicious editing may then be used to remove any bad data. In addition, self-calibration will fail to find phase-solutions for any telescope that appears on less than one un-flagged closure triangle at a given integration and similarly fail to find gain solutions for telescopes with no un-flagged closure amplitudes. The selfflag command may be used to specify how to deal with this. The default is to flag all corrections that can not be deduced. These correction flags are propagated to the visibilities that they effect when the corrections are applied to the data. To assign extra weight to a given telescope during self-calibration, or to prevent the complex gain of a given telescope from being changed see help on the selfant command. Application of selfcal to multiple sub-arrays, IFs and polarizations -------------------------------------------------------------------- Selfcal is applied separately to each sub-array of each IF. Corrections accumulated in this fashion are propogated to all spectral-line channels and polarizations in the same IFs and sub-arrays. Thus by careful mapping and self-calibration of a strong spectral-line channel you will also be correcting the phases and amplitudes of the weaker channels so that they can sub-sequently be cleaned. Agreement factors ----------------- Selfcal displays two agreement factors. The one denoted 'rms' is the root-mean-square difference between the model and observed visibilities, measured in Jy. This is formed as the mean of the squares of the amplitude difference vectors between the model and observed visibilities. The second, which is denoted by the name 'sigma', is the root mean square of the squared difference between the data and the model (taking the real and imaginary parts of visibilities as separate measurements) divided by the individual variances implied by the visibility weights. This would be equivalent to the square-root of the reduced chi-squared, except that no account is taken of the number of degrees of freedom implied by the number of gains being varied. RELATED COMMANDS ---------------- selfflag - Used to control the fate of un-correctable data. selftaper - Used to down-weight short baselines during selfcal. selflims - Used to set limits to amp and phase corrections in selfcal. selfant - Set antenna based constraints in selfcal. gscal - Calculate overall telescope amplitude corrections. startmod - Phase selfcal to a starting model then discard the model. corplot - Display and allow editing of corrections. uncalib - Remove selected parts of corrections.