The complex time-depdent gains for each antenna/spwid are determined from the ratio of the data column (raw data), divided by the model column, for the specified data selection.  The gains can be obtained for a specified solution interval for each spectral window, or by a spline fit to all spectral windows simultaneously.  Any specified prior calibrations (e.g., bandpass) will be applied on the fly.


The fundamental calibration to be done on your interferometer data is to calibrate the antenna-based gains as a function of time, using gaincal.  Systematic time-dependent complex gain errors are almost always the dominant calibration effect, and a solution for them is almost always necessary before proceeding with any other calibration solve. Traditionally, this calibration type has been a catch-all for a variety of similar effects, including: the relative amplitude and phase gain for each antenna/polarization, phase and amplitude drifts in the electronics of each antenna, amplitude response as a function of elevation (gain curve), and tropospheric amplitude and phase effects. In CASA, it is possible to handle many of these specific effects separately, as available information, circumstances, and required accuracy warrant, but if accuracy is not paramount it is still possible to solve for the net effect using a quick-and-dirty gaincal.  In fact, gaincal is often used for an initial exploration of a dataset, to find data problems, etc.  Also, a provisional gaincal solution can be used as prior calibration to optimize bandpass calibration.  Such gaincal solutions are typcially discarded.

It is best to have determined a (constant or slowly-varying) bandpass from the frequency channels by solving for the bandpass, and to include any other anciillary calibration that may be available via gencal (e.g., gaincurve, antenna position corrections, opacity, etc.).

See here for more information on the task parameters gaincal shares with all solving tasks, including data selection, general solving properties and arranging prior calibraiton.  Below we describe parameters unique to gaincal, and those common parameters with unique properties.

Gain calibration types: gaintype

The gaintype parameter selects the type of gain solution to compute.  For complex gain calibration, the choices are 'T', 'G', and 'GSPLINE'.  The gaincal task also supports rudimetary delay solutions using 'K' and 'KCROSS'.

Polarization-dependent sampled gain (gaintype='G')

Generally speaking, gaintype='G' can represent any multiplicative polarization- and time-dependent complex gain effect downstream of the polarizers. (Polarization- and time-independent effects upstream of the polarizers may also be treated implicitly with G.) Multi-channel data (per spectral window) will be averaged in frequency before solving (use calibration type B to solve for frequency-dependent effects within each spectral window).

Polarization-independent sampled gain (gaintype='T')

At high radio frequencies (>10 GHz), it is often the case that the most rapid time-dependent gain errors are introduced by the troposphere, and are polarization-independent. It is therefore unnecessary to solve for separate time-dependent solutions for both polarizations, as is the case for gaintype='G'. Thus gaintype='T' is available to calibrate such tropospheric effects, differing from G only in that a single common solution for both polarizations is determined. In cases where only one polarization is observed, gaintype='T' is adequate to describe the time-dependent complex multiplicative gain calibration entirely.  For the dual-polarization case, it is necessary to ensure that the two polarizations are, in fact, coherent by using a prior G or (unnormalized) bandpass calibration. 

Spline gains (gaintype='GSPLINE')

At high radio frequencies, where tropospheric phase fluctuates rapidly, it is often the case that there is insufficient signal-to-noise to obtain robust G or T solutions on timescales short enough to track the variation. In this case it is desirable to solve for a best-fit functional form for each antenna using the GSPLINE solver. This fits a time-series of cubic B-splines to the phase and/or amplitude of the calibrator visibilities.

The combine parameter can be used to combine data across spectral windows, scans, and fields. Note that if you want to use combine='field', then all fields used to obtain a GSPLINE amplitude solution must have models with accurate relative flux densities. Use of incorrect relative flux densities will introduce spurious variations in the GSPLINE amplitude solution.

The GSPLINE solver requires a number of unique additional parameters, compared to ordinary G and T solving. The sub-parameters are:

gaintype         =  'GSPLINE'   #   Type of solution (G, T, or GSPLINE)
     splinetime  =     3600.0   #   Spline (smooth) timescale (sec), default=1 hours
     npointaver  =          3   #   Points to average for phase wrap
     phasewrap   =        180   #   Wrap phase when greater than this

The duration of each spline segment is controlled by splinetime. The splinetime will be adjusted automatically such that an integral number of equal-length spline segments will fit within the overall range of data.

Phase splines require that cycle ambiguities be resolved prior to the fit; this operation is controlled by npointaver and phasewrap. The npointaver parameter controls how many contiguous points in the time-series are used to predict the cycle ambiguity of the next point in the time-series, and phasewrap sets the threshold phase jump (in degrees) that would indicate a cycle slip. Large values of npointaver improve the SNR of the cycle estimate, but tend to frustrate ambiguity detection if the phase rates are large. The phasewrap parameter may be adjusted to influence when cycles are detected. Generally speaking, large values (>180 degrees) are useful when SNR is high and phase rates are low. Smaller values for phasewrap can force cycle slip detection when low SNR conspires to obscure the jump, but the algorithm becomes significantly less robust. More robust algorithms for phase-tracking are under development (including traditional fringe-fitting).

GSPLINE solutions cannot be used in fluxscale. You should do at least some long-timescale G amplitude solutions to establish the flux scale, then do GSPLINE in phase before or after to fix up the short timescale variations. Note also that the phase tracking algorithm in GSPLINE needs some improvement.

Single- and multi-band delay (gaintype='K')

With gaintype='K' gaincal solves for simple antenna-based delays via Fourier transforms of the spectra on baselines to (only) the reference antenna. This is not a global fringe fit but will be useful for deriving delays from data of reasonable SNR. If combine includes 'spw', multi-band delays solved jointly from all selected spectral windows will be determined, and will be identified with the first spectral window id in the output caltable.  When applying a multi-band delay table, a non-trivial  spwmap is required to distribute the solutions to all spectral windows (fan-out is not automatic).

After solving for delays, a subsequent bandpass is recommended to describe higher-order channel-dependent variation in the phase and amplitude.

Cross-hand delays (gaintype='KCROSS')

With gaintype='KCROSS', gaincal solves for a global cross-hand delay.  This is used only when doing polarimetry.  Use parang=T and apply prior gain and bandpass solutions. This mode assumes that all cross-hand data (per spw) share the same cross-hand delay residual, which should be the case for a proper gain/bandpass calibration.  See sections on polarimetry for more information on use of this mode.

Multi-band delays are not yet supported for KCROSS solutions.