Copying, Concatenating and Averaging Visibility Datasets

The inputs are all fairly self-explanatory -- see the help file for more information. The default behaviour is to copy all the data, applying calibration corrections as it goes. If some averaging is being performed, vector averaging is normally used both for time and frequency. However an option ( options=ampscalar) switches the behaviour so that amplitude-scalar averaging is used for time averaging. Vector averaging is always used when averaging in frequency.

A brief summary of the inputs are:

• vis: a list of the input visibility datasets. Several can be given. Wildcards are supported.
• out: The name of the output visibility dataset. The output will be the concatenation of all the selected input data. No default.
• select: The standard visibility data selection parameter. See Section 5.5. The default is to select all data.
• stokes: The standard parameter to determine Stokes conversion. The default is to copy the polarisations or Stokes parameters as is (no conversion). See Section 5.8.
• interval: Time averaging interval, in minutes. The default is to perform no time averaging.
• line: The standard line parameter, which dictates channel selection and frequency averaging in multi-channel datasets. The default is to copy all channels unaveraged. See Section 5.4.
• options: Extra processing options. The nocal, nopass and nopol options are used to disable correcting for antenna gains, bandpass functions and polarisation leakage (see Section 5.3).

  There are also three options to determine the sort of averaging to be performed. Possible options are vector, scalar and scavec. Note these options only affect the averaging in time. Vector averaging is always use when averaging in frequency.

  • ``Vector'' averaging is just a straightforward average of the complex-values visibility data. This is the optimum thing to do if the data are corrupted by just Gaussian noise. However it is common for the data to be affected by fast-varying phase fluctuations. While the amplitude of the data are good before averaging, a plain vector average will result in some decorrelation.
  • ``Scalar'' averaging avoids this by using the average amplitude for the amplitude of the output quantity (the phase of the output is still the phase determined by vector averaging). The problem with scalar averaging is that it results in a noise bias when the signal-to-noise ratio is small (less than a few) -- you should not use scalar averaging in this regime.
  • For cases where phase stability is poor, and the signal-to-noise ratio on the parallel-hand correlations ( XX and YY ) is good, but the signal-to-noise ratio on the cross-hand correlations ( XY and YX is poor, uvaver provides the scavec options. This performs scalar averaging on the parallel-hand correlations, and vector averaging on the cross-hand correlations.

For example, the inputs below show how to form a channel 0 dataset from a data set with two spectral windows (IFs), each with 33 channels (selecting the middle 25 channels from each window). The two spectral windows are essentially treated as 66 contiguous channels. Thus, we must use the line keyword to select channels 4 to 28, and then channels 37 to 61. uvaver will correclty maintain the dual window status (use prthd to verify this).