The DiFX correlator has a capability to process many phase centers within a primary beam in one pass. It does this by cross correlating with high spectral resolution and short time integration, then splitting the data paths out for each desired phase center, shifting the delay and phase for the new center, and integrating in time and frequency. There is a price of about a factor of 2.5 to do the large transforms involved, but there is very little additional burden for each phase center. The difference in processing time between 20 phase centers and 200 is about 20%. Up to about 500 phase centers in a field have been tested. This mode is expected to be popular for surveys and for in-beam calibration at lower frequencies.
SCHED supports the multiple phase center mode by providing the details of all the desired phase centers to use for a given pointing center in the output .v2d file, which is used in correlator setup. In the future, this information could also go to the VEX file when the standard and the readers can receive the information.
To invoke multiple phase center processing, and specify the centers, the user should provide a list of all the sources to use with each pointing center. To do this a PCENTERS section should be added to the main SCHED input file. Within that section, each group of centers is given a name and a list of source names. The sources need to correspond to sources in the source catalogs. It is likely that the user would create a catalog of the offset pointing centers and invoke it, along with the standard catalog, using the new ability to use two source catalogs using SRCFILE and SRCFILE2.
To tell SCHED to use one of the named lists of pointing centers, specify the name of the group from PCENTERS using the input CENTERS for each scan. For now, the same list must be used for all scans on a pointing center and all scans on that pointing center must use the list. The internal structure of SCHED will allow that one-to-one correspondence between pointing centers and phase center lists to be relaxed if and when the information can be transmitted to the correlator.
When using this capability, the user should specify the size of the FFTs to do on each baseband channel before splitting the data for each phase center. This is done with the (new) second argument to SCHED input parameter CORCHAN. Normally that argument can be ignored and the FFT size will be set to the larger of 128 or the first argument. The FFT size needs to be large enough that there is is insignificant delay smearing in a single channel. The required size depends on the baseband bandwidth (B MHz) and the maximum distance between the pointing and phase centers (x arcsec). For the VLBA for a 5% loss of amplitude on an offset source, the number of channels should be near B*x/2.35. A somewhat more conservative criterion is to limit phase winding to 5 spectral points per turn of phase in the spectrum. By that, given that the maximum delay change for a 1 arcsecond shift is 139 ns on the longest VLBA baseline (8600 km), the number of channels should be near 5*0.139*B*x. The full width, half maximum beamwidth of a VLBA antenna (25m) is very close to 30 and 1 arcminutes at 1.4 and 43 GHz, respectively. For the 1.4 GHz case, the above two criteria give 3063 and 5004 channels for a 4 MHz baseband channel width. A specification of 4096 channels would likely be reasonable. With the RDBE, much wider baseband channels are possible requiring many more spectral channels.
See the Wide-Field Imaging section of the Observational Status Summary for more information on delay and fringe rate smearing.
There is an example, egcent.key, with an associated catalog egcentsrc.dat, with the SCHED examples to demonstrate the use of this mode.