The VLBA correlator accommodates the full range of scientific investigations for which the array was designed. The correlator supports wideband continuum, high-resolution spectroscopy, bandwidth synthesis, and polarimetric observations. Ultimately, more specialized techniques will be added: pulsar gating is currently under development; and the hardware design will accommodate simultaneous multiple frequencies or phase centers, and frequency or phase-center switching.
The correlator is designed to process all observations involving VLBA stations. With its 20-station capacity and sub-arraying capabilities, it can correlate an extended array combining the VLBA with as many as 10 foreign stations, or an extreme-wideband VLBA observation using both recorders at each of 10 stations, or two 10-station intra-VLBA observations, or virtually any combination of smaller sub-arrays, each in a single processing pass.
Each station input comprises 8 parallel ``channels'' (as defined in Section 5.13), which operate at a fixed rate of 32 Msamples per second, for either 1- or 2-bit samples. Observations at lower sample rates generally can be processed with a speed-up factor of 2 (for 16 Msamples per second) or 4 (for 8 Msamples per second or less) relative to observe time. Special modes are invoked automatically to enhance sensitivity when fewer than 8 channels are observed, or when correlating narrowband or oversampled data. The correlator accepts input data recorded in VLBA or Mark III longitudinal format. The correlator will also accept input data recorded in Mark IV format. However, Mark IV testing and verification is currently in progress and, so far, only a very limited set of modes and mode changes have been approved for use at the VLBA correlator. Section 18.1 provides further information concerning Mark IV recordings destined for the VLBA correlator.
Each input channel can be resolved into 1024, 512, 256, 128, 64, or 32 ``spectral points'', subject to a limit of 2048 points per baseline across all channels. Adjacent, oppositely polarized channels can be paired to produce all four Stokes parameters; in this case correlator constraints impose a maximum spectral resolution of 128 points per polarization state. The user may also specify a spectral smoothing function, or request an ``interpolated'' spectrum suitable for inversion to a cross-correlation function if further work is required in that domain.
The correlator forms cross-spectral power measurements on all
relevant baselines in a given sub-array, including individual antenna
``self-spectra''. These can be integrated over any integral multiple
of the basic integration cycle, 131.072 milliseconds (
microsec). Adjacent spectral points may be averaged while integrating
to reduce spectral resolution. A time-domain transversal filter is
available at the output from the integrator to maximize the
fringe-rate window while further reducing the data rate.
Correlator output is written in a ``FITS Binary Table'' format, and will eventually include amplitude and pulse calibration data obtained at observe time, and editing flags from both observing stations and the correlator (Diamond 1997). All results are archived on digital-audio-tape (DAT) cassettes. The output data rate islimited to 0.5 Mbytes per second, which must be shared among all simultaneous correlator sub-arrays. Data are copied from the archive for distribution to users on a variety of media, with DAT and Exabyte currently given primary support.
Operation of the correlator is governed primarily by information obtained from the VLBA control system's monitor data or from foreign stations' log files. A few additional items, all of which have been mentioned above, will be specified by the user prior to correlation. Supervision of the correlation process is the responsibility of VLBA operations personnel; user participation during correlation is not expected nor easily arranged, as explained below.
Scheduling of the correlator is currently done on a very short time-scale of days to optimize use of the correlator's resources and the array's stock of tapes. This makes it impractical, in general, to schedule visits by users during correlation of their data. As described in Section 23, however, users are encouraged to visit the AOC after correlation for post-processing analysis.
Consult Benson (1995) and Romney (1995), respectively, for more information on the VLBA correlator and on VLBI correlation in general.