The Correlator

The correlator is of an XF design which performs the cross correlation first, followed by a Fourier transform. More precisely, the IF signals, once they are at baseband, are rapidly sampled (see the section below on the lag spectrum) and then cross correlated for a range of lags leading to correlation coefficients R(u,v,$\tau$). These samples (``lags'') are then Fourier transformed to obtain r(u,v,$\nu$), the visibilities as a function of frequency. The correlator has four identical quadrants and the four IFs can be distributed over these quadrants in a number of ways. This leads to three basic groups of correlator modes:

• continuum: parallel and cross-products are produced allowing all four Stokes parameters to be measured for each IF pair.

• line: only RR and/or LL correlations are determined.
  • single-IF: one IF (A, B, C, or D) uses all four quadrants.
  • two-IF: each IF uses two quadrants.
  • four-IF: IF A, B, C, and D each use one quadrant.

• polarization line: one IF pair (IFs A and C or IFs B and D) uses all four quadrants to deliver parallel- and cross-products.

Continuum and line mode are treated differently in the sense that in line mode positive and negative lags (time domain) are produced by the XF correlator and processed to generate complex spectra (frequency domain). In continuum mode a quadrature network is used to generate cosine and sine correlation products.

The availability of the aforementioned correlator modes is summarized below:

• Continuum observations are fully supported.

• All one-IF line modes (1A, 1B, 1C, and 1D)^2.3, all two-IF modes (2AB, 2AC, 2AD, 2BC, 2BD, and 2CD), and mode 4 (allowing one spectrum for each of the four IFs) fully function.

• The spectral line polarization modes (PA and PB), producing RR, RL, LR and LL correlation products based on either IFs A and C or IFs B and D, are available. It should be noted that in spectral line mode the round-trip phase^2.4 correction for IFs B and C cannot be computed. Instead, the correction for IF A is applied to IF C and the correction for IF D is applied to IF B, leaving a residual phase drift. This small drift is normally calibrated out in the parallel hands but not in the cross-hand polarizations leading to a slow apparent rotation of a few degrees per hour of the plane of the polarization.

• Multiple subarrays in spectral line or continuum mode are supported, but mixed spectral line and continuum subarrays are not allowed.

The following is worth noting:

• The individual IFs may have independent bandwidths. If one IF pair is set at a narrow bandwidth to observe a spectral line using on-line Doppler tracking (see below), and the other pair uses a very wide bandwidth (e.g. 50 MHz) to sample over as wide a continuum as possible, Doppler tracking must be switched off for the latter IF pair ^2.5. This mode of observation is common at the higher frequencies (consult VLA Test Memo #189 by Doug Wood for further details).

• The individual IFs may have independent data selection criteria. One can select a starting channel and the number of channels for each IF independently. The number of selected channels should be larger than two.

• Autocorrelation spectra are produced for all active antennas. The autocorrelation spectra are processed in the same way as the cross-correlation data (i.e., the options for lag spectra, Hanning smoothing and data selection are applied). These spectra are available for inspection (e.g., to check for interference), but are not useful for total power measurements.

• Bandpass (autocorrelation) normalization and Hanning smoothing are available with all spectral line modes. The same options, such as Hanning smoothing or autocorrelation normalization, must be applied to each IF in the multiple IF modes.

• Channel 0 (the ``continuum'' channel) cannot be computed correctly in the case of a four channel spectrum (before Hanning smoothing). This will arise only in the case of correlator modes 4, PA and PB with a bandwidth of 50 MHz. If using a four channel spectrum, the observer will need to generate a new ``continuum channel'' for calibration purposes with the AIPS task AVSPC. This will not be a concern if a four channel spectrum is generated as a result of data selection (e.g. if the observer selects only 4 channels from an 8 channel spectrum).

• The same integration times must be used in all subarrays.