Aug. 31, 2000

A WIDAR-based, Second Generation Correlator Design for ALMA

P. Dewdney & B. Carlson, HIA, National Research Council, Canada

The “baseline” correlator for ALMA is now under construction at NRAO. This correlator is based on the “time-burst” XF design, developed and perfected at NRAO. We are developing an enhanced XF correlator for the Expanded VLA (EVLA), using a new signal processing architecture, WIDAR (Wideband Interferometric Digital ARchitecture). Compared with current wide-band XF designs, the WIDAR design achieves an order-of-magnitude increase in the number of spectral channels at full bandwidth – for a similar cost. XF correlators have a simplicity and regularity of design that leads to low cost, especially in the ALMA correlator, where internal data distribution bandwidths tend to dominate the cost.

Should a second-generation design be needed, the WIDAR design has unique features of interest to ALMA - very flexible deployment of spectral channels in programmable sub-bands, rejection of signals in the unwanted side-band, and fully digital sub-sample delay tracking. The latter capability provides improved aperture synthesis performance over the quasi-analog techniques usually employed. The WIDAR design also provides an order-of-magnitude increase in spectral dynamic range in the presence of interference or strong lines.

The essence of the technique is sub-dividing each baseband using digital filters. The narrower bandwidths can be processed at lower sample rates separately by downstream correlation. In a present EVLA design, there will be 16 sub-bands in each 2 GHz wide baseband. This leads to an increase by a factor of eight in the number of available spectral channels, compared with a time-burst XF design. In the simplest mode, anti-aliasing techniques permit sub-bands to be “stitched” together seamlessly to reconstruct the wide-band spectrum. Where flexibility is required, the sub-bands can be programmed, with minor restrictions, to cover any fraction of the original band. This can be used to target individual lines or groups of lines, each with its own spectral resolution. Where sub-bands are programmed at less than the maximum bandwidth, a re-circulation technique increases the spectral resolution as the inverse square of the bandwidth.

The EVLA version of this design has other features that would likely be retained for a second generation ALMA correlator - full polarization processing, fast dump capability (~10 ms), multiple overlapping sub-arrays, rapid on-the-fly re-configuration, 4-bit sampling (implying essentially no loss of sensitivity, and high spectral dynamic range in interference environments), multiple simultaneous phase centers, digital phased-antenna sum for VLBI, and independent sky frequencies for each baseband. The EVLA design will process 16 GHz of overall bandwidth (same as ALMA), with 16000 channels and more at narrower bandwidths.

If a second-generation design is started in (say) 2006-2007, further increases in performance are expected from technology improvements. The WIDAR architecture makes efficient use of hardware resources as bandwidth increases. Ignoring implications for receivers, even today it would not be outlandish to suggest doubling the baseband bandwidth (from 2 to 4 GHz) for only a modest cost increase (in a 2-bit correlator with the same number of channels). On the longer time scale, the number of correlator channels could be increased roughly by an order-of-magnitude. Alternatively, the total correlator cost could be substantially reduced.