The current VLA correlator is limited to a bandwidth of 
MHz. A new correlator is needed to process the 
2 GHz of
bandwidth and to achieve the increase in continuum sensitivity and
instantaneous spectral coverage. Moreover, the current correlator
limits the type of science which can be done due to its limited
spectral resolution at wide bandwidths. With a 50 MHz bandwidth now,
the VLA can only produce 8 (Hanning smoothed) spectral channels in
total. With so few channels, wide-field imaging at low frequencies,
and searches for redshifted spectral lines (e.g., H I) are extremely
inefficient. Similarly, a 50 MHz bandwidth at high frequencies (e.g.,
350 km/s at 43 GHz) makes observations of radio recombination lines
difficult, if not impossible. It also excludes many components of
those molecular lines which are split into multiple transitions.
Current specifications for the new correlator are demanding:
Current specifications call for processing 1 GHz of IF bandwidth in two orthogonal senses of polarization. The cost of the digital part of the correlator is minimized if the band is subdivided before digitization, but this increases the cost and complexity of the analog part. There is some number of baseband channels, presently undetermined, that will minimize the cost. It appears that the minimum number of channels is 2 per polarization (making 4 channels of 500 MHz each per antenna), and each of these should be separately tunable within the front end bandwidth. The present state of the art in digitizers allows channels up to about 2 GHz wide (4 GHz sampling rate, several bits per sample), so sampling does not constrain the channelization. Considerations other than cost may be important. In the case of strong in-band interference, a large number of channels is favored so that bad channels can be rejected before digitization.
If the IF channelization is such that two pairs are employed, a minimum of 512 channels in each of the four 500 MHz IF channels is specified. The bandwidth of each channel would be adjustable by factors of two over a wide range, allowing widely variable spectral resolution. The spectral resolution should, at least, be comparable to that of the Millimeter Array, of order 0.05 km/s at high frequencies. This represents a factor of three improvement in the spectral resolution on the widest bandwidth now available, and an order of magnitude improvement in the largest spectrometer bandwidth now available.
For bistatic planetary radar resolutions of 
Hz are desired
over a few kHz of bandwidth. This can be achieved by bandlimiting the
input channels with filters and then drastically reducing the
correlator clock rate, but might be better done by recording the
digitized signals on tape and correlating in software off-line.
Support of very short dump periods (<0.1 sec) is feasible for the correlator proper, but we know of no way to handle the data flow if all channels and baselines are used. We assume that each spectral channel of each baseline will have two integrating bins (say, noise calibration on/off or pulsar on/off) available to it; perhaps four bins could be provided, but not a large number. Observations requiring high time resolution (tens of msec) will have to make large sacrifices in other parameters, e.g., spectral channels or baselines.
We see no reason to change the present VLA design in which fringe rotation is done in an LO at each antenna. (Moving it to after the digitizers implies a large increase in correlator size.) To accommodate the addition of nearby VLBA antennas, which do not have fringe rotation, there is a consensus that this capability should be added in an LO that is part of a ``VLBA to VLA interface box." Similarly, the VLA phase switching scheme should be retained and that feature should be added to the VLBA antenna signals as part of the interface. It is noted that observations requiring high output dump rates might not be able to make effective use of the phase switching, thus limiting their dynamic range.
Careful consideration will be given to including hardware internal to the correlator for real-time detection and removal of short-time-scale interference. For example, it might be feasible (in an FX correlator) to examine each FFT for channels with excessive amplitude and delete those channels from the cross-correlation.
If possible, the development of a new VLA correlator should be combined with that of the Millimeter Array correlator. The overall similarity of the needs in terms of the antenna-bandwidth product should be exploited wherever possible. Unless the time scales on which the two correlators are required turn out to be significantly different, they should at least share the same basic architecture and the same ASIC(s).