Table 2.4 shows the capacity of the current VLA
correlator, whose properties were specified in the mid-1970's.
One measure of correlator size is the number of multiplications
required per second, which scales as:
(Number of antennas)
(Maximum value of: (Bandwidth Number of channels at that
bandwidth)).
Table 2.5 compares this figure of merit for
existing or proposed synthesis-array correlators.
| Bandwidth | Channels | Separation |
| (MHz) | (kHz) | |
| 50 | 16 | 3125 |
| 25 | 32 | 781.25 |
| 12.5 | 64 | 195.313 |
| 6.25 | 128 | 48.828 |
| 3.125 | 256 | 12.207 |
| 1.5625 | 512 | 3.052 |
| 0.78125 | 512 | 1.526 |
| 0.195313 | 512 | 0.381 |
| Correlator | Antennas N | Bandwidth B | Channels C | ``Size" |
| (MHz) | ( ) | |||
| VLA now | 27 | 50 | 16 | 5.8 E 5 |
| ATCA | 6 | 128 | 128 | 5.9 E 5 |
| VLBA | 20 | 128 | 128 | 6.6 E 6 |
| WSRT | 20 | 160 | 512 | 2.1 E 7 |
| SMA | 6 | 1968 | 3072 | 2.2 E 8 |
| MMA | 40 | 2000 | 1024 | 3.3 E 9 |
The scientific drivers for a new VLA correlator have been reviewed by Rupen (1997) in VLA Upgrade Memo #8. A new, and much larger, correlator is clearly required because (a) the maximum of 16 channels at 50 MHz bandwidth is inadequate for most spectroscopy, (b) much wider bandwidths (8 GHz per polarization, versus the current 100 MHz) are now proposed for the higher observing frequencies, and (c) the number of antennas whose outputs must be correlated will increase to forty or more from the current twenty-seven as we pursue increased angular resolution.
The new correlator must also provide flexibility in choice of bandwidths and frequency resolution (numbers of channels) that will allow observers to take full advantage of the improved sensitivity and broad-band capability of the instrument. This broad requirement of flexibility encompasses several other desiderata, notably high linearity, so that strong interfering signals can be removed in post-processing without corrupting nearby channels; the ability to recognize and remove time-variable interference on sub-integration timescales; the availability of temporal gating for pulsar observations and to blank out pulsed RFI. It also requires support for up to five independent subarrays, and most importantly the ability to use different frequency resolutions for each of several independently-tunable IF's (sub-bands). From the point of view of spectroscopy, a reasonable maximum requirement for the is 8192 channels (summing 4096 in each of RR and LL) over 125 MHz. As this replicates the MMA correlator's capacity in this key quantity, one option that may be attractive logistically is to duplicate the proposed MMA correlator (Escoffier 1997; Rupen and Escoffier 1998). This correlator is described in more detail elsewhere, and we discuss here only how it would appear to the observer. The 16 GHz bandwidth (8 GHz in each of two polarizations) available for each antenna is presented to the correlator in four pairs of baseband signals, each with a maximum bandwidth of 2 GHz. The possible modes for a single baseband pair (pair of oppositely polarized signals) are given in Table 2.6.
Single polarization product | Two polarization products | Four polarization products | ||||
| Bandwidth | No. | Freq. | No. | Freq. | No. | Freq. |
| MHz | Channels | Separation | Channels | Separation | Channels | Separation |
| kHz | kHz | kHz | ||||
| 2000 | 256 | 7812.5 | 128 | 15625 | 64 | 31250 |
| 1000 | 512 | 1953.1 | 256 | 3906 | 128 | 7812.5 |
| 500 | 1024 | 488.28 | 512 | 976.56 | 256 | 1953.1 |
| 250 | 2048 | 122.07 | 1024 | 244.14 | 512 | 488.28 |
| 125 | 4096 | 30.52 | 2048 | 61.04 | 1024 | 122.07 |
| 62.5 | 8192 | 7.629 | 4096 | 15.259 | 2048 | 30.518 |
| 31.25 | 8192 | 3.815 | 4096 | 7.629 | 2048 | 15.259 |
| 15.625 | 8192 | 1.907 | 4096 | 3.815 | 2048 | 7.629 |
| 12.0 | 8192 | 1.465 | 4096 | 2.930 | 2048 | 5.859 |
Each of the four baseband pairs can operate independently, giving a
great deal of flexibility. Most obviously, one can use the baseband
pairs together to cover a wider bandwidth at the same frequency
resolution: the widest-band mode uses all four baseband pairs to cover
4 2 = 8 GHz with 4 256 = 1024 channels, if only a
single polarization product were desired; or one could cover 4
0.25 = 1 GHz with 4 512 = 2048 channels and full
polarization information (four polarization products), yielding a
channel separation of 490 kHz. One could also use more than one
baseband pair to produce higher frequency resolution for the same
input bandwidth, to provide 32768 (single polarization product)
channels over 15.625 MHz, with a channel separation of 0.48 kHz. For
comparison, the current VLA correlator gives a channel separation of
195 kHz over 12.5 MHz with 64 channels (see
Table 2.4).
More complicated modes are also possible, since each baseband pair is independent. One might, for instance, use one baseband pair to produce a 512-point, full polarization spectrum covering a total of 250 MHz; employ another two to cover a total bandwidth of 62.5 MHz at 3.8 kHz resolution, RR only; and use the last baseband pair to ``zoom in" on 1 MHz of the spectrum with 4096 dual-polarization channels, giving a resolution of 0.24 kHz. The highest available frequency resolution is set by the IF/LO system rather than the correlator; if that system could provide a 32.8 kHz bandwidth to the samplers, the resulting channel separation could be as little as 1 Hz. Finally, this correlator design can handle up to four independently-tunable IF pairs if these are provided by the LO system. Even with only two independent LO systems, one could observe simultaneously two lines, or a narrow line and a broad-band continuum.
The MMA correlator requirements do differ from those of the in several respects, however. For the there is little scientific case for a contiguous bandwidth greater than 500 MHz, and only one scientific driver requires more than 16384 channels (full-field, full-band, full-polarization imaging at L, S, or C Bands in the A+ configuration). Deep continuum imaging experiments would use a total bandwidth of 16 GHz, as in the MMA, apportioned as two polarizations of 8 GHz each. As 500 MHz is the widest required individual bandwidth, sixteen separate input pairs of this width or eight pairs of 1 GHz would be appropriate for the correlator. For the the desire to do bistatic radar experiments sets a very small narrowest channel width of a few Hz, but this need only be available for about 2048 channels total; the lowest channel width driven by (extra-solar-system) spectroscopy is about 200 Hz.
The design of the correlator must also address two concerns driven by the current and expected future ``hostile environment". It is important to have considerable tuning flexibility to avoid harmful RFI. This is primarily of concern to designing the LO/IF system: very strong RFI must be blocked before reaching the correlator. Having more, but narrower, pairs of inputs to the correlator will, however, help strategies to avoid ``mid-strength" RFI. As there will always be some RFI within the bands input to the correlator, a high spectral dynamic range is also important. This requires more quantization levels than the MMA's planned 2-bit, 4-level system, especially at the narrower input bandwidths which would be used at lower frequencies.
A critical aspect of the design will be the ability to assign different numbers of spectral channels amongst the independently tunable IF's (``sub-bands''). This is needed for targeting up to 8 different spectral transitions with differing resolutions appropriate for each, and for permitting RFI avoidance by moving the sub-bands to cleaner parts of the spectrum with sufficient spectral resolution to permit efficient removal of weaker RFI. In a similar vein, the gating ability is needed both for scientific and practical purposes - observations of pulsars, and potentially, to blank the correlator against pulsed RFI.
The new correlator must also be designed to be ``VLBA-Enabled''. To include more distant antennas in the A+ configuration, it must be able to handle high fringe rates and delays, and include buffering to account for delays in signal transmission over commercial fiber optic lines. Ultimately, the correlator could become a 50-station unit, processing all the current VLA and VLBA antennas in real time, plus as many as eight new antennas, plus a number of ``foreign'' stations, for special experiments that require the ultimate in spatial frequency coverage.
The clearly requires a very large correlator to achieve the desired scientific productivity, but one that is technically feasible and in many respects similar to that required for the MMA. Plans for the development of the MMA and correlators will therefore be closely coordinated, whether or not the two are ultimately identical.