5 The Future

All the above numeric calculations apply to our current situation. Naturally, there is work ongoing both to reduce the minimum integration time and to increase the maximum number of frequency points. We list some of these efforts below to whet the appetite. All of these statements are based on my current understanding of the various projects, and not necessarily on any actual testing.

• The PCI (Post-correlator integrator) aims to allow integrations as short as 1/64s for the full correlator. Indeed, it has been the first stage of this project that has allowed us to drop the full-correlator minimum $t_{\rm int}$ to 0.25s; thus another factor of 16 remains. Promising tests have recently been conducted at $t_{\rm int}=1/16$s.
• Recirculation can increase $N_\nu$ for experiments that use $BW_{\rm SB} < 16$MHz (by the factor $\mathcal{R} = 16/BW_{\rm SB}$). However, the rules that a single baseline/subband/polarization product must fit onto a single correlator board (thus the maximum $N_\nu=2048$) and that $N_{\rm SB}N_{\rm pol_{\Vert}} \le 16$ would still apply. The type of experiment that would benefit most from recirculation from a practical viewpoint would be narrow-band global spectral line -- if $BW_{\rm sb} \le 4$MHz, then the factor-of-4 ``penalty" incurred in increasing $N_{\rm sta}$ from 8 to 16 (cf. equation 4 or table 1) can be absorbed by recirculation ( $\mathcal{R}\ge 4$), allowing a global experiment to achieve the same $N_\nu$ as an 8-station experiment can currently. However, should recirculation be used, the minimum integration time would be increased by a factor of $\mathcal{R}$ from its nominal value.
• We now operate using so-called ``local" validity. ``Global" validity would increase the number of possible frequency points (and also the output size of the data) by a factor of two . We don't currently use this because it problems ensue for MkIV-MkIV baselines (i.e., when both stations use a data-replacement frame-header format) when the a priori delay is such that portions of both frame headers correlate against each other. The data in affected time ranges must be flagged (and this time range increases with $N_\nu$, resulting in more data loss specifically in those cases you'd most want to take advantage of the extra factor of 2 -- for more information about this, see ftp://ftp.jive.nl/pub/rmc/dlyrt0.ps.gz). In the future, this problem should be avoided with Mark5B recordings, which when played back should output a non data-replacement frame-header format. This doubling of the effective correlator capacity would also increase the maximum $N_\nu$ to 4096. Until the full PCI is on-line, there may also be a corresponding factor of two penalty in the minimum integration time (because of the larger data-block size).