EVN MkIV DATA PROCESSOR AT JIVE Operational & Development Status Sheet 12 January 2010 OPERATIONAL HOURS: Standard operational hours 80/week; e-VLBI correlation manned throughout the observations. CURRENTLY OPERATIONAL SYSTEM: 16 Mark 5 systems, in an adjustable mix of 5A+ and 5B units, are available for use (all now housed in racks, both head-stacks from 5A+ units connected to SUs). The Mark 5A+ units can play back either 5A or 5B recordings. Playback of: MkIV, VLBA (including data modulation), Mark5A, Mark5B. MkIV, VLBA, 5A: Supported track rates: 4, 8, or 16 Mbps/track X up to 64 tracks multiplexing (fan-out) of 1:1, 1:2, or 1:4 5B: bit-streams of up to 32 Msamp/s Maximum total data rates (single pass) 1 Gbps: 5A: 64 tracks at 16 Mbps/track; demultiplexed to 32 bit-streams at 32 Mbps 5B: 32 bit-streams at 32 Mbps Bandwidths: 500kHz, 1 MHz (both via oversampling via 5A), 2, 4, 8, 16 MHz. All subbands correlated at the same time must have the same bandwidth. Sidebands: Upper, Lower, or Dual Polarization: Single or Dual or Cross (mixed single- and dual-pol channels in an experiment need to be correlated in separate passes) Sampling: 1 or 2 bits per sample (correlation of experiments in which some stations record at 2-bit sampling and some at 1-bit sampling is also possible -- providing the same channel set-up for stations having a factor of 2 difference in recording rates). CORRELATION: Spectral Capacity (full correlator, no recirculation): 9-16 telescopes: 512 spectral points/baseline 2-8 telescopes: 2048 " " These spectral points must be spread evenly over the product of the number of subbands times the number of polarizations correlated per baseline, with a minimum of 16 spectral points per individual Bsln/SB/Pol. There is a separate maximum of 16 (subbands)*(parallel polarizations) arising from the station units -- i.e., either 16 single-pol subbands or 8 dual-pol subbands (regardless of whether cross-pols are correlated). Oversampled data correlated by "decimating" the (demultiplexed) bit-streams down to Nyquist at the input to the correlator (thus is not a means to increase SNR of the correlation products). Recirculation can be used when individual subband (BBC-filter) bandwidths are less than 16MHz. Recirculation increases the amount of spectral points available by the factor 16/BW_sb, where the subband bandwidth is in MHz, up to a maximum factor of 8. The maximum number of spectral points per baseline/subband/polarization remains 2048 even with recirculation. At the time of writing, recirculation does not work successfully with oversampled data. Integration Times: Minimum integration time using the whole correlator is 0.25s. For correlation modes that use half or less of the correlator, 0.125s integration times seem possible. When recirculation is used, the minimum integration time increases by the recirculation factor. The maximum output rate from the correlator is 6 MB/s (the raw data is in lag-space, and includes ancillary header information). Exceeding Correlator Capacity via Multiple Passes: Correlation of multiple-subband experiments at a higher spectral resolution than permitted by the full set of N_sb*N_pol can be done in multiple passes, in which a smaller set of N_sb or N_pol may be correlated separately, each with higher number of spectral points. The "Spectral Capacity" numbers above represent the current maximum per SB/pol if done in separate passes. Cross-pol correlations can not be split by polarization. Correlation at multiple phase-centers currently supported by multiple passes. Correlation of projects with more than 16 stations recording simultaneously is possible by processing in multiple passes and merging the data to remove redundant visibilities. For 17-24 stations, 3 passes would be required to ensure correlation of all baselines. A maximum integration time of 1s may be require to ensure alignment of all visibility epochs. No automated handling of sub-netted observing schemes (but can be manually made to work in reasonable cases). OUTPUT DATA FORMAT: Data is transformed into aips++ measurement sets (MS version 2), in which stage the data are reviewed and various post-correlation corrections applied. The processed MS's are then transformed into IDI-FITS files, ready for analysis in classic AIPS or other packages. The data-size the user would receive at the maximum correlator-output rate is on the order of 7-12 GB per hour of observation (FITS files with higher numbers of frequency points seem to be larger for a given raw-correlator ouput size). The internal correlator archive comprises raw correlator-output data files in a JIVE-specific format. The EVN Archive provides access to the FITS files resuliting from the post-correlation review/processing, plus standard-plots, pipeline results, and station feedback. -------------------------o-------------------------- Status of the JIVE Software Correlator, SFXC Development of the software correlator SFXC at JIVE began as part of the FP6 EXPReS joint-research activity FABRIC, and continues under the related project SCARIe funded by the Dutch national science agency (NWO). SFXC has been used for correlating ftp fringe tests during EVN Network Monitoring Experiments since 2006, and is nearing operational readiness. The anticipated advantages of SFXC over the MkIV Data Processer would include correlating data at total rates above 1Gbps in a single pass, handling sampling rates higher than 32Mbps in a subband, achieving more than 2048 frequency points per baseline/subband/polarization, using integration times under 1/4 sec, and applying gating & binning to pulsar observations. Correlation and output capacities of SFXC are conceptually simpler than for the EVN MkIV Data Processor because it lacks the specific hardware limitations of the station units and correlator-board architecture. The number of subbands would be limited by the capabilities of the data-acquisition system used during observation at the stations, with no intrinsic limit to the subband bandwidth. Currently all subbands correlated together must have the same bandwidth. The number of frequency points per baseline/subband/polarization is limited only to a power of two, with a minimum of 32. Limitations within user analysis packages may limit the largest practical N_frq. The mimimum integration time would be set by the need to accumulate a correlation function within an integration: thus on the order of 2*N_frq/BW_SB for statistically independent integrations (parameterized via 0.256 * [N_frq/2048] * [16MHz/BW_SB] ms for Nyquist-sampled data), but would also currently be limited to powers of two. The size of the output data would scale as N_frq/t_int per baseline/subband/polarization. The higher possible limits to N_frq and lower possible limits to t_int mean that the size of the SFXC output data could explode much more rapidly than from the MkIV Data Processor. Pulsar Gating/Binning is designed to operate such that N bins can be placed within a fractional range of a period. Each bin can accumulate an independent correlation function, coherently averaging over an integration period as for normal correlations. N=1 with the fractional range covering the pulse profile would correspond to tradional pulsar gating. The minimum statistically-independent bin size would also be related to the number of frequency points and bandwidth subband; otherwise there is no intrinsic maximum value of N. The output data size would scale directly with N. The pulsar's phase ephemeris would typically be passed along by the user in the form of (geocenter) polyco files. The output path for SFXC correlations follows the aips++ MS route to IDI FITS files, as for the MkIV Data Processor.