The ALMA Correlator PDR was held on 20-21 Jan 2000 in Charlottesville. In attendance were: A. Bos (bos@nfra.nl), Torres (torres@iram.fr), A. Baudry (baudry@observ.u-bordeaux.fr), R. Dickman (rdickman@nsf.gov), J. Jackson (jjackson@nrao.edu), D. Edmans (dedmans@nrao.edu), D. Woody (dwoody@caltech.edu), External reviewer L. D'Addario (ldaddari@nrao.edu), Ray Escoffier (rescoffi@nrao.edu), Chuck Broadwell (cbroadwe@nrao.edu), Joe Greenberg (jgreenbe@nrao.edu), Jim Pisano (jpisano@nrao.edu), Bob Treacy rtreacy@nrao.edu), Chikada-san (chikada@optik.mtk.nao.ac.jp), Alan Whitney (awhitney@haystack.mit.edu), Peter Dewdney (peter.dewdney@hia.nrc.ca), Brent Carlson (Brent.Carlson@hia.nrc.ca), Al Wootten (awootten) John Webber (jwebber) Darrel Emerson (demerson) Dick Sramek (rsramek) Bob Brown (rbrown) The agenda was: ALMA Correlator PDR Agenda Final - 2000-Jan-19 January 20 ---------- Coffee, muffins, etc. 0800 30m Introduction 0830 15m John Webber Science requirements 0845 15m Al Wootten System overview 0900 30m Ray Escoffier Digitizer requirements 0930 10m Larry D'Addario Digitizer plans 0940 20m Alain Baudry BREAK 1000 20m Signal interfaces 1020 20m Ray Escoffier FIR filter 1040 40m Ray Escoffier Station cards 1120 40m Ray Escoffier LUNCH 1200 60m Fiber optics 1300 30m Dan Edmans Correlator chip 1330 30m Joe Greenberg Correlator card & rack 1400 30m Joe Greenberg Long term accumulator 1430 30m Chuck Broadwell BREAK 1500 20m Computer interface 1520 20m Chuck Broadwell Software issues 1540 20m Jim Pisano Schedule & budget 1600 30m John Webber Open discussion 1630 60m ADJOURN 1730 DINNER 1830 Particular issues to discuss: Filters at the antenna or the correlator? Interfaces with fiber optic cards Chip: 0.18 micron technology? 8K/4K lags? Monitor and control Control card/local microprocessor issues and programming language January 21 ---------- Coffee, bagels, etc. 0800 30m European plans 0830 120m Baudry, Torres, Bos BREAK 1030 30m Discussion 1100 30m Closed committee session 1130 60m LUNCH 1230 60m Lab tours, dispersal 1330 My notes: I was very impressed with the correlator and its ability to confront the science specifications. I think that the proposed correlator design can fulfull nearly all the science requirements which now exist or can be reasonably forseen, with good upgrade paths for future needs. I presented the science requirements. Dewdney asked about spectral dynamic range, a point not specifically addressed in my presentation. I replied that normally, the spec on this was 40 db. This is covered in the Rupen, Shepherd and Wright white paper. Woody inquired about the switching scheme, as the fact that the project will not be using Walsh functions apparently has not propagated very far. Bos mentioned that Westerbork has a tied array mode, whereby the interferometer is split into two phased subarrays. We have not specified this mode for ALMA. Baudry discussed ALMA/EU plans for the 3-bit sampler. The schedule is 30 April 2000 updated specification report 31 May 2000 Technical report on design of simple test autocorrelator. 31 Jan 2000 Final design of test unit 31 Mar 2001 Delivery of test unit ASIC production and testing Dec 2001 Report on test unit. This schedule does not mesh well with US plans, which produced discussion. The prototype correlator is to be at the VLA by 12/2002, with the final correlator released to production April 2003. There was discussion of the FIR filter input/output. At the URSI meeting, it was reported that Backer noted the nonlinearities resulting from digitization at input followed by digitization at output introduced problems. More bits might help. This may affect maser sources, especially. Larry suggested that the correction can be non-trivially calculated exactly, and then implemented. The correction need only be calculated once, so he did not believe this to be an insurmountable problem. Discussion of sampling ensued. The plan is 2-4 GHz 'bandwidth sampling' as opposed to 0-2 GHz 'baseband sampling'. This simplifies the electronics enormously. However, uncertainties remain as to whether this can be accomplished. D'Addario suggested that it was an intriguing but soluble problem. FIR filter: Basebands paired by polarization. Four independently tuned LOs will describe bandwidth and center frequency. There may be some between-band cracks, which the observer will address by going to the next larger bandwidth. The consequences of this should be investigated. Chips: Technology has progressed so that the chip may be realized in .18 micron design rather than .25 micron design. The .18 micron design will probably be available further into the future, should more chips be needed, an advantage. The smaller chips use less power, an advantage. The smaller chips will produce a correlator which is easier to cool, an advantage. They offer the possibility of increasing the number of lags from 4K to 8K, increasing the number of available channels. This is desirable for spectral surveys or redshift machine mode. Woody opined that the broad bandwidth many channel mode occupies 50% of the time at OVRO. Currently, the number of channels in the widest modes is seen as marginal and more would be better. The cost would be about $1M, Webber, estimated. The LTA can accommodate 4K or 8K, Broadwell reported. There was some discussion of binning. For autocorrelator mode, results are available in 1,2,4,8 msec bins. For autocorrelation mode, results are available in bins from 16 msec up to 65 sec. Woody suggested that the 'data chutes' be self-identified, as the complex correlator may need this for debugging; less data may be wanted at cal cycles. Pisano reminded us that the highest resolution/most channel mode may not be entirely supported. The first quadrant can handle all IFs of the first 32 antennas. Flexibility is provided by 'paddleboards' which are replaced as extra quadrants are brought on line. Baudry presented the ALMA/EU correlator ideas. This is a philosophy that the US proceeds while ALMA/EU begins studies. At a branch point 2003-2004, the US plan may be expanded beyond the first quadrant, of a 'new design' may be begun. The electronics release in 2Q 2003 may complicate this. In 2000, this will be in a feasibility phase, identifying an architectural concept, identifying critical technologies and estimating costs. The deliverable is a 2001 feasibility report. Japanese contributions may be to fulfill the 64 antenna array, if the US and EU cannot, or to expand it to perhaps an additional 32. Pisano noted that 100-1000 FLOP/pixel is the norm for military operations. He estimates the ALMA to be a 100 FLOP/pixel machine. If the array had less than 64 antennas, the extra capacity would probably be easiest used for extra spectral windows. There was discussion about whether perhaps pulsars required additional capabilities. The recommendation was to continue to pursue .18 micron technologies, with 4K channels. After some testing, moving to 8K may be considered.