Upgrade to Pro — share decks privately, control downloads, hide ads and more …

ARTEMIS

transientskp
December 04, 2012

 ARTEMIS

Aris Karastergiou

transientskp

December 04, 2012
Tweet

More Decks by transientskp

Other Decks in Science

Transcript

  1. The  Sun Pulsars Radio   Transients Interstellar   Medium Nanosecond

      Astronomy Ionosphere Cosmic  Rays The  Unknown LOFAR  staAon Beamformed Data
  2. What is ARTEMIS? A software/hardware backend to process streaming data,

    used for single-station science at LOFAR-UK, Nancay, Bonn + Onsala. Software: AMPP, the artemis modular pelican pipelines AMPP: prototype for streaming processing for the SKA and its pathfinders. 35 TB/day today. Tomorrow? Software: Pelican, a light and scalable framework for distributed streaming processing. Hardware: AMPP and Pelican have native CPU and GPU support AMPP is not a UDP packet recorder AMPP is not limited to LOFAR AMPP is documented and open to developers H5 data product that conforms to LOFAR standard
  3. LOFAR international stations • Large collecting area (96 coherently added

    dipoles), high sensitivity. 9 international stations = superterp • Very fast sampling rate (5 ns), reduced to 5.12 µs for the station beams. • Large fields of view, 10 to 100s of square degrees. • Not used for all LOFAR KSPs.
  4. Dispersion searches DM 0 N2 or N logN process per

    time sample (N > 1000 -- 10000+ samples/sec HPC computing
  5. Dell Poweredge C6100 Dell Poweredge C410x Stef Salvini Wes Armour

    Dalal Ait Allal Mike Giles Chris Williams Steve Roberts Fred Dulwich Ben Mort Alessio Magro Aris Karastergiou hardware
  6. Frequency (MHz) Time (ms) 8000 8200 8400 8600 8800 9000

    143 144 145 146 147 148 149 20 30 40 50 60 200 400 600 800 0 2000 4000 6000 8000 Time bin 40 60 100 200 300 400 500 600 700 800 900 dB Frequency channels Pulse
  7. Frequency (MHz) Time (ms) 8000 8200 8400 8600 8800 9000

    143 144 145 146 147 148 149 16 18 20 22 24 26 28 30 200 400 600 800 0 200 400 Time bin 22 24 26 100 200 300 400 500 600 700 800 900 dB Frequency channels
  8. frequency intensity frequency intensity intensity frequency 1. Model the bandpass

    2. Match model to data 3. Subtract model and filter 4. Store spectrum stats in ring buffer 5. Update model with running averages
  9. dynamic noise modelling for effective RFI removal and transient detection

    DM time (s) GPU dedispersion: 4000 dispersion measures searched, only pulses from B0329+54 detected, no spurious signals filterbank noise pre-filtering filterbank noise post-filtering
  10. Dispersion  measure  pc/cm3 Time  (s) Crab HAM SAT HAM SAT

    • Working  prototype  for  non-­‐image  processing • Real-­‐Bme  searches  for  Individual  Radio  Pulses  using  CPU/GPU • Dedispersion  over    4000  DMs  per  beam  in  real  Bme  with  GPUs • Pilot  survey  1:  6-­‐8  beams  tracking  circumpolar  targets • Pilot  survey  2:  6  beams  fixed  on  the  meridian  from  8o  to  28o  dec Pipeline  diagnosBc  plot  from  Pilot  survey  –  daily  summary  from  one  of  six  beams Receive  udp   stream Polyphase  filter Raw  to   Stokes RFI  removal Dedispersion output S/N DM Time  (s) Aris  Karastergiou  ([email protected]) Crab
  11. • Real time detection system which would easily detect RRATs

    and Lorimer burst as they happened. • Affordable and scalable hardware • Fastest ever dedispersion code on NVIDIA GPU • RFI rejection performing well in very contaminated environment Crab B1919+21 26 December 2011 drift scan with beam at Dec=+21º B1919+21 Crab time (s) ?? ? ?
  12. The hundred tied-array-beams realtime processor LOFAR APERTIF MeerKAT SKA-I SKA-II

    RFI-excision dedispersion folding detection CPU - GPU cluster