1. Ramesh Bhat Centre for Astrophysics & Supercomputing Swinburne University of Technology Time Domain Astronomy Meeting, Marsfield, 24 October 2011 Searching for Fast Transients with Interferometric Arrays

2. An Australia-India collaborative project  Developing new scientific capabilities for the GMRT Transient detection pipeline High time resolution pulsar science VLBI between GMRT and Australian LBA  Collaborating institutions: Swinburne, Curtin/ICRAR, CASS (Australia) National Centre for Radio Astrophysics (India)  Project team: Matthew Bailes (Swinburne)Ben Barsdell (Swinburne) Ramesh Bhat (Swinburne) Sarah Burke-Spolaor (JPL) Jayaram Chengalur (NCRA)Peter Cox (Swinburne) Yashwant Gupta (NCRA)Chris Phillips (CASS) Jayanti Prasad (IUCAA) Jayanta Roy (NCRA) Steven Tingay (Curtin)Tasso Tzioumis (CASS) W van Straten (Swinburne)Randall Wayth (Curtin)

3. In This Talk:  Searching for fast transients - important considerations  GMRT as a test bed instrument  Transient detection pipeline  Event analysis methodology

5. Searching for fast radio transients: Important considerations  Detection sensitivity, survey speed, and search volume -- Figure of Merit (FoM)  Propagation effects: e.g. dispersion, scattering, and scintillation due to the intervening media  Parameter space to search for: DM, time scale; computational requirements  Radio frequency interference (RFI) -- a major impediment in the detection of fast transients!  Detection algorithms; candidate identification and verification strategies

6. De-dispersion DM = Dispersion Measure (in units of pc cm -3 ) Dispersion smearing can be quite severe at low obs frequencies  Processing will involve searching over a large range of dispersion measure (DM)  Low frequencies will require very fine steps in DM (e.g. ~1000 trial DMs @325 MHz)  Incoherent dedispersion: channelise data, shift and align the channels, then sum

7. Searching for “events” in the time - DM parameter space Detections of single pulses from J0628+0909 Standard search strategy: Dedispersion + matched filtering Each “event” is characterised by its amplitude, width, time of arrival and dispersion measure (DM) Matched filtering Time domain clustering Matched filtering

8. Observational Parameter Space S (x, t,,  ) x : Location of the station  : Direction on sky t : Time domain : Radio frequency RFI is site-specific & direction dependent: function of x and  Effective use of “coincidence” or “anti-coincidence” filters Celestial transients vs. RFI: May have similar -t signature (e.g. swept-frequency radar and pulsars) Will have very different occupancy of x-  space:

9. Detecting fast transients: search algorithms and strategies PSR J1129-53 - an RRAT discovered by Burke-Spolaor & Bailes (2010)

10. Transient Exploration with GMRT 30 x 45m dishes, collecting area ~ 3% SKA Modest number of elements, long baselines Advent of GMRT software backend (GSB) Demonstration of multibeaming across FoV Superb event localisation capabilities (~5”) Computational requirements are significant, however affordable GMRT makes an excellent test-bed for developing the techniques and strategies applicable for next- generation (array type) instruments

11. Considerations for sub-arraying: False alarm probabilities N independent elementsMultiple sub-arrays, p = N/nIncoherent combination

12. 14 km 1 km x 1 km RFI environment is known to vary significantly across the array; e.g. between the arms; between the central square and the arms (east, west, south) Considerations for sub-arraying: RFI environment Local RFI sources: TV boosters Cell phone towers Power lines

13. Antenna locations are marked in red Locations of RFI sources are marked in blue courtesy: Ue-Li Pen

14. + GMRT software backend (GSB) GMRT + configurability

15. Transient Detection Pipeline for GMRT Real-time processing and Trigger generation + Local recording of Raw Data GMRT array GSB clusterTransient Detector Trigger Generator @ 2 GB/sec 512 MB/sec (Ndm/Nchan) x 64 MB/sec

16. Salient features of GMRT transient project  The GMRT + GSB combination offers some unique features for efficient transient surveys at low radio frequencies Long baselines: powerful discrimination between signals of RFI origin vs celestial origin (via effective coincidence filtering) High resolution imaging: event localization (~ 5”-10”) possible through imaging the field of view and/or full beam synthesis Software phasing (offline): sensitive phased array beams toward candidate directions (~5 x sensitivity); base-band data benefits (e.g. coherent de-dispersion)  Search strategy: commensal mode with other observing programs; real-time processing and local recording

17. Pilot transient surveys with the GMRT  Primary goals: Technical development Efficacies at low frequencies  Survey region: -10 o < l < 50 o, | b | < 1 o @ 610 -10 o < l < 50 o, 1 o < | b | 3 o @ 325  Data recording Software backend’s “raw dump” DR = 2 x 30 x (32 MHz) -1 x 4 bps Data from the surveys are used to develop the transient processing and the event analysis pipelines

18. Transient Detection Pipeline  RFI + quality checks  Form N Sub-arrays  De-dispersion  Transient detection  Event identification  Coincidence filter  Trigger generation  Data extraction  Event analysis

19. Examples from the pipeline: a real astrophysical signal

20. Examples from the pipeline: spurious signals (local RFI)

21. Spectral Kurtosis Filter for RFI excision: Implementation on CASPSR Andrew Jameson (Swinburne)

22. Need for high resolutions in time, frequency and DM space  Signals can be as short as tens of micro seconds at GMRT frequencies  Maximum achievable time resolution ~ 30 us with the current pipeline An example from the GMRT transient detection pipeline (mode: 7 sub-arrays) A Giant Pulse from Crab Pulsar at GMRT 610 MHz, Time duration ~ 50 us

23. Processing Requirements  Benchmark with current software: data at full resolution (30 us, 512 channel FB) 15 x real-time on a dual quad-core Dell PE1950 equivalent to 133 Gflops (theoretical)  Net processing requirement: 15 x 133 Gflops = 2 Tflops (per beam!)  Possible (practical) solutions: Data down sampling (degrading resolution in f-t) by a factor 4 4 machines per beam OR 16 machines for 4 subarray beams Alternatively, 4 x GPUs, each of 0.5 Tflops De-dispersion (searching in DM parameter space) is the most computationally intensive part of the pipeline  30 us, 512-channels  16 bit data samples  DM range: 0 - 500  tolerance level: T1.25 GPU dedispersion code by Ben Barsdell (Swinburne)

24. Considerations for the real-time system: false positives and RFI signals

25. Considerations for the real-time system: (false positives + RFI) + real signal

26. Event Analysis (offline) Pipeline Localisation of the event on sky + phasing up + further checks

27. FLAGCAL: A flagging and calibration package Description of the FLAGCAL pipeline in Prasad & Chengalur (2011)

28. Snapshot imaging for event localisation Currently FLAGCAL + AIPS; will soon be integrated into the main event analysis pipeline “Dirty” imageSingle pulse from J1752-2806 “dirty” image After cleaning and self-cal Signal peak ~ 0.27 Jy rms ~ 6 mJy; beam ~ 59” x 10”

29. Example from Event Analysis Pipeline On phasing up Detection Phase up the array

30. Summary and Concluding Remarks  Searching for fast transients with multi-element instruments involve several considerations and challenges; propagation effects, RFI, signal processing, etc.  The GMRT makes a powerful test bed for developing and demonstrating novel transient detection techniques and methodologies applicable for next-generation (LNSD type) instruments such as ASKAP  Transient detection pipeline for GMRT - development nearly complete; the commensal surveys to start by early 2012; the system will be extended to larger bandwidths  The VLBA and GMRT based efforts will help demonstrate the advantages of multiple stations and long baselines for transient exploration; valuable lessons for the SKA-era