Investigating RFI Flagging Techniques with LOFAR

Yvette Cendes

Transcript

1. Investigating RFI Flagging
   Techniques with LOFAR
   Yvette Cendes
   

   View Slide

2. The LOFAR RFI Environment
   Ü  RFI occupancy is 1.8% in the low band, 3.2% in
   the high band
   Ü  No difference between day or nighttime
   observations
   Plot Credit: A. Offringa
   

   View Slide

3. AOFlagger
   Ü  Default automated
   flagging method used by
   LOFAR
   Ü  Works with amplitude
   information of one
   polarization of a single
   sub-band
   Ü  It relies on thresholding,
   where cutoffs depend on
   the surrounding signal
   levels
   A LOFAR RFI pipeline A.R. Offringa
   2.1 Input
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   !"
   
   
   # !$
   
   %
   
   &
   '
   $(
   !
   ))*)+*+)*++&
   "
   %
   %
   
   
   +
   Figure 1: Overview of the RFI flag-
   The flagger is executed on the amplitude information of
   one polarisation of a single sub-band of a baseline. In LO-
   FAR’s common operation, a sub-band consists of 256 chan-
   nels of 0.8 kHz resolution. The full band has 248 sub-bands.
   LOFAR can observe in two bands: the 10-80 MHz low band
   and the 110-240 MHz high band, which are observed by phys-
   cally different antennae.
   If speed is essential, the algorithm can be executed once
   on the Stokes-I values. Otherwise, if accuracy is more impor-
   ant than speed, the algorithm can be executed on the individ-
   ual XX and YY or LL and RR polarisations, or on all polar-
   sations individually. We do see some RFI that manifests in
   only one of the polarisations, or rotates through the polarisa-
   ions, and some advantage is therefore seen when flagging all
   polarisations individually.
   2.2 Iterations
   A part of the algorithm is iterated a few times, depicted
   n Figure 1 by the “Continue iterating” block. This is nec- Image Credit: A. Offringa
   

   View Slide

4. Transients vs. AOFlagger
   Ü  The default flagger is designed to catch all the
   RFI, even if some non-contaminated data gets
   flagged
   Ü  AOFlagger uses time selection steps which
   compares RMS values, and automatically flags
   anything with a sigma > 3.5 in order to quickly
   reach convergence
   This may not be ideal for observations
   containing transients
   

   View Slide

5. AOFlagger
   Test signals in rfigui, flagged out by AOFlagger
   

   View Slide

6. Modified Flagger
   Ü  Modified AOFlagger to run more quickly, detect
   transient signals
   Ü  Changes include deleting time selection,
   decrease ‘sliding window’ resolution in time,
   ignore thresholding in frequency
   ✻  Tested on MSSS data which
   showed high RFI percentages
   flagged in processing- starting
   from raw data to re-flag, demix,
   calibrate, and image…
   

   View Slide

7. Test Data
   Default AOFlagger Modified AOFlagger
   

   View Slide

8. L44766, SB151: CS002LBA x
   CS004LBA
   Default Flagger Modified Flagger
   Polarization statistics: Polarization statistics:
   XX: 3.4%, XY: 3.5%, XX: 0.32%, XY: 0.33%,
   YX: 3.4%, YY: 3.4% YX: 0.49%, YY: 0.29%
   

   View Slide

9. Results
   (Frequency= 73.2 MHz)
   

   View Slide

10. Results
    The offset is fairly constant for both strategies
    

    View Slide

11. Results
    Computational time decreases, background RMS doesn’t
    

    View Slide

12. Results
    L21641 (Bell #1: Nov 28, 2010)
    

    View Slide

13. When No Flagging Occurs…
    Ü  When you do no flagging (including any post-
    BBS flagging) the RMS values increase
    dramatically including big spikes in amplitude
    Ü  When you do post-BBS flagging only (eg a simple
    amplitude cut), you get similar RMS values to a
    normal image
    Ü  How is post-BBS flagging in general affecting
    transient sources?
    

    View Slide

14. Inserting Fake Signals
    Ü  Work has begun on inserting artificial signals into
    MSSS data (starting with simple point sources)
    Ü  We need to figure out how a bright transient
    affects calibration and image quality when it
    isn’t in the sky model
    Image credit: Dan Calvelo
    

    View Slide

15. Summary & Future Work
    Ü  Modifying the default flagger for transient
    searches appears to work with automatic
    flagging
    Ü  Image quality does not appear to be
    compromised for faster computational speeds
    Ü  Future work needs to focus on testing the
    modified flagger on a wider range of data, eg.
    MSSS HBA
    Ü  See how this can apply to AARTFAAC
    Credit: ASTRON Daily Image
    

    View Slide

16. See You in Mauritius!
    

    View Slide