slide 1 25-jan-10huib intema recent low-frequency developments at nrao charlottesville and some...
TRANSCRIPT
25-Jan-10 Huib Intema Slide 1
Recent low-frequency developmentsat NRAO Charlottesville
and some other USA institutes
25-Jan-10 Huib Intema Slide 2
Overview
• Improvements / open issues of SPAM
• RFI mitigation
25-Jan-10 Huib Intema Slide 3
Short SPAM intro
• Source Peeling and Atmospheric Modeling
• Obtain initial sky model and calibration• Measure ionospheric phases through peeling bright FoV sources • Fit single / multi-layer turbulent ionosphere model to peeling phases• Calculate model phase corrections for grid of facets covering FoV• Apply corrections during imaging & deconvolution• Possibly repeat from step 2
• Mainly tested on archival VLA 74 MHz and GMRT 153 MHz data• Implemented in Python, using ParselTongue interface to AIPS
25-Jan-10 Huib Intema Slide 4
SPAM single layer geometry
25-Jan-10 Huib Intema Slide 5
SPAM issues
• Needed to adapt to latest AIPS and ParselTongue versions
• Several model fit convergence problems (TIDs)– Bad antennas / ‘bad’ sources
– More severe ionospheric conditions
– Re-sampling / unwrapping of lower S/N peeling phase solutions
• Performance of model fitting & generation of solutions
25-Jan-10 Huib Intema Slide 6
SPAM improvements
• Updated to latest AIPS & ParselTongue– Revealed bugs in AIPS: some solved, some work-arounds
• Improved model fit convergence– Phase interpolation: fixed problem in phase unwrapping routine
– Phase interpolation: re-reference to nearest antenna
– Initial fit: FT of phase solutions + weighted average near peak
– Initial fit: Additional 2nd order polynomial fit + projection onto KL base vectors
– Restricted criteria to reject excessive antennas and/or sources + re-fit
• Performance improvements– Better model re-use of previous time step
– Simplifications in calculation of ionospheric pierce points
– Better model convergence allows for single fit of all model parameters
25-Jan-10 Huib Intema Slide 7
Test results on VLA 74 MHz data in B-configuration
• Convergence of SPAM model fits has improved significantly– Post-fit phase RMS is roughly constant during quiet ionosphere
– Data loss has gone down significantly during TID periods
– S/N of peeling solutions biggest problem: VLA-B requires <1 minute solution intervals for good interpolation and model fits
• Performance of SPAM has improved significantly:– Despite using Python / AIPS, scripted data reduction of VLSS data sets
now matter of hours instead of days (on 3 GHz Linux desktop)
• Possible VLSS DR2 would require production version, e.g. SPAM implementation in faster C/C++-code
25-Jan-10 Huib Intema Slide 8
RFI mitigation
• Main idea from Athreya 2009 (ApJ, 696, 885):– Use fringe rate to identify and remove RFI signals from visibilities
– Subtract rather than flag
– Requires quasi-constant RFI signal
– Doesn’t work for near-zero fringe sources(e.g., near celestial poles, but also near u=0)
© Ramana Athreya
25-Jan-10 Huib Intema Slide 9
Similar developments based on Athreya scheme
• Implementation in AIPS by Leonid Kogan (NRAO Soc)– Fit of both circles and spirals
• Implementation in Python by Joe Helmboldt (NRL)– De-rotation and fitting
• Implementation in Obit by Bill Cotton (NRAO CV)– De-rotation and fitting
– Testing and bug-fixing by yours truly
• Issues:– Possible source flux subtraction bias
25-Jan-10 Huib Intema Slide 10
RFI subtraction source bias
• Artificial VLSS data of one 1 Jy source at varying position in field• After subtraction, source detection in UV plane
• No noise added
• Noise added
© Joe Helmboldt (NRL)
25-Jan-10 Huib Intema Slide 11
Example of Obit implementation on VLSS data (1)
• Combination of amplitude clipping and subsequent RFI subtraction
• 5x15 minutes VLA-B 74 MHz, BW 1.5 MHz, 127 ch, 10 sec integr.
Time ->
<-
fre
quen
cy
© Bill Cotton (NRAO)
25-Jan-10 Huib Intema Slide 12
Example of Obit implementation on VLSS data (2)
• Imaged using field-based calibration
Input image 78.4 mJy/beam Output image 69.3 mJy/beam
© Bill Cotton (NRAO)
25-Jan-10 Huib Intema Slide 13
Test results on VLA 74 MHz data in B-configuration
• Reduction of ripples & noise in image background– But improvements are moderate when considering magnitude of
removed RFI
• Scheme needs initial RFI flagging
• Scheme needs criteria on fringe rate range and vis. amplitude range
• Subtraction bias is little when using residual UV data– Requires good sky model and calibration
• Probably most useful for saving short baselines