1. Calibration WP 2.6.3 Calibration and Imaging techniques:
Ronald Nijboer (ASTRON) with input from Athol Kembal (TDP) Tim Cornwell / Maxim Voronkov (CSIRO)

2. Content Introduction Status Work to be done Challenges and risks
Real time calibration of stations and PAFs
Removing contaminating signals
RFI
A-team and / or other strong sources
Calibrating the FoV
Sky model
Bandpass
Clocks
Primary beam
Ionosphere
Troposphere
Other
Work to be done
Challenges and risks
Conclusion

3. SKA
Phase 1
Definition is work in progress: numbers from Calim2010 (Hall)
Dish SPF array
m dishes
450 MHz – 10 GHz
Baseline up to 180 km
~50000 channels
~0.1 s dump time
AA array (sparse)
m diameter stations
70 MHz – 450 MHz
~1 s dump time
~480 Independent beams on the sky
Demonstrators for PAFs and dense AAs
Phase 2
More …

4. Feasibility: practical community experience in high-DR imaging
Community survey of current DR limitations (EVLA, WSRT, GMRT, ATCA) (Chakraborty & Kemball 2009):
Hardware is destiny
Sky-mount vs rotating antennas
Closure errors
Direction-dependent errors:
Pointing
Beam pattern errors
RFI
Computational and software limits
Longitudinal study on 3C273, (Chakraborty & Kemball, 2010).
DR is tracking thermal sensitivity evolution, but at cost of calibration complexity.
Mean DR evolution over time
(Chakraborty & Kemball 2010)

5. Traditional Calibration
Calibration needed for
Astrometry: accurate (absolute or relative) positions
Photometry: accurate (absolute or relative) flux scale, spectral shape
Image / PSF quality and image fidelity / DR
Traditional method:
determine phase / gain (freq) on stable external (point) source
Good for 1) and 2)
Self-cal needed for 3)

6. Some observations External calibrator Self-cal
AA can use separate calibration beam and transfer cal solution
Repoint beam and extrapolate cal solution
If enough sensitivity “in-beam calibration” can be used
Provided the strongest sources in the FoV are known
Self-cal
Needed to obtain accurate (calibration) source model
Iterates through calibration and imaging / deconvolution steps
Hence needs to buffer all the uv-data
Complicates streaming processing

7. ASKAP: Global Sky Model
For calibration, we require a GSM
Threshold requires investigation, probably ~ 1mJy
Required accuracy not clearly understood yet
During commissioning, we first image the entire sky in continuum
We expect the processing will take all the computation resources
Necessary to deconvolve and self-calibrate with many cycles
Perform source detection
For each source detected in Stokes I, take spectral index information from MFS image
Repeat observations
During commissioning to converge on accurate GSM
During operations to update for variable sources
6km baselines
May be necessary to bootstrap from 2km baselines
CSIRO

8. ASKAP: Calibration during observations
Processing is I/O intensive, iteration over visibilities is costly
The number of iterations must be kept to minimum (ideally just 1)
ASKAP approach to solve this problem:
The instrument is stable and kept well-calibrated
Most significant PAF calibration is done up-stream
Only increments need be solved
Always use prediction-forward
Gains are fed back to correlator/beamformer/frontend in real time
We always have a fairly accurate continuum model at hand
No self-calibration is needed for spectral line and transients
Real-time calibration per synthetic beam is adequate
3-axis antenna mount takes care of the beam rotation issues
CP Applications / Calibration and Imaging

9. ASKAP: Calibration pipelines
A number of concurrent pipelines
One for beam-dependent gains, one for bandpasses, one for polarisation
Each accumulates data for a certain time interval
Then reports the solution back to the Telescope Control System
Solution is then applied to all future data closing the loop
Tropospheric phase absorbed into beam-dependent gains
Can add an antenna-dependent phase calibrated at a finer timescale, if necessary
Ionosphere is isoplanatic
Frequency dependence is absorbed into bandpass
Complex gains of PAF beams
Calibrate e.g. every 60s (or longer)
Always predict forward (no interpolation of calibration solution)
CSIRO

10. Initial LOFAR Calibration
The following initial approach is proposed on weak source fields ( < Jy peak in HBA)
Calibrate on a bright dominant source and solve for clocks
Transfer calibration per subband, or interpolate, to target field
Then inspect data quality, noise, X and Y
Flag bad stations/baselines
Make snapshot image (within ~15m from the calibrator), create a model
Solve for clock drifts which become important after >15m
Combine multiple bands (BBS-global) to improve solutions etc
Iterate
Start with core stations (<3 km baselines) and work from inside out
i.e. get a proper low resolution LSM with all the flux contained in the FOV
Process 5-15m snapshots individually and combine beam-corrected images
Then bring in remote stations which involve longer baselines

11. Content Introduction Status Work to be done Challenges and risks
Real time calibration of stations and PAFs
Removing contaminating signals
RFI
A-team and / or other strong sources
Calibrating the FoV
Sky model
Bandpass
Clocks
Primary beam
Ionosphere
Troposphere
Other
Work to be done
Challenges and risks
Conclusion

12. Station / PAF calibration
Calibration approach determined by availability of calibrator sources
Station of AAs
Random array of dipoles
All sky FoV;
Regular array of tiles or vivaldis
Tile FoV; repointing and / or redundancy
Station of Dishes
Dish FoV; repointing?
PAFs
Radiators on the antenna surface
Repointing

13. LOFAR LBA Calibration
Wijnholds

14. Embrace calibration Use Afristar satelite (or Sun) to calibrate
Wijnholds
Use Afristar satelite (or Sun) to calibrate

15. PAF Calibration
M. V. Ivashina, S.J. Wijnholds, R. Maaskant, K. F. Warnick, and B. Jeffs

16. Content Introduction Status Work to be done Challenges and risks
Real time calibration of stations and PAFs
Removing contaminating signals
RFI
A-team and / or other strong sources
Calibrating the FoV
Sky model
Bandpass
Clocks
Primary beam
Ionosphere
Troposphere
Other
Work to be done
Challenges and risks
Conclusion

17. Removing RFI Example: LOFAR low level RFI
Offringa, MNRS 405 (June 2010)
Example: LOFAR low level RFI
Developing a “streaming flagger” remains a challenge

18. WSRT 150 MHz; 3C196
CygA
Sun
CasA
NCP
VirA
TauA
De Bruyn

19. LOFAR Core Station 1: NCP
Yatawatta

20. LOFAR: 3C465
Jackson

21. Correlator Field-of-View Shaping
Background:
Use of small (~12m) antennas for SKA
⇒ large FOV
⇒ potentially excessive data volumes/rates (up to PB/s) and maps sizes
⇒ removal of dynamic range-limiting sidelobes from sources across the
full FOV will become an intractable problem using current techniques
One solution is to employ intelligent weighting in frequency/time to limit
the FOV for high-resolution work.
This can be accomplished using Correlator Field-of-View Shaping (Lonsdale et al & recent paper in prep.)

22. Sidelobe confusion levels at the phase center due to sources
beyond angular distance r
This plot summarizes our key results.
Green line represents integrand.
Thick red line its integral (values of integral are listed for different r values).
The yellow line shows the result without any correlator FOV shaping.
Vertical dash-dot lines indicate angles of 20% beam area (right) and 1000 times the resolution of the longest baseline (left).
from Lonsdale et al., in prep.
Bottom line: suppression levels below ~10-8 Jy can be achieved by this technique.

23. Content Introduction Status Work to be done Challenges and risks
Real time calibration of stations and PAFs
Removing contaminating signals
RFI
A-team and / or other strong sources
Calibrating the FoV
Sky model
Bandpass
Clocks
Primary beam
Ionosphere
Troposphere
Other
Work to be done
Challenges and risks
Conclusion

24. Sky model On long baselines sources become resolved
Accurate source models are needed for calibration
Van Weeren
3C 61.1, 60 hr, 20 HBA stations (16 split core + 4 remote),1 sub-band,
9.7 by 9.4 arcsec resolution

25. Sky model: shapelets EVLA 8.5 GHz 256 2 MHz channels
34 m – 1 km baselines
Peak flux 34 Jy
Residual 2.5 mJy
DR ~13000
Sarod Yatawatta (RuG/ASTRON), Tony Willis (DRAO) and Rick Perley (NRAO)

26. Bandpass Digital bandpasses can be corrected for without calibration
Romein

27. Bandpass Analogue bandpasses need calibration
For high DR imaging with WSRT per channel calibration of bandpass is needed (De Bruyn / Smirnov)
Smirnov, submitted

28. Clocks In LOFAR each station has a separate clock
Clock drifts complicate calibration
Make coherent beamforming more difficult (Pulsar observations)
Mix with ionospheric phase variations
To be considered in SKA system design

29. Primary beams Pointing errors can be treated with A Projection
But need full pol treatment
Only for known PB patterns
Reflections by e.g. struts for dishes
Projection and time dependence for AA beams
How to calibrate general PB variations?
Work by Smirnov
In field calibration beacons for calibration
Solving for deliberate mispointing

30. Simulations: DD effects
Accept WBSPF radiation patterns from antenna design programs.
Simulate direction-dependent calibration using these patterns in SKA array configurations to assess:
Impact on image quality.
Impact of errors in direction-dependent calibration models.

31. Differential gains
Differential phases for 14 telescopes in the direction of 7 sources
Interpretation is still hard
This may be the basis for a general PB calibration approach
Smirnov, submitted

32. Ionosphere Typically below ~2 GHz observing frequency Phase
SPAM by Huib Intema
Tested on VLSS data
Being implemented in LOFAR pipeline
But: SPAM tested on few 10s km baselines only
Faraday rotation
To be developed

33. SPAM

34. Troposphere Typically above ~2 GHz observing frequency
In general a DD antenna based effect
May need a SPAM like approach for the SKA

35. Feasibility: wide-band polarization calibration
Pathfinder ATA studies (Bower et al)
Typical values up to 10%.
Smooth changes with frequency.  Rate of ~3%/10 MHz at 1.4 GHz.
Spiral and loop patterns seen in real-imag space.
Period of loops similar to that of log-periodic feed (e.g., 3/4 of a turn at 1.4 GHz).
Leakages are contiguous between adjacent bands.  This suggests leakages originate in the feed.
1.4 GHz leakages for antennas (numbered) in real,imaginary space as a function of frequency.
Leakages for two adjacent correlator tunings.  The leakages change smoothly between bands.

36. Feasibility: DD polarization calibration
Pathfinder ATA studies (Bower et al)
Leakages are measured in offset pointings from calibrator.
Similar leakages throughout the primary beam.
Random contribution increases off-axis and toward higher frequencies.
Random component of order 3% within half the HP point at 1.4 GHz.
Leakages for ant 11x at center and half power points.  Color shows frequency dependence.  Axes show leakage scale.
Same leakages, but showing diff with center.

37. Polarisation: RM Synthesis
Scaife, Heald, de Bruyn, Trassati, …
3C66AB-NGC891-PSRJ0218 6h HBA

38. Polarisation: RM Synthesis
Scaife, Heald, de Bruyn, Trassati, …

39. Other effects “VLBI” BSR / beam squint Retarded baselines (Voronkov)
Continental drift
Effect of wheather fronts, oceanic loading, etc. on station positions
BSR / beam squint

40. Content Introduction Status Work to be done Challenges and risks
Real time calibration of stations and PAFs
Removing contaminating signals
RFI
A-team and / or other strong sources
Calibrating the FoV
Sky model
Bandpass
Clocks
Primary beam
Ionosphere
Troposphere
Other
Work to be done
Challenges and risks
Conclusion

41. Work to be done Requirements & system design
Feed back of initial findings to system design group
Further develop & implement algorithms / strategies
Integrate in a pipeline
Assessment of performance
Optimization of performance / finding alternative approaches

42. Challenges & Risks
Different approaches / pipelines will be needed for different receptor types / frequency regimes
Calibration approaches can only be fully validated when hardware is in the field
Total computational power needed is unclear
For current LOFAR calibration cost equals imaging cost, but is is unclear how this scales to full LOFAR, let alone SKA

43. Conclusion Ideas on how to calibrate the SKA exist
Software for some modules exist
Integration into end-to-end pipeline is underway for the pathfinders and precursors
Performance and scalability to SKA regime remains to be evaluated

44. Some final remarks
DR is tracking thermal sensitivity evolution, but at cost of calibration complexity.
SKA processing is likely to differ at points from precursor and pathfinder processing
Need for streaming processing has implications for system design and algorithms
Some of the current algorithms might (will) not scale up to SKA size
The sheer size of the SKA makes it a whole new regime
Incremental construction will help the learning process
Pathfinders, precursors, SKA1, SKA2, …