Special Considerations for Long Baseline Imaging with LOFAR


-- The long baseline pipeline is no longer offered --

Observing with the international baselines of LOFAR opens up the domain of sub-arcsecond imaging of the sky at metre wavelengths. Resolutions of order 0.3 arcseconds with sensitivity <<1mJy/beam are possible at the low end of the HBA frequency range. However, the present uv coverage of international LOFAR, combined with potentially large a-priori amplitude calibration uncertainties, mean that sources with significant extended structure are challenging to calibrate and map.

Calibration is the greatest challenge for long baseline LOFAR observations, because of the scarcity of bright, compact sources with simple structure. This situation is exacerbated by the large impact of differential ionospheric effects, which limits the amount of averaging that can be performed in frequency and in time before initial calibration.

Imaging can also present difficulties due to the wide field of view, which results in very large numbers of pixels if the entire FOV is to be sampled at the full resolution.  On the other hand, the high resolution means that individual sources are separated by very many beams, which can simplify the imaging process. Individual images of small fields centred on known sources can be made from averaged data. For bright sources imaged in this way it should be possible to achieve high dynamic range without imaging the whole field, as is needed for observations using only the Dutch component of LOFAR.

The information provided here is intended to aid correct scheduling of International LOFAR observations. If the observations are correctly set up, the available " Long Baseline Pipeline" can be run by the SDC before data is archived. This will perform RFI flagging and averaging, determine and apply initial calibration for the Dutch array, and convert the data into a form which is more suitable for subsequent long baseline calibration. This saves a large amount of compute time for the PI, and considerably reduces data volumes.  More details, including how to set up observations such that the long baseline pipeline can be used, are below. Otherwise users can ask the SDCO to only pre-process their data and later perform a more advanced calibration by using a novel Long Baseline pipeline under development.


The greatest possible sensitivity for long baseline calibration is obtained in two ways:

1)  Phasing the core stations using a post-correlation summation of the visibilities. This provides a single sensitive station which can be used as a reference for calibration.  In order to calibrate the core for phasing, "normal" LOFAR calibration approaches can be utilised (either beam switching with the full bandwidth over timescales of many minutes, or contemporaneous calibration using a fraction of the bandwidth in a second station beam positioned on a bright calibrator source).  A good source for calibration of the core stations will have a flux density 10 Jy (more in the LBA band) and be compact on arcminute scales. It need not be extremely close on the sky to the target.

2) Fringe fitting (solving for delay [derivative of phase with frequency] and rate[derivative of delay with time]) on the target itself or a very nearby compact calibrator source. This coherently adds visibilities across frequency and time, assuming the phase change with frequency and time is linear. In actuality, the ionospheric delay changes with frequency (quadratic change of phase), but it can can be assumed to be constant over a reasonably narrow bandwidth.

In general night time observing should be used if possible to ensure that the delays due to the differential ionosphere are minimised, although the specific ionospheric conditions at the time of the observation can vary much more than the average difference between night and day.

Assuming that the core stations are phased and the data is arranged into spectral windows of width ~3 MHz, the required flux density for calibration is of order 100 mJy (on the longest baselines) for HBA observations. This uses a solution interval of order 1 minute.  The LBA is of order 20x less sensitive, so a flux density of ~2 Jy is required.  These numbers are derived from current (July 2014) array performance and may improve somewhat if station calibration is improved further.  Rampadarath, Garrett & Polatidis (2009) contains a list of potentially suitable calibrators.  A calibrator survey is currently underway (LOBOS, PI N. Jackson) which will eventually identify of order 20,000 suitable calibrators across the Northern sky. Contact the SDC Operations helpdesk if you need help locating a suitable calibrator.

The above critical flux density must be present on all baselines within the target itself, or within a nearby calibrator with simple structure to whose phase solutions can be transferred to the potentially weaker / more complex target source. Given the present uncertainties in amplitude calibration, in order for fringe-fitting and imaging to work for long baseline LOFAR, the source used for fringe detection and calibration should either be simple (close to point-like) or have an existing good starting model at similar angular resolution.  If neither the target nor any nearby potential calibrator source fulfils the requirement for simple or known structure, a short scan should be included at some point in the observation on a more distant source which does satisfy this criterium, preferably at high elevation.  In this case, producing a lower resolution image of the source to be used as a calibrator using only the Dutch array as a starting point is also recommended.

Transfer of calibration phase has been demonstrated within a single station beam over angular separations of ~0.5 degrees in the HBA. Transfer of calibration over an angular separation of 0.75 degrees has been shown to work under good observing conditions at HBA frequencies, but this is not expected to be always successful.  Transfer of calibration over a separation of 1 degree or more will often not result in significant improvements in coherence time. The success of spatial calibration transfer will be heavily dependent on ionospheric conditions, and the expected conditions are not yet well characterised for international baselines. A separate station beam at identical frequencies can be placed on the calibrator object - transfer of calibration between station beams has been demonstrated. However, since the calibrator will necessarily fall within the station field of view if it is to be useful anyway, this approach is generally not useful for calibrating the international stations. The continuous monitoring of a calibrator is recommended -- beam switched phase referencing (i.e., time interpolation) has not yet been demonstrated, and is unlikely to be useful given the rapid variability solutions to the international stations.


The Long Baseline Pipeline


The Long Baseline Pipeline is no longer available, but has been used to reduce data in the past. It undertakes the following steps:

1) RFI flagging and averaging, as well as demixing if needed (rare; note the it is not currently possible to demix only a subset of the antennas and thus exclude the international stations.  Smart demixing, when available, will solve this problem)

2) Calibrate the Dutch array using the specified "core calibrator":

a) Make a copy of the data, average further

b) Run BBS to calibrate the amplitudes and phases of the Dutch stations. Note that since June 2015 only baselines between CS and RS are selected in the SOLVE step, while all baselines (i.e. involving CS, RS, and IS) are used for the CORRECT step

c) Apply the calibration to the Dutch stations in the previous dataset (on both the calibrator and the target), leaving the international stations unchanged (except for the application of beam corrections)

3) Perform the following steps 1 or more times:

a) optionally uv shift the target data to a new phase center

b) create a sensitive "tied station" ST001 in this shifted dataset by phasing up the core stations

c) optionally filter out the now redundant core stations from the dataset

d) convert this dataset to circular polarization

e) group every M subbands together (making fewer measurement sets, each of which covers a wider bandwidth)

f) create a virtual measurement set which holds all of the N newly created, wider bandwidth measurement sets in one place (as separate spectral windows)


The output of step 3) is a single measurement set containing all of the bandwidth on a given target source.  It will have been considerably reduced in size, and is ready for fringe fitting of the international stations.  The PI need only convert the measurement set to FITS (can be performed in CASA or with the tool ms2uvfits) and it is ready to be loaded into AIPS for further processing.  Note that the international station amplitudes have not yet been calibrated. 


*3) a) The phase shift option is not yet supported in the long baseline pipeline available at the moment. User should opt for the multi beam option or for the separate observation of a calibrator.  


In order to make use of the Long Baseline Pipeline, the following restrictions should be observed:

1) Subbands should be grouped into N contiguous chunks, each of M subbands.  (In practise, often many or even all of the groups of M subbands are actually adjacent).  When specifying the observation, the subband numbers (or at least the frequency range(s) and the value of M) needs to be supplied to the SDC Operations.

2) The calibration mode for the core stations (either 1) continuous calibration using a separate beam, or 2) alternating between calibrator and target) needs to be specified to the SDC Operations.  In either case, identical subband layouts need to be used for the "core calibrator" and the target; in mode 1), that means reducing the available bandwidth on the target, and the core calibrator must of course fit within the tile beam for HBA observations.  In mode 2), the only supported cycle type is C - T - C - T - C …, where C is the core calibrator and T is the target field.  It is not currently possible to (for instance) interleave observations of several different targets in between successive observations of the core calibrator.


The following information should also be supplied to the SDC Operations:

1) Source positions for all calibrators and targets (pointing positions and, if uv shifting is to be used, additional phase centres)

2) The desired number of channels per subband after averaging



Data for International LOFAR is generally taken with 64 channels/subband and 1 second averaging. For HBA this allows for a field of view of approximately 1 degree radius (predominantly limited by time smearing). Generally, the first step of any data processing will be flagging and averaging. Typical post-averaging values used with International LOFAR observations are 4 channels/subband and 4 seconds. This limits the field of view to about 10 arcminutes radius for the HBA (predominantly limited by bandwidth smearing). This is less restrictive than the "station beam" of the phased core, which with an effective diameter of 2 km has a beam width of around 4 arcminutes. To image sources further away from the phase centre, it is necessary to go back to the raw data, shift to a new tangent point using NDPPP, average again and then apply the calibration derived from the calibrator phase centre. The time taken for NDPPP to uvshift and average a dataset is considerably less than the time taken for NDPPP to flag a dataset, so if multiple widely separated tangent points will be required, a flagged but unaveraged dataset should be generated by once NDPPP and then shifted/averaged multiple times.

For HBA observations the required pixel size is approximately 0.1  arcseconds, and a 1024×1024 pixel image can be made without suffering from w-projection effects, so it is possible to directly image the whole smearing limited FOV in one pass using around 100 facets in e.g. AIPS. Alternatively, one can image using a w-projection capable  imager like CASA, and then it is possible to make a single large image (e.g., 8192×8192) but this is likely to be slow. Of course, it is also possible to uv shift to areas of interest within the 20'x20' FOV and image a smaller region. In this case, bright sources should be imaged first and subtracted.  Imaging a single facet is fast, taking a small fraction of observe time. Imaging the entire smearing-limited FOV of the averaged data using either faceting or w-projection techniques is considerably more computationally intensive. 


Computational Requirements for the Standard Long Baseline Pipeline

The SDC Operations has characterized the behaviour of the long-baseline pipeline on several typical settings. Tests indicate a systematic larger value of the P/O ratio for the amplitude calibration of the calibrator than the target and phase shift steps.The outcome of this testing is that for snaphot observations where the durations of the calibrator runs are comparable to those of the target, a P/O value of 4 should be assumed. For snaphot observations where the target runs are longer than those of the calibrator and for observations with a continuous multibeam setup, the PIs are advised to use a P/O value of 3.5 in their calculations. In case of simultaneous observations of target(s) and calibrator a P/O value of 5 should be assumed. The table below summarizes typical durations of the pipeline steps (calibrator, target, phase shift and concatenation) for a 3 min calibrator / 1 hour target observations and 240 SBs per beam:

59 min
  Calibrator Pipeline (CP) 0.98
124 min
 Target Pipeline (TP) 1.97
 10 min
 Phase shift & CS adding
 9 min
 Concatenation 0.15

We warn the observers that these values are indicative of the LB pipeline performance as tested so far. Further characterization will be performed to investigate a wider parameter space in the coming weeks.

Design: Kuenst.    Development: Dripl.    © 2022 ASTRON