BBS

This page describes BBS (BlackBoard Selfcal), a calibration package for LOFAR observations.

BBS has two modes of running. The first is to run as a stand-alone tool to calibrate one subband. If a single subband does not contain enough signal, BBS offers the possibility to use data from multiple subbands together for parameter estimation. This is called /global/ parameter estimation. On this page, we will focus on the stand-alone version; setting up a global solve is described elsewhere.

Currently this page is work in progress, see Configuration Syntax for the old version

To calibrate a single MS, execute the calibrate-stand-alone script on the command line:

calibrate-stand-alone -f <MS> <parset> <source catalog>

The important arguments provided to the calibrate-stand-alone script in this example are:

• <MS>: name of the MS to process.
• <parset>: description of the reduction (see below).
• <source catalog>: A list of sources that can be used for calibration (see below). When the MS already contains a sourcedb, it is not necessary to specify a new source catalog if you're calibrating on the same sourcedb.
• The -f option, which overwrites stale information from previous runs (remove old instrument table and old source model if they exist with the measurement set).
• The -t option, which provides some basic multithreading.

You can run the script without arguments for a description of all the options and the mandatory arguments, such as the -v option to get a more verbose output.

A source catalog is a plain text file containing information on your model. An example catalog file is shown below:

# (Name,Type,Ra,Dec,I, ReferenceFrequency='55.468e6', SpectralIndex) = format
3C196, POINT, 08:13:36.062300, +48.13.02.24900, 153.0, , [-0.56, -0.05212]

More information about the format is on the makesourcedb page. You can use gsm.py to get a source catalog from an existing catalog.

The parset file describes the steps in the reduction you want to perform. A very basic example parset is:

Strategy.ChunkSize = 100
Strategy.Steps = [predict]

Step.predict.Model.Sources     = []
Step.predict.Operation         = PREDICT
Step.predict.Output.Column     = MODEL_DATA
Step.predict.Model.Beam.Enable = False

An especially important key is Strategy.ChunkSize, which specifies the amount of data (in timeslots) that is read into main memory as a single block. If your measurement set is large, be sure to fill in a sensible value here. It should be at least as large as the largest Solve.CellSize.Time value in your reduction. Also, you do not want to make it too small, because repeatedly reading small chunks hurts performance. It is advisable to ensure all Solve.CellSize.Time values in your reduction are divisors of Strategy.ChunkSize, to avoid truncated solution cells.

• The tool parmdbplot.py can be used to plot the calibration solutions.
• Usually, the corrected data needs another round of flagging before imaging because of bad calibration solutions. Bad solutions may occur for various reasons, e.g. because most of the samples in a solution cell are flagged due to RFI or because some of the calibrators are not above the horizon (when not using a beam model).

Below are all the effects that can be modeled, in the order they are applied (from sky to ground).

The beam model tries to emulate all kinds of distortions to the signal that are caused by the beam of the station. These effects are split into two parts: the element beam, which is the response of a single dipole, and the array factor, which emulates the effect of combining the signal of the many dipoles in a station. In HBA, the array factor model also models the effect of the analog tile beam former.

To have a look at different elements of the beam, you can specifically use only the element beam or only the array factor (if you don't know the details, you need both the element beam and the array factor, which is the default). The options are:

Model.Beam.Mode = ELEMENT       # only element beam
Model.Beam.Mode = ARRAY_FACTOR  # only array factor
Model.Beam.Mode = DEFAULT       # both element beam and array factor (default)

The tile beam former in the HBA tiles forms a beam for a certain reference frequency. When modeling this beam, the beam model should of course do this for the same frequency. Usually, this works automatically: the reference frequency is stored in the measurement set. Things are different when you compress a number of subbands as channels into one measurement set (usually done with NDPPP). Then each `channel' was beamformed at a different reference frequency. In this case, the reference frequency is only right for one of the `channels'. To handle this case well, there is an option that tells the beam model to use the channel frequency (which is usually the center frequency of the compressed subband). This option is:

Step.Solve.Model.Beam.UseChannelFreq = T

Note that the beam model is a direction dependent effect like any other in BBS. That means that over a patch, the beam is assumed to be constant (it is evaluated at the centroid of the patch). This may change in the future.

• BBS parameter set documentation (old)
• makesourcedb, with the description of the sky catalog file format
• BBS Exercise from the LOFAR Data Processing School
• Initial report on BBS performance (Joris van Zwieten, Jan 7, 2008)
• Parameter database description (Ger van Diepen)
• Creating (distributed) MeasurementSets (Ger van Diepen)
• Ionospheric calibration in BBS (Maaijke Mevius)