fringefit(vis, caltable='', field='', spw='', intent='', selectdata=True, timerange='', antenna='', scan='', observation='', msselect='', solint='inf', combine='', refant='', minsnr=3.0, zerorates=False, globalsolve=True, niter=100, delaywindow='', ratewindow='', append=False, corrdepflags=False, docallib=False, callib='', gaintable='', gainfield='', interp='', spwmap='', paramactive='', parang=False)¶
Fringe fit delay and rates
vis (path) - Name of input visibility file
caltable (string=’’) - Name of output gain calibration table
field (string=’’) - Select field using field id(s) or field name(s)
spw (string=’’) - Select spectral window/channels
intent (string=’’) - Select observing intent
selectdata (bool=True) - Other data selection parameters
solint (variant=’inf’) - Solution interval: egs. 'inf', '60s' (see help)
combine (string=’’) - Data axes which to combine for solve (obs, scan, spw, and/or field)
refant (string=’’) - Reference antenna name(s)
minsnr (double=3.0) - Reject solutions below this signal-to-noise ratio (at the FFT stage)
zerorates (bool=False) - Zero delay-rates in solution table
globalsolve (bool=True) - Refine estimates of delay and rate with global least-squares solver
niter (int=100) - Maximum number of iterations for least-squares solver
delaywindow (doubleVec=’’) - Constrain FFT delay search to a window
ratewindow (doubleVec=’’) - Constrain FFT rate search to a window
append (bool=False) - Append solutions to the (existing) table
corrdepflags (bool=False) - Respect correlation-dependent flags
docallib (bool=False) - Use callib or traditional cal apply parameters
docallib = False
docallib = True
callib (string=’’) - Cal Library filename
paramactive (boolVec=’’) - Control which parameters are solved for
parang (bool=False) - Apply parallactic angle correction on the fly
calculates a fringe-fitting (phase/delay/rate) solution
WARNING: fringefit is currently an experimental task. Use with care and report issues back to the CASA team via the NRAO helpdesk. Note that calibration tables made with fringefit in CASA 5.5 will not work in CASA 5.6 and later.
The fringefit task determines phase, delay, delay-rate and (optionally) dispersive delay solutions as a function of time and spectral window, thus enabling correction of visibility phases for errors introduced by the atmosphere, the signal paths of the instrument, or other pre-calibration steps. It can correct a single scan on a bright source for delay offsets between spectral windows, which can result from instrumental signal paths (single-band delays), and it can correct multiple scans on a source to correct for errors that are variable in time (multi-band delay and delay rate). The task uses the model data column when present in the MeasurementSet. Fringe fitting is primarily useful for VLBI.
In Very Long Baseline Interferometry (VLBI), fringe-fitting is an essential step which is not typically used for connected element arrays such as JVLA and ALMA. The very long baselines make observations particularly sensitive to small errors in the correlator model (station positions), signal chain, and the temporally- and spatially-variable atmosphere. The errors manifest themselves as residual delays (which introduce a slope when plotting phase against frequency) and delay-rates (which cause the phase offset at, say, the center of the band to drift as a function of time). Additionally at low frequencies and large fractional bandwidths a dispersive delay can be observed proportional (in time) to the inverse of the square of frequency and therefore proportional to the reciprocal of the frequency in phase. These errors can be corrected by observations of a bright calibrator source and a phase reference source which is close to the target source in projection on the sky.
Prior to running fringefit it is recommended to calibrate the amplitudes, as the weights are used by fringefit to determine the reliability of the fringe detection. A tunable SNR-cutoff is implemented to allow weak fringes and non-detections to be discarded. Bandpass correction can be done either before or after fringe fitting using bandpass. ** ** Since the CASA bandpass is a simple complex gain correction, it will be more efficient after fringe fitting when VLBI-style geometrical errors are important.
The algorithm used in fringefit, based on the Schwab-Cotton algorithm (originally implemented in the AIPS task FRING), has two stages. The first step is to look for fringes on each baseline to a reference station separately and estimate the SNR value for each of these baselines using a Fast Fourier Transform. The fringes that pass the SNR requirement are passed through to the second step, a global least-squares solver that attempts to optimize the solutions and transform them into antenna-based solutions for phase, delay, rate and dispersive delay. All of the parameters except phase can be included or excluded in the solution at this phase using the paramactive keyword described below; the default is that phase, delay and rate are solved but dispersive delay is not, since this matches the historical default and the usage we expect to be most common in VLBI.
NOTE: For the multi-band delay, the solutions for the full combined spectral window are written for a notional spectral window 0. To account for this spectral window in the applycal step, use the spwmap parameter to ensure that the solutions are correctly applied to all spectral windows (see the Examples tab for details).
Common calibration solve parameters
See Solving for Calibration for more information on the task parameters fringefit shares with all solving tasks, including data selection, general solving properties and arranging prior calibration (i.e., specifying other caltables to pre-apply before solving). Also see the rerefant task documentation for the behavior of reference antenna application. Below we describe parameters unique to fringefit and those common parameters with unique properties.
NOTE: Like gaincal, fringefit supports passing a list of antennas through the refant parameter. However fringefit does not implement the refantmode parameter and effectively always operates in ‘flex’ mode, using the next antenna from the list as the reference antenna if an antenna is not available.
Solution interval: solint
The solution interval specified in solint (in seconds) is used to group data. It is important to make sure that for intervals shorter than the scan length, the scan is divided into roughly equal sized solution intervals. Avoid selecting a solution interval that will lead to intervals containing a single integration time, as this will cause an error.
Combining data for solutions: combine
As in other calibration solve tasks, data can be combined over different axes. To derive multi-band delay corrections, set combine=’spw’.
SNR control: minsnr
The minsnr parameter sets the threshold of the SNR value required for the baseline based fringe (FFT stage) to be included in the global least-squares minimization.
When correcting instrumental delays by solving for each spectral window separately, it is usual to apply the corrections derived to the entire dataset. Extrapolating the rates in time is undesirable, so use of zerorates=True will cause no time-dependent rate correction to be applied. Note that with this option the rates are still solved, and zeroed only when written to the table; paramactive can be used to turn off solution of rates altogether but this is currently not recommended; zerorates (and its equivalent in other software) has been the standard practice in VLBI for a long time, and is likely to remain so.
Prior correction for parallactic angle: parang
Although optional, it is is generally recomended that parang=True be used for VLBI observations, since parallactic angle causes differential phase rates among widely-separated antennas that usually should not be included within the fringefit solution.
Disabling the global least-squares solver: globalsolve
By default, fringe-fit solutions are refined by a global least-squares optimization algorithm after the FFT stage. For some purposes, it is desirable to use the estimates from the FFT stage directly; this can be done by setting globalsolve =False. (The default is True)
Setting a maximum number of iterations: niter
A maximum number of iterations for the global least squares solver can be set with the niter parameter. The default is 100; in cases with high signal-to-noise this limit is not reached.
Constrain the search window for delay: delaywindow
Sometimes a priori information is available to constrain the delays relative to the reference station at the FFT search step. The upper and lower bounds (in nanoseconds) can be provided as a two element list through the keyword delaywindow. The value None can be used to leave either the upper or lower limit unconstrained (setting both to None constrains neither; this is the default). Note that the same constraint is applied to all baselines in the FFT search step.
Constrain the search window for rate: ratewindow
Similarly to delaywindow, sometimes a priori information is available to constrain the delay rates relative to the reference station at the FFT search step. The upper and lower bounds (in units of seconds/second) can be provided as a two element list through the keyword ratewindow. The value None can be used to leave either the upper or lower limit unconstrained (setting both to None constrains neither; this is the default). Note that the same constraint is applied to all baselines in the FFT search step.
Select a weighting strategy for the least squares solver: weightfactor
It is common in VLBI practice for the user to choose how weights of visiblities should be used in the global stage of fringe-fitting. In any array such as the EVN with a very sensitive antenna (in the EVN’s case Effelsberg), the use of measurement set weights can mean that baselines to the sensitive antenna dominate and other baselines have neglibible impact. Choosing the square root of those weights gives, many users feel, a more balanced interpretation of the data.
The weightfactor parameter allows the user to chose between strategies:
0 => use a weight of 1 (i.e., ignore measurement set weights);
1 => use the square-root of measurement set weights;
2 => use the measurement set weights as they are (the default)
Select active parameters for least square solver: paramactive
As part of the inclusion of a dispersive component of delay we have added a parameter to control which model parameters are used in the least-squares part of the solver (the FFT stage is unaffected). The paramactive parameter takes a Python list of boolean arguments for the delay, rate and dispersive components, with a default value of [True, True, False] to match the historic default, which is also expected to be the most common future use-case. Note that we do not offer users an opportunity not to solve for phase offset (also known as “secular phase”).
Single-band delay: calibration of delay only for a single scan on a bright calibrator:
fringefit(vis='data.ms', caltable='data.sbd', # write solutions to this table on disk scan='30', # use only scan 30 solint='inf', # use all timestamps in the scan refant='EF', # a big antenna does well as reference antenna minsnr=50, # empirically proven to be a good value is anything over 25 zerorates=True, # for instrumental delay rates should not be used gaintable=['data.tsys','data.gc'], # apply the amplitude calibration on the fly parang=True) # always True for VLBI
Multi-band delay: calibration of time-dependent delay and delay-rate for a phase reference source, relative to single-band delay solution from above:
fringefit(vis='data.ms', caltable='data.mbd', # write solutions to this table disk solint='60', # anything shorter than the scan length will give more than 1 solution combine='spw', # combine spectral windows for this step, gives only a solution for spw0 field='1', # select the field that is your phase reference calibrator refant='EF', # pick a big antenna or one close to the geometric center of the array minsnr=50, # this seems to be a sensible value gaintable=['data.tys', 'data.gc', 'data.sbd'], # apply amplitude calibration and single band delay on the fly parang=True) # always set to True for VLBI
The calibration table data.mbd will contain phase, delay, and rate solutions per antenna, per polarization and per solution interval. For data with multiple spectral windows the solutions will be assigned to spectral window 0 in the output cal table. In the applycal step, the parameter spwmap needs to be set to apply the solutions to all spectral windows. For example, in a dataset with 8 spectral windows: spwmap=[8*]. Since the applycal step will include multiple calibration tables, this setting needs to correspond to the data.mbd table in the gaintable parameter:
applycal(vis='data.ms', field='0,1', # now select the phase calibrator AND the target source gaintable=['data.tsys', 'data.gc','data.sbd', 'data.mbd'], # include all the calibration tables interp=,spwmap=[, , , 8*], # map the spectral windows accordingly parang=True) # for VLBI this should always be True
In cases where it is necessary to constrain the search for group delay and fringe rates at the FFT stage, the parameters delaywindow and ratewindow can be used:
fringefit(vis='data.ms', caltable='data.mbd', # write solutions to this table disk solint='60', # anything shorter than the scan length will give more than 1 solution combine='spw', # combine spectral windows for this step, gives only a solution for spw0 field='1', # select the field that is your phase reference calibrator refant='EF', # pick a big antenna or one close to the geometric center of the array minsnr=5, # we're looking for weak detections, but we have a good a priori idea of # where they are to steer the FFT search delaywindow = [0,10], # FFT delay search range of 0 to 10 nanoseconds ratewindow = [-5e-9,5e-9], # FFT rate search range of -5 to 5 nanoseconds per second gaintable=['data.tys', 'data.gc', 'data.sbd'], # apply amplitude calibration and single band delay on # the fly parang=True) # always set to True for VLBI
No additional development details
- Parameter Details
Detailed descriptions of each function parameter
vis (path)- Name of input visibility file
caltable (string='')- Name of output gain calibration table
field (string='')- Select field using field id(s) or field name(s)
spw (string='')- Select spectral window/channels
intent (string='')- Select observing intent
selectdata (bool=True)- Other data selection parameters
timerange (string='')- Select data based on time range
antenna (string='')- Select data based on antenna/baseline
scan (string='')- Scan number range
observation (string='')- Select by observation ID(s)
msselect (string='')- Optional complex data selection (ignore for now)
solint (variant='inf')- Solution interval: egs. 'inf', '60s' (see help)
combine (string='')- Data axes which to combine for solve (obs, scan, spw, and/or field)
refant (string='')- Reference antenna name(s)
minsnr (double=3.0)- Reject solutions below this signal-to-noise ratio (at the FFT stage)
zerorates (bool=False)- Zero delay-rates in solution tableWrite a solution table with delay-rates zeroed, for the case of“manual phase calibration”, so that the calibration table can beapplied to the full dataset without the extrapolation of a non-zero delay-rate termaffecting the data
globalsolve (bool=True)- Refine estimates of delay and rate with global least-squares solver
niter (int=100)- Maximum number of iterations for least-squares solver
delaywindow (doubleVec='')- Constrain FFT delay search to a window specified as a two-element list with units of nanosecondsDefault: [None, None]Examples: [-10, 10]
ratewindow (doubleVec='')- Constrain FFT rate search to a window specified as a two-element list with units of seconds per secondDefault: [None, None]Examples: [-1e-13, 1e-13]
append (bool=False)- Append solutions to the (existing) tableDefault: False (overwrite existing table or makenew table)Appended solutions must be derived from the sameMS as the existing caltable, and solution spwsmust have the same meta-info (according to spwselection and solint) or be non-overlapping.
corrdepflags (bool=False)- If False (default), if any correlation is flagged, treat all correlations inthe visibility vector as flagged when solving (per channel, per baseline).If True, use unflagged correlations in a visibility vector, even if one or moreother correlations are flagged.Default: False (treat correlation vectors with one or more correlations flagged as entirely flagged)Traditionally, CASA has observed a strict interpretation ofcorrelation-dependent flags: if one or more correlations(for any baseline and channel) is flagged, then all availablecorrelations for the same baseline and channel aretreated as flagged. However, it is desirable in somecircumstances to relax this stricture, e.g., to preserve useof data from antennas with only one good polarization (e.g., one polarizationis bad or entirely absent). Solutions for the bad or missing polarizationwill be rendered as flagged.
docallib (bool=False)- Control means of specifying the caltablesDefault: False (Use gaintable, gainfield, interp,spwmap, calwt)Options: False|TrueIf True, specify a file containing cal library incallib
callib (string='')- Specify a file containing cal library directivesSubparameter of docallib=True
gaintable (stringVec='')- Gain calibration table(s) to apply on the flyDefault: ‘’ (none)Subparameter of docallib=FalseExamples:gaintable=’ngc5921.gcal’gaintable=[‘ngc5921.ampcal’,’ngc5921.phcal’]
gainfield (stringVec='')- Select a subset of calibrators from gaintable(s)Default: ‘’ (all sources on the sky)‘nearest’ ==> nearest (on sky) available field intable otherwise, same syntax as fieldExamples:gainfield=’0~2,5’ means use fields 0,1,2,5from gaintablegainfield=[‘0~3’,’4~6’] means use field 0through 3
interp (stringVec='')- Interpolation parameters (in time[,freq]) for each gaintable, as a list of strings.Default: ‘’ –> ‘linear,linear’ for all gaintable(s)Options: Time: ‘nearest’, ‘linear’Freq: ‘nearest’, ‘linear’, ‘cubic’,‘spline’Specify a list of strings, aligned with the list of caltable specifiedin gaintable, that contain the required interpolation parametersfor each caltable.* When frequency interpolation is relevant (B, Df,Xf), separate time-dependent and freq-dependentinterp types with a comma (freq after thecomma).* Specifications for frequency are ignored when thecalibration table has no channel-dependence.* Time-dependent interp options ending in ‘PD’enable a “phase delay” correction per spw fornon-channel-dependent calibration types.* For multi-obsId datasets, ‘perobs’ can beappended to the time-dependent interpolationspecification to enforce obsId boundaries wheninterpolating in time.* Freq-dependent interp options can have ‘flag’ appendedto enforce channel-dependent flagging, and/or ‘rel’appended to invoke relative frequency interpolationExamples:interp=’nearest’ (in time, freq-dep will belinear, if relevant)interp=’linear,cubic’ (linear in time, cubicin freq)interp=’linearperobs,splineflag’ (linear intime per obsId, spline in freq withchannelized flagging)interp=’nearest,linearflagrel’ (nearest intime, linear in freq with with channelizedflagging and relative-frequency interpolation)interp=’,spline’ (spline in freq; linear intime by default)interp=[‘nearest,spline’,’linear’] (formultiple gaintables)
spwmap (intVec='')- Spectral window mappings to form for gaintable(s)Only used if callib=Falsedefault:  (apply solutions from each calibration spw tothe same MS spw only)Any available calibration spw can be mechanically mapped to anyMS spw.Examples:spwmap=[0,0,1,1] means apply calibrationfrom cal spw = 0 to MS spw 0,1 and cal spw 1 to MS spws 2,3.spwmap=[[0,0,1,1],[0,1,0,1]] (use a list of lists for multiplegaintables)
paramactive (boolVec='')- Control which parameters are solved for; a vector of (exactly) three booleans for delay, delay-rate and dispersive delay (in that order)
parang (bool=False)- Apply parallactic angle correction on the fly.