Australia Telescope Compact Array

Off-line baseline fitting

With the move to observing at 3mm wavelength with the ATCA, it has become apparent that there are some instrumental phase terms that need to be carefully handled.

The so-called ``wrap-dependent phase'': This is an error where the phase of a source changed significantly depending on whether the source was observed in the north or south wrap. Very little signature of the error was noted in the round-trip measurements of both the 13 GHz and 160 MHz reference signals. It was found that the problem relates to a standing wave in the 160 MHz reference. The changes in the tension within the LO fibre as it goes through the azimuth wrap was causing changing attenuation in the fibre. Because the standing wave went through the azimuth wrap twice (forward and reverse) - unlike the reference signal - the standing waves contribution to the overall system phase varied with azimuth.
At its worst, the ``wrap-dependent phase'' could contribute the equivalent of 7mm of excess path to the signal.
A signature of the problem was that the phase error changed sign when going from 3cm to 12mm observing. Careful measurements also showed the phase error scaled with frequency between 6 and 3cm, but its magnitude was about half the frequency-scaled value at 12mm, and that the error seemed inconsistent with frequency-scaling at 3mm. The measurements were all consistent with the error scaling with the contribution of the 160 MHz reference signal to the downconversion process.
The solution to the problem has been to search for and remove the causes of the standing waves, and to instigate changes in the optical path to attenuate any standing waves that are generated.
Implicitly, CAOBS assumes that the displacement between the focus points on the different antennas is constant (for a given reconfiguration). Changing azimuth and elevation does not vary the distance between the focus points. Put another way, CAOBS implicitly assumes that the antennas are mechanically identical in the way the focus point moves with azimuth and elevation. Although this is good to first order, at some point this this approximation will break down.
Although this issue has not been investigated in depth, there appears to be an effect on CA01 which is equivalent of up to 1mm of excess path being introduced as a function of elevation. This is quite reproducible and detectable when the wrap-dependent phase is low and the atmospheric seeing is reasonable. There also appears to be a similar signature on CA05, but of a quarter the amplitude. The nature of the error suggests that it is a mechanical difference between the antennas. Because this error is readily predictable, it can be corrected in the on-line or off-line software.

An off-line Miriad-based baseline fitting program, blfit, was written to assist in exploring these issues. Blfit has a number of strengths over CABSLN.

It handles sets of observations at different frequencies or at different times. It solves in a fashion so that phase wrapping at higher frequencies is avoided. It handles the two polarisations as two separate sets of data.
It uses the seeing monitor data and a Kolmogorov model to estimate the rms contribution of the atmosphere to the phase. This allows data to be correctly weighted in a solution process.
It allows a number of plotting options to explore the residuals from the baseline fit.
It includes a model of the instrumental phase contrbuted by CA01 and CA05, and can correct for this.
It includes a simple model of the signature of the wrap-dependent phase, and can solve for a first-order term to eliminate the wrap-dependent phase.

Using blfit

The code, executable and standard Miriad help file for blfit in directory $OPER/blfit. To use blfit to solve for the the wrap-dependent phase term, you will need to merge the dial azimuth into the datasets used by blfit. Example scripts are present in the $OPER/blfit.

  GetAz.csh    Retrieved dial azimuth from MoniCA. The dates in this script
               will need to be monified.
  Load.csh     Atlod the data, split and bandpass calibrate the data. The
               RPFITS file and the name of the bandpass calibrator need
               to be set.
  DoFit.csh    Merge in the dial azimuth, and run blfit.

Output from blfit

blfit produces output as below

Frequency range:   4.17 -   5.43 GHz
Scans selected: 78

 Ant=1 Iter=1 Normalise rms=0.72
 Ant=2 Iter=1 Normalise rms=0.70
 Ant=3 Iter=1 Normalise rms=0.65
 Ant=4 Iter=1 Normalise rms=0.00
 Ant=5 Iter=1 Normalise rms=0.84
 Ant=6 Iter=1 Normalise rms=0.50

 Ant       X (mm)            Y (mm)            Z (mm)
 ---   --------------    ---------------   ---------------
   1  -0.063 +/- 0.442  -0.704 +/- 0.395  -0.722 +/- 1.624  -3.027 +/- 0.859
   2   0.183 +/- 0.357  -0.515 +/- 0.319   0.107 +/- 1.340  -1.751 +/- 0.693
   3   0.628 +/- 0.309   0.103 +/- 0.276   0.969 +/- 1.136  -2.373 +/- 0.600
   4   0.000 +/- 0.006   0.000 +/- 0.004   0.000 +/- 0.015   0.000 +/- 0.010
   5  -0.099 +/- 0.170  -0.389 +/- 0.151   1.520 +/- 0.623   2.099 +/- 0.330
   6  15.163 +/- 2.094 -11.261 +/- 1.872 -21.086 +/- 7.699  -2.273 +/- 4.064

It steps through increasingly large frequency spreads in its solution process, until the spread takes in all the frequencies of the available data. At each step, it list lines such as

   Ant=1 Iter=1 Normalise rms=0.72

This gives some information about the solution for each antenna. In this line, for CA01, it required just one iteration to find a solution, with the residual being 0.72 compared to the expected residual. The expected rms residual is estimated from the seeing monitor measurements, and so does rely on the accuracy of the Kolmogorov model in predicting the change of seeing with baseline. One expects the normalised residual to be near 1, but values from 0.5 to 1.5 are not unreasonable. The normalised residual for CA06 is usually on the low side, because the Kolmogorov model tends to be pessimistic at long baselines.

The solution is given as three or four numbers - the fourth number being the wrap-dependent phase term. Each has an associated 1-sigma error. The three coordinates are in an local x-y-z system corresponding to east, north and local vertical. The solution is the baseline error in millimeters.

At the end of the process, blfit prints out the eventual solution in two formats - that used by Miriad and that used by the ATCA on-line system.

The Miriad format is nanosec in a local equatorial system, and is the numbers that Miriad task uvedit uses to correct the data. These are the numbers that are saved in the dantpos.yymmdd file.

Miriad baseline convention:
Ant X (nsec) Y (nsec) Z (nsec)
--- -------- -------- --------
 1  0.07700 -0.05059 -0.00443
 2  0.07941 -0.04967 -0.00475
 3  0.08195 -0.04802 -0.00368
 4  0.08065 -0.05048 -0.00369
 5  0.08372 -0.05077 -0.00687
 6  0.00000  0.00000  0.00000
   
CABSLN convention:
Ant  X (mm)   Y (mm)   Z (mm)
--- -------- -------- --------
   
CA01  12.215 -24.774   1.329
CA02  12.976 -24.900   1.423
CA03  13.884 -24.860   1.103
CA04  13.174 -25.300   1.107
CA05  13.923 -25.842   2.059   
CA06   0.000   0.000   0.000

The ATCA on-line solver, CABSLN presents solutions in two ways: what I will call the CABSLN.TXT and CABSLN.DAT formats. blfit presents in the CABSLN.TXT format.

The CABSLN.DAT form is that stored in the file station.errors and that used by the on-line system. To convert from CABSLN.TXT and CABSLN.DAT, use

  olformat.pl blfit.log station.errors

where blfit.log is the output log file of blfit, and station.errors is the relevant station errors file to the observation.

Using non-standard continuum frequencies

To solve for the wrap-dependent phase, blfit needs to know what part of the LO downconversion frequency is produced by the 160 MHz reference. At centimetre wavelengths, the 13 GHz reference is not used, and the 160 MHz reference contributes the entire LO. This is not the case at 12mm and 3mm. The way to determine the contribution of the 160 MHz reference is to subtract the frequency generated from the 13 GHz reference from the observing frequency. Use LO_CHAIN to determine the frequency generated by the 13 GHz reference.

For example, when observing at 18496 and 19520 MHz (the standard 12mm continuum frequencies), LO_CHAIN indicates the 13 GHz reference contributes a tone at 28358 MHz. This is the Wiltron frequency less 160 MHz, and then multiplied by two. So the contribution of the 160 MHz reference is

  18496-28358
  19520-28358

Note these contributions are negative - which is fine. At 3mm, the 13 GHz reference contributes 8 times the (Wiltron frequency - 160 MHz). As well as at the centimetre frequencies, blfit can work out the contribution of the 160 MHz reference when using the 12mm and 3mm systems when using the standard continuum set-ups. Note this assumes the current LO algorithm.

If you use 3mm and 12mm systems at frequencies other than standard continuum ones, to let blfit know the contribution of the 160 MHz reference, you need to set a uv variable, f160 with this value. Use

  puthd in=set1.uv/f160 value=-9.862

This sets the contribution of the 160 MHz reference as -9.862 GHz.

Original: Bob Sault (27-Feb-2006)
Modified: Bob Sault (16-May-2006)