1934 6cm bow-ties

[keithBannister]

Hi All,

Another question (which I'm sure is very basic).

I've got 1934 data with the 6cm contiuum CABB data.

if I do
mfcal interval=3
gpcal interval=0.1 options=xyvary

Then plot real vs imaginary, I get bowties:

Bow ties

1934_reim.png (21.1 KiB) Viewed 76990 times

Also attached is a zip file containing the same plot over a range of frequencies. It shows what you'd expect (i.e. a spectral index from 6->4 with the scatter worst at the edge of the bands).

I guess I'm a little bit surprised that XX and YY have such bad scatter at the band edges. At first I convinced myself that this was just poorer system temperature at the band edges, but wouldn't that affect XY and YX too? If it's a bandpass effect, I'm surprised it doesn't affect the XY and YX in the same way....

[keithBannister]

Hiya,

I worked it out, I think.

End result: the 5GHz bandpass phase varies substantially in 2 hrs. Regular observations of a bandpass calibrator are recommended, and time-dependant bandpass calibration is important

Below (and attached) are 3 scans of 1934, at 2am, 5am and 7am (roughly).

The mfcal based on a single scan results (mf1) are calibrated based on a single (5am) scan:
mfcal vis = ./1934-638.5500.mf1 select = time(10NOV26:05:00:00,10NOV26:06:00:00 interval = 3

The tdb ones are based on a time dependant bandpass with
mfcal vis=1934-638.5500 interval=3,3,3

Both ones have gpcal interval=0.1 options=xyvary applied

The bandpass phase clearly changes over 2 hrs in a way that gpcal cannot compensate for. Time dependant bandpass calibration fixes the problem more or less, although there are still a few bow-ties (or peanuts) happening, which might be due to the frequency-dependant Tsys.

Anyhoo, I hope this is useful for someone. Those commas in mfcal interval are important.

K

[keithBannister]

Posting again so I can get over the 3 attachment limit.

[keithBannister]

Another thought:

Instead of solving for a time-dependent bandpass, one could split the data up into chunks in frequency (say, I dunno, 4 or 8?) and then calibrate them separately. That way gpcal will solve for (slightly) different phases on every chunk, which should get around this problem.

[keithBannister]

Oh wait, use gpcal with nfbin>1