Testing `delay_and_tec` on PKS1934

[rcyndie]

Consider some of the first attempts of delay and TEC separation on this MeerKAT UHF calibrator observation at around 7am (may be coinciding with winter sunrise in the Karoo). The table below summarises the different chains of gain_type selected, with G and B capturing the complex gain and bandpass variables respectively. Here, we are using K for delay and T for TEC corrections.

Figure	gain_chain	gain_type (K/T)
1	GKB	K $\leftarrow$ delay_and_offset with init. est.
2	GTB	T $\leftarrow$ tec_and_offset with init. est.
3	GKTB	KT $\leftarrow$ delay_and_tec with init. est.

(Figure: LHS: phase plot from gain_chain; RHS: phase of B)

Figure 1

Figure 2

Figure 3

No phase wrapping captured by the different gain chains suggest no highly varying phase delays in the observation period.

[landmanbester]

Interesting that they all look so similar. I think we need to look at the phase residuals when comparing to the complex (1,1) solution to test which chain is performing best

[JSKenyon]

I would also plot the phase on the B term - it will make it obvious if anything has gone wrong.

[rcyndie]

Figure	gain_chain	gain_type (K/T)
4	G	diag_complex (1, 1)

Figure 4

Figure 5 (difference phase plot GKB w.r.t. diag_complex)

Figure 6 (difference phase plot GTB w.r.t. diag_complex) Note: All difference phase plots have their colourbars set to $[-0.1, 0.1]$.

[rcyndie]

_About MAD flagging and SIGMA_SPECTRUM_ The following plots are not directly comparable with the previous ones because of different config settings for the calibration, but should help give an idea of the not MAD flagged plots, for example, Fig. 5. When implementing MAD flagging, only the threshold_bl is being considered (with the rest disabled including whitening). All plots shown are the difference phase plots from the diag_complex at the corresponding settings (unless specified otherwise).

Figure	gain_type	MAD threshold_bl	Whitening
7	GKB with K $\leftarrow$ delay	2.	Disabled

(colourbar limits are $[-0.04, 0.04]$) Figure 7a (LHS: SIGMA column not specified; RHS: SIGMA $\leftarrow$ SIGMA_SPECTRUM) [Full view]

Figure 7b (LHS: SIGMA column not specified; RHS: SIGMA $\leftarrow$ SIGMA_SPECTRUM) [showing only a few antennas]

[JSKenyon]

Ok, so it looks like the SIGMA_SPECTRUM column is throwing off the solutions. This doesn't exactly surprise me. I am a bit confused by your MAD threshold_bl setting. A two sigma threshold is extremely low. Could you possibly report the flagging statistics before and after the run?

[rcyndie]

Do you mean the 'percentage flagged' from for example, goquartical-summary?

[JSKenyon]

Do you mean the 'percentage flagged' from for example, goquartical-summary?

Exactly.

[rcyndie]

	Percentage Flagged
Initial flags	27.88
MAD flagging with threshold_bl$=2.$	70.40
MAD flagging with threshold_bl$=2.$ (with SIGMA_SPECTRUM)	70.40

With or without specifying the SIGMA column, the percentage slagged is the same.

[JSKenyon]

Yeah, I would dial that threshold back to 5. That is relatively conservative (and shouldn't overflag so heavily). The fact that the flagged percentage is the same with and without SIGMA_SPECTRUM is a little odd but not impossible.

[rcyndie]

Figure	gain_type	MAD threshold_bl	whitening
8	GKB with K $\leftarrow$ delay	5.	native
9	same	4.	native

(same as in Figure 7, colourbar limits are $[-0.04, 0.04]$) Figure 8a (LHS: SIGMA column not specified; RHS: SIGMA $\leftarrow$ SIGMA_SPECTRUM) [Full view]

Figure 8b (LHS: SIGMA column not specified; RHS: SIGMA $\leftarrow$ SIGMA_SPECTRUM) [showing only a few antennas]

Figure 9a (LHS: SIGMA column not specified; RHS: SIGMA $\leftarrow$ SIGMA_SPECTRUM) [Full view]

Figure 9b (LHS: SIGMA column not specified; RHS: SIGMA $\leftarrow$ SIGMA_SPECTRUM) [showing only a few antennas]

	Percentage Flagged
Initial flags	27.88
MAD flagging with threshold_bl$=5.$	30.87
MAD flagging with threshold_bl$=5.$ (with SIGMA_SPECTRUM)	29.11
MAD flagging with threshold_bl$=4.$	35.90
MAD flagging with threshold_bl$=4.$ (with SIGMA_SPECTRUM)	32.20

When comparing Figure 8b with Figure 9b, the residual structure is better observed with the threshold_bl$=4.$, and I believe it is on the lower end but should not be overflagging the results, as per the percentage flagged. But, it is difficult to definitely conclude a pattern in the residual at this point.