Getting Started

Documentation

Glish

Learn More

Programming

Contact Us

Version 1.9 Build 1556

News	FAQ
Search	Home

autoflag.settimemed - Function

	Package	synthesis
	Module	autoflag
	Tool	autoflag

Sets up the time-median flagging method

Synopsis
settimemed(thr, hw, rowthr, rowhw, norow, column, expr, fignore)

Description

The timemed method uses a median-in-time filter. It works by computing a sliding median of some value d_cbt (where c is frequency channel, b is baseline, and t is time slot), as specified by the expr parameter (see the general autoflag module description for details of how expr determines d.) The sliding median is computed in time, on a per-channel, per-baseline basis:

m_cbt = median{d_cbt', where | t' - t| $$\leq$$ $$\delta$$}

(where $\delta$ is given by the hw parameter.) Then, a median absolute deviation (MAD) of the data with respect to the median is computed on a per-row basis:

M_bt = median{| d_cbt - m_cbt|, for all c}

(For single-channel data, M_bt $\equiv$ m_0bt.) M_bt is then smoothed by applying a sliding median in time:

$$\tilde{M}_{bt}^{}$$ = median{M_bt', where | t' - t| $$\leq$$ $$\tilde{\delta}$$}

(where $\tilde{\delta}$ is given by the rowhw parameter.) All points where

| d_cbt - m_cbt| > $$\theta$$$$\tilde{M}_{bt}^{}$$

are then flagged. The threshold level $\theta$ is given by the thr parameter.

In addition, timemed can also flag entire rows based on their MAD. This is only performed for line data (specifically, when there are at least 10 channels per row), and may be explicitly disabled by specifying norow=T. To do this, it first computes a per-baseline MAD of the MAD:

$$\hat{M}_{b}^{}$$ = median{| M_bt - $$\tilde{M}_{bt}^{}$$|, for all t}

and flags rows where | M_bt - $\tilde{M}_{bt}^{}$| > $\hat{\theta}$$\hat{M}_{b}^{}$, where $\hat{\theta}$ is given by the rowthr parameter.

Timemed will iterate over the measurement set - and compute m_cbt - only once. Existing flags are taken into account at this time (unless fignore is specified). MADs are then computed and recomputed iteratively, dropping flagged points and rows, until convergence is achieved (i.e. no more flags are raised). Since m_cbt is not recomputed as new flags are raised, running this method a second time may sometimes be of use.

Arguments

thr		Flagging threshold
		Allowed:	numeric (real)
		Default:	5
hw		Half-window for the sliding median, if time slots is less than 4 times half window, ignoring this chunk
		Allowed:	Int
		Default:	10
rowthr		Row flagging threshold
		Allowed:	numeric (real)
		Default:	10
rowhw		Half-window for the sliding median of the MAD
		Allowed:	Int
		Default:	10
norow		Disable row flagging
		Allowed:	Boolean
		Default:	F
column		Which column to base flagging on
		Allowed:	"DATA", "MODEL" or "CORR"
		Default:	"DATA"
expr		Expression to use
		Allowed:	see text
		Default:	"ABS I"
fignore		Ignore existing flags
		Allowed:	Boolean
		Default:	F

Example

- af:=autoflag('3C273XC1.MS')
- af.setdata()
- af.settimemed(thr=5,hw=5,expr="- ABS RR LL")