The main advantage to averaging in time is to reduce the size of the data base. If you are working with large spectral-line data bases, this may be essential to reduce the data quantity to something manageable. Another reason you might like to average is so that any signal in very noisy visibilities can be seen more clearly when you are editing the data.
If you wish to average, use the task UVAVG, but make sure you do not incur significant loss of intensity owing to time smearing. This occurs when the source visibility structure changes significantly on the time scale over which you are averaging. It won't affect your calibrators much (as the visibility amplitudes should be roughly constant and the phases only slowly winding [assuming it is close to the phase centre]), if at all, but your program source might be badly affected. Do the sums correctly (see Appendix D for a more complete discussion) or don't average in time.
Another potential danger with time averaging is that you can't correct the antenna gains on a time scale shorter than the integration time with the self-calibration procedure. Atmospheric fluctuations that affect the antenna based gains may occur on time scales as short as a few seconds. However, you may not have sufficient signal-to-noise ratios in your data to benefit from this. Another way to put this is to say that if the atmospheric phase is winding systematically and significantly on a time scale of the order of 10 seconds, then averaging the data will degrade its quality irretrievably. You should inspect a piece of your observation first, and then average it if it is still desirable to do so. You could use UVPLT or IBLED for this inspection (see § 7).
As an example, consider a 6 cm observation with the 3 km array. If we demand no more than a 5% loss of intensity of a point source owing to time smearing at a distance of 6 arcminutes from the phase centre, then the integration time could be as long as almost 2 minutes (see Figure 2 in Appendix D). The acceptable averaging time would be longer for 20 cm, by 20/6.
UVAVG will not average across source or frequency changes. Note however, that it cannot read FG (flagging) tables (see § 7 below). Thus, if your data suffer from, say, intermittent but severe corruption (perhaps from interference), then averaging in time will corrupt what was previously uncorrupted data.
| UVAVG | |
| xinc =1 | Write all records |
| yinc =30 | Averaging time in seconds |
| opcode = ' ' | Average data |
Last update : 03/03/94
1