
Date: Fri, 27 Apr 2001 00:55:11 +0200
From: Stefan Larsson <stefan@astro.su.se>
MIME-Version: 1.0
To: Carol Anne Oxborrow <oxborrow@dsri.dk>
CC: allan@dsri.dk, njw@dsri.dk, peter.kretschmar@obs.unige.ch, stefan@astro.su.se, juhani.huovelin@astro.helsinki.fi, smaisala@astro.helsinki.fi
Subject: AI18.11 REST data binning
Content-Transfer-Encoding: 7bit

At our last meeting I got an AI on me to circulate a question
regarding the time values for the restricted mode data.
(AI 18.11, first among SDAST)

Here is my summary

-----------------------------------------------------------
RESTRICTED data property:

The restricted mode will produce raw data with 8 energy bins
and with time bins which are up to 32 seconds long. The
length of the time bins are not constant but can vary from bin
to bin.

Presently:

To simplify data handling after preprocessing, the restricted
events are stored in the same way as full events, BUT all
events from one restricted image are given the same time value.

Problem:

To correct for grey filter and deadtime the precis time 
intervals should be known. Note that a grey filter change
can occur inside a restricted image bin.

Forcing a rebinning into equally long time bins will
cause a smearing.

Possible solutions:

SL: Keep information about the restricted time bins. Then bin
these events in the actual recorded time bins.
+ Will keep the "true" bins (no smearing and better defined errors). 
+ More accurate grey and deadtime correction.
- A change in present data structures (and programs).

PK: REST events could get randomized times within their
respective integration intervals and then treated just
like all other events.
+ Little change in present software. No change in DS's.
- Does it easier to handle the corrections but does not
  remove the problem (I think).

An alternative to Peters approach is to distribute the
events evenly over the integration interval (faster). 
The random distribution introduces a little more noise 
when resampled, but its frequency distribution is more 
well behaved.


Question to users:

Should light curve bins be equal to the raw time
time bins with non-constant length (less or eq. 32 sec)
or should data be rebinned into equally long time bins?



Let your thought flow!

Stefan
