ASTERIX CONVENTIONS

Below are listed a number of conventions relating to ASTERIX data, whose consistent application throughout the system is important, so that users know what behaviour to expect, and so that incompatibility between the practices of different pieces of software is avoided.

• Location of a bin - a bin will be located by its centre value. Hence, for example when a subset of a binned dataset is selected it is the bin centre (not the boundaries) which will be tested against the specified range to decide whether a given bin qualifies for inclusion. The same applies for event data, where `primitive bins' (of width specified by QUANTUM) may already be implicitly present. Minimum and maximum values, as recorded in FIELD_MIN/MAX also refer to (primitive) bin centre values, not to boundaries.

• Behaviour on boundaries - in binning and selection processes the ASTERIX convention is that lower bounds are inclusive and upper bounds exclusive. This rule leads to rather `unnatural' behaviour in some circumstances (e.g. if one specifies a range of axis values 1 to 10 in a case where the values are integral, then one gets only the values 1 2 ... 9), however it has the great merit that one can select out subsets based on contiguous ranges (e.g. 0-10, 10-20, 20-30 etc.) without any risk of the same data being included in more than one subset. If there is a good reason for breaking the rule of the exclusive upper bound in some particular instance then the user should be explicitly warned.

  One result of the `upper exclusive' rule is that where software offers a default to the user corresponding to the full data range, it must (if the default is accepted) raise its upper bound slightly (AFTER the selection) in order to ensure that the uppermost point is not left out.

• Regular spacing - regular axis values should always be represented by a <SPACED_ARRAY> so that software can recognise its regular nature.

• Some axes (e.g. RA) are conventionally stored in reversed form. In such cases data ranges (for example, when subsetting) should be given in increasing index (i.e. decreasing value) order, for example 270:210. The first value will be inclusive and the second exclusive.

• Some applications may require a particular order of axes in a binned dataset. In this case the application should warn the user of this fact. AXFLIP and AXSWAP utilities are provided to allow axis order to be easily changed.

• Normalisation - the NORMALISED component in the AXIS structure in principle allows software to handle binned data in both normalised and unnormalised forms. However, it will be less confusing for users if a consistent practice is adopted with regard to normalisation. Normal practice is for binned data to be normalised with respect to time, but not with respect to the spectral dimension. In the spatial domain it is common to encounter both normalised and unnormalised data (e.g. counts per square arcmin and counts per pixel) so software will probably have to cope with both cases, however the `normal' case will be taken as unnormalised.

             Axis                  Typically
          --------------------------------------
              x                   unnormalised
              y                   unnormalised
             time                  normalised
            energy                unnormalised
  

• Default units - the following units should be assumed by default by ASTERIX applications:

             Dimension      Default unit       Comment
           ----------------------------------------------------------
             position        degree       relative to AXIS_RA/DEC/PA
             time            second       relative to BASE_TAI
             energy          keV