At the start of a raster the on board data store may be empty and so the data from the first few exposures can be stored there while at the same time data trickles out into the telemetry at the rate of ~1.5 Kb/sec.
If, however, the size of the data extraction windows is large, the store will quickly become full and the processing of data on board must wait until sufficient space is released by the telemetry trickle to allow new data in. This has the effect of postponing the start of new exposures in the raster and leads to a sudden increase in the inter-exposure time interval.
This is illustrated below in a plot produced by:
The sold circles represent the actual timings achieved during a raster in flight, while the line and open circles are the prediction from the routine raster_dur as used in mk_raster. Notice the sudden increase after 9 exposures once the on board buffer becomes full and everything slows down to the pace of the telemetry. The noise in the real data is caused by the interaction of many asynchronous tasks in the on board software.
Note, however, that if rasters of this kind are run consecutively, the second and subsequent rasters will start when the buffer is still almost full and so their inter-exposure intervals will start and remain at the higher value.
Even for rasters than include no mechanism movements (ie the sit and stare movie type study), owing to on board process interaction, there is still likely to be an rms of ~0.5 secs on the inter-exposure interval, as illustrated below.