Chronology Current Month Current Thread Current Date
[Year List] [Month List (current year)] [Date Index] [Thread Index] [Thread Prev] [Thread Next] [Date Prev] [Date Next]

[Phys-L] Timing Statistic

Hi --

A couple of hours ago I pushed a new version of the spindown

The most interesting thing is a new diagram:
which is zoomed-in enough to show the stair-steps associated
with the excessive roundoff error. The actual noise in the
sensor is /at least/ one order of magnitude smaller than the
roundoff error. This is the hallmark of a badly-designed

Also there are some lagged-residual plots, which also show
the effect of roundoff. Perhaps the most interesting of
these is:

In addition to the 20-per-second data analyzed previously, I
carried out the same sort of analysis for the 100-per-second
data. The diagrams for the latter are in the spreadsheet file
only; I didn't save them as separate .png files.

More finely-spaced t values are a step in the right direction,
but 100 per second does not solve the fundamental problem. Not
even close. The important concept is, we don't necessarily need
more points; what we urgently need is two or three more orders
of magnitude more resolution on the timestamps on the points
we already have.

More resolution is synonymous with less roundoff error.