[SystemSafety] Systematic and random error in systems

[Olwen Morgan]

All,

Picking up on something Martyn Thomas said a little way back;

In metrology, random error is that kind of error that causes repeated 
measurements of a quantity to be dispersed about the true value. A 
systematic error is the kind of error that causes repeated measurements 
of a quantity to be displaced from the true value. Both kinds of errors 
can coexist in a single measuring system.

Now, we know that the effects of software errors are systematic rather 
than random. Nevertheless, it is very easy to write, say as embedded 
code for a measuring instrument, software whose behaviour produces 
measurement errors that are indistinguishable by *tractable* experiment 
from random errors. This can be done by adding a pseudorandom quantity 
to the measured value before giving it as the output. If you choose the 
parameters of the pseudorandom number generator (PRANG) used, then you 
can make the output value of a measurement have virtually any form of 
distribution you like. Moreover, the only way to detect by experiment 
that the error is systematic rather than random is to run the PRANG 
through its whole cycle of values at least twice, so that the 
periodicity shows up.

I'm not suggesting that people do this but it is, at least to me, a 
striking example of how careful we have to be in adapting the principles 
of physical measurement to software used in measuring instruments.


Olwen

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.techfak.uni-bielefeld.de/mailman/private/systemsafety/attachments/20181104/c28a47c0/attachment.html>