Originally Posted by glanglois
I'm not sure I got this 
That's a great explanation of echo cancellation in a telecommunications system or perhaps of a digital frequency filter. But .....
In my hands, at any rate, the tremor
is quite irregular and therefore not necessarily predictable on a millisecond basis. My hands go back and forth but not evenly like a pendulum.
[...]
Pentax may still be using the Invensense IDG-300 for this purpose.
Ok, I gave a simplistic description (which is only correct for periodic shake) to say something about the "loose the brakes" idea for
SR ...
A more appropriate description would have been to compare the situation to that of noise cancellation where the phase is actually inverted (note that noise is irregular, too). Works only above a given frequency. Below this frequency (i.e., for frequencies of shake where only a few or less than one oscillation occur) you would counter-move in real time w/o waiting.
But this is no problem because at those low frequencies, the latency of the entire
SR system can be ignored. The tricky part is how to compensate for those higher frequencies of shake ... There is a chance that Pentax doesn't treat those higher frequencies at all (Human tremor is at frequencies between 4 Hz and 15 Hz). If so, then you are perfectly right and the system doesn't "anticipate" anything (no noise cancellation at all). Then however, the system would perform poorly for short shutter time at long focal lengths. Maybe, it does...
Pentax never published their exact algorithm but it was noted that it works best for exposure times of 1/15s. They don't use the IDG-300; but two similiar (smaller) chips, one per axis.
