On Fri, Dec 26, 2008 at 8:08 PM, Werner Almesberger <[email protected]> wrote: > Nelson Castillo wrote: >> That is because when we get an IRQ we gather as many samples >> as the filters need in order to produce a point, and only then we check >> if the touchscreen is still pressed. > > Hmm, but is this a bug or a feature ? :-) Samples you get when the > touch screen is no longer pressed don't seem to be particularly > useful, or are they ? What happens in this case ? Do you discard > the event ?
A feature I would say :-) This is even better than what we have in the stable kernel because we discard noise. Sometimes the last few events get discarded because they do not provide reliable data, for instance because not much pressure is applied. I tried to use a heuristic: discarding the first N and the last M events. And still that did not work. We could not guess the values of M and N because we don't know in advance much much noise we are going to get, it depends of how much pressure is applied, how fast the pen is moving, etc. The "group" filter discards a lot of noise. Sometimes it discards a lot of samples (say 50 adc conversions) in a click. This might sound crazy, but what is crazy is the data that we are discarding... Right now it seems we are not experiencing bogus clicks and this was an issue before. I had a few samples of what the data looks like but I made a mistake, so I removed the data. I should measure and upload the data again to illustrate this point. The hardware is reporting a lot of nonsense that we manage to discard. I'll do it... Group filter => discard noise (sometimes A LOT of noise that doesn't reach user-space). Median + mean => discard some noise, smooth samples _______________________________________________ devel mailing list [email protected] https://lists.openmoko.org/mailman/listinfo/devel
