mtdev smoothing/denoising

[Rafi]

When I try to write with my finger in xournal I get jagged lines, and when I hold a window, it bounces around a bit even when I hol my finger in place.

It would be so nice if something would filter out that jitter and smooth lines for me. Since both gesture events and normal xinput would benefit, for single and multi-touch and requires tracking (for multi-touch), it seems to me mtdev is a logical location to place to put this sort of processing.

btw, the script known as "ts_test" run without any arguments should demonstrate/measure the sorts of jitter and messy lines that are annoying me (at least for single touch).

[Henrik Rydberg (rydberg)]

Mtdev already contains a EWMA filter, with defaults sensible for devices with moderate signal-to-noise ratio. For this particular device, which is very noisy, it should suffice to add proper fuzz parameters in the kernel driver, and mtdev will pick them up.

[Rafi (rafi-seas)]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 09/03/10 12:51, Henrik Rydberg wrote:
> Mtdev already contains a EWMA filter, with defaults sensible for devices
> with moderate signal-to-noise ratio. For this particular device, which
> is very noisy, it should suffice to add proper fuzz parameters in the
> kernel driver, and mtdev will pick them up.
>
>
> ** Changed in: mtdev
> Status: New => Incomplete
>

I've used the fuzz parameters in unpushed versions of hid-ntrig. My
understanding was that it only works for the conventional events and the slotted
protocol. Are you saying you just read the fuzz from each device and then
provide the same filters in userspace/mtdev for mt events?

Ah, emwa== fir/low pass, makes sense. Does mtdev smooth single
touch/conventional events too, or only clean up the mt events?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAkyBL7oACgkQwuRiAT9o60/1WACfYtAecVHc9+pI/9hMlZUmw4j+
6qoAoL3LjRP7PHdIaEdUEOHdVZnkoKbH
=anqe
-----END PGP SIGNATURE-----

[Chase Douglas (chasedouglas)]

Rafi,

Yes, even though the fuzz parameters aren't used by the kernel for MT protocol A drivers, mtdev reads the fuzz params from the device and uses them if they are non-zero.

[Duncan McGreggor (oubiwann)]

I've assigned this to Rafi since the recommendation was to put proper fuzz parameters in the kernel.

[Henrik Rydberg (rydberg)]

Solved in the maverick kernel, so this is now an upstream issue only.