Improvements to Circular Scrolling in Synaptics driver

[Andrew]

(My previous post failed. Trying again with cleaned up text and a link instead of attachment...)

Hello. I'm new to this mailing list, and open source participation in general, so I welcome any constructive criticism.

The proprietary Synaptics touchpad driver provides a scrolling gesture called ChiralMotion (TM). It is the most efficient and effortless scrolling input I have ever used on any device, and I really miss it in the open source driver. Unfortunately, ChiralMotion is patented, and the proprietary Linux driver is only available to OEMs.

The open source driver does provide a very limited gesture of similar intent, called Circular Scrolling, which measures the difference in angular position relative to the center of the touchpad. However, this requires the user to divert attention and use a fairly large radius to avoid getting off-center, otherwise it will produce erratic jumps and oscillations. This also restricts both the precision and maximum speed of scrolling, so the usability of this system is substantially impaired.

As such, I was motivated to design some improvements to Circular Scrolling, to achieve usability on par with ChiralMotion, without infringing the patent (US7446754). Its output reliably detects clockwise versus counter-clockwise motion, anywhere on the touchpad, but the algorithm makes no such distinction of "chiral" state. The basic mechanism is still a difference in angular position, but relative to the previous point. In effect, it measures the change in the angles of the last two segments. This method is more similar to US patent 20110025618, but that patent specifically states that it does not apply to the use of arctangent, but rather to a simplified look-up table suitable for low-cost embedded systems. There are other differences from both of these patents, which I'll gladly defend if necessary. In my implementation, the angular difference is further processed with an adaptive low-pass filter to reduce noise while preserving responsiveness. Its
 computational complexity is negligible and shouldn't have any significant impact on performance.

Users already accustomed to the current implementation of Circular Scrolling shouldn't be adversely affected by the change. The main differences they will notice are:

* enhanced dynamic range, providing finer control
 on the low end, and greater speed on the high end

* greatly improved reliability for tight rotations, 
 allowing more efficient and effortless scrolling

As stated in the man page:

 "Use tight circles near the center of the pad for fast scrolling 
 and large circles for better control."

This recommendation is still perfectly appropriate, except that the circling doesn't need to be centered to the pad, it can be anywhere. I tried to find an optimal balance between latency and noise in response to radius and speed, consistent with that description. This is the main difference from ChiralMotion, in which the magnitude of scrolling is a function of distance moved. I believe my implementation is actually more intuitive and reliable than ChiralMotion. I'm sure the filter parameters could use a bit of tuning, but it's pretty good already.

I prototyped this in Javascript, and you can download it as a stand-alone HTML file (tested in Chrome and Firefox) at:

http://github.com/galundin/xorg-synaptics-circscroll-improved/blob/master/circscroll-prototype.html

If that link gets scrubbed, it's in github/galundin/xorg-synaptics-circscroll-improved

Before I begin hacking the driver in earnest, I'd like some review and feedback on the prototype. Please note that this is throw-away code, not representative of the style I would use in proper implementation.

I would also appreciate any suggestions that might be offered regarding efficient workflow for driver development.

Thanks!

- Andrew Lundin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.x.org/archives/xorg-devel/attachments/20130610/1dbe65d1/attachment.html>