On Friday 07 November 2014 01:30 PM, Richard Cochran wrote:

On Fri, Nov 07, 2014 at 11:04:05AM +0530, Vignesh R wrote:

quoted

Currently, there is too much noise in the TSC hardware that is being
removed by delta filtering.

The so called "filter" was only programmed because the fifo entries
were being mixed up. Sebastian fixed that.

quoted

I tested TSC unit by removing filtering
logic, the performance was not at all satisfactory. The cursor jumps
wayward and smooth circles cannot be drawn. Looks like delta filtering
cannot be removed as of now. May be I will try and address it in future.

The "filter" code is nonsensical. It picks the two values in seqeunce
that are closest to one and another. How is that supposed to work?

Did you look at the "noise"? What kind of properties did you see?

A median filter makes more sense. Or sort, remove outliers, and
average. But choosing the two closest in series is silly.

I was able to implement median filter as you described and achieve
reliable performance. I will append that to this series of patches in v3.

Regards
Vignesh

Thanks,
Richard