Re: [Algorithms] Truncate then average? Or average then truncate?

[Robin G. <rob...@gm...>]

On Mon, Jun 7, 2010 at 9:12 PM, Jon Watte <jw...@gm...> wrote:
> I don't get the AND thing -- you can't get instant key response that way,
> because you need to wait for all the words to have the bit set before it
> registers as a "key down." Is there more state there than the description?

The idea comes from Jack Ganssle's article on Software Debouncers:

   http://www.ganssle.com/debouncing-pt2.htm


> When it comes to the ADC, there are a few things I would think about, FWIW:
> 1) Put a capacitor and resistor on the analog input before the ADC. I don't
> care what they say; it's cheap, and very helpful!

An R/C Network on the input, isn't that just another form of
hysteresis? And one that's subject to the astonishing inaccuracies of
electrolytic caps?


> 2) You definitely want to sum first, and then divide. In fact, if you want a
> value from 0 .. 127, and have 10 bit numbers (0 .. 1023) coming in, I would
> sample four times at evenly spread intervals into a ring buffer, and
> calculate the output values as the sum of the last four samples, shifted
> right five bits. (your range will be 0 .. 4092, which divided by 32 ends up
> as 0 .. 127)

This was exactly the form that caused the problem around values with a
lot of bits set. You're trying to discard the random bottom bits, but
you include their influence by summing before removing them by
truncation.

The solution I ended up using records the previous ADC value in it's
full 10-bit value. A new ADC value has to exceed this old sample by
more than the Hysteresis amount before I allow it through to
truncation. The insight was to keep the comparison values at full
resolution, discard values below the H threshold and truncate the
samples only just before they are used. Values are now rock solid.

The problem was that the premature truncation was discarding important
information.

> 3) You probably also want to implement some dead space at both ends in the
> hardware. This means, if the summed range is 0 .. 4092, you might actually
> want to treat 0 .. 100 as 0, and 3093 .. 4092 as 127, and spread the rest of
> the interval between the two values, assuming you have efficient
> multiplication or division. If this is for a thumb stick (game controller
> style) you also want a dead space in the center, although that could
> potentially be applied in the software driver rather than firmware.

Linear POTs seem to have dead space at either end built in, but the
dead center is a good idea - but should it be circular or square?

- Robin Green.