[sdiy] Digital filtering of pot input

[Andrew Simper]

I've updated the efficient algorithm in the paper to not use the full
trapezoidal matched shapes, so that lowers the op count down to 6+-,
3* and one absolute value.

On 6 December 2016 at 14:08, Andrew Simper <andy at cytomic.com> wrote:
> Whoops, I didn't realise that the link I posted was wrong and I never
> publicly corrected it (several people emailed in private), here is the
> correct link:
>
> http://cytomic.com/files/dsp/dynamic-smoothing.png
>
> The basic idea is to use a two pole multimode filter with a low cutoff, then
> use the bandpass output as a measure of how much the signal is changing and
> increase the cutoff. The absolute value of the bandpass output is used since
> it doesn't matter if the change is upwards or downwards, and this is also
> scaled by a sensitivity factor, added with the initial low cutoff gain
> factor and used to filter the input. It's a very straight forward algorithm
> once you know it, but it did require some careful thought to come up with in
> the first place. I'm very pleased with how well it works and how
> conceptually elegant it is :)
>
> Looking back at the code I see I switched from using and SVF to using two
> cascaded one pole low passes since it was more stable, so the updated
> algorithm operation count is 8+- and 5*. Here is a technical paper with
> details on the algorithm:
>
> http://cytomic.com/files/dsp/DynamicSmoothing.pdf
>
> Cheers,
>
> Andrew
>
>
> On 3 December 2016 at 21:41, Tom Wiltshire <tom at electricdruid.net> wrote:
>>
>> That looks very interesting, Andrew. Can you tell us more? How does the
>> "dynamic" part work? (Since we know what two-pole filters are already!)
>>
>> Thanks,
>> Tom
>>
>> On 3 Dec 2016, at 09:40, Andrew Simper <andy at cytomic.com> wrote:
>>
>> > If you have enough DSP to throw at multiple boxcar filters and an IIR
>> > you may be interested in a dynamic 2 pole low pass filtering approach
>> > I've designed which gives these kinds of results:
>> >
>> > www.cytomic.com/files/dsp/dynamic-smooting.png
>> >
>> >
>> > It takes 2 multiples, 6 additions, 3 bit shifts and an absolute value
>> > for fixed point, or 5 multiplies, 6 additions and an absolute value
>> > for float. Please let me know if this sort of algorithm may be
>> > appropriate.
>> >
>> > All the best,
>> >
>> > Andrew
>>
>