[sdiy] An Improved Sine Shaper Circuit

[Donald Tillman]

> On Apr 17, 2020, at 1:56 AM, René Schmitz <synth at schmitzbits.de> wrote:
> 
> Interesting circuit, and a great article.
> 
> I'm pretty sure I have seen a similar technique before, because I have used it. (cusp canceling)

I am very familiar with cusp cancellation.  I've used it also.  And it's mentioned in the article.

This is not cusp cancellation.  While the circuit looks the same, I'm subtracting 1 or 2 orders of magnitude more of the original triangle signal than is necessary to cancel the triangle cusps.

I discovered this using some machine learning tools (Jupyter Notebook, Numpy) and some of my own software to optimize a classic sine shaper circuit... one with emitter resistors and with cusp cancellation.   And it kept pointing me to subtract more and more of the triangle signal.  I thought something had gone wrong, but what the heck, follow the data.  And sure enough, the harmonic spectrum really did get better as the transistor's tanh curve performed a different function.

Then I realized I was no longer correcting for a little nipple that got through the sine shaper, I had discovered a new way to approximate the sine function.   And that it was crazy accurate.  I couldn't find a reference to this zig-zagging "tanh(x) - beta x" curve mentioned anywhere before.

(I'll add some more about this in the article soon.)

  -- Don
--
Donald Tillman, Palo Alto, California
http://www.till.com