C[omp]ute

From: andrew cooke <andrew@...>

Date: Thu, 2 Oct 2025 18:13:39 -0300

[Maybe this is already known, but I failed to turn up anything with
Google and am pretty pleased with what I dreamt on my own.]

I'm writing software to generate audio signals with an 8 bit DAC and a
relatively weak CPU.  I wanted to add a compressor to the audio chain,
but my initial attempts used slow floating point arithmentic.  Then I
came up with this:

    def compress(n, depth):
      for bit = 7-depth to 7:
        thresh = 1 << bit
	if n > thresh: n = thresh + ((n - thresh) >> 1)
      return n

Basically: "divide by 2 above some knee" plus "iterate that for
increasingly large knees".  Where depth above selects the degree of
compression required.

In my experience it works pretty damn well.

Some plots of the response curve:

  https://acooke.org/img/compress-1.png
  https://acooke.org/img/compress-2.png
  https://acooke.org/img/compress-3.png

Andrew