junction noise (really), cont'd.

[Ian Fritz]

Next I looked at the pink noise filter. The original design consists of a
simple filter, implemented by the following three impedances in parallel
around an opamp:

(1) R = 330k
(2) R = 47k + C = .01u
(3) C = .001u

The overall response of this filter is 3dB/oct, but it has about a +/-3 dB
ripple about this slope. The resultant curve is peaked at 40 Hz and 4 kHz.
As a consequence, the response in the 70 Hz to 400 Hz region is quite steep,
whereas the response from 500 Hz to 2 kHz is rather flattish.

Having a decent math package on hand, I decided to see how much better could
be done by just adding another R+C combination in parallel with the three
above. By adjusting the R and C values and looking at computer simulation
results, I was quite quickly able to come up with a much better filter. The
new network is as follows:

(1) R = 330k
(2) R = 110k + C = .015u
(3) R = 24k + C = .0033u
(4) C = .001u

This design has about a +/- 1 dB ripple from10 Hz to 20 kHz. I did an AB
comparison of the two filters, and there is a big difference audibly. Of
course, you may well prefer the sound of the original filter. But if you
want pink, the above simple modification does a good job.

  Ian