Bugzilla – Bug 1026
Nyquist eq-lowshelf fails for low frequency high negative gain
Last modified: 2018-08-20 11:51:29 UTC
Initially marked as P2 because it causes a serious regression in a "flagship feature" of 2.1.0. If the frequency for a Nyquist low shelf filter is below about 3000 Hz and the gain less than -12 dB, there is an error in nyq:biquad (in dspprims.lsp): "a0 < 1 (unstable parameter) in biquad" This error will occur in Spectral Edit Shelves or any other Nyquist effect that uses a low-shelf filter with a half gain frequency below 3 kHz and more than 12 dB cut. I think that the sanity check in nyq:biquad is overly restrictive as instability does not appear in eq-lowshelf until a0 < 0.5024 Making the test less restrictive fixes this bug, but perhaps Roger has a better solution. (defun nyq:biquad (x b0 b1 b2 a0 a1 a2) (if (< a0 0.5024) (error (format t "a0 too small (unstable parameter) in biquad: ~a~%" a0))) (let ((a0r (/ 1.0 a0))) (snd-biquad x (* a0r b0) (* a0r b1) (* a0r b2) (* a0r a1) (* a0r a2) 0 0)))
Regression on 2.1.0. I can reproduce the problem fine, but am not clear why Low Frequency undefined, High Frequency 1000 Hz, Gain > +12 dB works, which I understand is a low shelving filter only, with half-gain frequency of 1000 Hz. Low Frequency 0 Hz, High Frequency 1000 Hz, Gain > +12 dB fails, which I understand is a low shelving filter with half-gain frequency of "minimally above 0 Hz", plus a high shelving filter with a half-gain frequency of 1000 Hz. Or am I not interpreting 0 Hz and undefined correctly? Is it a legitimate workaround for this issue if you want e.g. +14 dB gain to apply two passes at +7 dB gain? If not I would be tempted to say P1. Are you (Steve) happy with P2?
(In reply to Gale Andrews from comment #1) > I can reproduce the problem fine, but am not clear why Low Frequency undefined, > High Frequency 1000 Hz, Gain > +12 dB works, which I understand is a low > shelving filter only, with half-gain frequency of 1000 Hz. To reproduce the issue with a ('single') low shelf filter (low frequency undefined), set Gain to -18 dB ('minus' 18 dB).
(In reply to Gale Andrews from comment #1) > Is it a legitimate workaround for this issue if you want e.g. +14 dB gain to > apply two passes at +7 dB gain? I'd say that is a legitimate work-around, though the result will be 'slightly' different. > Are you (Steve) happy with P2? I think P2 is appropriate. Even without the "Spectral edit shelves" I'd be tempted to go for P3 because it will be a serious regression in an unknown number of (non-shipped) Nyquist plug-ins (those that use eq-lowshelf) and by inference is a regression in the (shipped) Nyquist Prompt effect.
Fixed committed (with agreement of RM) 07ca677.
Tested on Windows and Mac.
