Hi Bart, it seems that you are not convinced, let me try again.
On 04/16, Oleg Nesterov wrote: > > On 04/16, Bart Brouns wrote: > > > > Somehow the initial value of `u` got lost in the last commit to this > > function: > > https://github.com/grame-cncm/faustlibraries/commit/ba34ca657a8c98efc804061c5b2e1d4f7e6bc6ff. > > I don't think it was lost, this code deliberately tries to not use > "impulse" for initialization of delay line. > > See https://sourceforge.net/p/faudiostream/mailman/message/36700742/ > > (See also https://sourceforge.net/p/faudiostream/mailman/message/36534365/) lets look at the current code: quadosc(f) = tick ~ (_,_) : mem+1,mem with { k1 = tan(f * ma.PI/ma.SR); k2 = 2*k1 / (1 + k1 * k1); tick(u_0,v_0) = u_1,v_1 with { tmp = u_0 - k1 * v_0; v_1 = v_0 + k2 * (tmp + 1); u_1 = tmp - k1 * v_1; }; }; it is easy to show that mathematically (iow, ignoring the floating point issues) "tick" obeys to the following equation: [ u_1 + 1 ] [ cos(th) -sin(th) ] [ u_0 + 1 ] [ ] = [ ] . [ ] [ v_1 ] [ sin(th) cos(th) ] [ v_0 ] where th = th = f * 2 * ma.PI/ma.SR; So this is just a complex quadrature oscillator, but the " + 1" trick allows to avoid the initialization. > > Is that what you mean by "more precise"? > > Not at all, Lets look at the code after the patch: quadosc(f) = tick ~ (_,_) : (!,_*0.5) with { k1 = tan(f * ma.PI/ma.SR); k2 = 2*k1 / (1 + k1 * k1); tick(u_0,v_0) = u_1,v_1 with { tmp = u_0 - k1 * v_0; v_1 = v_0 + k2 * (tmp + 1); u_1 = (tmp - k1 * v_1)+impulse; }; }; This code obeys to the same equation above except at the very 1st step, and this "+ impulse" is just wrong. After your patch quadosc(f) no longer outputs "sine", it is equivalent (again, mathematically) to the following implementation: quadosc(f) = sin(n * th) + sin((n+1) * th) : /(2) with { th = 2 * f * ma.PI/ma.SR; n = ba.time; }; Not to mention it removes the 1st (cosine) output. To me this makes no sense. Please revert that commit? Oleg. _______________________________________________ Faudiostream-users mailing list Faudiostream-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/faudiostream-users
