Steve Harris wrote: >On Tue, Oct 22, 2002 at 09:38:26 +0200, Tim Goetze wrote: >> the problem with converging IIR is that either the IIR or the >> converging algorithm or both don't like impulses that are too >> lively. the converger will simply fall dead when it sees an >> exponentially decaying white noise impulse. it makes sense, > >I don't understand why. Is it just beacuse the kernle is a bit too long to >be stable, or is there someting more to it than that?
yes, the filter quickly becomes unstable during convergence. i think it's simply against the nature of the iir, which i take to prefer smooth oscillation. even if there existed some 'magical' set of coefficients that could emulate the noise impulse, finding it (i can only argument on the basis of my feeling for the algorithms involved) will require a different method than the one employed (foreach coefficient a small step, then combine what looked reasonable). >> what i don't get about the valve (the rect influence seems >> negligible) is that it only shapes the negative portion of a >> sine. without a LP, there's faint aliasing. it doesn't sound >> so bad though, in fact i like it. > >Thats because its only odd harmonics (or is it even?). It should be >follwed by a DC offset remover. This is pretty normal for a tube >simulation. The aliasing is because I never got round to oversampling the >valve. It needs some more optimisation to make that a good idea, and LP >filters are so much easier ;) yep. :) they should build hw lagrange- or sinc-interpolation into those fantastic 'multimedia' cpus. would save us a lot of work. still i'm not convinced that a sine should only be compressed if < 0 by valve_1209.so. >> however i'm into harsh distortion now, and have fed the fender >> three sines from zero to full gain (three octaves of 'c'). >> the shaping it does has some important properties that i'm only >> beginning to understand, but it really does look fundamentally >> different. it seems to self-oscillate at the zero-crossings, >> and otherwise compresses the sine much like your valve does, >> only the output signal is not a straight line where compressed, >> but rather another irregular but smooth oscillation about a DC >> value. a (realtime) electric circuit simulator would come in > >Hmmm... I'm quite out of my depth here, but could this be a class B effect >I'm not correctly modelling? I suspect a class B crossover discontinuity, >effected by the mostion and mass of the speaker cones (pretty high for a >guitar amp I'd guess) could look like that. Any chance of a short sample, >to make sure I'm thinking of the right shape. i've copied (950 + 50 silent) samples of each into http://quitte.de/fender-sine.wave -- the quitte.de quota is measured in k, not M :(. the two oscillations mentioned above seem to be one actually, their period of about 12 samples is independent from the sine frequency that is fed to the amp, i double-checked with a 'f#' sine (that got lost accidentally, it does not look substantially different however). actually i should record the output from the circuit not the speaker. >Infact, this should just happen, if you use the crossover plugin and >put the output through your speaker cone IIR you should see a similar >shape, as the LP characteristic of the IIR damps the abrupt 0 sticking >point from the crossover. yup, it definitely looks more like it on the scope. but the aliasing the crossover introduces is simply unbearable. i think i'll go for a ride and show some waves to the guy who fixed my fender amp recently, he knows a lot about the circuitry. tim ps: i've updated the 'unmatched.html' page to show graphs of both the original and the plugin response.
