[linux-audio-dev] [ann] unmatched - a LADSPA amp tone

[Tim Goetze]

Steve Harris wrote:

>On Thu, Oct 24, 2002 at 01:50:29 +0200, Tim Goetze wrote:
>> ah, now i dig it. maybe you could make a note about the intended
>> use in the valve doc?
>
>Yeah, I should probably do that. Documentation (well english in general!)
>is not my strong point.

yeah, i guess we'd all be writing poems instead of code if 
we did better in this field. ;)

>> good. do you think the 44100/12 oscillation is a speaker 
>> or output transformer effect?
>
>I'm 90% sure its from the crossover between the two halves of the class B
>amp (thats the power amp IIRC).

drat. i wish i was better at understanding electronics.
the amp guy pointed in this direction too.

>> afraid it still aliases a little, but it does look
>> like it's basically doing the right thing iiutc.
>
>I didn't think the aliasing would be audible. It uses a linear interp to
>go from the bulk of the sub lobe to the true zero corssing, but that should
>be OK. You do have smoothing set to 1.0 right? Otherwise it does something
>crazy (fix for that later).

the aliasing is not that strong, you're right. maybe i'm
over-sensitive.

>> the problem with both the valve and the xover i think
>> is that the further you drive them, the edgier the 
>> discontinuity gets. it would have to be computed at
>
>There definatly shouldn't be a discontinuity. If there is then its a bug.
>Also, the crossover effect is not amplitude dependent, maybe it should be,
>but the way it was described to me it isn't. Another thing to mention here
>is that many people claim the crossover effect isn't a desirable feature of
>amps.
>
>I think the valve needs to be applied gently (looking at the scope traces
>from yesterday its a fairly mild effect), and then some harder clipping
>needs to kick in.

what i gather from the scope shots and the analog distortion
guy's pages is that we want a range of clipping from none to
square. if we're going implement 'square' clipping -- which 
we'll need for the real rough tone -- we'll have to come up 
with a bandlimited solution. the current valve clipping is 
real soft compared to those scope shots, isn't it? if we
stack the valve to get to hard clipping, the net result will 
carry the sum of all aliasing i'm afraid.

>> my reasoning is that the problem we are facing here
>> is the same as in the generation of band-limited
>> square (or sawtooth, if you will) oscillation.
>
>Yes, but if the aliasing isn't too bad though it can probably just be
>solved with a bit of oversampling, 4x or 8x should be enough, though the
>CPU use wil go up 4x or 8x. The basic valve should be optimised.

my feeling is that oversampling alone will not do, although
you're probably right in that it will get rid of most of 
the aliasing. 

i've been working a little on a square oscillator recently
that integrates 32-sample windowed sincs centered at the 
wave discontinuities. the results are very promising, although
not perfect, because the straight parts of the wave are
simply that, and integrating the sinc actually is a crude
approximation. cpu usage is fairly low though.

i'm thinking about how to apply the technique to produce  
square clipping, or better a [0 .. 1] range of clipping, 
but the integration approach is awkward.

tim