[LAD] [ann] CAPS 0.4.0

Tim Goetze tim at quitte.de
Mon Jun 18 11:56:19 UTC 2007

[Giuseppe Zompatori]
> I have a couple of questions for you, namely:
> Could you please elaborate a little more on the new amp model? i.e:
> Is it based on real measured data as opposed to spice simulations?


David's page at CCRMA has his DAFX paper, I can only recommend you 
look into it, it's very well-written.  He comes from a circuits 
background so I assume he knows when to rely on simulation, and when 
not to.

> If it's based on a spice simulation did you implement a simple valve
> stage or a complete preamp made of multiple stages?
> The power amp section (valves & OT) doesn't seem to be there anyway
> AFAIK, but maybe you implemented it with a  transfer function of your
> real Fender amp that does from the input jack to the OT/speaker?

The original preamp was a two-stage model, as far as I remember.  
Unfortunately the netlist file got lost, so I have to work from memory 
here.  What I do remember is that one of the valve stages is always 
working in the linear window, regardless of gain settings, so it 
seemed safe to ignore its nonlinear contribution.

I did some comparisons between my Super 60 (input jack to speaker at 
low power amp gain) and the plugins back then, and the relative 
distribution of harmonics at increasing gain-induced distortion 
matched quite well.

However, I'll freely admit that the power amp and output transformer 
nonlinearity models in CAPS are very simple.  As far as I remember, 
these stages are fairly linear up to the point of clipping in real 
amps to start with.  This matches my experience with the Super 60.  
Turning up the power amp gain mainly produces more volume and 
compression, but not so much of a different tone.  And every time I 
finally had it turned up far enough to hear the effects of clipping in 
the power amp stage, the output transformer was quickly melting down 
(proving it a costly experiment).  

I think that the frequency response of the output transformer and 
speaker combo have a much more pronounced influence on the tone 
characteristics, and that a linear filtering operation (such as 
employed in CabinetII or an FFT-based convolver) can represent a very 
accurate model.

> I am browsing the source code right now and can't easily figure out
> which 12ax7 model you're finally happy with... there are multiple ones
> in TwelveAX7.h

The differences don't amount to much.  The linearly interpolating 
model is operating directly on data obtained from a spice simulation, 
so I think it's fair to say this is the most realistic.  The clipping 
characteristics (and its asymmetry) have a far more pronounced impact 
on the sound characteristics at high gain, and are about the same for 
all models.

> I am asking this as I'd like to experiment with it a little. I still
> think there are issues that prevent obtaining more "M*rshally sounds"
> or high gain sounds in general.

That may be true since neither has been the main focus of my work.  
To my ears, extreme high-gain scenarios sound quite credible with the 
current CAPS Amps, but then again I do not go there that often.

If you can come up with modifications that come closer to your goals, 
I'll be happy to include them in CAPS if you'd like me to.  After all, 
it's open source and meant for you to freely build on, or even 

Thanks, Tim

More information about the Linux-audio-dev mailing list