- The over-whole MT2 tone (and the bass variation)
- The MT2 pedal sections: analysis
- Analysis of the Boss MT2 Metal Zone pedal (2)
- How to model an opamp? (the implications on simulation)
- Should I invert my matrix or not?
- Analysis of the Boss MT2 Metal zone pedal (1)
- From netlist to code: strategies to implement schematics modelling results
- From netlist to code: strategies to implement schematics modelling
- Analog modelling: The Moog ladder filter emulation in Python
- Analog modelling: A prototype generic modeller in Python
- Comparing preamps
- Triode circuit
- SD1 vs TS9
In a previous post, I explained how I modeled the triode inverter circuit. I’ve decided to put it inside two different plugins, so I’d like to present in 4 pictures their differences.
The two plugins will start as the modeling of the Fender Bassman preamp (inverter circuit, followed by its associated tone stack) and the other will be the modeling of the inverter section of a Vox AC30 (followed by the tone stack of a JCM800). Compared to the default preamp of Audio ToolKit, the behaviors are quite different, all with just a few differences in the values of the components:
I will probably add the options of using a different triode model (Audio Toolkit has lots of options, and I definitely need to present the differences in terms of quality and performance), and perhaps also a way of selecting a different tone stack. But for now, the plugins will propose a single modeling of a triode inverter followed by a tone stack. To model a full amp, you still need to model the final stage and the loudspeaker.
The next picture displays the preamp behavior depending on the used triode function:
The Leach model and the original Koren model don’t behave as well the other models. It’s probably due to different parameters, but they give a good idea of the behavior of the tube. The modified Munro-Piazza is a personal modification of the tube function to make the derivative continuous as well. It helps the convergence when the state of the tube changes fast, even though it is clearly not enough to remove all discontinuities.
The following picture describes the cost with valgrind of the different models:
Obviously, the cost for the more complex functions, the time is spent trying to figure out the logarithm of a double number. This is because the fast math functions in ATK don’t support this function yet. If we use floating point numbers, then the cost is divided by 3 for the Koren model (for instance).
As the results are quite close, using floats or doubles is a matter of optimization and precision. In the plugins, I will use floats to maximize performance.
In the next two weeks, the two plugins will be released. And depending on feedback and comments, I’ll add more options.