What makes SEM filter so special :)

[mystran]

For a rough approximation of a "realistic" OTA in the spirit of CA3080 or LM13700, you could probably try something like:

Iout = Iabc*tanh(Vin/(2*Vt)) - Is*(exp((Vout+VdropPos-Vpos)/Vt)-exp((Vneg+VdropNeg-Vout)/Vt))

where Vin is the differential input voltage, Vout is the voltage of the output, Vpos and Vneg are the positive and negative supply voltages and VdropPos and VdropNeg are constants approximating a few junction voltage drops from the mirrors (eg. something like VdropPos=3V and VdropNeg=1.5V should be fairly close to LM13700).

This is obviously a couple of layers of approximations piled on top of each other (in order to make it cheap), but the idea is to shortcut the "ideal" current through a pair of diodes to voltages slightly below supply if we try to push the voltage too close to the rails.

As a practical note, I actually used a pair of diodes in series with voltage sources to implement the second term when testing this in LTSpice (something that might dramatically improve your convergence in a custom solver too, at the cost of having to solve two extra nodes), since it didn't seem to handle the above term in a behaviour source directly (which usually runs into converge problems if you try to model ideal diodes, so not really surprised).

When compared against this LM13700 model in LTSpice, it actually seems surprisingly reasonable. The negative rail in general seems fine, where as the current drop on the positive rail is a bit more gradual in the "proper model" so making the positive rail clamp diode worse (eg. maybe a larger ideality factor would fix that) might improve it further.

Overall the value for you money should be fairly decent though, given that the whole thing still costs less than a pair of Ebers-Moll transistors. I don't know if it improves the SEM filter (which might be too well-behaved for this to really matter), but it should (hopefully) handle more "abusive" circuits in a fairly reasonable way.

[ckoe]

I am new in plugin devlopment, but I will ask you guys if there is an solution out now of how I can make a good sounding SEM Filter model?