Improvement to DM's Generic Triode Model for 12AU7

[Joel]

By adjusting parameter values of Duncan Munro's "generic triode model" for the 12AU7, I was able to get it pretty close to the published plate curves. The base spice circuit before changes was WAY off. Like, not even close.
The AC performance seems pretty close too, but it's showing only about half the distortion than the Sylvania tube manual. No idea why.
Has anybody else done better? Or can you suggest which parameter to change to get the high current zone closer? I'm basically blindly changing things right now.

Thanks!
Joel


***********************************
12AU7 Test Circuit

.dc Vb 0 500 1 Vc 0 30 2

Vb 2 0 250v
Vk 3 0 0
Vc 0 1 8

x1 2 1 3 12au7

.subckt 12au7 a g k
* ANODE MODEL
BLIM LI 0 V=(URAMP(V(A)-V(K))^1)*0.0037
BGG GG 0 V=V(G)-V(K)-0
BRP1 RP1 0 V=URAMP(-V(GG)*0.024778659)
BRP2 RP2 0 V=V(RP1)-URAMP(V(RP1)-0.999)
BRPF RP 0 V=(1-V(RP2)^.68)+URAMP(V(GG))*0.1
BGR GR 0 V=URAMP(V(GG))-URAMP(-(V(GG)*(1+V(GG)*0.007)))
BEM EM 0 V=URAMP(V(A)-V(K)+V(GR)*16)
BEP EP 0 V=(V(EM)^1.35)*V(RP)*0.0000236
BEL1 EL1 0 V=URAMP(V(EP))
BEL EL 0 V=V(EL1)-URAMP(V(EL1)-V(LI))
BLD LD 0 V=URAMP(V(EP)-V(LI))
BAK A K I=V(EL)
* GRID MODEL
BGF GF 0 V=(URAMP(V(G)-V(K)-0)^1.5)*0.01
BG G K I=V(GF)+V(LD)
* CAPS
Cak a k 0.4pF
Cgk g k 1.6pF
Cga g a 1.5pF
.ends 12au7

.control
.endc
.probe
.end

[gorgon53]

why bother with a crappy tube like this?

[Robert McLean]

Joel
A fairly long reply here, you pushed one of my buttons.

The curves look fairly good now, they are inaccurate at high currents, but notice that is an area above the max power curve
so you dont run load lines or operate the tube there much anyway. They are also inaccurate at high negative grid voltages,
but again a typical load line may not pass through much of that area. So in practical terms it may not matter, depending on
what you want to do with the model.

But I agree, you want to get the model as accurate as possible, so here are 4 suggestions.

1)
If you havent already done so, go to the DuncanAmps site and download the pdf file of Generic Triode Model Documentation.

This will give you a better idea of what each parameter does, and you can change them a little less "blindly". I dont set
parameters manually anymore, but when I did I would start by setting Mu0, Erp and EMC. And CDO if the 0 volt grid does not go
through 0,0. Then I would leave those alone, and play with RAf, RAS, Rap, Eri, and Mur. The high negative grid voltage
curves are affected by RAf and Ras.

2)
There is only so much you can do setting the parameters by hand. It is pretty subjective. Sometimes it is hard to say which
of two graphs is better, one looks good in one area, the other looks better in another area. And you are never really sure
when to stop fiddling around with it. So I set parameters using a spreadsheet set up to do a regression or curvefitting of
the model equations and parameters onto the measured data. And mathematically compute the accuracy of fit, so you can decide
which of several possible models you like best. Regression or curvefitting is a fairly broad subject in itself, perhaps you
already know that.

3)
for better accuracy you may want to use Korens or Rydels models. Duncan Munro's models are good, and they can be adjusted manually more easily than Korens or Rydels, but for any triode I have ever modeled I have usually got the best fit with one of Korens, sometimes one of Rydels, but Duncans has never been the best. ( It is also far from being the worst ). Search
this forum for "Curvecaptor" and download the program. Somebody on this group wrote the program to do datacapture from spec sheets and then do the mathematical curve fitting to come up with the parameters for Korens and Rydels models. And some others if I remember correctly. You may want to compare the fits to what you get with Duncans Generic Triode.
Or make your own do-it-yourself spread sheets as I do.

4)
No matter which model you use, or what method of deriving the parameters, it is always better if at all possible to work from
data measured from a real tube. I know this is not always possible. The graphs in spec sheets are hand drawn lines joining points with smooth curves. When you pick a pont on one of those lines you never know if you are getting the real data or an interpolated point. And is it all from one tube or is it averaged out ? It was plenty accurate enough for its original intention, but slight discrepancies in the data can really through off the model. I have always found it much easier to get a good fit to data I measure myself compared to data taken from a spec sheet. My suspicion is that the model has trouble fitting the spec sheet data because the spec sheet data doesnt fit, or come from, an actual tube. It comes from an averaged out set of data from many tubes. but in any event real measured data seems to make for a better model.

[Gordy]

Edited. Ooops!