Spice models with polynomials
eppidei
I tested your model in a ciruit I use to test models by sweeping plate voltage and stepping grid voltage and comparing the results to the data. I swept plate voltage from 0 to 250 Volts and stepped grid voltage from +1 to -5 volts in .5 volt steps.
I also plotted a max power curve of 1.2 Watts.
Here is the result with your active model :
C:\downloads\12AX7A_dream_active.jpg
and here is the result using a Koren type model :
C:\downloads\12AX7_Koren.jpg
The problem with yours is that it is no good at all over about 4 or 5 milliamps. I suspect that this is because you have used polynomials fit to curves, and those curves only went to 4 or 5 milliamps. The problem with curve fitting using polynomials is that the result is only accurate over the range of data you have used, ie you can interpolate with polynomials, but you cannot extrapolate with polynomials. The polynomial just goes off wherevever it wants to outside the range of your data.
I dont like polynomials at all for spice models, but you may use them if you wish.
As the 1.2W power curve shows, 12AX7 tubes will frequently be biased in areas where your model is completely innaccurate.
So, if you are going to use polynomials then you must base them on a much wider range of data. I dont have any good 12AX7 data right now, but perhaps sometime in the next couple of weeks I will be able to take some measurements, and I will post the results
eppidei
I tested your model in a ciruit I use to test models by sweeping plate voltage and stepping grid voltage and comparing the results to the data. I swept plate voltage from 0 to 250 Volts and stepped grid voltage from +1 to -5 volts in .5 volt steps.
I also plotted a max power curve of 1.2 Watts.
Here is the result with your active model :
C:\downloads\12AX7A_dream_active.jpg
and here is the result using a Koren type model :
C:\downloads\12AX7_Koren.jpg
The problem with yours is that it is no good at all over about 4 or 5 milliamps. I suspect that this is because you have used polynomials fit to curves, and those curves only went to 4 or 5 milliamps. The problem with curve fitting using polynomials is that the result is only accurate over the range of data you have used, ie you can interpolate with polynomials, but you cannot extrapolate with polynomials. The polynomial just goes off wherevever it wants to outside the range of your data.
I dont like polynomials at all for spice models, but you may use them if you wish.
As the 1.2W power curve shows, 12AX7 tubes will frequently be biased in areas where your model is completely innaccurate.
So, if you are going to use polynomials then you must base them on a much wider range of data. I dont have any good 12AX7 data right now, but perhaps sometime in the next couple of weeks I will be able to take some measurements, and I will post the results
Hi Robert,
thanks for ur comment.I don't use pure polynomial interpolation but u're right the model fits only the RCA curves i Knew.let me know If u have other data .
However i'm preparing new semi-analytical models that will not suffer from this problem.
What do u think about the grid current?
thanks for ur comment.I don't use pure polynomial interpolation but u're right the model fits only the RCA curves i Knew.let me know If u have other data .
However i'm preparing new semi-analytical models that will not suffer from this problem.
What do u think about the grid current?
suggested changes
eppidei :
I havent had time yet to chart some real tube data, but perhaps Saturday or Sunday I will do this.
When I first looked at your model I did not understand what all the resistors and 0 volt sources were for, but then I figured it out why you were using them. You dont have to do it that way in SPICE, instead of using current sources (G) use voltage sources (E), and give names to the nodes which can then be referenced in functions later on, rather than the current through the 0 volt sources. I have attached a file showing what I mean, you will see radically reduces the size of the model.
eppidei :
I havent had time yet to chart some real tube data, but perhaps Saturday or Sunday I will do this.
When I first looked at your model I did not understand what all the resistors and 0 volt sources were for, but then I figured it out why you were using them. You dont have to do it that way in SPICE, instead of using current sources (G) use voltage sources (E), and give names to the nodes which can then be referenced in functions later on, rather than the current through the 0 volt sources. I have attached a file showing what I mean, you will see radically reduces the size of the model.
model explanation
Robert,
I know.It was my first choice but I had a lot of "Overflow" by Pspice.In the new versions this kind of error it's shown instead of simple convergence problem.I realized that it was due to the fact that interpolation coefficients were too low for voltage sensibility(usually 1 uV).So i decided to use current sources (1 pA) and it works.
If u can,try to sweep input current against grid voltage.Are the input current values meanly correct in your opinion?
Robert,
I know.It was my first choice but I had a lot of "Overflow" by Pspice.In the new versions this kind of error it's shown instead of simple convergence problem.I realized that it was due to the fact that interpolation coefficients were too low for voltage sensibility(usually 1 uV).So i decided to use current sources (1 pA) and it works.
If u can,try to sweep input current against grid voltage.Are the input current values meanly correct in your opinion?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.