Matlab (Student Version)

Matlab control file - go.m
data12ax7.m
coeff.m
interp.m
final.m
Matlab Results (stored coefficients)





Pspice files (used with Intusoft Demo Version 5.0/6.0)


I like Koren's effort a lot because it uses some interesting functions to accomplish
tube-like "transfer" trends ... except that it doesn't mimic the curvature of tube very
well in areas of standard operation, and the use of a diode to model the input circuit 
has been questioned by some ... his model certainly has great converging properties 
which can happen if strictly monotonic functions are used throughout (as composite 
functions) ... Koren's method is based on the insistance that, according to 
classic theory, the plate current characteristics are well approximated by the standard 
3/2 power expression ... the model I stumbled on is based on polynomial functions 
which are generated using "sampled" data points (RCA's and GE's in this case) ... all 
that is required to show is how (qualitatively and not just quantitatively) (i)a) 
data is well approximated on the sampled data and (i)b) on the curves they are taken 
from and (ii) to assure that there's a strong sense of continuity in how transfer 
curvature evolves in between the mutual characteristic curves (data) ... since 
small-signal simulations depend on modeling accuracy in the first order derivative on 
the data, it follows that strong "slope" accuracy is required of the models ... for 
example, if a device model generates a copy of a data curve that initially starts off 
a little low in comparison with data and then it ends a little higher against it then 
there's a good chance that where the two curves cross there will be different crossing 
slopes near and on that point ... this means that small-signal parameters, which are 
usually taken in and arond the weak spots of Koren's model, are likely to be off by a 
considerable amount as a result ... plotting the Koren model against RCA triode data 
shows that the modeling is not very good even in an absolute sense, but still it could 
be possible to produce a model that exhibits good absolute error but poor slope error 
... In this approach I've tested out the modeling can produce micro oscillations around 
a data curve simply by choosing interpolation "order" to a too high degree (versus the 
number of data points given ... I don't mean to knock Mr. Koren's efforts, in comparison 
my models might have a very hard time converging in circuits that his models would glide 
through ... the reason for this is the dual-nature of my models, both plate and grid
ports are modeled by bi-variable non-linear polynomial functions which may start 
fighting with one another if initial conditions are not set properly at the onset of a 
simulation ... even though this new model is to be taken as experimental I have a feeling 
that high accuracy levels can be reached using a generalized approach ... the Completeness
Theorem in the Space of Real valued Functions (in regards to polynomial function spaces) 
certainly adds support to using this kind of approach ... 



Model comparison between Koren's and Mine
Common Cathode Transfer Sim  *with older GE model

Non-linear Cascaded Circuit Sim



~:( HOME :)~


viva Analog /// jc -> lynx.net