Reply to all this time
On 9/29/07, Stefan Jahn <stefan@...> wrote:
> hello there,
> what do you think about a synthesis tool in qucs performing "vector
> fitting" in order to derive passive circuits (RLC) for frequency
> dependent transfer functions? does anyone have heard of it, or
> experiences with it?
I have heard of it. My thesis domain is non destructive control, and
therefore I need to fit data. Waht I can say is, in the general case
this problem is mathematically ill defined and very instable. In my
case I know by hand that is the model, and therefore I try to fit
parameter from measurement to my model.
Therefore, you need a really good measurement precision . For RLC
circuit a simple non linear least square fitting work pretty well if Q
is not high (<100), I used it during my scholarity. I used the non
linear least square fitting found in gnuplot :-)
By the way vector fitting is the algorithm described in
In the other hand this algorithm do not try to fit a model to
measurement,and could be really useful but in all the case keep in
- real pole (high Q) are difficult to approximate due to finite
- large imaginary pole (low Q) are difficult to approximate because
magnitude effect is quite low.
In conclusion: this algorithm is good for finding empirical models, or
fitting well behaved model.
> i recently heard about this topic regarding the ability to model
> frequency dependent transmission line losses using simple passive
> networks in the time domain... and i think this is a nice idea in fact.
> any comment is welcome.
> cheers, stefan.