Non-linear circuit modelling

Dear All,
I would like to discuss non-linear circuit modelling using commercial EDA tools. The component I am looking into is diode (in mixer, modulator design), and non-linear power amplifier. Please feel free to contribute your experience :arrow:

I think the most used method is harmonic balance.. It is an old method occured many years ago. And I want to know the recent advance of it.

And volterra series method is another choice for weak nonlinear circuit. AWR microwave office include it for some special problem.

Although envelop analyse method consider the nonlinearity, it main focus is for system performence simulate and some "transient like" analysis on high frequency.

Harmonic balance and volterra series method are widely used for non-linear circuit simulation and analysis. For simulation type, I have not used volterra before. I use HB in ADS most of the time. Most common problem faced is the convergence problem.

Actually I am refering to the component modelling, do anyone of you have experience on modelling the equivalent circuit of non-linear microwave components (such as diode, Power amplifier, etc)...

I have experience in VCO and diode modelling using FDD and SDD in ADS....

Hi,

just a few words. There are two classes of methods for nonlinear analysis: time domain methods, such is SPICE and frequency domain methods such are Harmonic Balance (HB) and Voltera series analysis.

Spice was the first method of nonliner analyis and stil is the mast for the low frequency designs where time domain models of components are simple (RLC circuits are straight forward to model by SPICE, transmission lines are a bit tricky but managable). The circuit is completely solved in time domain. Becomes more atractive because it is highly compatible with FDTD electromagnetic analysis so everything can be put into the same pot.

HB is a general frequency domain method that works well for RF/Microwave applications. The elements of the circuits are modeled in the frequency domain while device itself (dependent generators) is modeled in the time domain. Fourier transform is needed to bring the voltages in the frequency domain, etc., I don't want to give lecture on HB. It is very mature method and implemented (in slightly different forms) in all modern MW/RF simulators. Its universal and all of nonlinear circuits can be analized with the method (oscillators, mixers, multipliers).

Voltera series is the best suited for one exeption of nonlinear circuits where HB is usually failing to converge, this is the slight nonlinearity case, for example the linear amplifier if you are intrerested in its low level IMD products. It is implemented in MWO only.

At the end, just to mentioned that not all of the models available in the literature are applicable with comercally available nonlinear simulators. The problem is with some nonlinearities that cannot be described or handled by the solver due to convergence or charge preservation problems etc. It is best to use simulator built-in models. Also, some manufacturers give very good models for their products while the others are not really close to reality even for DC IV curves, so you might end up making your own models at the end.

I have very good experience with Ansoft Serenade/Designer models made by Ansoft even if the manufacturers model was available. The agreement between simulations and measurements was amazing!

flyhigh

Does the vco is so amazing?

How do you model it if manufacter's model is unavaible? please explain it in detail. thanks!

randyding said:

How do you model it if manufacter's model is unavaible? please explain it in detail. thanks!

Click to expand...

Hi,

well, don't get me wrong but that is a question like "what was at the beginning?" I think hundreds of PhD students are working world wide on this topic.

You should start reading and getting some help about this if you are really interested in making yours models (when I say yours I mean extracting e.g. Curtice model for the device you want to model all by yourself).

flyhigh

Volterra series are used for "mild nonlinearity" of active devices. HB is a general purpose simulation method as said..
I agree with that opinion that is Ansoft is more successful on HB simulations rather than MWO and others but not ADS...
In my opinion , ADS has great HB simulator even sometime it has convergence problems.

Modelling is not easy task and it should have hardworking on.
For bipolar transistors I suggest you to read this book mentioned below.

M. Reisch ,"High Frequency Bipolar Transistors",Springer-Advanced Microelectronics Series , 2003 , Heidelberg.

ISBN: 3-540-67702

It's quite theoritical book and expensive ( approx 110€ ) but to learn something it's worthy..isn't it ?

who may tell something about the RF behavious model.

To flyhigh: I truly agreed with your opinions on

"The problem is with some nonlinearities that cannot be described or handled by the solver due to convergence or charge preservation problems etc. It is best to use simulator built-in models. Also, some manufacturers give very good models for their products while the others are not really close to reality even for DC IV curves, so you might end up making your own models at the end."

But sometime, if you intend to use a component, which is cheap for mass production but the manufacturer do not provide the actual internal circuit model (even upon request), you have to use equivalent model via (DC IV curve for diode, and output frequency/power level vs tuning voltage for VCO, etc..) to simulate a loop circuit. This is why SDD and FDD is available in ADS.

Flyhigh, your experience in non-linear circuit modelling is based on manufacturer's parameters. Do you ever try to characterize a non-linear device using IC-CAP before? This is something I would like to try out but I am too busy to try it

To IICCEE: what is your interest? The VCO?

To create a user-defined nonlinear component which can simulate both the large-signal and small-signal behavior of a nonlinear device, without the use of source code, it can be done in 2 way: symbolically defined device(SDD) and Frequency Define device (FDD). Both model are available in ADS. You can define 1-10 port model in ADS (or even higher with some extra setting).

If you would like to know more, you can search ADS help file on Chapter 5: Custom Modeling with Symbolically-Defined Devices. Else I can upload the whole chapter here (1.73Mb).

activewei said:

Flyhigh, your experience in non-linear circuit modelling is based on manufacturer's parameters. Do you ever try to characterize a non-linear device using IC-CAP before? This is something I would like to try out but I am too busy to try it

Click to expand...

Hi activewei,

no, I don't have any non-linear modeling experience. We have dedicated engineers in my company so I could only take their experience for grant. Their experience about IC-CAP is that it works great for intrisic devices (bare chip) while for packaged devices it usage is limited to frequencies where packaging does not introduce significant influence. Modeling some packeged devices at K band was realy challenging, they said. For this particular device you will not get any model from manufacturer. Even linear Spar data were significantly deviated from measured.

The big trouble with manufacturer's model is that it tends to be general! A dedicated modeling person is capable to optimize model for a device to a particular circuit design. For example, we were using the same device for active mixer, oscillator and PA and we had 3 different models! Actually it was the same model optimized at 3 different operating points so paramaters of the models were different. If we use resistive mixer instead of transconductance one, we would require the fourth model!

So, for a serious nonlinear work, it is neccessary to have modeling expert(s) to support designers.

flyhigh

hi flyhigh, good to hear something of my interest. It is quite a sad thing that there is no dedicated person in my department to model the equivalent circuit.

Would you mind to share the experience on how your people model a non-linear circuit, for a given non-linear device after characterization using IC-CAP?

activewei said:

Would you mind to share the experience on how your people model a non-linear circuit, for a given non-linear device after characterization using IC-CAP?

Click to expand...

Hi activewei,

unfortunately, I am like a a customer to those guys, they are a bit closed for a trickey questions, so considering the modeling I don't have many secrets to share.

I can just acknowledge and admire their work and skills. Most of the time their models are excellent, from time to time need to be reworked. Anyway, don't ever expect the measurements on your circuit to match the simulation within the fraction of a dB, it will never happen.

I attended the @gilent workshop dealing with nonlinear network analyser they are trying to release in the future. It is now available for selected academic and research institution on a contract basis only. Once released it will be a great step forward in both theory and design of nonlinear circuits as we will be able to measure and use the nonlinear transfer function parameters straight in our designs. If I undersood correctly, not any specific modeling skills will be required.

flyhigh

That will be a good helper for non-linear circuit modeling with this equipment in future.

For the time being, I will still learning and struggling to understand non-linear circuit behaviour via literature, experiment, trial and error...

Anyone has actually dealing with non-linear circuit modeling here?

**broken link removed**

My 0.02. Have fun.

Hi uncle_urfi: the link is unavailable, could you give me the exact URL for the software downloading?

TO activewei:
I has find the document.
But have another question,is the model extract from the test in pratice?
and there is large signal model such as curtice model,why use the symbolically defined device.
In my knowlage,to get or extract a model is not easy thing.It must depend on the factory.

I know the ansoft co-operate mot to get the LDMOS model .Also to get the sigma-deta PLL model,In this way,to ensure the data correct.

yes, the data should be extracted from the practical measurement. But sometime, some parameter might be given by the manufacturer.

If there is a model available, it is better to use the model than using SDD.

As i said previously, I will only use SDD if the model is not available. Sometime the model available is not suitable, such as the response of a tunnel diode/step recovery diode is not suitable to use general model available in simulator....

I think to design a good nonlinearity microwave circuits needs excellent EM knowledge and good mathematics. FOr example if you want to design a broadband VCO, the first step is not to simulate,but to calculate and use the multi open stub to get the wideband match. Of course the method is HB with Voterra series. After you give the circuits topology and many calculations, you can design your circuit in ADS or ansoft designer to get what you want. And last the comparasion between calcuation and simulation and measurments are necessary.