Have a look at this reference: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1618575&url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel5%2F22%2F33917%2F01618575.pdf%3Farnumber%3D1618575

Can you clarify what non linearity you want to model? , what software you are using to model behaviour, frequency and bandwidth of operation and how accurate you want the answer?

There are a number of techniques for handling non-linear problems from "quick and dirty" solutions that are not too accurate through to some that take a lot of computational time. Its best to understand what effects the non-linearity has on your circuit and try to simplify the problem before tackling the maths. If you only want a ball park answer then simple is best. Usually the noise figures quoted (for transistors) are not very accurate +-20% if you are lucky, so there is no point in using mathematics that give a 0.1% answer.

For example, some problems can be solved using piecewise linear equations, i.e. a multiplicity of linear equations are used in the solution, each equation is selected depending on the value of an overall parameter. The overall parameter may be non-linear but over small excursions can be approximated to linear.

Best Regards

Paul

You can use EKV or even better ACM models. It is not clear for me what you mean when you say that you can get much better results. Do you mean more accurate with the real device? 

Best regards 

Santiago 

Thanks Paul and Santiago.

Paul: I am using Agilent ADS for simulations of transistors like FHX13LG, of which I have only S-parameter model which is a small signal model. But at large signals, the transistor behavior goes in non-linear region. I want a model which will give results considering the non-linear behavior of the transistor. For example, I can find out 1dB compression point and 3rd order intercept point. This can be achieved by only non-linear model. I am using it in 6 to 10 GHz frequency range with whatever NF I can get as of now (+/- 1 dB to that of datatsheet is perfectly fine). The non-linear model that can be used Advanced Curtice Model (which is what probably Santiago is suggesting) which is there is ADS but I don't have the parameter values like Rg, Lg,Ls,Ld,Cgs,Cds, Gm,etc.

Santiago: EKV can be used only for MOSFET I guess. I want it for HEMT. Advanced Curtice Model (do you mean this by ACM?) can be very much useful. But I don't have the parameter values. So, my question is can we use these S-parameter values to get a Advanced Curtice Model. 

Yes, more accurate results are more close to measured results. I have simulated an LNA with transistors with s-parameter models and Advanced Curtice model parameters available. On overlapping the S-parameter, ACM and measured results, I found that ACM results were much more close to measured results whereas with S-parameters, no measurement was close enough to measured other that gain.

Thanks again for your valuable replies. Please mention if I haven't answered your questions.

I do not think that linear characteristics can substitute nonlinear currents of the HEMT.

Please also consider the current hysteresis. Probably due to large signal inserted to a device the traps are charged and the behavior will be different from that derived from the linear s2p files..

Thank you Paul. I appreciate your reply.