ADS LSSP simulation non convergence errors

[bubuta77]

Hi
I simulate the PA using model from Infineon. I can't do the load-pull properly because of non- convergence errors. Additionally when i run LSSP simulation the results is the same. When i decrease the input power by 5-10dB from saturation level the problem dissapears. HB for single tone when i make some changes at setup (Krilov method) i can work at saturation levels.
I used to work with AWR and never see such problems. Is it model fault or i do something wrong?

 

[Manjunatha_hv]

Hello bubuta,

Yes, it may be due to inefficient model development ....Also change the convergence settings in MWO HB...
Try upload your MWO file here, we can help you better....



---manju---

 

[bubuta77]

Unfortunately i don't have AWR model it's a reason why i move to ADS. It's really start to look like model problem because when i make same simulation with different tranzistor (Freescale one) it works pretty fine without my changes at HB setup. Anyway thanks

 

[BigBoss]

in HB Controller block, there are different options targeted to specific problems.You'd better to read the principles of HB method and research in application notes of Agilent to resolve your problem.

 

[kspalla]

adding farther, under HB solver you can change solver type, increase iterations, and convergence mode and combination of these.

 

[pancho_hideboo]

First, increase harmonics of HB to 31.
Second, increse over sample number, such as 2, 4, 8, ....
See The Designer's Guide Community Forum - maxstep in shooting-envlp of Cadence Spectre

And try to set Transient-Assisted-HB manually.
Don't use automatic set for it.

 

[bubuta77]

adding farther, under HB solver you can change solver type, increase iterations, and convergence mode and combination of these.

I know, It converges when Krilov solver is set and at advanced settings the preconditioner set to SCP (it's for highly non-linear devices - i do read the manual). But still the behaviour is not stable change of some parameter can cause to non-convergence. Additionally the LSSP can't converge.
I move to freescale model - works pretty fine