(transient , TAHB , steady state response , initial guess file , harmonic balance , convergence )

Hello all,

Case (ADS cannot do this ?):

I believe this could be extrememly useful: being able to run TRANSIENT with a given setting (MaxTimeStep & Order), and specify an estimated "point in time: t_p" where TRANSIENT would switch to a different setting (different MaxTimeStep & Order). I think time spent in a TAHB/Harmonic_Balance degugging cycle for solving HB convergence problem could be greatly improved...

Purpose:

When using TRANSIENT for the **specific purpose** of getting an initial guess file for solving Harmonic Balance convergence, only that part of the TRANSIENT result that comes close to the steady state response (i.e the latest part of the time domain result) is helpful for HB convergence (...if I understand correctly from: page 21, section 4.11, last paragraph, Agilent document: A Comprehensive Guide to HB Simulation for ADS, release May 2003). Following this logic, we are not interested in a "very precise" transitory response...; thus, a bigger MaxTimeStep and lower Order could be specified for the earlier transitory part, and then starting from a user defined "t_p" point in time a much smaller MaxTimeStep and much higher Order could be used to "fine tune" the steady state response that is useful to help HB converge more easily.

Problem:

(assuming my reasoning is valid) It seems to me ADS TRANSIENT settings don't allow to do that. In my application where I am trying to use a TAHB cycle: TRANSIENT response as initial guess to solve really difficult HB convergence problems, I have to use a very small MaxTimeStep and large Order to account for strong harmonics and many cycles to reach steady state; as a result: the TRANSIENT simulation itself becomes very, very time consuming, and has to be repeated for every change in the circuit. ...And the TRANSIENT was supposed to be the "short cut" to get Harmonic Balance (which is my real simulation objective; not TRANSIENT) to converge more quickly...

Questions:

- am I right that TRANSIENT cannot allow to switch of MaxTimeStep/Order settings after a user defined point in time ?

- is there any way around (may be some kind of concatenation of data from 2 different TRANSIENT simulation runs? how to do it?) in order to "extract" more quickly that useful part of the TRANSIENT response that is closer to the steady response ?

Any help would be greatly appreciated. Thank you.

NicolasC

Hello all,

Case (ADS cannot do this ?):

I believe this could be extrememly useful: being able to run TRANSIENT with a given setting (MaxTimeStep & Order), and specify an estimated "point in time: t_p" where TRANSIENT would switch to a different setting (different MaxTimeStep & Order). I think time spent in a TAHB/Harmonic_Balance degugging cycle for solving HB convergence problem could be greatly improved...

Purpose:

When using TRANSIENT for the **specific purpose** of getting an initial guess file for solving Harmonic Balance convergence, only that part of the TRANSIENT result that comes close to the steady state response (i.e the latest part of the time domain result) is helpful for HB convergence (...if I understand correctly from: page 21, section 4.11, last paragraph, Agilent document: A Comprehensive Guide to HB Simulation for ADS, release May 2003). Following this logic, we are not interested in a "very precise" transitory response...; thus, a bigger MaxTimeStep and lower Order could be specified for the earlier transitory part, and then starting from a user defined "t_p" point in time a much smaller MaxTimeStep and much higher Order could be used to "fine tune" the steady state response that is useful to help HB converge more easily.

Problem:

(assuming my reasoning is valid) It seems to me ADS TRANSIENT settings don't allow to do that. In my application where I am trying to use a TAHB cycle: TRANSIENT response as initial guess to solve really difficult HB convergence problems, I have to use a very small MaxTimeStep and large Order to account for strong harmonics and many cycles to reach steady state; as a result: the TRANSIENT simulation itself becomes very, very time consuming, and has to be repeated for every change in the circuit. ...And the TRANSIENT was supposed to be the "short cut" to get Harmonic Balance (which is my real simulation objective; not TRANSIENT) to converge more quickly...

Questions:

- am I right that TRANSIENT cannot allow to switch of MaxTimeStep/Order settings after a user defined point in time ?

- is there any way around (may be some kind of concatenation of data from 2 different TRANSIENT simulation runs? how to do it?) in order to "extract" more quickly that useful part of the TRANSIENT response that is closer to the steady response ?

Any help would be greatly appreciated. Thank you.

NicolasC

One comment about your comment of re-running the intial guess every time you change the circuit. Unless you're fundamentally changing the output signals of the dividers, this shouldn't always be necessary. The InitialGuess file should be fairly robust to small tweaks, including adding and removing of components. [I will often "fine tune" the transient initial guess, save an HB solution, and then use this "more accurate" initial guess for further simulations. And if I'm manually tuning this design, re-writing this InitialGuess file with updated HB solutions will help by keeping it as current as possible.] Of course, this cannot be guaranteed, and if/when it doesn't properly converge anymore, a Transient initial guess is again required.