HB initial guess

[Alex Pacini]

The HB solver can use the transient to assist the initial guess for a transient assisted HB.

If TAHB=0, which is the initial guess? Can I set it manually, or example from a previous ad hoc solution?

Thanks!

[xyce-users]

Alex,

If TAHB=0, an all zero initial guess is used during HB solve. Currently, we do not support setting the initial guess manually. If TAHB=1, you can set the initial conditions on voltage nodes during the transient run to obtain an initial guess for HB. The HB can also work with continuation. In this case, the solutions from the previous step is used as the initial guess for the current step.

Thanks,

The Xyce Team

[Alex Pacini]

The HB can also work with continuation. In this case, the solutions from the previous step is used as the initial guess for the current step.

Do you mean using the .STEP analysis?

If I remember well, ADS has a similar feature If I sweep a parameter inside the HB analysis.

Is it planned to add a manual initial guess?

Thanks a lot,

Alex

[Anon]

On Thursday, December 20, 2018 at 9:02:35 AM UTC-7, Alex Pacini wrote:

The HB can also work with continuation. In this case, the solutions from the previous step is used as the initial guess for the current step.

Do you mean using the .STEP analysis?

If I remember well, ADS has a similar feature If I sweep a parameter inside the HB analysis.

Sorry, "continuation" was a little vague. We're specifically referring to "Natural Parameter Continuation"; but HB can also work with .STEP analysis. 

There's an example of direct continuation in the Test Suite (source stepping in this case): Xyce_Regression/Netlists/Certification_Tests/BUG_573_SON

The full netlist follows:

.global_param source_fac=0.1

.print HB V(Ve) V(out)

.print homotopy V(Vcc) V(1)

.hb 1MEG

.options hbint numfreq=50

.options linsol-hb type=belos 

.options nonlin-hb continuation=1

.options loca stepper=0 predictor=0 stepcontrol=0

+ conparam=source_fac

+ initialvalue=0.1 minvalue=0.0 maxvalue=1

+ initialstepsize=0.1 minstepsize=1.0e-6 maxstepsize=0.3

+ aggressiveness=1

+ maxsteps=80 maxnliters=50

.op

R5 0 OUT 1MEG  

V1 1 0 sin 0 {source_fac*1} 1meg AC 1

V2 Vcc 0 {source_fac*12}

C2 Vc OUT 10u  

R4 0 Ve 270  

R3 Vc Vcc 1.5k  

Q1 Vc Vb Ve 2N2222 

.MODEL 2N2222 NPN (Is=14.34f Xti=3 Eg=1.11 Vaf=74.03 Bf=255.9 Ne=1.307

+ Ise=14.34f Ikf=.2847 Xtb=1.5 Br=6.092 Nc=2 Isc=0 Ikr=0 Rc=1 Cjc=7.306p

+ Mjc=.3416 Vjc=.75 Fc=.5 Cje=22.01p Mje=.377 Vje=.75 Tr=46.91n Tf=411.1p

+ Itf=.6 Vtf=1.7 Xtf=3 Rb=10)

R2 0 Vb 6.8k  

R1 Vb Vcc 39k  

C1 1 Vb 10u  

.end

If you're curious, an example of HB and .STEP is also in the test suite: Xyce_Regression/Netlists/Certification_Tests/BUG_340

 

Is it planned to add a manual initial guess?

 The manual initial guess is in our to-do list, but we aren't sure when it will be available.

[Alex Pacini]

Thanks for the detailed reply and for the references.

There are lot of features I was not aware of...

Thanks again,

Alex

Alex Pacini

email alexpa...@gmail.com

about me alexpacini.gitlab.io

Google+ google.com/+alexpacini

mobile +39 338 3692711

OrcID 0000-0002-0316-4989

[xyce-users]

More information about Xyce's continuation methods can also be found in chapter 8 of the Xyce Users' Guide, starting around page 117.