hfss convergence

Im having trouble with resonance and converging solutions. I've designed a wave cavity that consists of a full cell and a partial cell, seperated by irises, that are meant to set up a 0 mode (All E fields in the same direction) and a pi mode (opposing directions). The main cell has a wave guidfe and port created to transport the microwaves into the cavity. The problem is, when the field strength in the cavity is strongest the solutions stop converging.
My task is to tune the full cavity by adjusting the radius of it until the field strength inside the cavity is a maximum, but as stated above, when the field strenght is a max in the cavity and almost 0 in the waveguide, the solutions stop converging. The smaller cavity is completely detuned to assure only the main cell is being tuned.

My questions are:
What is it trying to converge?
Are there methods for tweeking solution setups to get convergence?
I've tried changing the mesh refinement, what now?

Its been a few weeks and am having no luck with trying to tune this bad boy! Any ideas would be great. I've included a graph and a picture of the cavity that might help you folks understand what im on about.

Hello,

I do not understand - are you using driven modal solution or eigen mode solver?

Itai

Driven, sorry.

I still dont understand this model and design objective.
Is it a cavity mode converter or power divider or phase controller?
I assume three wave ports involved in your model, (upper,left and right), you are tuning the size of cavity to get what you want (actually I dont know what you do want). The convergence criteria of HFSS is based on S-parameter at the ports. S-parameter is the integration of field intensity on the port surface. So, S-parameter is considered convergent by the criteria we set up, the field intensity actually is not convergent, since the convegence of field is slower than convegence of S-parameter.

Wish this can answer your 1st question.

Best Regards.

The upper port is a waveguide, where the microwaves enter the system. The left port (the larger of the two remaining) is not a port at all, it is set finite conductivity like the rest of the object, save the perfect H for symmetry. The port on the right is left blank, because the object itself doesn't have any material at this point. It is a system for trapping electrons in electric fields for acceleration. The piece was modeled successfully in SUPERFISH, but not by me, and I can't seem to replicate the results because around resonant frequencies my solutions stop converging. I could set the number of attempts greater but after 12 or so tries it takes a very long time per solution, so im curious as to why for resonant frequencies the solution stops converging.

Please upload your structure, so some people can examine it for you!

The upper port is the input waveguide port, no problem, left port is perfect-H plane, means you are using symmetry, so this structure is just half of your model. It should be ok also, How about the right port? it is left blank? means this port open to air? I think your objective is to study the periodic structure, I remembered I should have a slide from AnSoft about design cavity accelerator, if you need it, let me know. I still suspect your model has some problems.

Best Regards,

Hi,

I am not fammiliar with this model, but here are a few guidelines:

1) HFSS needs energy inside the model in the convergence frequency. For example, if you try to converge a low-pass filter in the high frequencies it will be much slower than converging in the lower frequency. In the higher frequency it will only adapt the mesh near the ports (since there is no energy in the model)
Before continuing plot your mesh (select objects, right click > plot mesh).

2) If you are adapting in mag(S11(f_adapt)) > 0.92 you are in total reflection, and the statement above holds.

3) Another option is to change the initial mesh. Either decrease the lambda refignment factor (default is 0.33 lambda) or select the imporatnt objects and assign a mesh operation inside selection. Set the size of the maximum tetra size of the initial mesh. Note that these methods may result in an unnecessary large matrix for the solver. In this case mesh only the important objects in your model.

4)If the convergence is going good, but not fast enough, you can increase the refignment per pass from 20% to 30%.

Hope this helps,
Itai

1) Energy is getting inside the main cavity from the waveguide, Im not sure what else you mean. Energy is getting inside the system because I can plot the electric field inside the cavity and there is field inside.

2) Im not sure what this means

3) I've set the lambda to .2 I'll post results when I have them.

4) Convergence is not good, the "Max Mag Delta S" is all over the place, between .10 and .90 sometimes higher than 1.

The left 'port' isnt perfect H it is finite conductive, the big flat surface closest to the viewer (including the waveguide and all cells) is the symmetry plane, and so its assigned the perfect H. Yes the smaller port on the right is left open to air, its how the physical system is built so is required to be left 'open' for accuracy.

what is the value of S11 ?

0.87

Another problem I am having is the following error:

[info] Parametric Analysis is done. (6:18 PM Nov 09, 2004)
[error] Adaptive solution, process solver : Write failure in subprocess %3. Possible causes are: out of disk space, no file access privileges, or hardware and/or network problems. (6:18 PM Nov 09, 2004)
[info] A variation (d1='0mm' d2='0.19mm' l1='0mm' l2='0mm') has been requested (5:59 PM Nov 09, 2004)
[error] Adaptive solution, process solver : Write failure in subprocess %3. Possible causes are: out of disk space, no file access privileges, or hardware and/or network problems. (5:59 PM Nov 09, 2004)
[info] A variation (d1='0mm' d2='0.18mm' l1='0mm' l2='0mm') has been requested (5:41 PM Nov 09, 2004)
[info] A variation (d1='0mm' d2='0.17mm' l1='0mm' l2='0mm') has been requested (5:38 PM Nov 09, 2004)

Im certain the file privileges are ok and there is plenty of disk space, but what hardware/network problems should i suspect? The simulation computer has a gig of ram, should I find some more?

Yeah, the solutions stop converging, it looks like i need more ram or something. When I check out the convergence data its all over the place:

1 11748 N/A
2 12925 1.1523
3 14222 0.46266
4 15649 0.51254
5 17216 1.1641
6 18945 0.15373
7 20852 0.80858
8 22945 1.1884
9 25247 0.16375
10 27775 0.83216

The convergence just sort of jumps around. If i try to do step 11 then it crashes and gives the %3 memory error. So it looks like i need more memory or methods for cutting down on the amount of memory per attempt, or ways to get convergence faster, or anything really.

Hi, 27k unknowns are not big deal for HFSS, i forgot the exact maxium number of unkonws for HFSS. Take a look at the S-parameter of the convegence rate, it seems this structure is osilating, I have no preknowlege of this kind of structure, I can not tell which part is the most critical part, but you can try to put some dummy object in the critical part, and specify manual meshing, that would speed up the convegence.
Best Regards,

its not the number of tetrahedrons, its the memory the number takes...its giving a memory. I'll try adding dumb meshing objects.

1 11748 N/A
2 12925 1.1523
3 14222 0.46266
4 15649 0.51254
5 17216 1.1641
6 18945 0.15373
7 20852 0.80858
8 22945 1.1884
9 25247 0.16375
10 27775 0.83216

it IS the number of tetrahedrons.

Best Regards,

Hi nadnerb,

since you have some much trouble with HFSS: why don't you post the model here and some of us give it a try with MWS....
All what we would need is the SAT file any maybe some additonal information on units, freq. range. parameter you would like to model....


F.

Hi nadnerb,

I think the reason the model does not converge is very simple.
27K tetras are far from enough for this model. It shows you are not using correctly the software.

Therefore, I would suggest you would do one (or all) of the following:
1) Select the cavity and add a mesh operation (start with W/3, where W is the width of the cavity). (Adding dummy objects are irrelevand as far as I can see from the model - if you want to speed up convergence use mesh operation).
2) Increase the number of passes to 20 or 30 or until it converges. The first ten passes should not take more than a few minutes for HFSS. These are small matrices, and they do not reflect the true complexity of the model
3) Increase refignment per pass to 30%.

If this does not work it means that there is something interesting going in your adaptation frequency. Stop after 15 passes and do a fast sweep around your convergence frequency. Study the graph or post it here.

Best Regards,
Itai

Ok, in order to do all of these things, I am going to need more ram, at 27k i get the %3 memory error. Thanks, I'll see what I can do.

(I only have a gig now)

nadnerb