We did already discuss that for electrical large problems MoM and Time Domain Methods are more suitable then Frequency Domain FEM Techniques.
I’m just curious: What is the larges numbers of Tets you ever solved with HFSS before running into memory problems? What kind of computer was used for this.
CPU: Intel Pentium 4 HT 3.0GHz
Memory: 2*512MB DDR266 (1GB Total)
OS: Windows XP SP2
EM Solver and Problem Info
Solver: Ansoft HFSS 8.0.25
Type: Eigenmode Analysis
Model Size: 2mm (L) * 2mm (W) * 1mm (H)
No. of Eigenmodes: 4
Largest Eigenmodes value: 50 GHz
Total no. of Tetrahedrals at 5th Adaptive Passes: 32535 (Simulation Successful)
Total no. of Tetrahedrals at 6th Adaptive Passes: 37567 (Simulation died due to insufficient memory)
Simulation Time after 5th Adaptive Passes: 4 hours 18 minutes
It depends. If you let HFSS seed the mesh automatically then 1ghz is enough. But in certain radiation problems, users seed the mesh maually, and they take the tetra length too small which is just a waste of computer resources. So its better to let HFSS seed mesh automatically. Even in case u want to do manually keep the tetra length no smaller than lambda by 6.
I use P4 3.2CG , but my friend use K8 3200+ (754)
Simulate the HFSS
He said that his 3200+ is saving more time than P4 2.8CG about 40-50mins.
(He designs the RFID)
What is the larges numbers of Tets you ever solved with HFSS ....
The answer strongly depends on geometry (relation between numbers of outer and inner cells) and boundary conditions.
For the simplest case - (2 single-moded ports, no losses, disk-loaded circular structure) I reached more then 300.000 tetras with HFSS v8.5.19
The simulation time at one freq. point was 3 hours at P4-2.6, 2Gb. Of course, "Fast Sweep" option is not possible...
Now with HFSS v9.2.1 Ansoft enhances the memory usage and the only limitation is your physical amount of RAM
(do not foget to use "/3GB" option at your server start up config, if you have >4 Gb memory installed !").
The max tetras I solved with HFSS v9 is about 200.000 ( single-mode port + 5-mode port, lossy walls)
It took about 1.5 hour at XEON-Dual 3 Ghz, 4 Gb
thanks for the feedback. Of course you are right. The memory usage depends on the on the problem type /boundaries….
However, I was just looking for some estimations to get an idea.
Another question: By default HFSS uses second order elements. As far as know, there is the possibility to switch back to “first “order elements in order to save memory. If you do this you certainly need much more tets to sample the fields. Does this possibility still exists in HFSS9.2? Any experience with this option?
I've been an ardent user of HFSS till 2 months ago when I got into MicroStripes. Time-domain Transmission Line Matrix (TLM) based tool. I've solved a 288 element patch antenna array with it! 61 million cells (equivalent with HFSS' tethraedras). Ran it on a PentiumM 2GHz laptop with 2MB RAM. Competitors don't even dare to speak about problems this size. And the GUI is pretty cool and easy to use. You can find more details on their website at www.microstripes.com
You are right, Ansoft v8.5 has a lot of officially undocumented so-called environmental variables. I attached the "pdf" with description of them.
In HFSS v9 - part of theese variables included to GUI itself.
For example you can mark /YOU Project/Analysis/Setup/Advansed/Use Low Order Solution Basis and chose tetras with less nodes.
I never tried it, because I never needed to solve a geometry with size less than wavelength.
I have one particular model that is run on HFSS v9.1. It is very fine detail throughout and crashes out just over 110,000 tet. Machine is a 4GB running XP.
The dielectric substrate is in square form. (6λ*6λ)
Thickness of the substrate is th=0.1λ
Height of the airbox is 1.0λ (airbox also contains the substrate)
Operating frequency is 30GHz, (λ=1cm)
The infinitesimal dipole length=0.02λ, width=0.002λ.(+z directed 1A current excitation is used along the dipole)
In this problem at adaptive pass#7 I get the famous " Adaptive solution, process solver : Insufficient memory for subprocess "%3" " error. And execution stops.
In reality I need to increase the width of the substrate up to 15-20λ. Now I'm step by step trying to increase the problem geometry and try to see what happens.
Now what do you advise me? I use automatic meshing of HFSS and don't do any meshing operations. I sure need to manually adjust mesh properties but have no idea about how to adjust these mesh properties. Any material on mesh adjustments is welcome.
On the other hand, as far as I know, I may use a symmetry plane etc. but again have no idea about how this is done. So again I need some good sources about these thing.
Finally, I don't know how to adjust the Analysis setup, how many adaptive passes are actually required, which delta S (E) gives acceptable nearly accurate results.
It seems to me that these kind of important keypoints or tricks will take time to learn, so i will be very glad if you share your ideas.
I'm also attaching the Solution data (convergence plot) and the HFSS project file.
There is no other choice but to use symmetry if the problem size is huge.
Having said that, that has been explained a lot in the examples for HFSS 8.0, but not for the new version. Seems that the guys from Ansoft believe that everybody has unlimited RAM.
To reduce size, try to decrease the number of segments when placing an arc or circle. Default value is 12, but can be decreased. In some cases the best is to use only rectangular structures.
Manual meshing may help, but I have no experience with that.
Reducing the frequency also helps. If you are setting the circuit to extract a SPICE mode, set the namber of steps per rise time to less than 5, that decrease the bandwidth.
Other than that, FEM as a numerical method tends to result in a large sparse matix, nothing much you can do there. In case of planar structures, try using Designer.
:D
LinkBack URL
About LinkBacks
Reply With Quote
View Profile
Private Message
