how to obtain multiple resonances in hfss
I have a question regarding HFSS. I am designing a quad-band stacked patch antenna. I have resonances of 4, 5, 6, and 6.5 GHz. So far, I have set a solution frequency of 10 GHz in thinking that this would achieve a more refined mesh and a more accurate solution (such as in IE3D). Is this true in HFSS, or should I set the solution frequency to another value?
I am noticing a deviation in the lowest frequency resonance from the simulated antenna and the fabricated antenna. IE3D also better predicts the lowest resonance (4 GHz) which baffles me.
Does anyone have any ideas to increase the accuracy of my solution without significantly adding simulation time? Thank you in advance!
From my experience you need to make an adaptive mesh refinement at each frequency of interest, so I would say that solution freq of 10GHz would not be the best case for your 4, 5, 6 and 6,5GHz frequencies.
So, I essentially need to run 4 different solution setups (and thus simulations) in order to do this? This isn't ideal, but I'll give it a shot and compare the results.
Thanks!
If the simulation is wide band then it would be advisable to do more than one simulation.
Adaptive meshing is done in the field domain, i.e. in the 3D space where the fields are calculated. So, it is important where the maxima and minima of the fields are located, because that's where the changes of the fields are more dynamic then in the other areas. So, the areas where you will have the finest meshing will not be the areas where you really need it, if you change the working frequency a lot.
I hope I was clear enough.
Regards,
Drasko
What I'd do in your case is to start the simulation at the lowest solution frequency (4GHz in your case), then add a 5GHz solution frequency which uses the mesh from the converged 4GHz solution, then add a 6GHz which uses the converged mesh from the 5GHz and so on.
Then I add a broadband (4-6,5GHz) frequency sweep to the last solution frequency (6,5GHz in your case).
I would expect that the mesh is good at every solution freq, so broadband sweep (discrete of course) would give good results.
good idea... btw, how do we re-use meshes/solutions from other analysis setups? is there a setting for that? i never tried that...
@Zangetsu57: what i would do in this scenario is: set up the solution freq: 6.5GHz. add a fast sweep for 4-6.5GHz. for the airbox (vacuum) set the box size to be at least λ/4 far from the radiator at 4GHz, so that your airbox will be safe to be used at 6.5GHz.
"good idea... btw, how do we re-use meshes/solutions from other analysis setups? is there a setting for that? i never tried that... "
Under "solution setup" there is a tab "advanced" where you can set "use current mesh from:" ...
Thank you for the help everyone!
@rfmw, I will definitely try that. Is HFSS smart enough to mesh the 4 GHz patch more finely than the other 3 higher resonant patches at a 4 GHz solution frequency, though? Either way, I'll check out the results.
Right now I am assuming the patches to be infinitely thin planes. Do you think accounting for the thickness would significantly increase the simulation accuracy without adding lots of simulation time?
Another thing I was thinking about was I have not accounted for the solder ball on the top patch where the probe feed connects. In everyone's experience, might this affect the response?
Thanks again!
Added after 33 minutes:
Also, I just realized that you recommend a discrete frequency sweep. Is the fast sweep just too inaccurate? My worry is that the discrete sweep will take far too long to compute, and isn't very time efficient if I want to tune my design.
"Is HFSS smart enough to mesh the 4 GHz patch more finely than the other 3 higher resonant patches at a 4 GHz solution frequency, though? Either way, I'll check out the results."
Might be, that the 4GHz patch will have the greatest delta-S, so the mesher should improve the mesh mostly around it. Others might add something more useful here.
"Right now I am assuming the patches to be infinitely thin planes. Do you think accounting for the thickness would significantly increase the simulation accuracy without adding lots of simulation time?"
No, as long as you have relatively thick substrate and thin copper sheet.
"Another thing I was thinking about was I have not accounted for the solder ball on the top patch where the probe feed connects. In everyone's experience, might this affect the response?"
This is a good question. It definitely has effect on the antenna, but how much?
"Also, I just realized that you recommend a discrete frequency sweep. Is the fast sweep just too inaccurate? My worry is that the discrete sweep will take far too long to compute, and isn't very time efficient if I want to tune my design."
I have very bad experience with the fast sweep. Usually, from my experience, the results were totally off. Discrete sweep does take long, but it gives accurate results. The comparison between the MWS TD CST solver and discrete sweep in HFSS is very good. I can confirm that myself. Sometimes a discrete sweep of my projects lasts for 2 days, at least.
When you use fast frequency sweep, the meshing is done at the solution frequency, and an extrapolation is done. Hence, you would want your solution frequency to be in the middle of your band of interest.
Now if your antenna has multiple resonances, then you may have to shy away from using fast frequency sweep.
When you do discrete sweep, the meshing is done only once at the solution frequency, and the remaining frequencies are solved based on the solution frequency mesh. Hence, for discrete sweep, it is advisable to you put the maximum frequency as the solution frequency.
When using a discrete sweep, typically how many points would you simulate? Say I have a 4GHz patch (like my case) and am interested from 3.9 to 4.1 GHz as I expect the resonance to be within that range. Should I use a spacing of 0.01 GHz, 0.1 GHz, or what?
My experience is only with the fast sweep.
Added after 13 minutes:
After doing some reading, it looks like an interpolation sweep might be best suited for what I am trying to do. It also only requires meshing at one frequency. I'll try this as well.
Luckily with HFSS, you can manually add frequency points after a frequency sweep.
That is, if you do a simulation with 5 frequency points in your band, and you do not see any resonance peak, you can add more points to the sweep. When you increase it to 10 or 15 points, HFSS automatically solves for the unsolved frequencies using the old mesh.
I would suggest you start with 100MHz.
actually, peleda is right. in hfss help, i always read similar things to what he has said above...
now, what about the interpolating sweep? interpolating sweep also selects frequencies automatically, and it may be useful to find the resonant frequencies. the bad thing about the interpolating sweep is, it doesnt' save fields at your sweep frequencies (you may not need fields maybe, then that's fine). what do you guys think?
So I found out some interesting information after running some tests. When keeping a single solution frequency of 7 Ghz, I did both an interpolation and fast sweep, and got exactly the same results even after the interpolation converged. This was unexpected, but good.
I did notice, however, that my results are slightly more accurate for each resonance when I do a separate mesh for each patch. The difference isn't huge, though. I am unsure whether or not it is worth the trouble and extra time as it takes 4 times the time to simulate (about 40 minutes as opposed to 10 minutes) since 4 meshes are required.
How long does discrete sweep take? 40 minutes?
I believe meshing should only be done once for discrete sweep.
The entire simulation (solution and frequency sweep) takes about 40 minutes total for 4 meshes and 4 sweeps. I simulate to a max mag delta S of appx. 0.02, which seems sufficient for my case. This ends up being about 65,000 tetrahedra.
On the other hand, a single mesh at 6.5 or 7 GHz and one large sweep takes about 10 minutes.
I also found that if I set the max mag delta S lower (to say 0.005) I get results that are very close to the independent meshes. This adds only a small amount of simulation time as well. I'll most likely continue with this route.
Does this answer your question, peleda?
This is where some art comes into the picture. I would be inclined to run a solution frequency at 6 Ghz or so and then do a fast sweep. I assume that you want to look at both radiation and s-parameters. 4 GHz to 6.5 GHz isn't so wide a bandwidth that I would anticipate problems with a fast sweep.
The results are best closest to the solution frequency so if you have particular concerns pick additional frequencies.
For good radiation pattern fidelity you might also want to use a virtual object and seed it to force the mesh to refine near the edge of your antenna.
I think the bottom line is:
Don't ever think that you can predict for sure how you deal with a structure with which you have no prior experience using the simulator.
What I am saying is: you must start the design, get an initial result and then verify it in some way by increasing the accuracy of the model/algorithm. If you don't do this verification, you just won't be sure how reliable your results are.
I would recommend this approach for any simulator, because this is good engineering practice and common sense. When you acquire experience with some type of models, than you will be able to predict how you need to approach those models, and you will verify results more quickly... That' s the process.
Knowing more about the algorithms behind the program helps you learn how to check and improve your results. The software is just a tool with a character of its own.
Regards,
drasko