I encounted one problem concerning the S parameter simulation frequency step.
To check the simulation with experimental results, I have done two adaptive sweep type simulations consecutively, with all other parameters being the same.
1) frequency range 0.1~8GHz. Results agree quite well, but 0.7~0.9GHz range has relatively big difference. So I
change and narrow down to the desired range, and (2) simulation is conducted.
2) frequency range 0.7~0.9GHz.
In the rational fitted raw S parameter results (touchstone file), I found out that, the case (1) has coarser discrete points
than case (1). Which puzzled me quite a bit, as I thought the narrowed down frequency range would result in denser points.
Then, I tried to run the linear discrete sweep type rather than adaptive sweep type in case (2) simulation, however, due to memory problem, simulation failed.
My questions are
(a) In the adaptive sweep type, if I want to get denser points result, what should I do? Reduce the delta error, or increase the sample points limit? Is it possible to set the desired frequency points in the adaptive sweep?
(b) In the discrete sweep type, using iterative rather than direct solver would consume less memory, are there any
other solutions to reduce the memory consumption?
The best way to force more points in an adaptive sweep is to add a linear sweep segment to the same simulation (you can have multiple sweeps defined in a single simulation).
One thing to watch out for is to make sure the mesh is refined at the proper frequency. The default is the highest frequency in the sweep, which is usually good for broadband applications, but not the best for applications where there are resonances. In your case you may have refined the mesh at 8GHz (the highest frequency in your first simulation). Then when you ran the next simulation (0.7-0.9GHz) the mesh may have been refined at 0.9GHz (assuming you left the mesh refinement settings at the default). But if this was near a resonance the optimum mesh might be quite different (and much more dense) than the first mesh at 8GHz, and this is what may have caused you to run out of memory in the second simulation. But this is probably a better mesh to use if you want to accurately capture a resonance. So in general you may want to try the mesh refinement frequency setting "Chosen automatically after initial pass", which will look for resonances during the initial mesh refinement pass and then use a more appropriate frequency to refine the mesh than the highest frequency.
If you are running out of memory during mesh refinement, the best approach is to switch to the iterative solver (which by the way is much faster in recent releases and is now multithreaded). You could also increase delta error but this will reduce your accuracy of course.
thanks a lot for your detailed reply. Actually I tried what you suggested and unfortunately, no improvement.
I added the linear sweep into the adaptive sweep, 0.7G to 0.9G, having 401 points. Using the iterative
solver. Also the mesh is refined "after initial pass". Delta error is set the same with previous simulation.
This time, the memory overflow problem occurred in direct solver did not happen. However, it really took
150 hours to do the calculation. adaptive only sweep took around 2 hours for the same simulation setting.
When I checked the results, viewing the exported touchstone file. I found the frequency step is 4.4 MHz, not the expected 0.5 MHz, as in the linear sweep (401 points for frequency range 0.7~0.9GHz).
Do you have any ideas about this?
p.s., what I need to add is that, I am doing the package simulation for ladder filter (comprised of resonators).
However, resonator part is not included in the EM simulation. In other words, in EMPro simulation, there is
no resonance part in the whole geometry.
When you select the mesh refinement option "after initial pass" it will create a mesh at every defined point in the first adaptive pass and then choose the refinement frequencies. For an adaptive sweep the number of points is usually small, but if you defined a linear list of 401 points, it will create a mesh at all 401 points. So this explains why the meshing process took so long. If you are creating a frequency plan with this many points, it would be better to use the mesh refinement option "manual selection". Then select the highest frequency, plus one or more points of interest within the narrow band (0.7-0.9G). Then it will only refine the mesh at these frequencies and save a lot of time. After the mesh refinement, it will still solve at each frequency, so you may also want to consider reducing the number of points (say, 51 instead of 401).
When you plot or export the results you have 2 choices. If you plot/export the "discrete frequencies" data (filter on "domain" in the results browser), you will get just the points in the frequency plan (the adaptive sweep, plus the linear sweep). If you plot/export the "frequency" data, you will also get the interpolated data points that are calculated after the simulation is complete. In either case the exported data should have inlcluded the 401 points from your linear sweep, did this not happen for you?