Adaptive meshing in FEKO
Altair EmployeeIs there a way to switch the meshing conditions between the frequency bands? I need to simulate a 5mm radius dielectric hemisphere in 1GHz to 370GHz. My current mesh is adapted to ensure correct results in upper frequency. Therefore, it is taking longer to simulate in-spite of taking advantage of symmetry option. I would prefer to use a coarse mesh at lower frequencies and then use a finer mesh at higher frequencies. Is this possible either in CADFEKO or EDITFEKO?
Answers
-
Hello,
The easiest way that I can think of is for you to try the parameter sweep plugin. You'll find it under the 'Macro library' in CADFEKO. How it would work is that you set your model up to work at a single frequency. The parameter sweep can then vary over frequency. This will result in a model that is meshed as specified for the specific simulation frequency only. Each permutation will result in a different model, which is then combined in POSTFEKO using another plugin.
At the higher frequencies, you can start to look at different solver techniques as well. This might allow you to get away with much coarser meshing without compromising accuracy. My point here is that there's no silver bullet that will guarantee good results. SEP/FEM should work well lower down in your frequency range but have a look at RL-GO at higher frequencies. Also, FDTD might be an interesting one to try - it's usually one of the more expedient ways to get wideband data.
Let me know if that helps!
0 -
As Andries mentioned, there is no easy way to do this and in general we would recommend splitting the model over a few frequency ranges so that you can adjust the meshing and possibly the solution method so that it is optimal for that frequency range.
You can create multiple models using the parameter sweep plugin as Andries mentioned. Alternative options are:
- Use the grid search (this is similar to the parameter sweep plugin, but the parameter sweep plugin is more user friendly)
- You can use EDITFEKO and ADAPTFEKO (continuous frequency sampling) with the #adaptfreq variable (see '22.3 The *.pre input file for adaptive sampling') with either the IP card (if you mesh in EDITFEKO) or RM card (if you meshed in CADFEKO and only want to refine the existing mesh). This is quite an advanced option to try.
I suspect that parameter sweep plugin that Andries suggested is the best option (given the limited info that we have about your model).
0 -
Hey @andries
Thank you for your input. Physical optics seems like the way out. It reduced my computation time and allowed me to compute near fields. However, I am not able to understand the difference between PO-full ray tracing and PO-always illuminated. There seems no difference in the result. I guess in my case where I am looking at the propagation of Gaussian pulse through lens medium it works the same way but not sure.
0 -
Great! I'm glad you're getting closer. When you're on the property dialogue where you set the solution method to PO, you can press 'F1' and the manual will open up to that position. They'll explain the differences there.
In a nutshell, though, 'full ray tracing' is the 'most correct' of the options. The 'always illuminated' option can be used if you know a little bit more about the shadow regions of your model but can give incorrect answers if you make an incorrect assumption. My advice is if full ray tracing is working for you, then stick to it.
0 -
0
-
Full ray tracing is the safest option and since ray tracing is not that time consuming, rather stick to it. This use to be a feature many years ago when computers were much slower. There are a few special cases where one of the other options can be used to improve the results, but no need to go into that here. Below is an illustrative example of what the different options mean and what the implications of the options are on results (in this case currents, but all field calculations are affected).
PO ray tracing options available in FEKO
The PO ray tracing options available are as follows:
- Full ray tracing - Slowest, rays are traced from each MoM element to each PO element. PO mesh normals can be mixed, with some pointing inward and some outward.
- Only illuminate from front - Rays will be traced only to the front of the PO mesh (implying that fields incident on the inside of PO triangles will not cause current flow). This will be faster than full ray tracing and is well suitable for closed bodies where all mesh normals point outward.
- Always illuminated - This is the fastest of all the ray tracing options. FEKO does not check for any shadowing and assumes that all PO triangles are illuminated by the source and MoM area.
Example 1: Applying the ray tracing options to a dipole and triangular shaped box
Consider the geometry below. A triangular shaped box is illuminated by a dipole antenna. The box is solved with the PO and the dipole is solved with the MoM.
Figure 2: A triangular shaped box and dipole solved with the MoM/PO hybrid 
Now consider the results if, in turn, the three ray tracing options are applied to the model. We compare these to the MLFMM solution which can be taken as the reference. In all cases the current distribution is displayed over a magnitude range of -35 dB to -70 dB.
Figure 3: Current distribution (with MLFMM as reference) for the different ray tracing options (a) MLFMM (b) always illuminated <?xml version="1.0" encoding="UTF-8"?> 
<?xml version="1.0" encoding="UTF-8"?> 
(c) only illuminated from front (d) full ray tracing <?xml version="1.0" encoding="UTF-8"?> 
<?xml version="1.0" encoding="UTF-8"?> 
We see that the cases 'only illuminate from front' and 'full ray tracing' yield identical results, but the 'only illuminate from front' will be a more computationally efficient solution.
For the case 'always illuminated' we see that the incident field causes current flow on all the PO triangles, including those that are 'shadowed' (i.e. not in direct line of sight) from the dipole. The MLFMM predicts some current flow on the back faces of the box, but the 'always illuminated' case shows a much stronger current flow on the back compared to the MLFMM.
0 -
@JIF thank you for the explanation and the attached example. However, I have one more question not related to the posted problem though. In FEKO, I know we can look at the time domain results in POSTFEKO. I also understand that it works by taking the inverse Fourier of multiplication of frequency response and frequency response of time signal. I guess this means in time domain it is a convolution of Sinc function (as frequency response would be similar to rectangular function) and say Gaussian. Therefore, we would get a response looking similar to Sinc. But in the time domain example given in the manual, the software is showing a plot with the only incident wave and reflected wave. Any idea how those results were plotted or am I thinking wrong?
0 -
The process in FEKO is as follows:
- User performs simulation in frequency domain (need to ensure that the correct frequencies are simulated - determined by time signal in next step)
- User defines a time signal in POSTFEKO and then time domain results become available.
- User can display the time signal or the system's response to the time signal in the time domain. POSTFEKO does this by:
- Calculating the FFT of the time signal
- Multiplying the spectrum of the time signal with the spectrum of the model (calculated in the first step)
- Taking the IFFT of the result (after multiplication) to produce the time response of the system with the particular time signal
Does that answer your question? Note that this topic seems to have diverged and is no longer about 'Adaptive meshing'. Please try to log new questions for new topics in the future (it makes it easier for other users to find and use the information in the future).
/emoticons/default_wink.png' srcset='/emoticons/wink@2x.png 2x' title=';)' width='20'> I'm also guilty - I should not have added all the info about PO on this thread.
0 -
@JIF Got it thanks
0
