HFSS User Experience (Excerpt)

tmily

HFSS User Experience (Excerpt) [Copy link]

Similar to most large-scale numerical analysis software, Ansoft HFSS based on the finite element method is not a fool-proof software. For most problems, in order to obtain fast and accurate results, human intervention is necessary. In addition to being very clear about the details of the model, the modeler himself is also better to have a certain foundation in electromagnetic theory. It is assumed here that the reader has used HFSS, so some basic operations are not mentioned.
1. The use of symmetry
For a specific high-frequency electromagnetic field simulation problem, you should first see whether it can use symmetry surfaces. The constraints here are mainly in the requirements of geometric symmetry and excitation symmetry. If the excitation of a problem does not require to be changeable, such as an array with all in-phase feeds, it is best to use symmetry at this time, and even two symmetries (E and H symmetry) can be used, which will greatly save time and equipment resources.
2. The use of surfaces
In practical problems, many structures can be replaced by 2D surfaces. The advantage of using 2D surfaces is that it can greatly reduce the amount of calculation and the results are almost the same as using 3D entities. For example, when calculating a microstrip branch line coupler, the microstrip and ground of the printed circuit board can specify certain surfaces as ideal electrical surfaces instead, so that the required physical size and performance can be quickly obtained. Taking the calculation of dipole as an example, if the dipole is a cylinder made of an ideal conductor, then under the same other conditions, its calculation time is about 4 to 5 times that of using an equivalent surface as a dipole. This shows that in general.
3. The use of Lump Port (centralized port)
provides a new excitation in HFSS8: Lump Port. This excitation avoids the establishment of a coaxial or waveguide excitation, thereby reducing the model volume and calculation time to a certain extent. LumpPort can also be represented by a surface. It should be noted that the calibration line and impedance line of the port must be set accurately, and the port must be connected to other metals (or electrical surfaces) in space, otherwise the result is very easy to make mistakes.
4. Regarding the problem of radiation boundary,
when there is no need to solve the near (far) field problem, such as the filter sealed in the metal box and other closed problems, there is no need to set the radiation boundary. When it is necessary to solve the field distribution or directional pattern, the radiation boundary must be set. There are some issues that need attention here: when calculating large bandwidth periodic structures, such as 3 octaves, it is best to calculate in segments, for example, one octave as a segment, that is to say, set radiation boundaries of different sizes when calculating in different frequency bands, otherwise it is difficult to ensure the calculation accuracy at the frequency edge of the calculation; secondly, the size of the radiation boundary is closely related to the specific shape of the problem. If the outer contour of the object can be contained in a sphere or an ellipsoid that is not too large, it is advisable to use a cubic boundary - simple and effective. If the outer contour of the problem is more complex or the difference between the two axes of the ellipsoid is too large, a similar boundary or a cylindrical boundary can be used. For radiation problems, if the estimated gain of the problem is low (for example, 2dB), then the boundary should be spherical. At this time, in order to obtain accurate results, time has to be sacrificed; in addition, a new absorption boundary is provided in HFSS 8 - PML Boundary conditions. I am not very satisfied with this boundary. Although its effective distance is one eighth of the central wavelength, which is half of the old boundary, it can reduce the amount of calculation. However, this boundary is generated by the program itself and is a complex structure of a cube. For some special and complex problems, there is a lot of useless space inside this boundary. At this time, it is better to use the old boundary.
5. About opening holes
. Some problems require opening holes on the wall. At this time, there are two ways to use. One is to hollow out the model honestly; the other is to use H/Natrue boundary conditions. Usually, if the hole is opened on the surface, the latter will be used. It is simple and easy to modify.
6. About meshing.
After the model is established, enter the calculation module. The first step is to mesh the problem. For general problems, it is more worry-free to let the software automatically divide, but for large and complex problems, letting the software divide itself may require a lot of patience to wait. According to practical experience, manual division can achieve good results in densely structured areas or field-sensitive areas of large models. The calculation results of some problems begin to converge, but further improve the accuracy, but rebound. The problem is that the mesh division of the field-sensitive area is not careful enough at the beginning, resulting in deviations in the calculation results.
7. When calculating the problem with the required accuracy
, it is generally necessary to give a convergence accuracy and number of calculations to avoid the program "getting stuck in calculations". When you are familiar with the model, you can rely on the given number of times. At the beginning of the problem, it is recommended that the calculation accuracy should not be too high. In practice, some operators have set the S parameter accuracy of the problem to 0.00001. In fact, this is completely unnecessary. Generally, the accuracy of the S parameter is set to about 0.02, which can meet the needs of most problems (at this time, you should pay attention to whether there is convergence rebound). If it is the number of calculations, for closed problems, it is recommended to set it to 8 to 12 times. For radiation problems, generally 6 to 8 calculations can be performed to observe the results. If it is not enough, decide whether to continue the calculation.
8. About scanning
HFSS provides a scanning function, which is divided into 3 modes: fast, discrete and interpolation. Discrete scanning only retains the field results of the last frequency point, and its calculation time is the sum of the calculation time of a single frequency point; for fast scanning, all frequency field results within the calculated frequency range can be obtained, but its calculation speed has little to do with the number of frequency points, and is basically proportional to the complexity of the model. Sometimes the scanning calculation time is very long. If you do not need to care about all the field conditions, it is recommended to use discrete scanning. For particularly large problems, fast scanning is appropriate. The interpolation method is less used.
9. Regarding the scale of the problem
The scale of the problem that HFSS can calculate is closely related to the computer hardware, followed by the operating system used. In HFSS8, the order of the problem description matrix is basically proportional to the number of grids, and it can easily handle 100,000 problems on tetrahedrons (as long as the machine is good enough). However, this does not refer to the electrical size of the actual problem. In fact, the time and resources required to accurately calculate a computer network cable connector (RJ45) are not much less than those required to calculate a general radiation problem with an electrical size of one wavelength. Therefore, the main constraint on its calculation scale is the complexity of the problem, and the complexity includes factors such as electrical size and structural complexity. This reminds us that we should try to simplify the model when modeling. Generally speaking, except in the excitation area, when the electrical size of the structure is smaller than one twentieth of the wavelength, its existence can be ignored without introducing obvious errors. This is very effective at the beginning of solving the problem and can quickly find the key to the problem. When the main requirements of the problem are met
, the model can be refined to obtain more accurate results.
The above summarizes some experiences in various aspects of HFSS use for your reference. If there is anything inappropriate, please communicate and correct me. Thank you in advance.