When a result is obtained with Feko a regular consideration arises as to the accuracy of the results. This article provides some guidelines for consideration.
Results obtained with FEKO can be confirmed in the following ways:
Compare with a measured result
Compare with an analytical result
Compare with another solution method
Perform a mesh convergence study
When Feko is compared with a measurement, it is assumed that the conditions and environment of the measurement is the same as the Feko model. For example, a trustworthy comparison would be comparing the gain of a horn antenna in Feko with that obtained by a three antenna measurement of the horn inside an anechoic chamber.
The horn antenna in Feko is usually modelled in free space, meaning there is no reflecting/returned EM radiation. This condition is mostly true inside an anechoic chamber where the frequency of interest is well within the specification of the chamber.
Measurement and simulation conditions could be mismatched in any or a combination of ways (note the below lists only a few examples and is by no means exhaustive)
It is assumed that cables or measuring equipment do not affect the measurement. Common mode currents are often a culprit in sensitive measurement environments.
If a ground plane was used, then the same size and shape ground plane may need to be used in the Feko model – an infinite ground plane is computationally more efficient than a finite one, but ignores edge effects.
Do the losses (dielectric, metallic) of the measurement match those of the Feko model?
If a wire port is used to feed, for example, an antenna or microstrip, the wire radius is important. Firstly the wire radius determines the reactance of the feed. For example, if the physical model is fed by an SMA connector, the radius should be set to 0.65 mm. Secondly, the solver requires a certain ratio between the radius of the wire and the meshed wire segment length, as well as between the radius and the size of the meshed triangle element it connects to, if applicable. For more information, see the Meshing chapter in the Appendix of the Altair Feko User Guide.
If warnings are given for the wire radius and/or the connection, consider changing the wire to a cylinder, meshed into triangles. In this case an edge port could be used instead of a wire port. It may also be wise to model a section of coaxial line which can be excited with a waveguide port.
Analytical results exist for some problem types. One example is the bistatic RCS of a PEC sphere that is exactly known and represented by a Mie series.
A stripline feed network can be compared with analytically computed results using non-radiating transmission line theory due to the low radiation properties of stripline.
Confidence in Feko results can be increased by using another solution method. For example, in antenna applications or bio-EM, the solver or method by which the dielectrics are modelled can be interchanged between the default SEP method, FEM or VEP.
When using iterative solvers, such as the MLFMM, sometimes the iterative solvers fails to converge. For example, when solving monostatic RCS, sometimes one of the angles in the angle range to be solved only converges partially. Then a good verification would be to solve the aforementioned angle with a full wave method (MoM with HOBF or CBFM).
In general, once a result is obtained, it should be compared with a model that contains 50% more mesh elements. If the results agree well, then the mesh can be regarded as stable/converged. If a frequency loop was used, then in most cases the mesh can be regarded as stable/converged for any frequency up to the maximum frequency.
If computational resources are limited, or the runtime is already impractically high, a mesh convergence study can also be done by using 50% less mesh elements. However, if results then disagrees, no conclusion can be made regarding the original mesh.
When starting to use Feko for a new application, it is always recommended to build confidence with Feko by starting with simple examples, for which the results are known. Complex models could also be split into different parts than can initially be solved separately, to gain confidence. The added advantage is that smaller models always solve faster, allowing to make multiple adjustments in a short time (feed network + printed antenna array, or vehicle + cable harness + antenna systems).