Next: Models generation Up: Frequently asked questions for Previous: Contents

Subsections


General

1.1. What does the abbreviation FEKO stands for?

FEKO is an abbreviation derived from the German phrase ''FEldberechnung bei Körpern mit beliebiger Oberfläche'' (Field computations involving bodies of arbitrary shape)

1.2. What numerical formulations are implemented in FEKO?

FEKO implements a frequency domain Method of Moments (MoM) solution of the Electric Field Integral Equation (EFIE). It allows both wire segments and triangular surface current elements. In addition it is fully hybridised with the Physical Optics (PO) and Uniform Geometrical Theory of Diffraction (UTD) -- flat plates only -- formulations at the level of the interaction matrix.

1.3. Is the licence a one-off purchase?

Yes, once bought the FEKO licence will not expire. The purchase also includes one year of support. After this there is a maintenance fee for support and updates. Note that, in general, the passwords do expire and have to be renewed each year.

1.4. Do EMSS offer a training course?

Yes. The course covers basic discussions on numerical electromagnetics as well as how to set up and solve problems in the FEKO environment. It may be presented at EMSS in Stellenbosch or at the clients own location.

1.5. Is FEKO Y2K compliant?

FEKO is a numerical analysis program and are, in itself, not sensitive to computer dates (apart from the obvious dates associated with password expiration, etc.) The operating systems it runs on might, however, be sensitive to computer dates.

1.6. What languages are supported?

The user manual and program output is available in both German and English. Support is generally only available in English.

1.7. What platforms are supported?

The current user interface runs only under MS-Windows (both 95/98 and NT, but it is more stable under NT). The preprocessor and the solution engine runs on a number of platforms:
PC versions
MS-DOS, MS-Windows (both 95/98 and NT), Linux, and OS/2 (version 2.0 and higher).
Unix versions
DEC, HP (PA 1.1 and PA 2.0 architectures) IBM, SGI (32- and 64-bit versions), SUN
Parallel version
Cray T3E, HP, Parallel Linux clusters, SGI
Currently Windows NT on Dec Alpha platforms is not supported.

1.8. What are the system requirements?

The user interface requires a screen resolution of at least 800x600 and preferably more than 256 colours. The memory depends on the size of the model under consideration, but one should have at least 15 MByte of allocatable memory. For the solution engine, the memory requirement is entirely dependent on the problem size. In general, a MoM problem with N basis functions requires 16 N2 Bytes of memory. For metallic surfaces, one unknown is associated with each internal edge between two triangular patches. Double this for dielectric surfaces and divided by two for each plane of symmetry. (For closed surfaces the number of internal edges are 1.5 times the number of triangles. For more complex structures it is less than this.) For the PO formulation the memory required is proportional to the number of triangles rather than its square and is generally less of a concern.

1.9. What are the memory limits.

All 32-bit operating systems should support up to 2 GByte. The HP versions are limited to 1.9 GByte. The 64-bit SGI and DEC versions support 263 bit -- practically all the memory and disk space that is available. If the problem size is larger than the amount of allocatable RAM -- up to the limits described above -- a very efficient out-of-core solution is used. (Note that this is much more efficient than the system paging.) In the parallel versions, each process can address the amount of memory specified for the sequential versions.

1.10. How much will I gain by the using parallel version of FEKO?

Many practical examples on different parallel computer systems have shown that the scaling of FEKO is very good. Typical parallel efficiencies are 90% or higher. The out-of-core solution used by the sequential version is very efficient and can be used to solve large problems. However, it cannot make optimum use of symmetry during matrix fill, but this is usually not the largest contributer to the solution time. Secondly it cannot treat multiple right hand sides -- this is a big disadvantage for calculations, such as monostatic RCS computation, where a large number of excitations are used. On 32-bit systems the parallel version also have the advantage that each process can address the same amount of memory as the sequential version.