Re: NEC-LIST: Antenna pattern distortion modelling

"Chatterjee, Deb" schrieb:
>
> I have run the NEC-BSC3 (containing UTD formulations). The code is
> superbly written by Ron Marhefka and his students (and the earlier
> versions by Ron Marhefka & Walter Burnside). All from ESL, OSU.
>
> The code seems, in my view, the only solution to modeling of "antenna
> pattern distorions by neighboring structures" for structures about a
> wavelength or greater. The greater the better the accuracy.

hmmm, there is another great code, called FEKO by Uni Stuttgart and
EMSS South Africa (http://www.emss.co.za) which is able to do very
nice things. In contrast to NEC, it uses patches also in the EFIE, so
open structures are possible without having the need to do wiregrid
modeling. Included in the code is physical optics, geometrical optics
and MOM, you can use them in a mixed simulation.

The above might seem like a commercial ad, however, I do not have any
interests in FEKO. I have used it several times (also NEC), I feel it
is great software, and it deserves a good review also in this group. I
very much recommend having a look at that software. I'm just enchated
by the code, especially after having seen some other numerical EM
codes.

> The advantage ? Well, the computer resources required as frequency
> goes higher remains constant for NEC-BSC, while it does increase
> almost exponentially for NEC-MOM codes.

again hmmmm. Mom increases the memory with the square of the unknowns,
with a wire code that is also with frequency. Solution time is N cube,
however I think that this part is greatly overestimated on modern
machines AND WELL PROGRAMMED CODES. NEC in its original version needs
380 s for 1944 unknowns on a particular machine. Using a well
programmed and machine optimized version of Gauss, the computation
time goes down to 72s. Filling for the same was 44s (see JvHagen,
Mittra, Werner, ACES conference this year). On my tests here I almost
acheive max performance of the HP machines which is around 800 MFLOPS
(this means that a 4000 unknowns matrix should be factored in a couple
of minutes). If your software needs more time for similar
computations, send it back to the programmer because he/she did not
his/her work.

so it does not increase exponentially, but only in a polynomial
way.
 
> This is not a problem with NEC-BSC or NEC-MOM, but a fundamental
> numerical limitation with "exact" (MOM) and "asymptotic" (BSC)
> formulations themselves.

which remains true.
 
> So if you are working around 1 GHz and structures around you have
> mean dimensions equal to or greater than 1*lambda, use NEC-BSC.
> However, you can use the NEC-MOM code about this frequency (1 GHz)
> to verify the results.

or look at FEKO where you can mix PO, GTD and MOM for even enhanced
accuracy.

cheers
juergen
Received on Thu Sep 23 1999 - 02:01:07 EDT