Hello all the FDTD experts over there ,

I am starting out on my class project which requires me to write
a 2D FDTD code.I am thinking of doing it in Matlab or C.Matlab would offer
good user friendliness but I have heard that Matlab being an interpreted
language is quite slow when it comes to loops.My requirement is about 4000
cells utmost ( 200 x 200 ) and time step of the order of 1 Pico second.Will
Matlab be very slow that I should consider coding in C or will Matlab do
okay with respect to the execution time.



-Warm regards
--Arun S V

"Arun S V" wrote in message
...
Hello all the FDTD experts over there ,

I am starting out on my class project which requires me to
write
a 2D FDTD code.I am thinking of doing it in Matlab or C.Matlab would offer
good user friendliness but I have heard that Matlab being an interpreted
language is quite slow when it comes to loops.My requirement is about 4000
cells utmost ( 200 x 200 ) and time step of the order of 1 Pico
second.Will
Matlab be very slow that I should consider coding in C or will Matlab do
okay with respect to the execution time.



-Warm regards
--Arun S V



Don't think about it, do it. You'll find out soon enough if it's too slow.

I did one several years ago using the interpreted Visual Basic that comes
with Excel, letting Excel do the plotting. It seemed to update pretty
quickly. Matlab would probably be fine. C is great if you're comfortable
with plotting, otherwise you might spend all your time fighting the
interface. Here are some references, if you're interested:

References:



1. R. C. Booton, "Computational Methods for Electromagnetics and
Microwaves" John Wiley & Sons, Inc. 1992. This text gives an uncommonly
clear presentation of the basic theory, although it lacks enough detail to
set up practical problems of the type that we selected. The FDTD section
discusses primarily 1- and 2-dimensional problems.



2. K. S. Yee, "Numerical Solution of Initial Boundary Value Problems
Involving Maxwell's Equations in Isotropic Media," IEEE AP-14, 1966. This
is the classic paper to which all subsequent writers refer.



3. Sadiku, "Numerical Techniques in Electromagnetics." This book
provided a good level of detail for setting up FDTD problems. It provides a
good indexing scheme that makes for efficient implementation of Yee's
algorithm. This book also gives a limited discussion of absorbing boundary
conditions and briefly discusses options for feed waveforms. There is also a
brief note about reducing run times by filling the space with a low
conductivity, therefore reducing the effects of multiple reflections.



4. Omiya et. al. "Design of Cavity-Backed Slot Antennas Using the
Finite-Difference Time-Domain Technique," IEEE AP-46 Dec. 1998. This paper
provided some more detail on the problem setup, showed sensitivity of
accuracy to the number of time steps, and discussed a windowing technique
that could be used to shorten run time. It seems that this windowing in the
time domain ends up being quite similar to Sadiku's introduction of loss in
the cavity. The mathematical operations are very similar, even though the
justifications are different.



5. The Internet site "FTDT.org" has some good information and
programming examples. The content matches the information in the above
references, with more effort put into formatted output.



6. Jeong Phill Kim, "FDTD Basics and Programming Tips," March 22,
1999. This is a very clear, concise treatment of the subject with
sufficient detail to begin a practical analysis.



7. R. J. Luebbers and H. S. Langdon, "A simple feed model that
reduces time steps needed for FDTD antenna and microstrip calculations,"
IEEE AP-11, July 1996. This paper describes a feed loading technique, used
by Kim in reference 6 that greatly reduces the convergence time.



8. Cheng, Dib, and Katehi, "Theoretical Modeling of Cavity-Backed
Patch Antennas Using a Hybrid Technique. The configuration studied in this
paper is similar to the one studied in the final configuration of this
study.



9. Balanis text, "Antenna Theory Analysis and Design," 1982, Harper
and Row. There is a good description of the application of Huygen's
Principle on pages 447-456.

In message , Arun S V
writes
Hello all the FDTD experts over there ,

I am starting out on my class project which requires me to write
a 2D FDTD code.I am thinking of doing it in Matlab or C.Matlab would offer
good user friendliness but I have heard that Matlab being an interpreted
language is quite slow when it comes to loops.My requirement is about 4000
cells utmost ( 200 x 200 ) and time step of the order of 1 Pico second.Will
Matlab be very slow that I should consider coding in C or will Matlab do
okay with respect to the execution time.

You can compile Matlab code as one of the options at least in Version 6
That speeds it up considerably. You can handle matrix maths without
element by element addressing - I don't think C does that.
I'd use FORTRAN myself


Brian GM4DIJ
--
Brian Howie

I'm not sure what calcs you need to run, but wouldn't it be trivial to
do a test of Matlab's speed with respect to the sorts of things you
need to do? It's one of the very nice things about an interpreted
matrix language. I use Scilab (freeware), and just ran a little test:
create a=rand(200,200), b=rand(200,200), c=a*b, ainv=inv(a), and then
evaluate ainv*c-b, to make sure it comes out close to zero. It did,
and the whole thing was fast enough that I couldn't time it accurately
till I put it in a loop to repeat ten times. It's about 1.4 seconds
per iteration. Guess I could have let it time itself, too. That's on
a 700MHz Pentium machine. (Considering the work I put into inverting
matrices in my linear algebra class years ago, the speed with which it
can invert a large matrix puts a wry smile on my face.) Scilab, as
Matlab, can compile functions, but if your math is very compact
because of the matrix notation, it may not really matter much; most of
the time is spent doing the matrix math anyway. And if the matrix
notation lets you do things in a way that's more readable, that's a
huge plus, and also consider the time you'll save programming (unless
you already have good matrix libraries available, I suppose).

(By the way, last time I checked, 200*200 = 40000, not 4000. ;-)

Cheers,
Tom



"Arun S V" wrote in message ...
Hello all the FDTD experts over there ,

I am starting out on my class project which requires me to write
a 2D FDTD code.I am thinking of doing it in Matlab or C.Matlab would offer
good user friendliness but I have heard that Matlab being an interpreted
language is quite slow when it comes to loops.My requirement is about 4000
cells utmost ( 200 x 200 ) and time step of the order of 1 Pico second.Will
Matlab be very slow that I should consider coding in C or will Matlab do
okay with respect to the execution time.



-Warm regards
--Arun S V

Thanks a lot to Dave Brian and Tom.I wrote out a simple 1D FDTD code with
PEC boundary condition , Gaussian excitation in Matlab. ( Well , I did not
have to worry about multiplying the no. of cells in each dimension to get
the total number of cells as Tom pointed out !! I will keep in mind that
200x200 is 40000 and not 4000 :-) ). It came out rather nicely. .I guess I
will try to develop my code in Matlab and try to port it to C for later
versions , if needed.Thanks for your comments and Dave , thanks for
suggesting the papers to study.

-Warm regards
-Arun S V

Arun S V wrote:
I am starting out on my class project which requires me to
write a 2D FDTD code.I am thinking of doing it in Matlab or C.Matlab
would offer good user friendliness but I have heard that Matlab being an
interpreted language is quite slow when it comes to loops.

We (myself and other classmates) did an entire numerical electromagnetics
class in Matlab. It's true that Matlab 'for' loops are slow... usually the
execution time of the programs was completely dominated by actually 'filling
up' the matrix (or cells) rather than, say, inverting the matrix, updating
cells, etc. Even so, for a 200x200 FDTD program on any contemporary PC
(1GHz), Matlab is plenty fast and I'd argue the ease of
plotting/manipulating the results more than outsweighs the generally
ignoranble speed penalty.

Heck, I did a FDTD program in C back in 1994 on some PA-RISC HPUX
workstation, and I'm _sure_ Matlab is a lot faster on a modern PC than C was
back then, and back then it was certainly fast enough!

BTW, look into parallelizing your code... sometimes it's a lot more trouble
than it's worth, but often it's very easy to get Matlab to fill in the
results of an entire row or an entire matrix in a single statement if you
code it correctly.

---Joel Kolstad