Richard Clark wrote:

As for automated optimization, NASA spent huge bucks on this stuff to
design twisted paper clips to replace Walt's work of 30 years ago. I
can well bet that license runs pages. If the testimonials to Beezley
are any indicator, the utility of the software is in inverse
proportion to the length of its license.

73's
Richard Clark, KB7QHC

However some of it does work very well. YO, when used correctly can
just barely beat K1FO's designs, which were done on a PDP11, using a
special version of BASIC, as I remember from conversations with him long
ago.

He got the designs as right as possible, using an EME'rs version of
right. He hit max gain for boomlength within less than 1dB, pattern is
wonderful, SWR BW is astonishing, and pattern and gain are all fairly
constant across the usable SWR BW. Input impedance is not too low, at
about 20-25 ohms, and efficient match can be had with a T-match. And it
handles ice and rain detuning perfectly. Build sensitivity is nice; you
can skew the design by induced errors of +-2mm element length and +-5mm
vertical off the boom and +-2mm element position on the boom with no
significant change. Gain not off by .1dB, pattern not off by 2dB,
normally less. I ran a lot of tests. And I could be misremembering a
bit, but probably by too high rather than too low.

And YO could beat K1FO by only hudredths of a dB. If he'd had more CPU
power to do more runs per day...

tom
K0TAR

On Wed, 12 Jul 2006 21:16:29 -0500, Tom Ring
wrote:

He got the designs as right as possible, using an EME'rs version of
right. He hit max gain for boomlength within less than 1dB, pattern is
wonderful, SWR BW is astonishing, and pattern and gain are all fairly
constant across the usable SWR BW. Input impedance is not too low, at
about 20-25 ohms, and efficient match can be had with a T-match. And it
handles ice and rain detuning perfectly. Build sensitivity is nice; you
can skew the design by induced errors of +-2mm element length and +-5mm
vertical off the boom and +-2mm element position on the boom with no
significant change. Gain not off by .1dB, pattern not off by 2dB,
normally less. I ran a lot of tests. And I could be misremembering a
bit, but probably by too high rather than too low.

Hi Tom,

This is all pretty significant stuff. Its success probably ties in
with what Reggie had to say about the quality of automated software
being tied to the competence of the user/designer (Reggie may wish to
distance himself from my paraphrase however).

As a negative example, some half decade or more ago we had a fractal
designer who threw as much computational horsepower at this as his
budget would allow in hiring eager, bright faced graduates building
parallel processors. They perhaps knew Genetic Algorithms (the hot
topic in academia whose bloom had long faded in cut-throat industry),
but certainly nothing about the bajillion degrees of freedom in
antenna design. Well, that stack of computers was more a marketing
paper weight than a design producer - I've never seen an independent
headline announcing the dawn of a new age of fractals in Boston. In
fact, it would seem that same NASA program stole their thunder - and
it is still a yawn.

73's
Richard Clark, KB7QHC

Tom Ring wrote:

And YO could beat K1FO by only hudredths of a dB. If he'd had more CPU
power to do more runs per day...

YO7 includes a model of K1FO's 40 element 70 cm Yagi. In YO7, it
measures:

Gain 20.93
F/R 24.01
Z 21.8 + j5.4
SWR 1.0
Gain FOM -0.4 (versus theoretical limit for a given boom length)

In 10k iterations (minutes on a 250 MHz Pentium II), YO7 produced:

21.24
19.54
20.9 + j46.1
1.0
0.0 (I stopped it when it reached this)

You can tweak for Gain, F/R, etc depending on how you weight
performance tradeoffs. Looking at the current distribution, it appears
fewer elements might result in a better design. YO7 does not optimize
for number of elements but it doesn't take much effort to remove
elements and see what happens on the same boom length.

Regarding K1FO's design using Basic on a PDP11, here's Brian's
description:

************************************************** *********************************
YO includes an automatic optimizer that can maximize
forward gain and input resistance, and minimize backlobes,
sidelobes, and SWR. The optimizer iteratively adjusts element
lengths and spacings to optimize performance objectives you
specify using parameter tradeoffs you decide. It can perform
both local and global optimization.

YO is calibrated to NEC, the reference-accuracy
Numerical Electromagnetics Code from the Lawrence Livermore
National Laboratory. YO and NEC results normally differ by less
than 0.05 dB in forward gain, a dB or two in F/B, and a couple
ohms in input impedance. You can invoke NEC from within YO to
verify results.

YO's analysis and graphics engines use assembly language
with pipelined floating-point code optimized for Pentium processors.
************************************************** **********************************

The entire yo.exe program is only 82k (and DSP Blaster is only 16k).
The major change from YO6 to YO7 was the addition of global
optimization, so it will not get stalled on local maxima.

The other thing I must give Brian credit for is his excellent command
of English. I have *never* seen a spelling or grammatical error in any
of his documentation, which is very unusual these days.

73, Bill W4ZV

wrote:

Tom Ring wrote:

And YO could beat K1FO by only hudredths of a dB. If he'd had more CPU
power to do more runs per day...
snip

The other thing I must give Brian credit for is his excellent command
of English. I have *never* seen a spelling or grammatical error in any
of his documentation, which is very unusual these days.

73, Bill W4ZV


Oh, I give Brian lots of credit. He did a fantastic job. And there
were times his developments happened at a blistering pace.

I've had more than a little experience with YO and AO. I was an alpha
tester for him. I once ran an optimization for 3 weeks straight in AO.
Unfortunately it was a dead end design idea for an odd dual band
yagi. I digress.

When I said you could only get hundredths of a dB, I meant it. I did
hundreds of models and thousands of runs trying. You are probably not
optimizing to the specs that an EME'r would desire. Setting the
tradeoffs in YO to get the balance right is touchy, and changes as boom
length increases. You also need to partially or completely turn off
optimization on a few elements for it to do its best job on a long yagi.

The K1FO yagi designs are still the best around in my opinion. For one
reason above the fact that they have great specs - they are a very high
performance design that is easy to reproduce. And when in an array they
do not detune as much as "better" designs.

tom
K0TAR