[balm]

its Ok now, solved, thanks ....i had the 2x files in the wrong folder

 

[snowman53]

Segment Size vs Calc Results

When I first started using 4NEC2 (Nec2dXS engine), I asked in the 4NEC2 forum for a recommendation on segment size and it was suggested that I refer to the EzNEC manual. EzNEC recommends between 10 and 20 segments per ½ wave. I did a series of tests with simple dipoles to see where models converged on a “good” answer and found that NEC seemed to converge around 10 segments. So I started using Auto Segment =15 for modeling.

However, in a recent posting of a modified DBGH, others got significantly different results compared to mine. They had manually segmented the model with a segment size between 14 and 30 depending on element, with the mean being around 22.

So I did another series of runs using different segment sizes to find the segment size at which the solution converged to a valid solution. I discovered that Auto Segment = 15 significantly overstated the gain. As best I can tell it never converges to a single solution within 5 to 45 segments per ½ wave. See charts below.

Thinking it might be a problem with the DBGH model, I also ran a SBGH and a CM4228 model by Hollands. Both showed similar problems, i.e. never converging on a single value solution, with significant step changes in the results.

For comparison I ran Mmana-gal, using its auto segmentation, the gain comes closest to matching when NEC uses between 20 & 30 segments. There was not corresponding match in SWR, however the values were close enough not to see big difference in Net Gain.

In summary I think NEC is OK for comparing variations of the same antenna only if a consistent segment sizes are used.

I am not sure one can trust NEC to provide consistent values between different types of antennas which may use different segment sizes.
I would welcome a discourse on this topic with the resident modeling wizards.

 

[300ohm]

EzNEC recommends between 10 and 20 segments per ½ wave.

Yep, thats what I recall reading too, but that was about the minimum recommended. From what I understand about the method of moments theory, the more segments the better. But it becomes marginally pointless after too many and then can run into the other NEC rule violations about segments too short for the diameter, etc. And on mesh models with tons of wires, its easy to run past the 11,000 segment limit.

From your charts above it looks like Mmana-gal gives conservative figures, which is good. (but being based on mini-nec, Mmana-gal has other limitations compared to the full NEC2, from what Ive read. Looks like simplistic averaging on the conservative side.)
And it looks like about 17 to 35 segments per 1/2 wave is the magic number for uhf antennas (subject to wire diameter).

 

[snowman53]

.... From what I understand about the method of moments theory, the more segments the better. But it becomes marginally pointless after too many and then can run into the other NEC rule violations about segments too short for the diameter, etc. And on mesh models with tons of wires, its easy to run past the 11,000 segment limit.

I have experience using Finite Element Analysis typically the smaller the mesh size the better. But you reach a point of diminishing returns. So it is a tradeoff between accuracy and computing resources.

My problem with what I see here is that NEC’s solution is not necessarily getting better with increasing segments. You should not get more than 1.5 Db shift with just a small change in segment size and almost a 3 Db change over the range possible values.

It would appear that you can get almost any answer you want by manipulating the segment size.

That gives me pause when it comes to believing NEC’s data.

 

[300ohm]

EzNEC recommends between 10 and 20 segments per ½ wave.

Heh, found it also in the 4nec2 help file, under Auto-segmentation. (I thought I had read about the 10 - 20 figure in a couple of places. Probably Cebrik has something on it too. )

When enabling the Auto-Segmentation feature, you are asked to specify the number of segments per half-wave. Although, mostly accuracy improves if more segments are specified, also the computation time is increased. For ‘simple’ structures, without sharp angles, a value from 10 to 20 will be sufficient. If precise impedance values are required or if wires joining at sharp angles are used more segments could be required. If you are not sure what amount to specify, please use increasing values and compare the results. (You could do this using the Sweeper function). Also abrupt current changes on the structure might indicate an insufficient number of segments. One place more segments aren’t better is if wires of different diameters are connected.

If computer performance is an issue, you could consider not to the use the auto-segmentation feature. If this feature is disabled, the processing especially for large input files with many wires is speed-up.

One place more segments aren’t better is if wires of different diameters are connected.

Thats where it gets tricky when adjusting the Source wire to get an AGT of 1.0 (0db).
The GHs and the Bowties have the sharp angles.

BTW, thats an excellent presentation above. I sometimes had a gut feel something like that was going on, but never bothered to plot it all out.
It looks like you hit upon the unlucky number 15 where the gain really overstates the most on the DBGH and CM4228. Thats the benefit of Area 51, it allows for peer review to catch weird stuff.

That gives me pause when it comes to believing NEC’s data.

Well, I never thought the NEC data was 100% correct in the absolute sense (probably have to get segmentation down to about the molecular level), but it is correct in the relative sense for comparisons. Using the same 15 segmentation on both with the same NEC engine, you still show the DBGH significantly beating the CM4228 by pretty close to the amounts of the higher segmentations.

But you reach a point of diminishing returns.

Yep, the old rule of thumb of about 1/10 wavelength hole size still seems to apply. Any smaller, and its just squat of a difference.

 

[snowman53]

Prof

Heh, found it also in the 4nec2 help file, under Auto-segmentation. (I thought I had read about the 10 - 20 figure in a couple of places. Probably Cebrik has something on it too. )




Thats where it gets tricky when adjusting the Source wire to get an AGT of 1.0 (0db).
The GHs and the Bowties have the sharp angles.

BTW, thats an excellent presentation above. I sometimes had a gut feel something like that was going on, but never bothered to plot it all out.
It looks like you hit upon the unlucky number 15 where the gain really overstates the most on the DBGH and CM4228. Thats the benefit of Area 51, it allows for peer review to catch weird stuff.

Well, I never thought the NEC data was 100% correct in the absolute sense (probably have to get segmentation down to about the molecular level), but it is correct in the relative sense for comparisons.

I view the issue as not one of absolute accuracy (which no one should expect from any FEA analysis), but that the results should be consistent and not subject to jump shifts in results due to small changes in segment size.

In order to compare results there must be a consistent methodology, especially when the tools are this inconsistent. I would guess that more than one hand segmented model has similar problem due to the auto segmentation issues, but they are effectively hidden.

I might suggest we agree to use Auto Segmentation to create consistent segmentation and use say 20 or 25 segments per ½ wave length. Or at a minimum publish the auto segmentation value used.

But then that’s just my 2 cents.

(When did the cent key disappear from the keyboard?)

 

[300ohm]

I might suggest we agree to use Auto Segmentation to create consistent segmentation and use say 20 or 25 segments per ½ wave length.

That sounds good to me for UHF. But unfortunately on some large models we may not have that luxury.

Or at a minimum publish the auto segmentation value used.

Thats a good idea, and also note if manual segmentation was used. It not only helps others to duplicate the results, but also the modeler himself in the future to do so.

Arie Voors also noted that using a 2 degree resolution when testing for AGT is better than the 5 degree default setting. At 5 degrees, some nulls can be missed.

(When did the cent key disappear from the keyboard?)

I think it was on the Apple II. It definately was on the IBM and Burroughs terminals, heh.

¢

¢

hold down alt and on the number keypad (not the top row number) type in 0162 (alt+0162) and then another +

 

[nikiml]

Segmentation question

My question is about the "Odd number of segments" rule.
Is this necessary for every wire or it is just for the source wire?

If one wants to keep the segment lengths as close as possible, sometimes an even number may work better on some wires.

Also, when symmetry is used, a physical wire may get constructed from 2 model wires and thus have even number of segments, even though the model wires have odd. Should this be avoided?

It seems the 4nec2 Auto-segmentation assigns odd numbers always...

 

[300ohm]

My question is about the "Odd number of segments" rule.
Is this necessary for every wire or it is just for the source wire?

Definately for the source wire and preferable for the other wires, because then you always have a center segment. Also important is same # of segements for parallel wires.

The main problem I have with auto segmentation is modeling real life commercial antennas. I model them as what I see. Sometimes unless I manually segment, I run afoul of the of the other NEC rules, heh.

 

[balm]

after entering my symbol parameters, and start run optimizer I get the message: NEC ***card error invalid number at integer position 3???

 

[300ohm]

I get the message: NEC ***card error invalid number at integer position 3???

You may have an invalid, non displayable, character in that card. Delete that line and re-enter it.

 

[balm]

the only cards I have basically are SY cards, and GW cards, I used the SY card parameters in the GW lines, and the geometry view works out correctly

the only other cards are the usual ones EX, LD, EK, CM, CE, GE, GS, GN


since the geometry works, I assume its not a problem in SY, or GW cards or parameters

as for the others is there any required order

 

[holl_ands]

Yes, there is a required order for some but not all cards.
NEC documentation discusses order for some....but not all cases:
http://www.wuala.com/300ohm/Documents/

Just the other day, I had to move LD cards AFTER EX Card to avoid NEC error.
Try to follow examples known to work.

Why not post what you have so far....two eyes are better than one....

 

[balm]

Ok thanks I read up, and had to move the GS before the GE card!

 

[300ohm]

Just the other day, I had to move LD cards AFTER EX Card to avoid NEC error.
Try to follow examples known to work.

Hmm, most examples I used have the LD card before the EX card and after the EK card, and Ive been using that order without problems (that I know of).

But, thanks for the heads up.

and had to move the GS before the GE card!

That one I knew. The order that I WAS using:

CM
CE
GW
GS
GE
EK
LD
EX
GN
FR
RP

EN (seems optional as 4nec2 removes it after some runs)

 

[balm]

if I understand correctly, to adjust AGT to1, I need to adjust the source wire, either larger or smaller depending on +/- AGT result, and this should first be done for the center frequency of the band concerned


once I do this, and throws the say the extremes of the band off (AGT), then what else can be done to improve the overall AGT?

thanks

 

[balm]

if I run the regular optimizer enough, will I arrive at the same ultimate results as if I had used the slower evolver?

 

[300ohm]

once I do this, and throws the say the extremes of the band off (AGT), then what else can be done to improve the overall AGT?

Basically you should do smaller/narrower frequency sweeps so the AGT difference stays minimal, ie hundreds or at most a couple of tenths of a db.
Generally youll notice big differences in AGT between vhf and uhf bands, even between vhf-low and vhf-hi. So you dont want to run a frequency sweep from channel 2 to 51, heh.
I feel safer when running vhf channels one by one, and also get to notice the pattern. Keep in mind the NEC engine was developed for the Navy, which only has to deal with many, but relatively narrow bandwidths. The uhf-tv band is pushing it.

 

[TaylorTech]

Hi everyone, I need some help. I've been working on this antenna and 4nec2 keeps asking for 4nec2dxs because I have too many segments. I'm on Windows 7 and can't seem to get it working. I also can't seem to get it to use nec2dxs.

How can I get this to work?

Is there a problem with my nec file? http://jump.fm/VJVGU




Thanks!
Taylor

 

[nikiml]

SWR question

I'd like to now how is the SWR calculated based on the nec2 output.
Any pointers are welcome.