LLNL-SM-791163

NEC-5 Validation Manual

G. J Burke

September 24, 2019

Disclaimer

This document was prepared as an account of work sponsored by an agency of the United States
government. Neither the United States government nor Lawrence Livermore National Security, LLC,
nor any of their employees makes any warranty, expressed or implied, or assumes any legal liability or
responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or
process disclosed, or represents that its use would not infringe privately owned rights. Reference herein
to any specific commercial product, process, or service by trade name, trademark, manufacturer, or
otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the
United States government or Lawrence Livermore National Security, LLC. The views and opinions of
authors expressed herein do not necessarily state or reflect those of the United States government or
Lawrence Livermore National Security, LLC, and shall not be used for advertising or product
endorsement purposes.

This work performed under the auspices of the U.S. Department of Energy by Lawrence Livermore
National Laboratory under Contract DE-AC52-07NA27344.
 1. Introduction
The NEC antenna modeling code has been in use since about 1975 in various versions
[1, 2]. NEC–5 [3] uses an algorithm new to NEC, solving the electric field integral equation
(EFIE) for wires and surfaces using the mixed potential method developed by Rao, Wilton
and Glisson [4]. This method is widely used in moment method codes and overcomes many of
the limitations of the point-matched algorithm in previous NEC codes. In addition to NEC–
5 providing new or more accurate modeling capabilities, comparing results from NEC–5 and
4 can serve as validation of both codes or reveal limitations, since the two codes are mostly
independent in their numerical solution methods. If they agree it is pretty likely the result
is accurate.

NEC–4 and previous codes solve the EFIE for wires and the magnetic field integral
equation (MFIE) for surfaces with point matching of the boundary conditions. The MFIE
surface model is limited to closed, perfectly conducting surfaces, and connection of an EFIE
wire to a MFIE surface o↵ers limited accuracy at best.

In the mixed-potential method the boundary condition is enforced on approximations of

path integrals of the electric field between adjacent element centroids. The integrals of E
reduce the order of the singularity of the kernel permitting the EFIE to be used for both
surfaces and wires. The integrals of E are approximated in terms of potentials, with the
integral of the vector potential A~ approximated using the average of A
~ at the centroids and
the scalar potential term r approximated from the di↵erence of at the centroids.

Many people have contributed to the development of the mixed-potential solution method.
Throughout its history D. R. Wilton at the University of Houston has been the major coor-
dinator and developer of the techniques. An early code for modeling surfaces was the Patch
Code [5], which was followed by the Junction Code [6, 7] allowing connection of wires to
surfaces. The method of connecting a wire to a surface comes from the work of Rao [8]
and Costa and Harrington [9, 10]. A large e↵ort went into the code EIGER [11], which
was essentially a test bed for abstracting ideas and developing object oriented programing
methods. The EIGER project resulted in a code that could be extended to higher-order
basis functions, curvilinear elements, dielectrics, two-sided surfaces for greater isolation and
alternate Green’s functions. NEC–5 does not use the developments from EIGER, but starts
at the level of wires, patches and junctions with a ground model from the work of Michalski
[12, 13].

This validation manual contains results for a number of modeling situations to demon-
strate the accuracy or numerical modeling errors of the codes NEC–4 and 5. A separate issue
is physical modeling errors which result from simplifying a real antenna in its environment to
an idealized form that can be modeled, simplifying parts or neglecting large nearby objects
to limit the model size. The significance of physical modeling errors must usually be judged
from experience.

Numerical modeling errors result from limitations of the modeling code in enforcing

1
the boundary conditions, usually related to convergence in the solution. For an accurate
numerical model the solution must satisfy Maxwell’s equations and the boundary conditions
on the fields for the idealized model. Maxwell’s equations are pretty much guaranteed, since
the Green’s function satisfies Maxwell’s equations. NEC–4 and 5 use di↵erent methods of
enforcing the boundary condition on E, so comparison of the results can reveal errors or give
good confidence that the results are accurate.

NEC–5 uses a linear current expansion, with triangular and roof-top basis functions and
a singular junction basis function to join surface current to a wire current. This linear
current expansion may converge somewhat more slowly than the sinusoidal expansion used
in previous NEC codes, as shown in this manual for dipoles.

NEC–5 is essentially a complete rewrite of NEC. The input and output have been kept as
much as possible the same as in earlier NEC codes. Some di↵erences were unavoidable, since
voltage sources in the mixed-potential method are at the ends of segments and on patch
edges, while in NEC–4 they are at the segment centers. Patch input is di↵erent since patch
edges must coincide to be connected. These di↵erences make it somewhat more difficult to
build a model file for NEC–5, so graphic assistance is provided, an optional interactive mode
while running the code and a separate GUI, NEC5GI [14], to build the model file and view
results.

The results in this manual were mostly produced in checking and debugging NEC–5, with
comparisons made with NEC–4 for validation. In some cases comparisons are made with
classic analytic approximations. Comparisons with experimental data is of course valuable
but difficult to come by. Sometimes we receive reports of results supporting NEC, particularly
for antennas with ground, but we have not been conscientious in documenting such reports.
Also, in experimental results, unless they are done very carefully, the physical modeling
errors often exceed the numerical modeling errors which, as seen in results in this manual,
are often fairly small.

Sections of this manual cover accuracy of the thin-wire kernel, voltage source models,
convergence for a dipole, small loops, surface meshing, wire-surface junctions and wires and
loops over ground. Validations for some of these cases have been used in conference papers
or the Newsletter of the Applied Computational Electromagnetics Society (ACES) but are
probably lost, so this will bring them together. Additional validation results of general
interest may be added to this manual as they become available or are contributed by users.

2
 2. Wire Modeling

2.1 The Thin-wire Kernel

A number of approximations are possible in evaluating the kernel for the electric field
integral equation for wires. In some early work the current was treated as a filament on the
wire axis with the boundary condition enforced on electric field at the wire surface. This
convention had the advantage that the current could remain continuous and free of charge
singularities at wire bends and steps in radius. However, locating the current on the axis
with basis functions continuous to the first derivative in NEC–2 resulted in convergence to
the wrong solution at steps in radius. In NEC–4 the current is treated as a filament on the
wire surface and the boundary condition is matched on the wire axis. With that change
the solution does converge toward the correct behavior at a step in radius. Both of these
conventions are called the reduced kernel.

More recently [15, 16] people have used exact thin-wire kernels, where integrals are eval-
uated over the current on the wire surface and the field is also evaluated on the wire surface.
The exact kernel formulations have the advantage that wire segments can have arbitrarily
small length relative to their radius. With the reduced kernel the accuracy degrades for
segment lengths less than about five times the radius, and the solution for current blows up
when the segment length is much shorter than the radius.

A solution with the current on the wire surface and the field matched on the axis can also
be exact as an extended boundary condition [17] if the current forms a completely closed
surface. However, closing wire ends and the gaps at voltage sources can be difficult. It can be
done by assigning additional basis functions and unknowns at the ends and voltage sources
[18] or with an approximation of the end-cap current. The approximate end-cap treatment
in NEC–4 extended the minimum segment length relative to radius, but it was still limited.
All thin wire approximations allow no variation of current around the wire, as would occur
when a wire is near another conductor or at a bend.

NEC–5 uses an approximation of the exact kernel as was done in [6]. For demonstration
we will consider the scalar potential due to a segment with constant charge density on the z
axis, centered at the origin. With the exact kernel the potential at coordinates ⇢, z can be
written Z /2 Z ⇡ jkR
j e
(⇢, z) = d dz 0 (1)
4⇡!✏ /2 ⇡ R
where p
R= ⇢2 + a2 + (z z 0 )2 2a⇢ cos (2)
for segment length and radius a and constant unit charge density on the segment. The
inner integral over in (1) is sometimes called the exact kernel.

Following the treatment in [6], the 1/R term is removed from a series expansion of the
exponential in (1) and integrated accurately, while the reduced kernel approximation is ap-

3
plied to the reminder. This form, which will be called an extended thin-wire approximation,
is Z /2 jkR0 Z /2 Z ⇡ !
j e 1 0 1
1 (⇢, z) = 2⇡ dz + d dz 0 (3)
4⇡!✏ /2 R 0 /2 ⇡ R
where p
R0 = ⇢2 + a2 + (z z 0 )2
is R evaluated for = ⇡/2. The reduced kernel evaluation involves the single integral
Z /2 jkR0
j e
0 (⇢, z) = dz 0 . (4)
2!✏ /2 R0

The accuracy of the reduced and extended kernels relative to the exact kernel were
compared using numerical evaluation of the integrals in the program Mathematica [19]. The
NIntegrate function in Mathematica can numerically evaluate integrals including integrable
singularities, so can be used to evaluate the integrals in (1), (3) and (4) including when the
evaluation point is on the wire surface. Interpretation of the error is somewhat problematic,
due to the choices in locating the evaluation points. As an extended boundary condition,
evaluation points on the axis may yield a solution as accurate as with evaluation on the
surface, so di↵erences in potential evaluations for individual segments or matrix elements
may not always translate to errors in the solution when all of the matrix elements are taken
together. However, the comparison still provides an indication of the limitations of the
approximations.

The errors of the extended and reduced kernels over a range of segment lengths and
radius a are plotted in Fig. 1 for the self-term evaluation at the midpoint of the segment.
The extended kernel was evaluated at the wire surface, 1 (a, 0). The reduced kernel was
evaluated at the center of the segment, 0 (0, 0), since this is the way it is used, and it resulted
in smaller errors relative to the exact kernel than evaluating on the surface at 90 degrees
from the current filament. The error in the extended kernel is seen to be nearly independent
of and to increase with increasing radius a due to the approximate evaluation of the first
integral in (3) containing the phase information. The error in the reduced kernel depends
on the thickness ratio, and is poor when /a is on the order of 1 or less.

Similar plots are included in Appendix A for ⇢ equal to 3, 10 and 100 times a and for
increasing z. With increasing ⇢ the extended kernel accuracy remains limited by the phase
error. For the reduced kernel the error related to /a decreases with increasing ⇢, and the
phase error becomes the limiting factor at large a/ . With increasing z the extended kernel
is limited only by the phase error which decreases with increasing z. For the reduced kernel
both the error related to /a and the phase error decrease with increasing z.

4
 Extended kernel Reduced Kernel
10-1 10-1
10-2
0.7
Dêa=0.001 10-3 Dêa=0.001

10-2 10-4 10-2 0.5

0.2
Dêa=0.01 10-5 Dêa=0.01
10-1
a/λ

a/λ
10-3 10-6 10-3 10-2
10-3
Dêa=0.1 10-7 Dêa=0.1 10-4
10-8 10-5
10-4 10-4
10-6
10-9
10-7 -8
10
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. 10-10 Dêa=1 Dêa=10. Dêa=100. Dêa=1000.
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. 2.1.1 Errors in the extended and reduced kernels relative to the exact kernel for the evaluation
point at the center of the segment (self-term) with ⇢ = a for the extended kernel, ⇢ = 0 for the
reduced kernel.

In the NEC-5 code the second integral in (3) is evaluated first over z 0 in terms of log
functions. The log function result can be written so that the part that becomes singular
as the evaluation point approaches the surface can be integrated exactly. The remaining
smoothly varying terms are integrated with a low order Gauss-Legendre rule. Singularity
extraction is needed only when the evaluation point approaches near one of the segment
ends. The numerical errors in this evaluation appear not to be significant compared to the
approximations in the kernel.

Results Testing the Thin-wire Kernel

The e↵ectiveness of the extended kernel can be demonstrated by modeling a dipole an-
tenna with extended and reduced kernels. Solutions for current on a dipole with length
L = 0.5 and radius a = 0.01 (thickness parameter ⌦ = 2 log(L/a) = 7.8) using the re-
duced kernel in NEC–5 are shown in Fig. 2. The current blows up to large values near the
source and ends. Instability also occurs with 80 and 160 segments but is covered in the
plot. Solutions obtained with the NEC–5 extended kernel are plotted in Fig. 7 and show no
instability up to 640 segments.

5
Fig. 2.1.2 Current on a dipole with length L = 0.5 , radius a = 0.01 modeled with N segments
and segment length = L/N with the reduced kernel in NEC–5.

Fig. 2.1.3 Current on the same dipole as Fig. 2 modeled with the extended kernel in NEC–5.

The input admittance computed for this dipole with the NEC–5 extended and reduced
kernels and the NEC–4 reduced kernel are shown in Fig. 4 The reduced kernel results blow up
to large values for /a less than about 0.16, while the extended kernel converges smoothly
to /a = 0.01 with 5000 segments on the dipole. The linear increase in Bin is due to the
increasing e↵ective shunt capacitance as the source gap becomes narrower, and is normally
seen in such solutions. It could be eliminated by dividing the source voltage between multiple
segments to keep the width of the source gap constant. In a plot of impedance this changing
Bin would a↵ect both real and imaginary parts, so that the result would appear not to
converge. The relatively large di↵erence between NEC–4 and NEC–5 for large /a is due to
convergence, since the 0.5 dipole is on the steep slope of the admittance curve resonance.

6
Fig. 2.1.4 Input admittance of a dipole with length L = 0.5 and radius a = 0.01 for varying
segment length = L/N .

The computed input admittance of a square loop antenna with side length 0.1 and wire
radius 0.01 is shown in Fig. 5. Again, the NEC–4 and NEC–5 results using the reduced
kernel fail for /a less than about 1, while the NEC–5 extended kernel result converges
smoothly. For small segment lengths there are many segments buried within other segments
at the corners.

Fig. 2.1.5 Input admittance of a square loop antenna with side length 0.1 , radius a = 0.01 for
varying segment length .

A test for modeling closely spaced wires is provided by a two wire transmission line.
Transmission lines are usually modeled with parallel wires connected with short wires at the
ends for the source and termination. This model would combine wires that may be closely
spaced compared to their radius and segments with small /a at the terminations. To
separate closely spaced wires from the small /a problem, two parallel wires were modeled
with lengths of 10 m and radius 0.001 m driven with voltage sources of 1 volt and 1 volt at
their centers. This model can be cut in half with a PEC plane to have a single open-circuited
transmission line with a net 1 volt source.

The input admittance of an ideal open circuited line of length l at frequency f is

Yin = (j/Z0 ) tan(2⇡f l/c)

7
where c the velocity of light and Z0 the characteristic impedance. Since the line was modeled
around 5 length (5 m at 299.8 MHz) the slope of Yin with frequency was used to determine
Z0 . The derivative of Yin is ✓ ◆
dYin 2⇡jl 2 2⇡f l
= sec .
df Z0 c c
When the line length is a multiple of a half wavelength the characteristic impedance can be
estimated as
2⇡l
Z0 = /(dBin /df )
c
where Bin is the imaginary part of Yin and the derivative is approximated from numerical
values about 299.8 MHz.

Results for Z0 from solutions with NEC–5 extended and reduced kernels and NEC–4
reduced kernel are shown in Fig. 6 and compared with the characteristic impedance of an
ideal parallel-wire transmission line
⌘0 ⇣ ⌘
1 s
Z0 = cosh .
⇡ d
Results from the NEC–5 extended and reduced kernels essentially coincide, and the NEC–4
result is close. None of the thin-wire kernel results track the ideal characteristic impedance
for s/d less than about 2. The transmission line was also modeled in NEC–5 with cylindrical
conductors of rectangular patches, with 400 patches in the 10 m length and 8 patches around
the conductor. The patch result in Fig. 6 does track the steeper slope of Z0 down to s/d = 1.1.

8
Fig. 2.1.6 Characteristic impedance of a two-wire transmission line for wire separation s and
diameter d determined by processing the computed input impedance of an open circuited line
modeled with reduced and extended thin-wire kernels and a patch model.

The wire radius in this transmission line is about 0.001 , so from Fig. 1 the relative error
in the self term should be around 10 6 with the extended kernel and 10 3 with the reduced
kernel. For closely spaced adjacent wires the reduced kernel error would be considerably
larger. So more accuracy in the extended thin-wire kernel probably would not change the
result. The current on a section of one conductor of the patch transmission line model is
shown in Fig. 7 for varying wire separation. The current is seen to concentrate between the
conductors as the wires become close together, so that the separation of currents is less than
the separation of the conductor centers.

It has also been observed with a code using a 2D periodic Green’s function that the
shielding by a screen of parallel wires reaches a maximum when the wire spacing is 2⇡a for
wire radius a with either the reduced or essentially exact thin-wire kernel. So the “equal
area rule” should still hold with the NEC–5 extended kernel. Of course, there should not be
much need for wire grid modeling with the patch model.

9
Fig. 2.1.7 Current on the upper conductor of a section of two-wire transmission line modeled as
a cylinder with rectangular patches as the separation s is varied for wire diameter d = 0.002 .
Transverse dimensions have been multiplied by 1000.

10
2.2 Voltage Source Models

Accurate modeling of voltage sources is critical in modeling antennas to get accurate

input impedances and antenna gains. The voltage source models in the point-matched
NEC–4 solution and the mixed-potential NEC–5 solution di↵er substantially, so di↵erent
modeling rules will apply. In NEC–4 the source is located on a segment, and the voltage
drop is approximately spread over the width of the segment. In NEC–5 the voltage source
is located at the end of a wire segment or at the edge of a patch. Current sources in NEC-
5 are voltage sources with the voltage determined to produce the desired current, so the
same modeling rules apply. Also, lumped loads and networks use the voltage source model.
The results in this section demonstrate some e↵ects of modeling changes on voltage source
accuracy and the e↵ective width of the source.

In NEC–4 a voltage source with strength Vs is modeled by setting the electric field
value at the center match point on a segment to Vs / for segment length . With a
small wire radius and uniform segment lengths in the source region, the electric field has an
approximately rectangular distribution over the segment, so that the integral of the field is
approximately the voltage Vs . The axial field distribution from a NEC–4 solution for a /2
dipole modeled with 9 segments is shown in Fig. 1. The field distribution over the source
segment is somewhat rounded due to the moderate wire radius of 0.001 . The field away
from the source segment is down by nearly two orders of magnitude, and would be smaller
with more segments on the dipole. The field shows sharp downward spikes near the match
points, although the plot points do not necessarily coincide with match points. Since in
NEC–4 the current at segment junctions is continuous to the first derivative, the field is
reasonably well behaved at the junctions.

Fig. 2.2.1 Magnitude of electric field along the axis of a dipole antenna with length 0.5 and
radius 0.001 modeled with 9 segments in NEC–4. Vertical lines mark the ends of segments.

11
Fig. 2.2.2 Electric field Ez and the potential from integrating Ez along the axis of a dipole
antenna with length 0.5 and radius 0.001 modeled with 9 segments in NEC–4.

The real and imaginary parts of Ez from the NEC–4 model of the /2 dipole are plotted
in Fig. 2, and also the potential from integrating Ez . The change in potential of 1 volt is
seen to take place almost linearly over the source segment.

The field along the /2 dipole modeled with NEC–5 with 8 segments is plotted in Fig. 3.
Since NEC–5 uses a piecewise-linear current expansion, the charge density is constant over
each segment and discontinuous at the junctions. If an exact kernel were used in calculating
the field there would be delta functions of E at the junctions. While NEC–5 uses an extended
kernel for the potentials in the solution for current, it uses the reduced kernel for the fields,
so the spikes between segments are spread over a width on the order of the wire radius.
There are no match points in the mixed-potential solution, so no points where the error in
field must be very small. The potential from integrating Ez in Fig. 3 is seen to oscillate
around the correct values. While the field and potential plots in Fig. 3 do not look so nice,
the mixed-potential solution for antenna impedance and fields away from the antenna can
be quite accurate, and in many case as good or better than with NEC–4. Plots of the field
and potential when the dipole is modeled with 16 and 32 segments in NEC–5 are shown in
Fig. 4 and 5. The field is seen to converge toward the correct values.

Fig. 2.2.3 Electric field Ez and the potential from integrating Ez along the axis of a dipole
antenna with length 0.5 and radius 0.001 modeled with 8 segments in NEC–5.

12
Fig. 2.2.4 Electric field Ez and the potential from integrating Ez along the axis of a dipole
antenna with length 0.5 and radius 0.001 modeled with 16 segments in NEC–5.

Fig. 2.2.5 Electric field Ez and the potential from integrating Ez along the axis of a dipole
antenna with length 0.5 and radius 0.001 modeled with 32 segments in NEC–5.

Since the NEC–4 source is located on a segment, a center fed dipole is usually modeled
with an odd number of segments. In NEC–5 the dipole would be modeled with an even
number of segments to have a junction at the center. This di↵erence is a barrier to complete
compatibility between NEC–4 and 5 models. With NEC–4 the dipole can be modeled with
an even number of segments and the center two segments each excited with half the total
voltage. The source width is then doubled. Alternately, a dipole can be modeled in NEC–
5 with an odd number of segments and the voltage source split between the ends of the
center segment. The potential distributions from splitting the source in NEC–5 are shown
in Fig. 6 for dipoles with 9 and 17 segments. The split NEC-5 source has a somewhat wider
distribution than the NEC–4 source on a single segment, but it provides a reasonable model
for a source on a segment. An option could be added to NEC–5 to split all voltage sources,
loads and network connections for compatibility with NEC–4 models, but that option is not
in the present code.

13
Fig. 2.2.6 Potential along the /2 dipole antenna with radius 0.001 modeled with 9 and 17
segments. In NEC–4 the 1 volt source was on the center segment and in NEC–5 it was split
between the ends of the center segment.

Fig. 2.2.7 Input admittance of a 0.475 dipole with radius 10 6 computed with NEC–4. Curve 1:
source segment length s , other segment lengths ; Curve 2: center 3 segments s with source on
the center one, other segment lengths ; Curve 3: all segments have uniform length approximately
s.

E↵ects of nonuniform segment lengths

It has always been recommended that a voltage source in NEC–4 should be located on a
segment with equal length segments adjacent to it. The result of violating this rule is shown
in Fig. 7. For the top curve the length of the center source segment s is varied with 50
segments with length on either side of the source. The total number of segments is 101.
For the second curve in Fig. 7 the center three segments have equal lengths of s with the
source at the center, and the 50 segments on either side have lengths , for a total of 103
segments. For the final curve the dipole is modeled with segments of uniform length s .
Source segments with length di↵ering from the surrounding segments are seen to result in
an error in input admittance, with shorter source segments causing larger errors. Having
segments of equal length adjacent to the source reduces the error, but it is still a problem.

The power balance, obtained from the “average gain” calculation in NEC–4 and 5 pro-
vides another check on solution accuracy. The ratio of radiated power to input power should

14
be 1.0 when there are no losses involved. The error in power balance is shown in Fig. 8
for varying source segment length. The error is due to error in input resistance, since the
integral of radiated power is insensitive to errors in the current. The floor in the error with
uniform segmentation is probably due to the accuracy of integrating the radiated power with
2 degree steps in ✓.

Fig. 2.2.8 Error in the ratio Prad /Pin for the NEC–4 dipole models in Fig. 7.

Fig. 2.2.9 Field Ez and integrated potential over the source region of the dipole when the ratio of
source segment length to adjacent segment lengths s / is 0.1 in NEC–4. Vertical lines mark the
segment ends.

The electric fields in the source region of the dipole and the potentials from integrating
Ez are shown in Fig. 9 through 11 for s / of 0.1, 1.0 and 10. With s / = 0.1 the
source field is mostly localized on the source segment, but large skirts result in an error in
the total voltage drop. The error in potential is not as obvious with s / = 10, but the
drop is somewhat too small.

For the NEC–5 model the two center segments were given lengths s and the dipole was
completed with 50 segments on either side with lengths . The total number of segments
was 102. The result for input impedance in Fig. 12 is seen to be very insensitive to s / .

15
The power balance error in Fig. 13 also shows low sensitivity to s / . Plots of the field
distribution Ez and potential from integrating Ez are shown in Fig. 14 through 19. The
wire radius has been increased to 10 4 to make it easier to sample the field. The e↵ective
source width is narrower than the segment length, especially for the wide source segment,
but the voltage drop remains correct.

As a final test of the NEC–5 source model half of the dipole was modeled with 50 segments
with lengths 1 = 0.00475 and the number of segments on the other half was varied for
segment length 2 with the voltage source at the junction where the segment length changes.
The input admittance shown in Fig. 20 is fairly insensitive to the ratio 2 / 1 . The change
that is seen may be due more to convergence with large segments than to error in the source
model.

Fig. 2.2.10 Field Ez and integrated potential over the source region of the dipole when the ratio
of source segment length to adjacent segment lengths s / is 1.0 in NEC–4.

Fig. 2.2.11 Field Ez and integrated potential over the source region of the dipole when the ratio
of source segment length to adjacent segment lengths s / is 10.0 in NEC–4.

16
Fig. 2.2.12 Input admittance of a 0.475 dipole with radius 10 6 computed with NEC–5 with
the source segment length s and other segment lengths , and also with uniform segment length
s.

Fig. 2.2.13 Error in power balance Prad /Pin for a 0.475 dipole with radius 10 6 computed with
NEC–5 with the source segment length s and other segment lengths , and also with uniform
segment length s .

Fig. 2.2.14 Field Ez over the source region of a 0.5 dipole with radius 10 4 from the NEC–5
solution with s / = 0.1.

17
Fig. 2.2.15 Potential from integrating Ez in Fig. 14 for s/ = 0.1.

Fig. 2.2.16 Field Ez over the source region of a 0.5 dipole with radius 10 4 from the NEC–5
solution with s / = 1.

Fig. 2.2.17 Potential from integrating Ez in Fig. 16 for s/ = 1.

18
Fig. 2.2.18 Field Ez over the source region of a 0.5 dipole with radius 10 4 from the NEC–5
solution with s / = 10.

Fig. 2.2.19 Potential from integrating Ez in Fig. 18 for s/ = 10.

Fig. 2.2.20 Input admittance of a 0.475 dipole with (10 6 ) radius. Half of the dipole is modeled
with 50 segments of length 1 = 0.00475 and the other half has segments of length 2 , with the
change taking place across the voltage source..

19
2.3 Convergence for a Dipole Antenna

Results demonstrating the convergence of the solution for dipole antennas of varying
thickness modeled with varying numbers of segments are shown in this section. Results from
the King-Middleton second-order solution from [20] are included for reference.

Fig. 1 through 5 are the input admittance obtained with NEC–5. The admittance is
seen to converge rapidly at the first resonance and more slowly at the second resonance. For
thicker wires the susceptance becomes more capacitive with increasing number of segments
N due to the increased e↵ective shunt capacitance of the source gap. The input impedance,
plotted in Fig. 6 through 10, shows slower convergence at the peaks since the change in
susceptance due to gap width a↵ects both real and imaginary parts of impedance. These
anti-resonant points typically show slow convergence in method of moments solutions.

The same sequence of plots is repeated in Fig. 11 through 20 for NEC–4 results. The
admittance peaks are seen to converge very rapidly in NEC–4, probably due to the sinusoidal
current expansion. For higher numbers of segments on thick wires the NEC–4 solution starts
to fail due to use of the reduced kernel.

Fig. 2.3.1 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 20 (L/a = 22026)
modeled with NEC–5 with N segments.

Fig. 2.3.2 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 15 (L/a = 1808)
modeled with NEC–5 with N segments.

20
Fig. 2.3.3 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 10 (L/a = 148.41)
modeled with NEC–5 with N segments.

Fig. 2.3.4 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 8 (L/a = 54.60)
modeled with NEC–5 with N segments.

Fig. 2.3.5 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 7 (L/a = 33.12)
modeled with NEC–5 with N segments.

21
Fig. 2.3.6 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 20 (L/a = 22026)
modeled with NEC–5 with N segments.

Fig. 2.3.7 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 15 (L/a = 1808)
modeled with NEC–5 with N segments.

Fig. 2.3.8 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 10 (L/a = 148.41)
modeled with NEC–5 with N segments.

22
Fig. 2.3.9 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 8 (L/a = 54.60)
modeled with NEC–5 with N segments.

Fig. 2.3.10 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 7 (L/a = 33.12)
modeled with NEC–5 with N segments.

Fig. 2.3.11 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 20 (L/a = 22026)
modeled with NEC–4 with N segments.

23
Fig. 2.3.12 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 15 (L/a = 1808)
modeled with NEC–4 with N segments.

Fig. 2.3.13 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 10 (L/a = 148.41)
modeled with NEC–4 with N segments.

Fig. 2.3.14 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 8 (L/a = 54.60)
modeled with NEC–4 with N segments.

24
Fig. 2.3.15 Input admittance of a dipole with thickness factor ⌦ = 2 ln L/a = 7 (L/a = 33.12)
modeled with NEC–4 with N segments.

Fig. 2.3.16 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 20 (L/a = 22026)
modeled with NEC–4 with N segments.

Fig. 2.3.17 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 15 (L/a = 1808)
modeled with NEC–4 with N segments.

25
Fig. 2.3.18 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 10 (L/a = 148.41)
modeled with NEC–4 with N segments.

Fig. 2.3.19 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 8 (L/a = 54.60)
modeled with NEC–4 with N segments.

Fig. 2.3.20 Input impedance of a dipole with thickness factor ⌦ = 2 ln L/a = 7 (L/a = 33.12)
modeled with NEC–4 with N segments.

26
2.4 Wire Loops

Electrically small loops are a problem with both NEC–4 and 5 and any code that uses
localized basis and weighting functions, since the matrix becomes ill-conditioned at low
frequencies as the current becomes uniform around the loop. The input admittances of
a loop modeled in NEC–4 and 5 down to C/ = 10 8 are shown in Fig. 1. The input
susceptance Bin should increase as the inverse of C/ at low frequencies. Bin fails in NEC–4
at around C/ = 10 6 . NEC–5 goes a little further with susceptance, but fails also. The
Gin result fails in NEC–5 above C/ = 10 4 when it is about 10 12 below Bin , while NEC–4
goes to almost C/ = 10 6 . A considerable e↵ort was put into NEC–4 to preserve the real
part when it gets small relative to the imaginary part. A similar e↵ort might extend the
NEC–5 solution.

Fig. 2.4.1 Input admittance of a loop antenna with circumference C and wire radius 1.59(10 4 )C

modeled with 22 segments in NEC–4 and 5.

A weakness of NEC–4 is the instability of currents in small loops that are coupled to
driven structures. A typical problem is illustrated in Fig. 2, where a driven dipole is near
a loop that does not contain a source. At low frequencies the electric field of the dipole is
dominated by the gradient of scalar potential (r ) from the charge at its ends. Since r
has zero curl, the integral of this term around the loop must be zero. NEC–4 evaluates the
electric field accurately at the match points on the segments, but does not take into account
the variation between match points. The resulting error in sampling r appears as a voltage
in the loop and can drive a current proportional to one over frequency, while the loop current
should go to zero with frequency.

27
 L

1V

Fig. 2.4.2 NEC–5 result for the imaginary part of current on a square loop coupled to a dipole
antenna with s/L = 0.2 and C = 4L = 4(10 5 ) .

In the mixed-potential solution in NEC–5 the potentials are also evaluated only at the
segment centers. However, the r term is approximated as di↵erences between at suc-
cessive center points, so the sum of the r approximations around a closed loop is zero
to machine accuracy. The mixed-potential solution will also fail at a low enough frequency
as the matrix becomes ill-conditioned. But the mixed-potential method can go orders of
magnitude lower in frequency than NEC–4 in this type of problem.

The accuracy of the NEC–4 solution can be improved by evaluating the field at more
than one point on each segment and approximating the integral of E over the segments. A
version of NEC–4 was set up to evaluate the field at a variable number of points on each
segment and apply a Gauss-Legendre rule for the integral.

Results for the current on the loop in Fig. 2 are plotted in Fig. 3 as frequency is decreased
until the circumference C of the loop is about 4(10 5 ) . The NEC–4 solution is seen to blow
up for C/ just less than 1. Integrating with up to 9 points per segment in the modified
NEC–4 reduces the frequency where it fails by about an order of magnitude, but at the
expense of increased time to fill the matrix. The NEC–5 solution shows the correct decrease
of loop current with frequency through the range of the plot.

NEC-SL [21] is a version of NEC–4 that was modified to find loops and replace one basis
function in each loop with a constant current around the loop and replace one match point
with a loop weighting function. The NEC-SL result also remains stable through the range
of the plot. However, it is difficult to implement this solution for complex structures, since
all loops must be found and care is needed in handling loop that share common sides. The
NEC–SL code was never developed to the point that it could be released.

28
Fig. 2.4.3 Current at the center of the near side of the loop in Fig. 2 versus frequency, showing
the low frequency blow up of current with NEC–4 and the stable solution with NEC–5.

29
 3. Wire-Surface Junctions
Modeling surfaces and wire-surface junctions has been a major weakness of NEC–4 where
surface models use the MFIE which is limited to closed, perfectly conducting surfaces. The
MFIE model as implemented in NEC–4 works well for scattering by a sphere and reasonably
well for a box, but attempts to model even a moderately thin plate as a box with one thin
dimension may not be handled accurately. NEC–4 is fast in filling the MFIE matrix due
to the use of point matching and single-point integration. In modeling ship hulls it was
found to give reasonable accuracy in radiation patterns, but is not accurate for impedances
of connected wire antennas.

Connection of a wire modeled with the EFIE in NEC–4 to a MFIE surface is particularly
problematic. A wire is connected at the center of a patch which is split into four sub-patches
and a singular basis function is used to join the surface current to the wire current using the
method in [22]. As seen in the results that follow, the impedance at a wire-surface junction
may have reasonable accuracy if the wire segment length is in the right range relative to
the patch size, but the solution is not reliable. As a result, wire-surface junctions are often
modeled using a wire grid surface in NEC–4. This approach has its own problems with small
loops and other issues.

NEC–5 uses the mixed potential formulation for both wires and surfaces with triangular
or roof-top basis functions as developed by Rao,Wilton and Glisson [4] and a singularl basis
function at a wire-surface junction as developed by Rao [8] and Costa and Harrington [9, 10].
This treatment appears to be stable and accurate. NEC–5 results are compared here with
measurements by Bhattacharya, Long and Wilton [23]. Subsequent plots compare NEC–4
results for varying segmentation using NEC-5 as a standard.

3.1 The Monopole on a Box

An extensive set of measurements has been reported in [23, 24] for impedance and radi-
ation patterns of monopoles on a 10 cm box. These results have often been used in the past
in validating wire-surface junction models. The basic model, shown in Fig. 1, is a cube 10
cm on a side with monopoles of various lengths and radius 0.8 mm located at positions on
the top. NEC–5 has been tested for a 6 cm monopole at the center and near one corner.
In the measurements the box was on a 78 ⇥ 78 cm ground plane, while the model used an
infinite ground plane.

Fig. 1 shows the currents computed with NEC–5 for the 6 cm monopole at the center of
the top at 1.15 GHz. The arrows show the real part of current over three orders of magnitude
range, and the color represents the magnitude of current at the center of each element on
a log scale covering two orders of magnitude from red to violet. Current is shown on wires
and surface current density on patches, so they are not directly comparable. The currents
at 1.15 GHz with the 6 cm monopole near a corner, displaced 5.15 cm from the center, are
shown in Fig. 2.

30
 10 cm

cm
10

Fig. 3.1.1 Current on a 10 cm box with a 6 cm monopole at the center of the top at 1.15 GHz.
Arrows show the real part of wire current or surface current density and color represents the
magnitude.

10 cm

cm
10

Fig. 3.1.2 Current on a 10 cm box with a 6 cm monopole near the corner, displaced 5.15 cm from
the center, at 1.15 GHz.

31
 The input admittances from NEC–5 for the 6 cm monopole at the center position are
plotted in Fig. 3. The measured results were read as accurately as possible from [23]. The
NEC–5 model 1 had 10 by 10 divisions of triangular patches on the top and 10 by 10
rectangular patches on the sides, with 8 segments on the monopole, as shown in Fig. 1.
Model 2 had 20 by 20 divisions on top and sides and 16 segments on the monopole. The
convergence is seen to be good in the NEC–5 results, with a slight frequency shift from
measurements. The admittance with the monopole at the corner position, 5.15 cm from the
center, is plotted in Fig. 4. Model 1 was as shown in Fig. 2, and model 2 had 20 by 20
divisions on continuous sides with 18 between the wire and the far sides and 3 between the
wire and the near sides with 16 segments on the wire. Again, there is a slight frequency shift
from measured results but good convergence.

The monopole at the center of the box was also modeled with the NEC–4 patch model.
The first model had 11 by 11 patches on each side and the top, and the monopole was
modeled with Nm segments. The results in Fig. 5 show a strong dependence on the wire
segment length. The best agreement is obtained with segments about equal in length to the
patch widths, although the agreement is not great. A di↵erent view of the result is seen in
the impedance plots in Fig. 6. At low frequencies the agreement in resistance is not bad, but
a large capacitive reactance error is added as the wire segments are made shorter. Another
NEC–4 model was run with 11 by 11 patches on the four sides and 21 by 21 on the top. This
can be done in NEC–4, since the patches do not have to have matching edges as in the NEC–
5 mixed-potential solution. The admittance and impedance results for this higher resolution
model are shown in Fig. 7 and 8. The best agreement is now obtained with 10 segments
on the monopole. It may be best in NEC–4 models to make the wire segments as long as
possible at the connection point. These results are consistent with previous observations that
the NEC–4 patch model may give useable accuracy for some purposes with the advantage
of a fast matrix fill for the MFIE matrix elements.

The radiation patterns from NEC–5 for the 6 cm monopole at the center of the box at
1.138 GHz are shown in Fig. 9. The horizontal radiation (E ) would be zero for the monopole
on an infinite ground plane, but a small horizontal component is introduced by the box. Since
it is down by 42.7 dB from the vertical gain, the horizontal pattern is sensitive to asymmetries
in the mesh. Fig. 10 shows the mesh patterns and resulting current magnitudes for values
of the IALT parameter on the QP command from 0 through 3. The azimuthal radiation
patterns resulting from these meshes are shown in Fig. 11. In this case the symmetric mesh
patterns with IALT of 2 or 3 are the better choice. The power balance from integrating the
radiated power was Prad /Pin = 0.990 for IALT of 0 or 1, 0.986 for IALT = 2 and 0.994 for
IALT = 3. Elevation patterns of vertical and horizontal gain in the = 22.5 plane are
shown in Fig. 12. These patterns agree with the computed patterns reported by Chu et al.
in Fig. 15 of [24] and agree as well as can be expected with the measured patterns in [24].

32
Fig. 3.1.3 Input admittance of the 6 cm monopole on the 10 cm box at the center position.
Measured values were read from [23].

Fig. 3.1.4 Input admittance of the 6 cm monopole on the 10 cm box at the corner position, 5.15
cm from the center. Measured values were read from [23].

Fig. 3.1.5 Input admittance of the 6 cm monopole at the center position on the box from the
NEC–4 patch model with Nm segments on the monopole and 11 by 11 patches on top and sides.
The NEC–5 result from the model 2 mesh is included for comparison.

33
Fig. 3.1.6 Input impedance of the 6 cm monopole at the center position on the box from the
NEC–4 patch model with Nm segments on the monopole and 11 by 11 patches on top and sides.
The NEC–5 result from the model 2 mesh is included for comparison.

Fig. 3.1.7 Input admittance of the 6 cm monopole at the center position on the box from the
NEC–4 patch model with Nm segments on the monopole, 21 by 21 patches on top and 11 by 11 on
the sides. The NEC–5 result from the model 2 mesh is included for comparison.

Fig. 3.1.8 Input impedance of the 6 cm monopole at the center position modeled with the NEC–4
patch model with Nm segments on the monopole, 21 by 21 patches on top and 11 by 11 on the
sides. The NEC–5 result from the model 2 mesh is included for comparison.

34
 Vertical Gain Horizontal Gain
20

20

20
15
20 10
10 5
0
0 0
20

20
0
0 20
20 20

Fig. 3.1.9 Normalized gain patterns for the monopole at the center of the box at 1.138 GHz. The
maximum directive gains are 5.85 dB for vertical polarization (E✓ ) and 36.9 dB for horizontal
polarization (E ).

IALT = 0 IALT = 1

IALT = 2 IALT = 3
Fig. 3.1.10 Mesh patterns and current magnitudes obtained for the triangular patches on the top
of the box with di↵erent values of IALT on the QP command.

35
Fig. 3.1.11 Vertical and horizontal gain patterns in azimuth for the monopole at the center of the
box at 1.15 GHz, showing the e↵ect of the mesh patterns in Fig. 10.

Vertical Gain
Horizontal Gain

10 0 -10 -20 -30 -40 -50 -40 -30 -20 -10 0 10

(dB)

Fig. 3.1.12 Elevation patterns of vertical and horizontal gain in the = 22.5 plane for the
monopole at the center of the box at 1.138 GHz.

The radiation patterns for the 6 cm monopole at the corner position on the box, 5.15 cm
from the center, at 1.192 GHz are shown in Fig. 13. The power balance from integrating the
radiated power was Prad /Pin = 0.987. Elevation patterns of gain in the = 45 and 45
planes are shown in Fig. 14. The pattern in the = 45 plane matches the computed
pattern in Fig. 17 of [24] and is in reasonable agreement with the measured pattern in that
paper.

36
 Vertical Gain Horizontal Gain 20

20
10

0
0

20 10

20 20

10
10

0
0
10
20 0
0 10
20 20

Fig. 3.1.13 Normalized gain patterns for the monopole at the corner position on the box at 1.192
GHz. The maximum directive gains are 5.14 dB for vertical polarization (E✓ ) and 4.59 dB for
horizontal polarization (E ).

Fig. 3.1.14 Elevation patterns of vertical and horizontal gain in the = 45 and 45 planes for
the monopole at the center of the box at 1.192 GHz. The monopole is in the = 45 plane.

37
3.2 The Half-Loop on a Box

A model with a half-loop on the 10 cm box on an infinite ground plane is shown in Fig.
1. There are no measurements for this case, but the results can be compared for consistency.
The current is shown in Fig. 1 for a voltage source at the center of the horizontal wire
at 1 GHz, when the source location is near a current antinode. Color represents current
magnitude over a two-decade log scale from red to violet and arrows show the imaginary
part of current over three decades.

The input admittance with the voltage source at the center of the top wire is plotted
in Fig. 2. The NEC–5 model was run with 3 and 6 segments on each vertical section of
the loop to test convergence. The NEC–5 results show good convergence and a reasonable
comparison with the admittance of the half-loop on an infinite ground plane. A NEC–4
model with 3 segments on the vertical sections also is in good agreement with other results.
Fig. 3 and 4 show the result of increasing the number of segments on the vertical wires in
the NEC–4 model. As with the monopole, a large capacitive reactance error is introduced
with shorter segments at the junction in the NEC–4 model.

m
10 c
2 cm
10 cm

Fig. 3.2.1 Half-loop on a box. Loop height is 2 cm, length 9 cm and radius 0.9 mm. Currents are
shown for a voltage source at the center of the top wire at 1 GHz.

38
Fig. 3.2.2 Input admittance of the half-loop on a 10 cm box with voltage source at the center of
the top wire. Nwv is the number of segments on the vertical sections of the half-loop.

Fig. 3.2.3 Input admittance of the half-loop on a 10 cm box with voltage source at the center of
the top wire, showing the error in the NEC–4 result as the number of vertical wire segments Nwv
is increased.

Fig. 3.2.4 Input impedance for the cases in Fig. 3.

Currents for the voltage source at the end of the half-loop at 200 MHz are shown in Fig. 5.
At this lower frequency the current is nearly constant over the loop and is concentrated on the
top surface of the box. Radiation patterns at 200 MHz are shown in Fig. 6. The maximum
vertical gain is o↵ the end of the loop and maximum horizontal gain is broadside above the
ground plane. The power balance from integrating the radiated power was Prad /Pin = 1.0007.

39
 voltage
source

Fig. 3.2.5 Half-loop on a box showing currents with the voltage source at the base of the vertical
wire at 200 MHz. Arrows show the imaginary part of current.

horizontal gain 20
vertical gain 20
10
0
0 10
20 20
30

20
20
voltage
source 10

0 10

20
0
0
20
10
0
10

Fig. 3.2.6 Normalized gain patterns for the half-loop on the 10 cm box at 200 MHz with voltage
source at the end. The maximum directive gains are 5.23 dB for vertical polarization (E✓ ) and 2.20
dB for horizontal polarization (E ).

40
 4. Antennas over Ground
The ground model in NEC-5 uses interpolation in tables of potentials for distances up to
3 in ⇢ and about 1.5 in height and asymptotic approximations at larger distances. NEC-4
also uses interpolation for E. The interpolation in NEC-5 should be more accurate than in
NEC-4, and similar to the GN3 ground model in NEC-4.2. The NEC-4 and 5 ground models
are sufficiently di↵erent that useful validation information can be obtained by comparing the
two. Where possible other results have also been used for validation.

4.1 Horizontal Wires over Ground

A simple test for the ground model is a horizontal dipole lowered toward the interface.
NEC-4 and 5 results for a dipole with length 0.4 and radius 10 8 are shown in Fig. 1 for
height h from 0.1 to 10 7 . The models had 15 segments for NEC-4 and 16 segments for
NEC-5, so the voltage source was in the center in each model. Results are included for a
NEC-5 strip model with 16 sections of triangular elements (32 elements) with strip width
4(10 8 ) . Ground interaction tunes the dipole from capacitive to inductive over the height
range. The three results are in good agreement over this range.

250. 1000
NEC-4 wire NEC-4 wire
NEC-5 wire NEC-5 wire
Input Reactance (ohms)
Input Resistance (ohms)

200.
NEC-5 strip NEC-5 strip
500
150.

100.
0

50.

-500
0.
10-7 10-6 10-5 10-4 10-3 10-2 10-1 10-7 10-6 10-5 10-4 10-3 10-2 10-1
h/λ h/λ

Fig. 4.1.1 Input impedance of a dipole with length 0.4 and radius 10 8 for varying height h
over dielectric ground with ✏r = 16, comparing NEC-4, NEC-5 wire and NEC-5 triangular patch
models.

On a long wire over ground the current driven by a voltage source has the form of an
exponentially decaying mode, known as the transmission line mode or structure-attached
mode and a slower decaying surface-attached mode [25]. The current magnitude on a 40
wire is plotted in Fig. 2 from NEC-4 and NEC-5 for height h from 0.1 to 10 4 . The wire
had a radius of 10 6 and was modeled with 1600 segments with a voltage source on segment
10, or on the inside end of segment 10 for NEC-5, and a matched load on the opposite wire
end to reduce the standing wave from reflection. The transmission line mode dominates near
the source, and the surface-attached mode becomes dominant for larger distances for heights
of 0.01 or less. While the two modes have comparable amplitudes an interference e↵ect is
seen. NEC-4 and 5 results are in good agreement over this range of heights and distances.

41
 1
h = 0.1λ
10-1

Current Magnitude
NEC-4
10-2 NEC-5

h=0
10-3 .01λ

h = 0.00
10-4 1λ

h = 10 -4λ
10-5
0 10 20 30 40
x/λ

Fig. 4.1.2 Current on a 40 wire at height h over ground with complex relative permittivity
✏˜r = 16 j16 ( = 0.00267 S/m at 3 MHz) showing the exponential structure-attached modes and
the slower decaying surface-attached modes.

An analytic result for the propagation constant of the transmission line mode can be
obtained by solving for zeros of a modal equation in the complex plane [25]. To extract the
phase and attenuation constants from the NEC solutions a 10 wire with radius 10 6 and
400 segments was modeled over ground. The source was located on segment 4, or the inside
end for NEC-5, for a position on the wire of about x = 0.1 . A matched load at the opposite
end of the wire was determined to minimize the standing wave so that the propagation
constant could be determined accurately. The matched load can be determined numerically
with one solution with NEC by saving the current I(x). The current is reversed and added
to the driven current as Im (x) = I(x) + VL I(L x) for wire length L. A measure of the
standing wave
P is obtained as a mean-squared estimate of the second derivative of magnitude
2
as ISW = i [ |Im (xi 1 )| + 2|Im (xi )| |Im (xi+1 )|] , summing over values of i away from
the source and load regions, and this value is minimized by adjusting the complex VL in a
minimization routine. A result is shown in Fig. 3, showing the original current with standing
wave and the current with matched load.

The current is assumed to propagate as exp( ik0 ↵x) where ↵ is a normalized propagation
constant. The normalized phase constant, Re[↵], and normalized attenuation constant,
Im[↵], are found by linear regression on log[|Im (x)|] and the phase of Im (x) after unwrapping
the phase.

42
 1.0

Current Magnitude
0.5

0.2

0.1
0 2 4 6 8 10
x/λ
Fig. 4.1.3 Current on a 10 wire with radius 10 6 at height h = 0.1 over ground with ✏r = 16
showing the current with no termination (red) and with numerically matched termination (green).

The propagation constants determined in this way for NEC-4 and 5 models of wires
and also NEC-5 strip models using triangular and rectangular patches are plotted in Fig.
4 and are in good agreement with the solution of the modal equation from [25]. There is
increased di↵erence for the strip models near the ground, but this may be a result of the
model di↵erence.

1.7 0.10
Normalized Atten. Const. (−Im[α])

Chang and Olsen

Normalized Phase Const. (Re[α])

1.6
NEC-4
0.08
1.5 NEC-5 wire

1.4 NEC-5 triangles 0.06

NEC-5 rectangles Chang and Olsen
1.3
0.04 NEC-4
1.2 NEC-5 wire
0.02 NEC-5 triangles
1.1
NEC-5 rectangles
1.0 0.00
10-5 10-4 10-3 10-2 10-1 1 10-5 10-4 10-3 10-2 10-1 1
h/λ h/λ

Fig. 4.1.4 Normalized propagation constant on a wire at height h over ground with ✏r = 16 for
the structure-attached mode exp( jk0 ↵x).

4.2 Loop Antennas over Ground

Modeling loop antennas over ground has always been problematic with NEC-4. Electri-
cally small loops, at least in the vertical orientation, often give negative input resistance.
This may result from the limited accuracy of the interpolated Sommerfeld integrals (about
3 to 4 places) and the treatment of scalar potential in the point-matched solution. Since the
mixed-potential solution ensures that the approximated integral of scalar potential around
the loop is zero, whether the potential is accurate or not, it might be expected to do better
than NEC-4.

43
 The first test is for an electrically small horizontal loop over ground. An analytic result
for the impedance of a small horizontal loop with uniform current was developed by Wait
[26]. For a small horizontal loop with radius a and wire radius c at height h0 over ground
Wait gets the input impedance as

Zin = iµ0 !a(P + S) (1)

where
Z 1
P =⇡ [J1 (x)]2 exp( B0 x) dx (2)
0
Z 1
x (x2 + j2A2 )1/2
S=⇡ [J1 (x)]2 exp( Bx) dx (3)
0 x + (x2 + j2A2 )1/2

with J1 (x) the Bessel function, A = a/ for skin depth = (2/ µ0 !)1/2 , B0 = c/a and
B = (2h0 + c)/a. The term P represents the primary inductance of the loop and S is due to
ground.

Wait [26] has the evaluation the integral for P as

✓ ◆
2 k02
P = 1 K(k0 ) E(k0 ) (4)
k0 2

where k0 = 2(B02 + 4) 1/2 and E and K are the elliptic integrals. However, Mathematica
[19] evaluates the integral in (2) as
2 + B02
P = K( 4/B02 ) B0 E( 4/B02 ). (5)
B0
These results are not the same, and (5) agrees with the numerical integration of (2), so that
value was used. It also agrees with numerical evaluation of the integral in Eq. (1) in [26].
The integral in Eq. (3) must be evaluated numerically. S does not include the radiation
resistance of the loop, so the radiation resistance of a small loop, Rrad = 320⇡ 6 a4 / 4 was
added.
175
1.0 NEC-4
Input Resistance (ohms)

Input Reactance (ohms)

174 NEC-5
Wait
0.1 173

172
0.01
NEC-4
NEC-5 171

0.001 Wait
170
10-5 10-4 10-3 10-2 10-1 1 10-5 10-4 10-3 10-2 10-1 1
h/λ h/λ

Fig. 4.2.1 Input impedance of a horizontal loop over ground comparing NEC–4, NEC–5 and Wait
[26].

44
 The input impedances from NEC-4 and 5 for a horizontal loop with ka = 0.05 and wire
radius 10 6 at height h over ground with ✏˜r = 16 j160 ( = 0.0267 S/m at 3 MHz) are
plotted in Fig. 1 along with the solution from[26] using Eq. (5) for P . The agreement for
resistance seems very good with NEC-5 the closest to the analytic solution. For reactance
NEC-4 and 5 are in good agreement. The analytic solution does not get the trend for small
h, but overall is within a percent. Using Eq. (4) from [26] for P the reactance goes from
177.2 ohms at h/ = 1 to 176.7 ohms at h/ = 10 5 , so that is not an improvement. A
smaller loop might be better for the [26] solution, but possibly more difficult for NEC. May
try that sometime.

The input impedances from NEC-4 and 5 for a vertical loop over ground are plotted in
Fig. 2 versus ka. The thickness parameter of the loop was ⌦ = 2 log(2⇡a/c) = 15 for loop
radius a and wire radius c. The lower edge of the loop is at height h over the dielectric
ground with ✏r = 16. The results are in good agreement, with the typical downward shift in
the resonant frequency for the mixed-potential solution with linear basis functions relative to
NEC-4. The input resistances are plotted in Fig. 3 and 4 for small ka and h/a varying from
0.1 to 10. NEC-4 produces negative input resistance for small ka while NEC-5 results appear
reasonable. The NEC-4 results are in reasonable agreement with NEC-5 for ka greater than
about 0.16 or h/a of 10 or greater.
15
NEC-4 NEC-4
25
NEC-5 NEC-5
Input Resistance (ohms)

10
Input Reactance (ohms)

20
5
15
0
10
-5

5 -10

0 -15
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
ka ka

Fig. 4.2.2 Input impedance versus ks of a vertical loop over ground with thickness parameter
⌦ = 2 log(2⇡a/c) = 15 for loop radius a and wire radius c. The lower edge of the loop is at 0.1a
above the ground with ✏r = 16.

45
 0.6
h/a NEC-4
0.1
Input Resistance (ohms)
0.4 0.2
0.5
1.0
2.0
0.2 5.0
10.0

0.0

-0.2
0.00 0.05 0.10 0.15
ka

Fig. 4.2.3 Input resistance versus ka from NEC-4 for a small vertical loop over ground with loop
radius a and wire radius c for thickness parameter ⌦ = 2 log(2⇡a/c) = 15. Ground permittivity is
✏r = 16 and height of the lower loop edge h/a is varied.

0.6
h/a NEC-5
0.5 0.1
0.2
Input Resistance (ohms)

0.5
0.4
1.0
2.0
0.3 5.0
10.0
0.2

0.1

0.0
0.00 0.05 0.10 0.15
ka
Fig. 4.2.4 Input resistance from NEC-5 for a small vertical loop over ground with loop radius a
and wire radius c for thickness parameter ⌦ = 2 log(2⇡a/c) = 15. Ground permittivity is ✏r = 16
and height of the lower loop edge h/a is varied.

46
4.3 Monopole on a Radial-Wire Ground Screen

A monopole on a buried radial-wire screen can be modeled, as in NEC–4, with the base of
the monopole at the ground plane and a segment sloping down from that point to the desired
depth and then a horizontal wire to the required radius. Results are shown in Fig. 1 for a /4
monopole on a screen of 32 radials at a depth of 10 4 with wire radius 10 5 . The ground
had a complex relative permittivity of 15 j15. The radiation efficiency was determined
by evaluating radiated field for ✓ from 0 to 90 degrees and at 0 and 5.63 degrees, halfway
between radials, and computing average gain. The value for average gain assumes that
radiation into the region not integrated repeats the pattern of the region integrated, where
really there is no radiation into the lower half space, so the average gain value is divided by
two to get radiation efficiency into the upper half space. NEC–4 and 5 are in close agreement
except for very small screen radius, possibly because the di↵ering representations of charge
on the short screens have an e↵ect. NEC–4 evaluated the Sommerfeld integrals for E while
NEC–5 evaluated the integrals for potentials, so when they agree it is very likely the result
is correct for the ideal ground screen model.
35
30
Radiation Efficiency (%)

25
20
15
NEC-5
10
NEC-4
5
0
0.01 0.02 0.05 0.1 0.2 0.5
Screen Radius/λ
Fig. 4.3.1 Radiation efficiency of a /4 monopole on a screen of 32 radial wires with varying screen
radius.
In entering the model the sloping segment and horizontal radial are entered first and
rotated for 32 radials with a GR 0, 32 command, then the monopole is entered so that
NEC–5 can take advantage of the partial symmetry. With 32 radials and 23 segments on
each radial the time for NEC–5 to fill the matrix was 14 times faster than NEC–4.
A source of uncertainty in modeling antennas with ground is the ground parameters
permittivity and conductivity. Often ground parameters are not measured, but average
values for the ground type are assumed. Plots are included in Appendix B in an attempt
to show the sensitivity to ground parameters for a monopole on a radial-wire screen. The
antenna is seen to become less sensitive to ground parameters as the number of radials is
increased.

47
 5. Conclusion
Many of the results here were chosen to illustrate known limitations of NEC–4, and NEC-
5 is seen to be more accurate for most situations. One case where NEC–4 beats NEC–5 is in
convergence for a dipole, since NEC–5 uses linear, triangular basis functions while NEC–4
has a higher order sinusoidal basis. The mixed-potential code should converge faster with
a higher-order basis, but there are no plans to add that to NEC–5. So NEC–4 may remain
the preferable code for modeling antennas such as Yagis where accurate resonant frequencies
are critical.

In cases where NEC–4 and 5 agree we can have considerable confidence in the accuracy of
the results. This is particularly valuable for antennas with ground, where it is difficult to find
independent results, and measurements usually involve uncertainties in physical modeling
errors, such as ground parameters and the contact of wires and surfaces with the ground.
When NEC–4 and 5 agree it seems very likely that the results are correct for the idealized
model, since NEC–4 is evaluating electric field while NEC–5 is evaluating potentials and
di↵erent methods are used to enforce the boundary conditions. In cases such as a small loop
over ground it is obvious that NEC–4 fails.

Results in this manual were generated in debugging and validating NEC–5. Other com-
parisons may be added in the future as they are developed or if contributed or suggested by
users.

48
 References

[1] G. J. Burke and A. J. Poggio, Numerical Electromagnetics Code (NEC) – Method of

Moments, UCID-18834, Lawrence Livermore National Laboratory, CA, January 1981.

[2] G. J. Burke and A. J. Poggio, Numerical Electromagnetics Code (NEC–4) – Method of

Moments, UCRL-MA-109338, Lawrence Livermore National Laboratory, CA, January
1981.

[3] G. J. Burke, A. J. Poggio, Numerical Electromagnetics Code NEC–5 – Method of Mo-

ments, User’s Manual, LLNL-SM-742937, Lawrence Livermore National Laboratory,
CA, November 22, 2017.

[4] S. M. Rao, D. R. Wilton, A. W. Glisson, “Electromagnetic Scattering by Surfaces of

Arbitrary Shape,” IEEE Trans. Antennas and Propagation, Vol. AP-30, No. 3, pp.
409-418, May 1982.

[5] W. A. Johnson, D. R. Wilton, R. M. Sharpe, Patch Code Users’ Manual, Report

SAND87-2991, Sandia National Laboratories, Albuquerque NM, May 1988.

[6] S. U. Hwu and D. R. Wilton, Electromagnetic Scattering and Radiation by Arbitrary

Configurations of Conducting Bodies and Wires, Tech. Rept. 87-17, University of Hous-
ton, May 23, 1988.

[7] D. R. Wilton, S. U. Hwu, Junction Code User’s Manual, Report 87-18, University of
Houston, May 30, 1989.

[8] S. M. Rao, “Electromagnetic Scattering and Radiation of Arbitrary Shape Surfaces by

Triangular Patch Modeling,” Ph.D. dissertation, University of Mississippi, Aug. 1980.

[9] M. F. Costa and R. F. Harrington, “Minimization of Radiation from Computer Sys-

tems,” International Elecrical Electronics Conference and Exposition Proceedings,
Toronto, Canada, pp. 660-665, Sept. 1983.

[10] M. F. Costa and R. F. Harrington, “Electromagnetic Radiation and Scattering from

a System of Conducting Bodies Interconnected by Wires,” Report TR-83-8, Syracuse
University, April 1983.

[11] R. M. Sharpe, J. B. Grant, N. J. Champagne, W. A. Johnson, R. E. Jorgenson, D. R.

Wilton, W. J. Brown, J. W. Rockway, “EIGER: Electromagnetic Interactions GEneR-
alized”, IEEE Antennas and Propagation Society International Symposium Proceed-
ings, 1997, Montreal, Quebec, Canada, 13-18 July 1997.

[12] K. A. Michalski, D. Zheng,“Electromagnetic Scattering and Radiation by Surfaces

of Arbitrary Shape in Layered Media, Part I: Theory”, IEEE Trans. Antennas and
Propagation, Vol. AP-38, No. 3, pp. 335-344, March 1990.

49
[13] K. A. Michalski, D. Zheng,“Electromagnetic Scattering and Radiation by Surfaces of
Arbitrary Shape in Layered Media, Part II: Implementation and Results for Contiguous
Half-Spaces”, IEEE Trans. Antennas and Propagation, Vol. AP-38, No. 3, pp. 345-352,
March 1990.

[14] G. J. Burke, NEC5GI Graphical Interface for NEC–5, LLNL-SM-747061, Lawrence

Livermore National Labotratory, CA, March 1, 2018.

[15] D. H. Werner, “An Exact Formulation for the Vector Potential of a Cylindrical Antenna
with Uniformly Distributed Current and Arbitrary Radius,” IEEE Trans. Antennas
and Propagation, Vol. 41, No. 8, pp. 1009-1018, Aug. 1993.

[16] D. H. Werner, “A Method of Moments Approach for the Efficient and Accurate Mod-
eling of Moderately Thick Cylindrical Wire Antennas,” IEEE Trans. Antennas and
Propagation, Vol. 46, No. 3, pp. 373-381, March 1998.

[17] P. C. Waterman, “Matrix Formulation of Electromagnetic Scattering,” Proceedings of

the IEEE, Vol. 53, pp. 805-812, August 1965.

[18] B. D. Popović, M. B. Dragović and A. R. Djordjević, Analysis and Synthesis of Wire

Antennas, Research Studies Press, New York, 1982.

[19] Wolfram Research, Inc., Mathematica, V. 10, Champaign, IL, 2014.

[20] R. W. P. King, The Theory of Linear Antennas, Harvard Univ. Press, Cambridge, MA,
1956.

[21] G. J. Burke, Treatment of Small Wire Loops in the Method of Moments Code NEC,
UCID-21196, Lawrence Livermore National Laboratory, CA, October, 1987.

[23] S. Bhattacharya, S. A. Long and D. R. Wilton, “The Input Impedance of a Monopole

Antenna Mounted on a Cubical Conducting Box,” IEEE Trans. Antennas and Propa-
gation, Vol. 35, No. 7, pp. 756-762, July 1987.

[24] A. W. C. Chu, S. A. Long and D. R. Wilton, “The Radiation Pattern of a Monopole

Antenna Attached to a Conducting Box,” IEEE Trans. Antennas and Propagation,
Vol. 38, No. 12, pp. 1907-1912, Dec. 1990.

[25] D. C. Chang and R. G. Olsen, “Excitation of an Infinite Antenna Above a Dissipative

Earth,” Radio Science, Vol. 10, No. 8, 9, pp. 823-831, Aug.-Sept. 1975.

[26] J. R. Wait and K. P. Spies, “Low-Frequency Impedance of a Circular Loop over a

Conducting Ground,” Electronics Letters, Vol. 9, No. 15, pp. 346-348, July. 1973.

50
 Appendix A

A.1 Accuracy of Kernel Evaluation

Errors in evaluating a potential integral on a wire segment using the extended kernel
and the reduced kernel relative to the exact kernel are shown in Fig. 3.1.1. This appendix
includes plots of the errors as ⇢ and z are increased away from the source segment

For the self-term, Fig. 1, the error in the extended kernel is seen to be nearly independent
of and to increase with increasing radius a due to the approximate evaluation of the first
integral in (3) containing the phase information. The error in the reduced kernel depends on
the thickness ratio, and is poor when /a is on the order of 1 or less. Similar plots are shown
in Fig. 2 through 4 for ⇢ equal to 3, 10 and 100 times a. With increasing ⇢ the extended
kernel accuracy remains limited by the phase error. For the reduced kernel the error related
to /a decreases with increasing ⇢, and the phase error becomes the limiting factor at large
a/ .
Extended kernel Reduced Kernel
10-1 10-1
10-2
0.7
Dêa=0.001 10-3 Dêa=0.001

10-2 10-4 10-2 0.5

0.2
Dêa=0.01 10-5 Dêa=0.01
10-1
a/λ

a/λ

10-3 10-6 10-3 10-2

10-3
Dêa=0.1 10-7 Dêa=0.1 10-4
10-8 10-5
10-4 10-4
10-6
10-9
10-7 -8
10
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. 10-10 Dêa=1 Dêa=10. Dêa=100. Dêa=1000.
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. A.1 Errors in the extended and reduced kernels relative to the exact kernel for the evaluation
point at the center of the segment (self-term) with ⇢ = a for the extended kernel, ⇢ = 0 for the
reduced kernel.

51
 Extended kernel Reduced Kernel
10-1 10-1
10-2 0.1
Dêa=0.001 Dêa=0.001
10-3
10-2 0.08
10-2
10-4
Dêa=0.01 Dêa=0.01
10-5 0.078 0.07
a/λ

a/λ
10-3 10-3 0.05
10-6
0.02
Dêa=0.1 10-7 Dêa=0.1
10-4 10-8 10-4 0.01
10-9
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. 10-10 Dêa=1 Dêa=10. Dêa=100. Dêa=1000.
0.007
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. A.2 Errors in the extended and reduced kernels at ⇢ = 3a and z = 0.

Extended kernel Reduced Kernel

10-1 10-1
0.1
10-2
Dêa=0.001 Dêa=0.001 0.01
10-3
10-2 10-2 0.008
10-4
Dêa=0.01 Dêa=0.01 0.0075
10-5
a/λ

a/λ

10-3 10-3
10-6 0.007
Dêa=0.1 Dêa=0.1 0.005
10-7
10-4 10-4 0.002
10-8
10-9 0.001
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. 10-10 Dêa=1 Dêa=10. Dêa=100. Dêa=1000.
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. A.3 Errors in the extended and reduced kernels at ⇢ = 10a and z = 0.

Errors with increasing z are shown in Fig. 5 through 7. For these plots ⇢ was equal to a
for both extended and reduced kernels. In the reduced kernel solution the evaluation point
would be on the axis, but putting it on the surface gave a smaller error and is the way that
it is used in evaluating the first integral in (3). Again, the extended kernel is limited only
by the phase error which decreases with increasing z. For the reduced kernel both the error
related to /a and the phase error decrease with increasing z.

52
 Extended kernel Reduced Kernel
10-1 10-1
10-2 10-2
Dêa=0.001 Dêa=0.001

10-2 10-3 10-3

10-2
10-4 3(10-4)
Dêa=0.01 Dêa=0.01
10-5 10-4
a/λ

10-3

a/λ
10-3
10-6 8(10-5)

Dêa=0.1 Dêa=0.1
10-7 7(10-5)
10-4 10-4
10-8 5(10-5)
10-9 2(10-5)
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. Dêa=1 Dêa=10. Dêa=100. Dêa=1000.
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. A.4 Errors in the extended and reduced kernels at ⇢ = 100a and z = 0.

Extended kernel Reduced Kernel

10-1 10-2 10-1
0.13
10-3 0.11
Dêa=0.001 Dêa=0.001

10-2 10-4 10-2

0.1
10-5
Dêa=0.01 Dêa=0.01
a/λ

a/λ

10-3 10-6 10-3 0.05

10-2
Dêa=0.1
10-7 10-3
Dêa=0.1
10-1110-12 10-4
10-4 10-8 10-4
10-10 10-5
10-9 10-6
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. Dêa=1 Dêa=10. Dêa=100. Dêa=1000. 10-7
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. A.5 Errors in the extended and reduced kernels at ⇢ = a and z = Max( , a).

53
 Extended kernel Reduced Kernel
10-1 10-1
10-3 0.05
Dêa=0.001 10-4 Dêa=0.001
0.02
10-2 10-5 10-2

Dêa=0.01
10-6
Dêa=0.01
10-2
a/λ

a/λ
10-3 10-7 10-3
10-3
10-8 10-4
Dêa=0.1 Dêa=0.1
10-5
10-4 10-9 10-11
10-4 10-6
10-7
10-10
10-8
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. Dêa=1 Dêa=10. Dêa=100. Dêa=1000.
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. A.6 Errors in the extended and reduced kernels at ⇢ = a and z = 5 ⇥ Max( , a).

Extended kernel Reduced Kernel

10-1 10-1
10-3 0.02
10-4 0.01
Dêa=0.001 Dêa=0.001
10-5
10-2 10-2
10-6 0.005
Dêa=0.01 Dêa=0.01
10-7
a/λ

a/λ

10-3 10-3
10-8 10-3
10-4 -5
Dêa=0.1 Dêa=0.1 10
10-9
10-4 10-4 10-6
10-10 10-7
10-11 10-8
Dêa=1 Dêa=10. Dêa=100. Dêa=1000. Dêa=1 Dêa=10. Dêa=100. Dêa=1000.
10-5 10-5
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
Δ/λ Δ/λ

Fig. A.7 Errors in the extended and reduced kernels at ⇢ = a and z = 10 ⇥ Max( , a).

54
 Appendix B

B. E↵ect of Ground Parameters for a Monopole on a Radial-Wire Screen

Sources of uncertainty in modeling antennas that involve ground are the ground pa-
rameters and condition of the ground. Often ground permittivity and conductivity are not
measured and just average values for the ground type are assumed. Plots in this appendix
demonstrate the sensitivity of a monopole on a radial-wire ground screen to ground permit-
tivity and conductivity for ground screens with 8, 16, 32 nd 128 radials.

The optimum efficiency of a radial-wire ground screen near a quarter wave in radius
occurs when the screen radius is about a quarter wave in electrical length in the ground.
Since this length would depend on the ground parameters and proximity of the wires to the
ground surface, the free-space wavelength is used for these plots, with the screen radius 9 /4,
somewhat longer than optimum, and monopole height also 9 /4. The screens are buried at
10 4 0 in the ground and all wire radii are 10 5 0 .

Other conditions a↵ecting an antenna in ground include ground parameters varying with
depth, irregular ground surface and contact conditions and chemical reactions between the
wires and the ground. These are factors for real antennas rather than the ideal antennas
modeled with NEC and are not considered here.

Fig. B.1 shows radiation efficiencies of a monopole on a radial-wire ground screen for
varying ground parameters, relative permittivity and conductivity, Scales are shown for
both conductivity at 1 MHz and the imaginary part of the complex relative permittivity
✏˜r = ✏r j /(!✏0 ). Some people find these efficiencies to be surprisingly low. Much of the
power may enter the ground beyond the radius of the ground screen, so only higher ground
conductivity can increase efficiency.

Fig. B.2 and B.3 show the input resistance and reactance of the antennas for varying
ground permittivity and conductivity. As expected, more radials in the screen reduces the
variation of antenna impedance with ground parameters. The color scale in these plots
remains constant between antennas but the contour lines change.

55
 104 104
8 Radials 80 16 Radials 80
75
-1 75
70 10 70 10-1
103 103

Conductivity (S/m) at 1 MHz

Conductivity (S/m) at 1 MHz

10-2 10-2
100 45 100
40 45

10-3 10-3
−Im( ε~r)

−Im( ε~r)
10 35 10 15 40
30 20 35
10 15 20 25 30
25
-4
20 10 10-4
1 25 1
30
30
10-5 10-5
0.1 0.1

10-6 10-6
0.01 0.01
1 20 40 60 80 1 20 40 60 80
Re( ~
εr) Re( ~
εr)

104 104
32 Radials 80
128 Radials 80
75
10-1 75 10-1
3 70
10 103 70
Conductivity (S/m) at 1 MHz

Conductivity (S/m) at 1 MHz

10-2 10-2
100 50 100 55

50
10-3
−Im( ε~r)

45
10-3
−Im( ε~r)

40
10 20 25 10 40 45
30 35 20 25 30 35

-4
10 10-4
1 15 1

10-5 10-5
0.1 0.1

10-6 10-6
0.01 0.01
1 20 40 60 80 1 20 40 60 80
Re( ~
εr) Re( ~
εr)

Fig. B.1 Radiation efficiency in percent for a /4 monopole on a radial wire ground screen for
varying complex relative permittivity ✏˜r = ✏r j /(!✏0 )

56
 Resistance (ohms) Resistance (ohms)
30 40 50 60 70 80 90 100 24 28 32 36 40 44 48 52

104 104
8 Radials 40
16 Radials 40
10-1 10-1
103 103
42

Conductivity (S/m) at 1 MHz

Conductivity (S/m) at 1 MHz

10-2 44 10-2
100 100
50 46
10-3 10-3
−Im( ε~r)

−Im( ε~r)
10 10
48
70 60
10-4 50 10-4
1 60 50 1

10-5 10-5
0.1 0.1

10-6 10-6
0.01 0.01
1 20 40 60 80 1 20 40 60 80
Re( ~
εr) Re( ~
εr)

Resistance (ohms) Resistance (ohms)

24 28 32 36 40 44 24 26 28 30 32 34 36 38 40
4
10 104
32 Radials 128 Radials

40 10-1 10-1
103 103 39
Conductivity (S/m) at 1 MHz

Conductivity (S/m) at 1 MHz

10-2 39 10-2
100 42 100
40 38
10-3 10-3
−Im( ε~r)

−Im( ε~r)

10 42 10
40 37
10-4 10-4
1 1

10-5 10-5
0.1 0.1

10-6 10-6
0.01 0.01
1 20 40 60 80 1 20 40 60 80
Re( ~
ε r) Re( ~
εr )

Fig. B.2 Input resistance for a /4 monopole on a radial wire ground screen for varying complex
relative permittivity ✏˜r = ✏r j /(!✏0 )

57
 Reactance (ohms) Reactance (ohms)
-10 0 10 20 30 40 50 60 70 80 -4 0 4 8 12 16 20 24 28 32
4
10 104
8 Radials 16 Radials
10-1 17 10-1
103 103
20 18

Conductivity (S/m) at 1 MHz

Conductivity (S/m) at 1 MHz

10-2 19 10-2
100 100 20
18
10-3
−Im( ε~r)

10-3

−Im( ε~r)
20
10 30 10
17
10-4 10-4
1 1
20

10-5 10-5
0.1 10 0.1

10-6 10-6
0.01 0.01
1 20 40 60 80 1 20 40 60 80
Re( ~
εr) Re( ~
εr)

Reactance (ohms) Reactance (ohms)

0 16 19 0 2 4 6 8 10 12 14
4 8 12
4
104 10
32 Radials 128 Radials
10-1 10-1
103 103
17
15

Conductivity (S/m) at 1 MHz

Conductivity (S/m) at 1 MHz

10-2 10-2
100 100
17 14
10-3
−Im( ε~r)

10-3
−Im( ε~r)

16 13
10 10
12

10-4 10-4
18 1
1

10-5 10-5
0.1 0.1

10-6 10-6
0.01 0.01
1 20 40 60 80 1 20 40 60 80
Re( ~
εr) Re( ~
εr)

Fig. B.3 Input reactance for a /4 monopole on a radial wire ground screen for varying complex
relative permittivity ✏˜r = ✏r j /(!✏0 )

58