Professional Documents
Culture Documents
Volakis
David B. Davidson Dept. E&E Engineering University of Stellencosch Stellenbosch 7MK). South Africo (t27)21 808 4458 (127) 21 808 4981 (Fox)
dovidsonOflrgo.sun,oc,za (e-moil)
volokls@umlch.edu (emoil)
1. Introduction
tandard Maflab@ has all the necessary ingredients in order to develop a simple and interactive antenna toolbox. These include: (i) built-in surface-mesh generators in two and three dimensions; (ii) highly efficient matrix solvers; (iii) Fourier analysis tools; and (iv) two-dimensional and three-dimensional plots with rotation, zoom, and scroll options.
100
To date, there are two EM antenna-related packages built on Matlab: The FemlabB software of the COMSOL group (Sweden, Finland, US), and SuperNEC from Poynting Software Ltd., South Africa [l]. In SuperNEC, an executable C++ file is still employed in order to perform the key MOM calculations, whereas Mallab essentially plays the role of a GUI. Femlab includes a number of finite-element modules, among which is an Electromagnetics Module developed to perform finite-element simulations of quasistatic and high-frequency fields in Maflab. The emphasis here is on
IEEE AntennosandPropogofionMagazine, Vol, 43. NO. 5 .October 2M1
-mAAA-I
.1
4.8
4.8
4.4
4.2
0.2
0.4
0.6
0.8
xm
Figure l a . A surface mesh created with the PDE toolbox for a thin strip with 131 triangles.
parameter. To export the mesh to the main workspace, select the Export Mesh option from the Mesh menu. The Matlab workspace will then have three arrays: p . an array of Cartesian-node coordinates x and y; I , an array of node numbers for each triangle; and e, an array of boundary edges. In a similar fashion, a variety of planar meshes can be created, including strip and slot antennas. Three-dimensional surface meshes require either a custom formula for the z coordinate, i.e., z = f( x , y ) , or the use of three-dimensional Delaunay tessellation, implemented in the Mutlub function delaunay3. Now, adding a monopole to a finite-sized ground plane will be considered. In Figure lb, a 0.2 m long and 0.0134 m wide strip is connected to a 2 Y 2 m square plate. Figure 3b shows the monopole area at a higher resolution. The structure can also be generated using the Maflub PDE toolbox, as follows. The plate rectangle is created using rectangle tool.Another small rectangle is then drawn exactly in the middle of the plate. There are two reasons to do so. First, we will refine the mesh close to the anticipated monopole. Second, the position of the common edge and its size can be controlled by the position and size of the small enclosed rectangle. After that, a strip is attached to the common edge using mesh rotation (coordinate y is replaced by 2). The script
Figure l b . A surface mesh created with the PDE toolbox for a monopole on a finite ground plane (734 triangles)
Figure le. A surface mesh created with the PDE toolbox for slot antenna (550 triangles).
quasi-statics and microwaves, but not on the antenna analysis. The complete Femlub EM module is also rather expensive. This paper discusses the use of the standard Muflub package, including the partial-differential-equation (PDE) toolbox, in order to simulate radiation and scattering of simple metal antennas. The electric-field integral equation [ 2 ] , Rao-Wilton-Glisson (RWG) basis functions [3], and the feeding-edge model (cf. [4]) constitute the underlying theoretical background. A few working examples are considered, including a thin-strip dipole model, a monopole on a finite ground plane, and a slot broadband antenna. All simulations were done using Marlab 6 Releose 12.
where N is the number of inner edges. The MOM equation relates the vector of expansion coefficients, I = { I , ) , and the applied voltby age vector, V = {Vm),
101
Z.I=V,
(2)
where an N x N complex impedance matrix 2 has the same form for scattering or radiation problems [3]. For impedance matrixes larger than approximately 500 x 500. a preliminary Muflub memory allocation must he made in the form
function generator [ I l l or the feeding-edge model [4, 12, 131. This model assumes a gap of negligible width, A (Figure 2b). When A tends to zero, an infinite electric field, E, is predicted within the gap; that is the delta-function approximation
E = VG(y)ny.
(4)
z e r o s ( N ,N)+j*zeros(N, N I ;
prior to matrix filling, otherwise the corresponding Muflub script will he terminated. Memory allocation makes it possible to use direct solvers for complex matrixes (double precision) as large as 5000 x 5000 in size (on a Pentium Ill, 933 MHz, with 512 Mb RAM and a 30 Gigabyte hard drive). Iterative solvers, which require only one row of the impedance matrix at a time, are practically unlimited in size, hut consume a large amount of CPU time. A number of methods to calculate the impedance matrix exist (cf. [7]), starting with the classic paper [3]. Mostly, they differ in how the integrals over the triangles are calculated. Analytical approaches (line integrals and potential integrals) are accurate and fast, hut require extensive preliminary mathematical work. An altemative is to use numerical integration over a triangle [SI. If quadrature points do not coincide with the triangle midpoint, no separate formulas for diagonal elements of the impedance matrix are necessary. All elements of the impedance matrix can he calculated in a straightforward fashion, using the same formula. We employ the so-called harycentric subdivision of an arbitrary triangle [9]. Any primary triangle is divided into nine equal small suhtriangles by the use of the 113 rule. Further, we assume that the integrand is constant within each small triangle. Then, the integral of a function g over triangle T, is equal to
Equation (4) sim y states that the integral of the electric field over the gap is equa;;o / the applied voltage, V. It is very convenient to associate the gap with an inner edge, m, of length l, (Figure 2h). The voltage vector in Equation (2) will have all zeros, except for only one RWG element, m,i.e.,
gles, and A, is the area of the primary triangle. The system of Equations (2) is solved using one of the builtin Marlub matrix solvers. The simplest method is the matrix inversion, i.e., I = i n v ( 2 ) *v. In practice however, it is u n n e c e s s q to form the explicit inverse of a matrix and, indeed, this is generally regarded as bad numerical practice. A better way, from both execution-time and numerical-accuracy standpoints, is to use the matrix division operator, I=z\v.This produces the solution using Gaussian elimination, without forming the inverse. If the matrix size is larger than 3000 x 3000, using an iterative procedure like the Generalized Minimum Residual method (GMRES) [IO] may he considered. The built-in Muflub function gmres is faster than Gaussian elimination when the number of iterations is relatively small (10-30).
2 W 1
Table 1. The input impedance of the strip model and of the wire model of a half-wavelength, 2-m-long dipole a t 75 MHz. T h e strip width is 0.02 m; the equivalent wire radius is 0.005 m.
where indices 1, 2 label two distinct RWG junction elements. The input impedance of the monopole is again in good agreement with the corresponding NEC simulation and with the prediction of the theory (one-half that of the equivalent center-driven dipole, of total length twice the monopole).
Zh-U
since a component of the RWG basis function f,n normal to the edge is always equal to one [3]. In the following, we assume that the feeding voltage, V, is equal to 1 V. Using RWG basis functions (Equation (I)) only, the basis function f, will contribute to the impedance calculation, since no other basis functions have a component normal to edge m. Since the component of the RWG basis function f, normal to that edge is always equal to one, the total normal current through the edge is given by /,I,. The antenna impedance is therefore
4 0
4.8
4.5
4.4
4.2
0.0
0.2
0.4
0.6
0.8
Rm
1.0
Figure 3a. The surface-current distribution for a center-fed thin strip (dipole) antenna a t different frequencies. The gray scale extends from minimum to maximum current magnitude.
Table 1 compares the input impedance and far-field parameters of different strip models, assuming center-feeding. For the purposes of comparison, we present the corresponding ,result obtained by a NEC thin-wire solver. SuperNEC, from Poynting Software Pty. Lid. [I], is used to model a 2 m long wire with an equivalent radius of 0.005 m (cf. [I41 for the equivalence formula). The wire is divided into 39 segments. Table 1 indicates that coarse strip meshes (strip0 and stripl) give relatively large errors in the input impedance, compared to the result of the wire solver with 39 segments. However, finer meshes (strip2 and strip3) agree well with the wire solver, as well as with each other. There is also a good agreement of the reported data with the data measured by King [15] for similar antenna parameters. For the base-driven monopole, the feeding edge at the junction requires special consideration. It is common to both the plate and the strip. Therefore, there are two RWG elements that correspond to the same edge. To separate these two elements, it is convenient to "double" the junction edge, i.e., just repeat it twice in the mesh code. The model of the delta-function generator is then applied, similar to that for the dipole. Since the feeding edge is doubled, the contribution of each single edge should he taken into account. Equations ( 5 ) and (6) are transformed to
Figure 3b. The surface-current distribution for a base-driven quarter-wavelength monopole on a ground plane. The gray scale extends from minimum to maximum current magnitude.
17)
2
TI
Figure 3c. The surface-current distribution for a slot antenna a t 1 GHz. T h e gray scale extends from minimum to maximum current magnitude.
103
yz-plane, gain=2.2 dB n
ow
,m
50
Figure 4c. The dipole radiation intensity over a spherical surface (with a radius of 100 m). The gray scale extends from minimum to maximum intensity magnitude.
180
Figure 4a. The directivity pattern of modeled by a thin strip 0.2 plane).
half-wavelength dipole
yz-plane, gain=4.0 dB
Figure 4d. The monopole radiation intensity over a spherical surface (with a radius of 100 m). The gray scale extends from minimum to maximum intensity magnitude.
180
Figure 4b. The directivity pattern of a quarter-wavelength monopole on a finite ground plane (vr plane).
104
Figure 3a, a thin strip of length 2h = 2 m with M = 244 triangles is considered at four different frequencies. Since the strip has a finite width, the current at the feeding point is not exactly zero at 2h = 1 or 2 1 . For the monopole (Figure 3b), the frequency is 375 MHz, and the monopole length is 0.2 m (a quarter-wavelength monopole). The slot antenna in Figure 3c is calculated at 1 GHz. Once the surface current is known, the electric and magnetic fields everywhere in space can be found, either using surface inte/FEE AntennasondPropaQationMoQazine, Vol. 43. No. 5. October 2cO1
grals [2, 31 or a more elegant dipole model [4j. The dipole model assumes that every RWG edge element behaves like a dipole of constant current, located between the centroids of the triangles adjacent to each edge. The dipole moment, m, which is the product of an effective dipole current and an effective dipole length, is obtained by integration of the surface current, corresponding to edge element m,over the element surface, i.e.,
E m
1'1
m=
I
T,f+T;
I ~ f , ( r ) ~ ~ = ~ ~ ~ , ,( r ~ - - r ~
(8)
i
Jm
4
dw a
05 i
where the index c denotes the triangle's midpoint. The derivation of this integral is discussed in [3]. A misprint in Equation (8) of reference [3] should be noted: it should be r:- - ! r : , instead of r," - ri' . The product /,I, is associated with the dipole current,
l. whereas the effective dipole length, i, is given by l r ~ - - r ~ The radiated magnetic and electric field of a short dipole at a point r are (cf. [W) H(r) = *(m 47r x r)CCrh.
i5
2
f.
25
3.5
4.5
Figure 6a. The impedance of the slot antenna as a function of frequency. The real part of the impedance is shown by the solid line; the dots indicate the imaginary part
Equations (9) hold in both the near and far fields of an antenna. Figures 4a and 4h show two radiation patterns: for a halfwavelength dipole (a 1:lOO strip with 244 triangles) and for a quarter-wavelength monopole (734 triangles), respectively. The built-in Matlab function p o l a r has a well-known problem: it is unable to handle negative directivity values in dB. To avoid these negative values, an offset should be introduced. The polar diagram is then rescaled similar to that in Figures4a and 4b. Another
Figure 6h. The power received by the second antenna at 1 m as a function of frequency. The transmitted voltage amplitude is 1 V. The circles are the result from the Friis transmission formula; the solid line is from the scattering calculations. inviting possibility is a plot of the radiation intensity over a large sphere, shown in Figures 4c and 4d for the dipole and monopole, respectively. If the figure is obtained using function fi113, as explained in this section, the Matlab command r o t a t e 3 d on turns on mouse-based three-dimensional rotation of the, figure. Rotation can be used to examine the intensity behavior in different directions.
frequency band by simply converting the corresponding Matlab scripts to functions of frequency. (To reduce the number of sampling frequencies, an interpolation procedure for the impedance matrixes may he been employed [17]). Here, the frequency response is calculated directly, over the hand 50 MHz - 5 GHz, at a sampling interval of 50 MHz. Figure 6a shows the input impedance of the antenna (the solid line is the real p a t , the dots are the imaginary part). The unbalanced antenna has a resonance at approximately 1 GHz. Otherwise, the input impedance is predominantly real, and has a relatively small variation between I GHz and 3 GHz. Figure 6h shows the total power received by a second identical antenna at 1 m, if the feeding voltage of the first antenna is I V. The power is calculated using two different methods. One of them is based on the Friis transmission formula [I61 (the circles). Another is based on a direc: calculation of the received voltage, when the incident field is the field radiated by the first antenna (the solid line). The corresponding scattering analysis is done with
1
0.8
0.6 o4
i 1
.
.
02
_ - -\
. .
.
L.
The received voltage over a frequency hand is a transfer function of the antenna system. If the transmitted voltage is a short pulse, the received voltage spectrum is obtained as a product of the transfer function and the input pulse spectrum. The Matlab functions f f t , ifft, f f t s h i f t are used to perform direct and inverse discrete Fourier transforms. Figure 6 shows the received pulse (the dotted line) if the transmitted voltage is a Gaussian pulse with zero DC content (the solid line). In the case of Figure 6a, the pulse duration is 1 us (the center frequency is 1.2 GHz). In the case of Figure 6b, a wider pulse, of 2 ns duration, is considered, The conclusion is that the present type of antenna is capable of transmitting a 1 ns pulse, hut fails if substantially wider pulses are considered.
7. Conclusions
.. Although the results discussed primarily have an educational character, they also demonstrate the capabilities of Marlab for solving some practical antenna problems. An extension ofthe present study to an array of dipoles or monopoles is very straightforward. Array optimization might involve existing Matlab tools (the optimization toolbox and the neural network toolbox). Two major problems associated with Matlab are the limitation on the size of the impedance matrix, and the relatively large running times. Whereas the first problem involves the design of an appropriate iterative solver, the second problem has two potential solutions. One of them is to avoid large loops, and to try to vectorize the Marlab code. Another solution is to use the Matlah compiler (the C/C++ compiler suite).
0.8
: I f0:
> 02
02
0.4
I
7
\,
8. References
1. A. Fourie and D. Nitch, SuperNEC: Antenna and IndoorPropagation Simulation Program, IEEE Antennas and Propagation Magazine, 42, 3, June 2000, pp. 31-48.
I v
A -
2. A. I. Poggio and E. K. Miller, Integral Equation Solutions of Three-Dimensional Scattering Problems, in R. Mittra (ed.), Compuler Techniques for Electromagnetics, Second Edition, Oxford, Pergamon Press Ltd., 1973, pp. 169-264.
3. S. M. Rao, D. R. Wilton, and A. W. Glisson, Electromagnetic Scattering by Surfaces of Arbitrary Shape, IEEE Transactions on Antennas andPropagation, AP-30,3, May 1982, pp. 409-418.
0.6 0.8
-1
6. G. J . Burke and A. J . Poggio, Numerical Electromagnetic Code NEC Method of Moments. Part 11: Program Description - Code, Lawrence Livermore National Laboratory Report UCID-18834, January 1981.
~
7. R. D. Graglia, On the Numerical Integration of the Linear Shape Functions Times the 3-D Greens Function or its Gradient on a Plane Triangle, IEEE Transactions on Antennas and Propagation, AP-41, 10, October 1993, pp. 1448-1455. 8. J. S. Savage and A. F. Peterson, Quadrature Rules for Numerical Integration over Triangles and Tetrahedra, IEEE Antennas and Propagation Magazine, 38,3, lune 1996, pp. 100-102.
9. Y. Kamen and L. Shirman, Triangle Rendering Using Adaptive Subdivision, IEEE Computer Graphics and Applications, MarchiApril 1998,pp. 95-103.
role in the world-wide communication of ideas. We are also a world-wide group of intelligent people. As such, we can - and I believe, should - play a role in how the societies in which we live react to what has occurred. The terrorist attacks of September I 1 and subsequent weeks have affected not just the US, but the world, and that most certainly includes all of us. I hope 1 will be forgiven if I share some personal thoughts. These arent intended as political comments: they are simply one engineers poor attempts to grapple with what has happened, and what should perhaps happen f r o m here. Part of the reason the effects of the attacks go far beyond the US is that they were not made just against the US: it is estimated that over one-third of those killed or missing in the World Trade Center were non-US citizens, from perhaps 65 other countries. That number alone is far greater than the total number of people killed in the worlds 10 worst prior temorist attacks. It has also become apparent - particularly, from the weapons used in the recent anthrax attacks - that these terrorists are not acting simply as groups of individuals, but hava at least some support from one or more nations. It is for these reasons, and others, that the war that the US and the many countries that have joined it have declared against the terrorists is, and will continue to be, a war involving much of our world. Terrorism is irrational: trying to force a country to change its political views by killing innocent citizens simply doesnt work. In fact, the recent attacks have had exactly the opposite effect: there is now a degree of unity of belief and purpose, both in the US and around the world, that probably hasnt existed since World War 11. Given modem communications and the degree to which countries .economies are intertwined, this may prove to be the greatest unity the world has yet known. In the case of the US, terrorism involves even an additional irrationality. The US was founded on the belief that all people ...are created equal, that they are endowed by their creator with certain unalienable rights, that among these are life, liberty, and the pursuit of happiness (those words are from the US Declaration of Independence, the founding document of the country). That belief transcends politics. Its even in common with some of what the terrorists claim they want. But beliefs are at the core of this terrorism, too. Unfortunately, those bcliefs have been twisted and molded into hatred and prejudice. People arent bom with hatred and prejudice: they are taught it. That is, perhaps, where we can make a difference. We have to educate people that racial, ethnic, religious, and political hatred are fundamentally wrong. We have to stop the teaching of such hatred in the guise of education. And where we are ourselves guilty of such hatred, we need to find the strength to leam otherwise. We must also recognize that people lacking the necessities for survival are perhaps easier prey for those who teach hatred. Its hard to hate someone who feeds you. Terrorism is irrational: thats why you cant negotiate with temorists. It is also probably impossible to re-educate a terrorist away from hatred. But through education, we might be able to prevent the creation of a future generation of terrorists. The greatest weapon of terrorism is fear. I pray that you will feel safe, and be safe.
10. Y. Saad, Iterative Methodsfor Sparse Linear Systems, Second Edition, Boston, PWS Publishing, 2000.
1I . B. D. PopoviC, M. B. DragoviC, and A. R. Djordjevit, Analysis and Synthesis of Wire Antennas, New York, John Wiley & Sons, 1982.
12. M. Davidovitz and Y. T. Lo, Rigorous Analysis of a Circular Patch Antenna Excited by a Microstrip Transmission Line, IEEE Transactions Antennas and Propagation, AP-37, 8, August 1989, pp. 949-958.
13. B. G. Salman and A. McCowen, The CFIE Technique Applied to Finite-Size Planar and Non-Planar Microstrip Antennas, Computation in Electromagnetics, (Conf. Publ. No. 420), 1996,pp. 338-341. 14. R. C. Johnson (ed.), Antenna Engineering Handbook, Third Edirion, New York, McGraw-Hill, 1993, p. 4-10.
15. R. W. P. King, Tables ofAntenna Characteristics, New York, IF1 Plenum Data Corporation, 1971, p. 39.
16. C . A. Balanis, Antenna Theory. Ana/ysis and Design, Second Edition, New York, John Wiley & Sons, 1997, pp. 133-136.
17. K. L. Virga and Y. Rahmat-Samii, Efficient Wide-Band Evaluation of Mobile Communications Antennas Using [Z] or [Y] Matrix Interpolation with the Method of Moments, IEEE Transactions on Antennas and Propagalion, AP-47, 1, February 1999, pp. 65-76. $2
Terror
The IEEE is a transnational organization of electrical engineers. It is, by definition, non-political (except for a semi-separate entity, the IEEE-USA). However, our technology plays a major
I07