Engineering Perspective, Part Discrete Wavelet Techniques': Tutorial Wavelets From An Electrical 1
Engineering Perspective, Part Discrete Wavelet Techniques': Tutorial Wavelets From An Electrical 1
Engineering Perspective, Part Discrete Wavelet Techniques': Tutorial Wavelets From An Electrical 1
IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998 IEEE
1045-9243/98/$10.0001998 49
One of the goals of this paper is to illustrate how one can generate
the scaling functions and the wavelets, specially tailored to one's
needs.
50 /€€E Antennas and Propagation Magazine, Vol. 40, No. 5 , October 1998
Next, the signal is reconstructed from the decimated trans- H ( z ) = h(O)+ h(l)z-'+ ...+~ ( N ) z - ~ . (14)
mitted signals (.) and .(.). The signals are up-sampled by a
factor of two to produce ~ " ( n and) ~ " ( n )The
. principle of up- Without loss of generality and for convenience we choose
sampling is shown in Appendix 2 (Section 9). Then, they are
filtered by two receiving filters, g'(n) and h'(n) . The outputs are El'(.)= z-"H(z-') so that h ' ( n ) = h ( N - n ) . (15)
combined to form ."(n). Now, let us see how x(n) is related to its
estimate X(n) , and then the methodology to extract x(n) will be The factor Z-" is used to guarantee causality of the filters [Le.,
obvious. A'(.) = 0 for n < 01. The high-pass filter g'(n) is chosen in such a
way that
Please observe that
V ' ( z )= G ( z ) X ( z ) . (6)
In addition, we define the high-pass filter g ' ( n ) in terms of the
From Appendix 1 (since ~ ( n and
) v(n) have been decimated by a low-pass filter coefficients h'(n) by choosing
factor of two)
g'(n)=-(-l)"h'(N -.)=-(-l)"h(n),
1
U ( Z ) = -[ut(&)
2 + V'(-&)], (7)
g(.) = (-l)"h(N- n) = (-l)%'(n),
U"(z)=u ( z 2 ) = 21 [ u ~ ( z ) + z i ' ( - z ) ] ,
(9)
Substitution of Equations (18) into Equation (13) demonstrates that
all four equations are consistent, and the aliased component due to
V " ( z )= V ( z 2 )= ,[V'(z)+
1 V+)] (10) X(-z) is zero.
'[
k ( z ) = - z- N G ( ~ ) G ( ~ - z' )- "+H ( z ) H ( z - ' ) ] X ( z )
2
1
= -2- { [ G ( z ) X ( z +
) G(-z)X(-z)]G'(z) (19)
= -[H(-z-')H(-z)
1 + H(z)H(z-')]z-"X(z).
+[ H ( z ) x (2) + N(- z ) X ( -z)]H'( z ) } (1 1 ) 2
Let H ( z ) be a FIR (finite-impulse-response) filter of order N + 1. The filters H ( z ) and G ( z ) are called quadrature mirror fil-
Then, h(n) will have N +1 terms. We consider N to always be ters (QMF), because they have symmetry around the point 7d2, as
odd. Then, shown in Figure 2. Ideally, we would prefer H ( z ) and G ( z ) to be
IEEE Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998 51
So, if we solve only for H ( z ) , the other three filters of Figure 1
will then be given by Equations (24),(25), and (26).
d n 2 7t '
0 h(O)h(2)+ h(l)h(3)= 0 (for z 2 and z - ~ ) . (30)
then from Equation (18), H(e") = H(1) = h(0)+ h(1) + h(2)+ h(3)= A . (34)
H ' ( z ) = z-"H(z-')
In addition one has Equations (29) and (30). We need one more
= h(3) + h(2)z-1 + h ( ~ ) +~h -( ~~ ) ~ - ~ , equation, as these four equations [Equations (29), (30), (35), and
(36)] are linearly dependent, since Equations (35) and (36), in
and finally from Equation ( 1 8), conjunction with Equation (30), lead to Equation (29). The
question is how to find the fourth equation. Without the fourth
G ' ( z )= z-NG(z-*)= -H(-z) equation it is not possible to get the complete solution. Here, the
various methodologies differ, and different researchers have come
= -h(O) + h(1)z-I - h(2)z-Z + h(3)z-3. up with different procedures.
I 7-1
Hence, one can represent a dilated version of &)-namely,
&2)-by a combination of the functions 4(t - m ) [or dm ( t ) ]with
some coefficients h'(nz), resulting in the dilation equation
4 4
N N
4(t)=~ C h f ( m ) 4 ( 2 1 - m ) = 1 / 2 C h ( N - m ) ~ ( 2 t - m )(41)
.
m=O nz=o
54 IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998
Here, " d t "can be represented as the incremental interval length.
Therefore, from Equation (41), utilizing Equation (42), we get 4(t)= f i [ * 2 - ' / 2 h ' [ 2 k t ) ]where
, * denotes a convolution. Note
k=l
that the sequence of convolutions is carried out by various com-
j C ( t - k)dt = 1 =
I
Jic m=O
hQ)j
1
4(2t-2k-m)
2
JiN
d(2t) = - h'(m)
2 m=O
pressed versions of the same signal. When the sequence of convo-
lutions converges, it yields the function # ( t ) .
where N + 1 is the order of the filter (which is always even). Here, The wavelet is generated from Equation (45), and is given by
the filter coefficients h'(m) and g'("z) have been assumed to be
, a given g ' ( m ), is called the mother of
real. The function ~ ( t )for -1 f o r O < t < l / 2
wavelets. (This might be viewed as the traditional Judeo Christian
concept of mother: where the mother is generated from the father!)
0 otherwise
So, from an electrical-engineering perspective, if we have the
filter coefficients h ' ( m ) , then the scaling function 4(t) can be since y/(t) = -4(2t)+ 4(2t - 1). This is shown in Figure 8. This is
obtained by iteratively solving the dilation Equation (41). This is the pulse doublet, and it is related to the Haar wavelet. Please note
carried out by starting with letting #(t) be a pulse function, and that the scaling function @(t)is in this case orthogonal with respect
then filtering by h ' ( m ) , and continuing until convergence is to its own translates, i.e.,
reached. Once the scaling function is known, the wavelets are a)
given by Equation (45).
p(t)+(t - m)dt = 0. (52)
-m
The scaling function can also be solved for in the transform
domain. By taking the Fourier transform of Equation (41), one can
This is also true for the wavelets:
write the dilation equation in the w domain as
m
A
= --N
$(a>
I
) $(w/2),
-JW/2
(e (46) j ~ ( t ) l y (-
t m)dt = 0 (53)
2 -m
where i ( w ) is the Fourier transform of &), and H ' ( z ) is the z For N = 3 , we obtain the case presented earlier. Note that for this
transform of h'(n) (see Equation (15)). One can repeatedly apply case, the filter coefficients hi are given by Equation (38). Hence,
Equation (46) to obtain
IEEE Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998 55
the scaling function can be obtained from the solution of the The confusing part here is what to use as the starting point for
dilation equation, a scientific endeavor to carry out a numerical analysis utilizing
wavelets? There are two choices:
&)= $[(1+.11)~(21)+(3+.11)~(2t-l)
1. Do we start with + ( t ) ,construct h ' ( n ) ,and then generate y ( t )
+ (3 - &)&t - 2)+ (1 -.11)4(2t - 3)]. (54a) [from Equation (45)]?; or
The wavelets are generated in an analogous fashion by using 2 . Do we start with h ' ( n ) , and then create $(t) and ~ ( t? )
Equations (45) and (38), resulting in
For the discrete case that we arc dealing with, the answer is
v(t)=$[(1-&)4(2t)-(3-&j$(2t-l) straightforward: that is, we design h'(n), and then obtain # ( t ) and
~ ( t. This
) is also much simpler in practice. However, for the dis-
+ (3 + &)4(2t - 2)- (1 + &)4(2t - 3)]. (54b) crete wavelet transform, as we shall see, 4(t) and ~ ( t are
) not at
all required in the numerical computation!
We are not going to delve further into the solution of the dilation
equation, since for the discrete wavelet transform and for the
In summary, the mathematical basis of the wavelets has been
problems that we are interested in electromagnetics-namely, solu-
presented from a filter-theory perspective. How to construct scal-
tions of large matrix equations-the scaling functions and the
ing €unctions 4 and wavelets y has been shown, starting from the
wavelets are really not necessary. This is because the discrete
wavelet representation can be carried out from the knowledge of filters h'(m) , and utilizing the perfect-reconstruction argument
only h(m)! However, we present some other wavelets for illustra- presented with subband-filtering techniques. Once h'(m) i s known,
tion purposes. 4 can be generated from Equation (41), and v , from Equa-
tion (45).
For example, consider the Shannon wavelet, which is the
dual of the Haar wavelet. The scaling function is given by
4. Approximation of a function by wavelets
H ' ( z )= - c
1 + ' h;(2n - 1)[z-**+' + 2 - 1 1 , The scale factor 2?12, appearing in Equation (56), is there so as to
make the functions ~ ~ , ~orthonormal,
( t ) Le.,
2 n=l
56 IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998
and these are the discrete wavelet coefficients of the function ~ ( t. ) and from Equation (45),
These are the same values shown in Figure 5. Our objective in this
section is to establish that isomorphism.
) 2-"24(2-'t - j ) .
@ l , j ( t= (60)
m m
2 ( n )= j p ( t ) @ (-
t n)dt = j p ( t ) ChF(t)&@(2t - 2n - k)dt
-m -m k
We now define the coefficients q nthrough
(68)
= Ch'(k)a(-')(2n+k)=Co(-l)(k)h'(k-2n),
k k
and
It is now shown how the dk,n are evaluated recursively through
the C k , n .
bo(.)= Ca(-l)(k)g'(k-2?2).
CI
We have, from Equations (58), (61), and (62), that the fol-
lowing orthonormal set Given the existence of relationships like Equations (68) and (69),
and drawing the isomorphism between
/€E€ Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998 57
it is not necessary to even know what the scaling functions and
wavelets are, as one can directly use Equations (70a) and (70b)
without going through the mathematical derivations, as Figures 5
and 6 illustrate. One starts with c ~ , - which
~ , is the coefficient gen-
erated by correlating 4(2t) with the function ~ ( ,t and
) then recur-
sively computes the discrete wavelet transform mathematically
through Equations (70a) and (70b), and graphically using Figure 5 ,
which is easier to visualize from a filter-theory perspective. The
methodology is the same. The process described so far is similar to
the transmitter part as labeled in Figure 1. If the 4(2t) are the
impulse scaling functions, then q - 1 will be equivalent to the
sampled version of ~ ( t namely
), x(n).
58 IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998
By utilizing Equations (66) and (67) in Equation (65b), we bits are utilized to code the compressed image, the reconstruction
is still better than the conventional JPEG algorithm [21].
In the following examples, we consider the image to be a There are essentially two different ways to solve operator
two-dimensional array. We take the discrete wavelet transform in equations utilizing wavelet techniques. The first possibility is to
two dimensions by utilizing a recursive relationship similar to use the wavelets as basis functions, in a conventional Method-of-
Equations (70a) and (70b). We utilize an eighth-order filter that has Moments formulation. The second approach is to use traditional
been designed to match the signal [21]. Once the discrete wavelet sub-sectional basis functions, and obtain a dense complex matrix.
coefficients are obtained, they are thresholded, and then the origi- The wavelet techniques are then used to compress the elements of
nal image is reconstructed utilizing the recursive relation of Equa- the matrix, and either a direct sparse solver or an iterative method
tion (74). The objective is to illustrate that even though only a few is used to obtain the solution of the sparse-matrix system. The
n=l
(a, cos nx + b, sin nx) , (79)
generates a locally finer grid, are usually used when the solutions
of the integral equations or the differential equations are known to
vary widely in different domains. By non-uniform gridding, one in which the Fourier bases are linearly independent and orthogonal.
can reduce the size of the problem and improve the accuracy. The This yields a solution y = -sinx . Hence, it does not provide the
multi-level or the multi-grid technique has been widely used in d2Y formed
solving differential equations and integral equations [28-321. correction solution. This is because the series for -,
dx
Kalbasi and Demarest [33, 361 applied multilevel concepts to solve after the double differentiation of Equation (79), is not complete,
the integral equation by the Moment Method on different levels, as the a. term is missing. So, Ae, does not form a complete set
which has been called the Multilevel Moment Method. No matter [13-171. Hence, Equation (79) does not and cannot provide the cor-
what the multi-grid technique is, the basis functions for an rect solution. To rectify the error, one should choose the basis
improved approximation have to be reconstructed again. By using functions as [ 15,161
the multi-scale technique in one dimension, the basis functions for
the new scale have to be reconstructed. The new approximation
grids formed by the multi-scale technique are the same as those for
the multilevel technique; however, the constructions for the func-
y = co + c p + c2x2 + c
m
U=1
(a, cos nx + b,, sin nx) . (80)
tions are different.
These basis functions are not even linearly independent in the
In addition to these numerical concems, there are some philo- interval 0 to 2z,as both x and x2 can be approximated by the
sophical concems that need to be addressed in the selection of an remainder of the h c t i o n s representing the Fourier series. How-
appropriate set of basis functions. Consider the solution of an
operator equation ever, if we form __,
d2Y we obtain
dx2
AX=Y, (75) m
IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998 61
finite elements has been illustrated. Initial results illustrate how to g‘(n) (see Equation (17) for the relationship between h‘(n) and
use the wavelet techniques to generate a sparse matrix, utilizing a g‘(n)). Therefore, multiplying a vector (say [Y]) by [PI is
wavelet-like basis. A similar methodology can also be used in equivalent to filtering (in the transmitted portion) in Figure 1, fol-
making sparse matrices-arising in the implementation of finite lowed by a sub-sampling of two. The matrix-vector product yields
elements in the solution of differential forms of Maxwell’s equa- u(n) and v(n) . The first four elements are equivalent to u(n) and
tions-still sparser.
the last four are v(n). The matnx-vector product has already
In the next Section 5.2, how to use the discrete wavelet tech- incorporated the sub-sampling by a factor of two. The sub-sam-
niques to transform a complex dense matrix, arising in conven- pling by a factor of two is accomplished by the shift between the
tional MOM, into a sparse matrix, utilizing a set of orthogonal elements of each row of the matrix [PI. Now, for [PI to be an
transformations, is illustrated. Some work [35-381 has already been orthogonal matrix, it is necessary that the following three equa-
done to address this topic. This is described next. tions, similar to Equation (29) [the normalization constant is set to
unity] and Equation (30) [the filter is orthogonal to its two-shifted
version] hold:
5.2 Solution of large matrix equations by the discrete wavelet
transform C h 2 ( i )= 1,
1
-O*h(O)+lh(l)-4h(2)+9h(3)-16h(4)+25h(5)= 0 . (87)
I J
0 0 -h, +h, -h2 +h3 4 4 +h5
-h4 +A5 0 0 -h, +h, -h, +h3 If x ( t ) has p derivatives, its wavelet coefficients decay like
-h2 +h3 -h4 +hs 0 0 -h, +hi 2-np [38]:
62 IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998
However, since we are dealing with a finite number of terms,
the drop-off rate of the wavelet coefficients is of little significance.
The solution of the above equations can be obtained analytically,
and has been given by Daubechies [ 11 as
l + f i + J X
h(5) = h’(0) = -
16&
[p2116x16[y~~16x1 [y~116x1
wavelet coefficients
5 +f i +3 & Z T
h(4) = h’(1)= -
16&
[[y,’l1+ scaling function contribution
10 - 2 4 1 3 + 2 4 5 + 2&0
h(3) = h’(2) = -
16&
where
IEEE Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998 63
dimensional discrete wavelet transform is applied to the system
matrix [A], the resulting system matrix will be sparse if all the
elements below a threshold are set to zero. Typically, one would
Even though the sizes of the various [Pl]s are not the same, we
make them the same by supplementing, say, [P2] by a diagonal
unity matrix, [I], to make it the same size as [P,]. Thus,
Next, take that sparse matrix [Ba], and try to reconstruct [A]
where [I] is the identity matrix. So, we take the wavelet transform by computing
of [XI and [Y]to form [X'] and [Y'], and we take the wavelet
transform of [A] to form [B] . Hence, Equation (93) reduces to
[B][X'] = [Y'] with [B] = [S][A][S]T. (95) Define an average value, 6, of the error between the elements of
[ A a l and [AI by
The unknown [XI is solved for fi-om this by
[B] is the two-dimensional wavelet transform of [A], and has From Tables 1-6, it is seen that the matrix [A] can be recovered
been computed by a series of one-dimensional transforms to its from [B,] to provide [A,] with an average error that is lower than
integer part of’). Please note that the wavelet transform of a real
o:-[ 1 Error in
Reconstruction 1 Cond[B,] function is always real.
oi=
: I Error in
Reconstruction 1 Cond[B,] matrix equation of size Q. This i s essentially the contribution of
Beylkin, Coifman, and Rokhlin [18].
IEEE Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998 65
As a final example, consider the electromagnetic scattering
from an array of wires, randomly spaced. We considered 56 thin-
wire antennas. Six were of length 2.7A and radius 0.0051. The
remaining 50 were 32 long and of the same radius. The 56 wires
were located inside a parallelepiped of dimensions
271xX5/2x21/2. The usual MOM application led to a
2096 x 2096 matrix. The matrix was compressed, utilizing a filter
h ( n ) of length 16. The compression for the real part of the imped-
ance matrix was 17.8%, Le. 749289 of the elements of the matrix
c x (4 YD(4
Figure 12. The symbol used to represent decimation by a factor
of two.
There are a few points that are worth mentioning. First of all,
if the compression was not done on the real and the imaginary
4 3
>
parts of the matrix separately, then the degree of compression was
merely 35%, as opposed to 17.8%. This is significant. Secondly,
the size of the impedance matrix has to be a power of two. Thirdly,
the conjugate-gradient method takes the same number of iterations
(95) to converge to the solution when applied to the original dense
matrix, or to the sparse matrix, as the transformation is presumably
orthogonal, from a strictly theoretical point of view. However, now
as the entire compressed matrix is in memory, the number of page
faults is small, and so the result can be obtained quite efficiently.
8. Acknowledgment
66 IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998
= ‘[x(S)
2
+ Xj-&)I.
If we observe the spectrum, then one observes that the original sig- Figure 18. The spectrum of the original signal.
nal has the spectrum given in Figure 14. Once the signal is down-
sampled, the spectrum is as given in Figure 15. Hence, the spec-
trum of YD(e’”) is aliased, and it is the sum of
Pictorially, up-sampling can be represented by the symbol or, in the transform domain
shown in Figure 16. In the sampled domain, this is equivalent to
inserting a zero between the sampled signals, as shown in Figure
17. Mathematically, this is equivalent to
11. References
/E€€ Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998 67
tion,” Applied and Computational Harmonic Analysis, 2, 1995, pp. 25. C. Su and T. K. Sarkar, “A Multiscale Moment Method for
61-67. Solving Fredholm Integral Equation of the First Kind,” in J. A.
E=mg (ed.), Reports on Progress in Electromagnetic Research,
11. S. Schweid and T. K. Sarkar, “Iterative Calculating and Fac- Volume 17, Cambridge, Massachusetts, EMW Publishing, 1997,
torization of the Autocorrelation Function of Orthogonal Wavelets pp. 237-264.
with Maximal Vanishing Moments,” IEEE Transactions on Cir-
cuits and Systems, CAS-42, 11, November 1995, pp. 694-701.
26. T. K. Sarkar, R. S. Adve, L. Castillo and M. Salazar, “Utiliza-
tion of Wavelet Concepts in Finite Elements for an Efficient Solu-
12. R. Ansari, C. Guillemot and J. F. Kaiser, “Wavelet Construc-
tion of Maxwell’s Equations,” Radio Science, 29, July 1994, pp.
tion Using Lagrange Half-Band Filter,” IEEE Transactions on Cir-
cuits andSystems, CAS-38, 1991, pp. 1116-1118. 965-977.
13. M. A. Krasnoselskii, G. N. Vainikko, P. P. Zabreiko, Ya B. 27. C. Su and T. K. Sarkar, “Electromagnetic Scattering from
Rutitskii and V. Ya Stetsenko, Approximate Solution of Operator Coated Strips Utilizing the Adaptive Multiscale Moment Method,”
Equations, Groningen, The Netherlands, Walters-Noordhoff, 1972. in J. A. King (ed.), Reports on Progress in Electromagnetic
Research, Volume 18, Cambridge, Massachusetts, EMW Publish-
14. S. G. Mikhlin, The Numerical Performance of Variational ing, 1998, pp. 173-208.
Methods, Groningen, The Netherlands; Walters Noordhoff, 1971.
28. A. Brandt, “Multi-level Adaptive Solutions to Boundary Value
15. K. Rektorys, VariationalMethods in Mathematics, Science and Problems,” Mathematics of Computation, 31, 1997, pp. 330-390.
Engineering, Dordrecht, Holland, D. Reidel Publishing Company,
1975. 29. W. Hackbusch, Multigrid Methods and Applications, New
York, Springer-Verlag, 1985.
16. T. K. Sarkar, “A Note on the Choice of Weighting Functional
in the Method of Moments,” IEEE Transactions on Antennas and 30. S. F. McCormick, Multigrid Methods: Theoly, Applications
Propagation, AP-33,4, April 1985, pp. 436-441. and Supercomputing, New York, Marcel Dekker, 1988.
17. T. K. Sarkar, A. R. Djordjevic and E. h a s , “On the Choice of 3 1. J. Mandel, “On Multilevel Iterative Methods for Integral Equa-
Expansion and Weighting Functions in the Numerical Solution of tions of the Second Kind and Related Problems,” Numer. Math.,
Operator Equation,” IEEE Transactions on Antennas and Propa- 46, 1985, pp. 147-157.
gation, AP-33, 9, September 1985, pp. 988-996.
32. P.W. Hemker and H. Schippers, “Multiple Grid Methods for
IS. G. Beylkin, R. Coifman and V. Rokhlin, “Fast Wave Trans- the Solution of Fredholm Integral Equations of the Second Kind,”
forms and Numerical Algorithms,” Communications on Pure and Mathematics of Computation, 36, 153, 1981.
AppliedMathematics, 44, pp. 141-183.
33. K. Kalbasi and K. R. Demarest, “A Multilevel Enhancement of
19. A. P. Calderon, “Intermediate Spaces and Interpolation, the the Method of Moments,” in Seventh Annual Review of Progress in
Complex Method,”Stud. Math., 24, 1964, pp. 113-190. Applied Computational Electromagnetics, March 199 1, Monterey,
California, Naval Postgraduate School, pp. 254-263.
20. W. H. Press, S, Teukolsky, W. T. Vetterling and B. P. Flemery,
Numerical Recipes-The Art of Scientijic Computing, Cambridge, 34. K. Kalbasi and K. R. Demarest, “A Multilevel Formulation of
Cambridge University Press, 1992. the Method of Moments,” IEEE Transactions on Antennas Propa-
gation, AP-41, 5, May 1993, pp. 589-599.
21. S. C. Schweid, “Projection Method Minimization Techniques
for Smooth Multistage QMF Filter Decompositions,” PhD thesis, 35. Z. Baharav, Y . Leviatan, “Impedance Matrix Compressing
Syracuse University, May 1994. Using Adaptively Constructed Basis Functions,” IEEE Transac-
tions on Antennas and Propagation, AP-44, 9, 1996, pp. 1231-
22. B. Z.Steinberg and Y. Leviatan, “On the Use of Wavelet 1238.
Expansions in the Method of Moments,” IEEE Transactions on
Antennas and Propagation, AP-41, 5, 1993, pp. 610-619. 36. R. L. Wagner and W. C. Chew, “A Study of Wavelets for the
Solution of Electromagnetic Integral Equations,” IEEE Transac-
23. J. C. Goswami, A. K. Chan and C. K. Chui, “On Solving First- tions on Antennas and Propagation, AP-43, 8, 1995, pp. 802-810.
Kind Integral Equations Using Wavelets 011 a Bounded Interval,”
IEEE Transactions on Antennas and Propagation, AP-43, 6, June 37. F. X. Canning, J. F. School, “Diagonal Preconditioners for the
1995, pp. 614-622. EFIE Using a Wavelet Basis,” IEEE Transactions on Antennas and
Propagation, AP-44, 9, 1996, pp. 1239-1246.
24. G. Wang, “A Hybrid Wavelet Expansion and Boundary Ele-
ment Analysis of Electromagnetic Scattering from Conducting 38. G. Strang and T. Nguyen, Wavelet and Filter Banks, Cam-
Object,” IEEE Transactions on Antennas and Propagation, AP-43, bridge, Wellesley-Cambridge Press, 1996.
2, February 1995, pp. 170-178.
68 IEEE Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998
Introducing Feature Article Authors PhD degree at Syracuse University. His research interests include
numerical electromagnetics and the use of signal-processing tech-
niques in numerical electromagnetics.
IEEE Antennas and Propagation Magazine, Vol. 40, No. 5,October 1998 69
nicaci6n from the Universidad PolitCcnica de Madrid (Spain). In
1993, he became a Research Assistant in the Departaniento de
Seiiales, Sistemas y Radiocomunicaciones (Signals, Systems and
Radiocommunications Department) in the Escuela TBcnica Supe-
rior de Ingenieros de Telecomunicaci6n of the same university.
Since 1997, he has been an Associate Professor of the Departa-
mento de Ingenieria Audiovisual y de Comunicaciones (Audiovis-
ual and Communications Engineering Department) in the Escuela
Univesitaria de Ingenieria TCcnica de Telecomunicaci6n of the
same university, where he teaches microwave theory and the
related laboratory. His research activities and interests are focused
on the application of numerical methods, mainly finite elements, to
electromagnetic problems, including research on curl-conforming
finite elements, the characterization of multiconductor and Rafael Rodriguez Boix received the BSc, MSc, and PhD
waveguide structures, analysis of scattering and radiation prob- degrees in physics from the University of Seville, Spain, in 1985,
lems, and the use of wavelet theory in computational electromag- 1986, and 1990, respectively. Since 1985, he has been with the
netics. Other research areas of his interest are network theory and Electronics and Electromagnetics Department at the University of
filter design. Seville, where he became an Associate Professor in 1994. Durmg
the summers of 1991 and 1992, he was at the Electrical Engineer-
He has authored four contributions for chapters and articles ing Department of UCLA as a Visiting Researcher Also, d u n g
in books, six papers in international joumals, and 29 papers in the summer of 1996, he was at the Computer and Electrical Engi-
intemational symposiums, plus a number of national publications neering Department of Syracuse University as a Visiting
and reports. He has participated in seven projects and contracts Researcher. His current research activities are focused on the
financed by international, European, and national institutions and numerical electromagnetic analysis of planar microwave circuits
companies. and printed circuit antennas. %’
The first announcement booklet and call for papers for the XXVIth General Assembly of the
Intemational Union of Radio Science is now available. It includes the schedule for the sessions
of the 10 URSI Commissions, as well as the instructions and format for submitting papers. There
is also information on the Young Scientists Program and a Canadian Student Competition.
The information in this first announcement booklet is essential for anyone wishing to submit an
abstract of a paper to be presented at the General Assembly. The booklet can be obtained by
sending a request with full address and contact information to
Those interested can also obtain the information, and request being placed on the mailing list, at
70 IEEE Antennas and Propagation Magazine, Vol. 40, No. 5, October 1998