Antenna Pattern Synthesis: A New General

Approach
OVIDIO MARIO BUCCI, FELLOW, IEEE, GIUSEPPE D’ELIA, GIUSEPPE MAZZARELLA,
MEMBER, IEEE, AND GAETANO PANARIELLO

The antenna pattern synthesis problem is of the utmost impor- determine the geometrical and electromagnetic structure of
tance in almost every kind of antenna applications. Therefore, a the antenna so that the latter fulfills the required specifi-
very large number of contributions have appeared on this subject.
cations.
But virtually all of them deal with simplijied versions of the com-
plete synthesis problem, wherein the degrees of freedom available Due to their ever-increasing relevance, many efforts have
in principle are strongly reduced, andlor idealized design criteria been devoted to the developement of reliable and efficient
or requirements are considered. design procedures for high-performance antennas, and a
In this paper we present a formulation which allows us to very large number of (partial) answers to the synthesis
overcome this fragmentation of the synthesis problem. A clear
and direct description of the performance actually required by problem have been proposed. To the best of our knowledge,
the antenna and a representation of the radiating properties of all adopt, from the very beginning, a restrictive meaning
the antenna as a system allows us to formulate the synthesis for the problem.
problem as an intersectionsfinding problem, i.e., rofind a common The first kind of specification concems the choice of
element between a number of sets, each one containing elements
fulfilling part of the requirements. This allows a completely general
the radiating structure: a reflector, an aperture, an array,
and flexible formulation of the problem, independent of the actual etc. This kind of division has been assumed virtually
structure of the antenna. Then the practical implementation of this in all the works on antenna pattem synthesis. Hence,
formulation is widely discussed, showing how an efJicient solution we have array or aperture synthesis methods [1]-[25]
procedure can be devised. The implications of the well-known ill-
and reflector synthesis methods [26]-1411. Due to the
conditioning of the synthesis problem are also discussed.
In order to show how the approach works and to assess its particularly simple relationship between source and radiated
flexibility and power, a couple of signif cant examples are included, field, the former methods are more sound and sophisticated
namely, a phase-only reconfigurable array and a shaped reflector from the mathematical point of view. As a matter of fact,
synthesis. These examples are unconventional since no a priori most of the theoretically oriented papers deal with linear or
choice of the intensity distribution (for the array case) or of thefeed
cluster (jor the refrector case) is required. The method presented planar sources [31, [5]-[91, [151, [161, 1201, [421-[441. On
is able to exploit all the available degrees of freedom in order to the other hand, most reflector antennas synthesis methods
fuljill the design requirements. are more heuristic, and rely on the availability of efficient
numerical codes for the evaluation of the radiated fields.
I. INTRODUCTION By using some kind of optimization algorithm, the system
The antenna synthesis problem can be quite generally parameters are varied until the design specifications are met
stated as the inverse of the analysis problem, i.e., given a [261-[281, [311, [331, P71.
set of design specification conceming Furthermore, because in all cases of practical interest the
relationship between the primary sources and the far field is
a) the required far-field pattem (or pattems, in the case
of scanning or reconfigurable beam antennas), not explicitly available, some kind of approximation must
b) the antenna structure and geometry, be made to obtain a reasonably simple, closed-form expres-
c) the feeding system sion for it. According to the required accuracy, Geometrical
Optics (GO), Physical Optics (PO), and Geometrical Theory
Manuscript received June 15, 1992; revised November 10, 1993. of Diffraction (GTD) are the classical tools used for this
0. M. Bucci, G. D’Elia, and G. Panariello are with Dipartimento di purpose.
Ingegneria Elettronica, Universith “Federico 11” di Napoli, Via Claudio,
2 1, 80 125 Napoli, Italy. Apart from the specification of the radiating structure,
G. Mazzarella was with Dipartimento di Ingegneria Elettronica, Univer- another restriction is usually made by choosing the part of
sitA “Federico 11” di Napoli, 80125 Napoli, Italy. He is now with Istituto the antenna allowed to be varied. Hence, in the avalilable
di Elettrotecnica, Universith di Cagliari, Piazza d’Ami, 09123 Cagliari,
Italy. literature, we have fuced structure procedures [3], [7], [ 101,
IEEE Log Number 9215535. [13], [14], [171, [22]-[28], [32]-[34], [38], whereinone tries

0018-9219/94$04.00 0 1994 IEEE

358 PROCEEDlNGS OF THE IEEE, VOL. 82, NO. 3. MARCH 1994

to fulfill the pattem specifications by varying the source for use in selected steps, where their efficiency can improve
excitation, and varying structure or jixed source techniques the overall performances without being affected by their
[29]-[31], [35]-[37], [40], wherein, for a given source, drawbacks.
the fulfillment of the far-field specifications is sought by This program is possible, since the mathematical differ-
changing the geometric and/or electromagnetic structure of ences between the various antenna systems and synthe-
the antenna. sis problems, while obviously relevant for the practical
The far-field requirements, with some notable exceptions implementation of a synthesis algorithm, are not very
([lo], [211-[23], [25], [391, [41], [45] among others), are significant in principle, and affect only the last stages of
expressed either in terms of performance indexes (gain, the development. Therefore, we can include in the same
efficiency, etc.) to be maximized, or in-terms of a “desired” framework problems which seem to have little in common,
pattem, often not a realizable one, to which the syn- such as the field and the power synthesis problems, or
thetized pattem should be as “near” as possible. However, reflector and array synthesis, as we are able to deal with
in all practical instances, the required antenna performance the full synthesis problem, i.e., the determination of both
is not strictly specified. Instead, we are given certain the antenna geometry and excitation at the same time.
tolerance limits, within which the performance indexes The approach we propose is developed in the following
(gain, pattem shape, sidelobe, cross-polar levels, etc.) must sections. However, some remarks are useful to put it in
lie. For this reason, it is not natural and satisfactory perspective, to highlight its novelty, and to help catch its
to formulate the synthesis problem as the search for a essence before entering into technical details.
pattem closest to a “desired” one. Last, but not least, A better understanding of the questions involved in a
many of the available procedures require the prescription synthesis problem can be gained if one considers that it can
of both amplitude and phase of the sought pattem and be embedded in a class of inverse problems. More precisely,
therefore can be labeled as field synthesis procedures, in it constitutes one of its two subclasses, the other one being
contrast to the power synthesis ones, in which only the that of the “identification” problem. This aspect has been
amplitude is prescribed. The latter problem is by far more thoroughly investigated in [46], [47] with reference to the
complicated but it is also the most important from a field synthesis problem. What really matters in our context
practical point of view. Pattem specifications for radar is the sensitivity of the solution, at variance with the case
and telecommunication applications often involve only the of “identification” problems, wherein the stability and the
amplitude of the pattem, so that the phase is a degree of uniqueness are the main instances.
freedom available to the designer. Nevertheless, the field The ill-posedness of the synthesis problem [46], [47]
synthesis problem is widely considered in literature, mainly makes any sufficiently general algorithm intrinsically ill-
because of its simplicity (it is inherently a linear problem in conditioned, and this, apart from numerical problems, leads
the fixed structure case) and also because it is amenable to a to solutions which can be very sensitive. Therefore, some
general and elegant formulation as a quadratic optimization kind of implicit or explicit “regularization” must be in-
problem in Hilbert functional spaces (see, e.g., [46]-[49]). cluded in order to obtain practically acceptable solutions.
On the other hand, the power synthesis problems can be The simplest way to regularize the problem is to require
explicitly solved only in particular and/or very idealized [46] that the solution has a small “norm” or satisfies
cases [3], [5], [7], [8], [13], [34].Hence, they are usually other constraints (possibly connected to suitable quality
formulated as nonlinear optimization problems with a finite criteria @I, [91, [ W , 1161, [201, W I , [381, [461-[481, [501,
but large number of parameters, which are solved by brute [5 11). This introduces the discussion about the inclusion
force, using more or less standard numerical techniques, of constraints into the synthesis problem. Constraining the
sometimes very time-consuming, and implemented without solution in some ways is very often desirable also from
any deep discussion of ill-conditioning and convergence the realization point of view. Without constraints, we can
properties [24], [26]-[28], [31], [33]. get a solution which is hard, or even impossible, to realize.
From this very short examination, it appears that the large For instance, in the array case, an excitation with too large
number of available synthesis procedures, even the most variations between neighboring elements can lead to such
useful ones, do not meet the need for a unified framework, an amount of mutual coupling that a corresponding feeding
general enough to accommodate the formulation of all network cannot be realized. For a shaped reflector, we
synthesis problems and capable of being translated into a can easily get irregular, or even discontinuous, surfaces,
flexible and efficient algorithm. The aim of this paper is a problem which is usually managed by some kind of
to present such a unified framework, as well as a general a posteriori smoothing. The relevance and necessity of
and flexible algorithm for the actual solution of the full a constrained synthesis has been already recognized [52]
synthesis problem. The approach relies on a functional but, nevertheless, it has been hardly considered in detail.
analysis setting, but to achieve the required generality, Constraints expressible in terms of quadratic functionals
the involved sets and operators cannot enjoy the useful [7], [13] or convex sets [21] have been considered in liter-
properties (mainly convexity and linearity, respectively) ature. Moreover, in the most popular synthesis procedures
which makes so elegant (and conceptually easy) the solution and software packages, explicit constraints on either the
of (fixed-structure) field synthesis problems. However, this excitation or the antenna geometry are not enforced. The
approach is able to retain existing optimization procedures exceptions are the obvious and mandatory ones stemming

BUCCI et al.: ANTENNA PA’ITERN SYNTHESIS 359

from the finite number of primary feeds (in the case of
fixed reflector synthesis) or array elements (in the case of
array synthesis), or also the finite number of basis functions
adopted to represent the reflector surface (in the case of
shaped reflector synthesis).
1
To the best of our knowledge, the first paper that broadly Fig. 1. System view of the radiating process. .z is the primary
field, s is the radiating object, and y is related to the radiated field.
discusses the constrained synthesis problem (for the m a y
case) is [53]. There, a distinction is made between the
constraints that a solution must strictly fulfill (e.g.. the the case of a reflector antenna, we can consider as input
number of array elements) and those which can be more or either the primary field or the current induced on the
less relaxed. In this paper, this distinction is still enforced, reflector surface; consequently, in the system description
leading to an explicit definition of two different sets: the the relation between primary field and induced currents is
solution must belong to the first of them and should belong to be included or not included. Concerning the far field,
to the other or, at least, be as near as possible to it. we can adopt different sets of independent variables to
Each requirement or constraint on the radiated field as individuate the observation direction, as well as different
well as on the antenna geometry and excitation, defines domains of definition for the field itself, according to the
a subset of all possible antenna systems, i.e., the subset angular extension of interest and/or some mathematical
of the antennas satisfying that requirement. The solution or practical convenience. Moreover, the system S can be
of the synthesis problem therefore amounts to searching split in one or more subsystems, which can be described
for an antenna which is acceptable according to all the separately, in order to avoid useless complications in the
requirements, i.e., an element of the intersection of all description. Of course, also this splitting can be made in
those subsets. This point of view first proposed in [54], has different ways. For instance, in the case of array synthesis,
been considered in [21], [23] and pursued by the authors one usually considers an external problem, wherein an ex-
as a basis for the formulation of the synthesis problem citation corresponding to the required far field is searched,
(see [25] with reference to array and [39], [55] for the and an internal problem wherein the design of the network
reflector antenna case). It is not only natural but also very feeding the array elements is considered.
powerful as shown in the following sections. It allows all Once X, S, and Y have been defined, the synthesis
particular synthesis problems to be seen under a unique problem amounts to determine an input z and a system,
general framework and, above all, to formulate and solve i.e., an s, in order to get a radiated field y with the required
for the first time the synthesis problems in its full generality. characteristics.
Letting A be the Cartesian product3
11. MATHEMATICAL FORMULATION OF THE
A = X X S X Y (1)
SYNTHESISPROBLEM
The logical process leading from the sources to the which is also a Hilbert space under the induced norm
far field radiated by a generic antenna can be described
according to the scheme of Fig. 1. The input z is the
primary excitation, the output y the corresponding radiated the object of a synthesis problem is a point of A, i.e., a
far field, and the system S represents the structure of the triple (z, s, y).
radiating system. From the mathematical viewpoint, S is Of course, not all the elements of A can be solutions
a continuous [56] operator, acting on the input z and of the synthesis problem. First of all it is required that
is identified by a suitable family, s, of parameters. This y is the actual response of the system s to the input
operator is linear with respect to z, but usually nonlinear z. Then we must consider that, in general, both z and
with respect to s. To be definite, the description requires s cannot be arbitrary. For instance, requirements on the
the choice of the sets X , S, and Y to which the input antenna size and/or primary excitations are always present
z, the parameter family s and the output y belong. In the and only those z and s which fulfill these requirements
following, we assume the sets X, S, Y to be (separable) can be accepted. Let D be a set of all the couples (z,s)
Hilbert spaces’ [56]. For the sake of simplicity we denote fulfilling the “realizability requirements.” We are therefore
the norm in each of these spaces with the same symbol led to define the set
11 . 11; when this can lead to confusion, we append a
suffix to the norm symbol. It must be stressed that the G = {(z,s,S(z,s) ) D)}c
: ( z , ~E A (3)
choice of the input and output spaces X and Y and of i.e., the graph of the function S(z,s) restricted to
the system space2 S is not univocal. For instance, in
D-pictorially shown in Fig. 2-as the set characterizing
--
‘ A more general setting could be obtained by letting X, S , and Y to all the realizable radiating structures. But the goal of our
be Banachspaces [56]. However, such a generality is scarcely required in synthesis must also fulfill the design requirements, which
practical applications. - allows to pick up the synthesized antenna from G. The
2As mentioned above, S is the space of the parameters, however, since
a point in this space identifies an operator S and vice versa, we will refer 3The Cartesian product X x S x Y is the set of all ordered pairs ( ~ , s , y ) ,
to it interchangeably as the space of systems or the space of parameters. wherein I is a point of X ,s is a point of S , and y is a point of Y.

360 PROCEEDINGS OF THE IEEE, VOL. 82. NO. 3. MARCH 1994

 wherein d(z, W) denotes the distance between the point z
and the generic set W , i.e.,
d ( z , W ) = inf llz - wII.
W EW

The sets G and AC can always be assumed to be closed.

As a matter of fact, the set AC is closed, except when
one or more constraints involve strong inequalities. In this
Fig. 2. Graph of the function S(r,s). case, we can get a closed set by simply replacing these
inequalities with “not larger than,” “not smaller than,”
which do not change the problem. Analogously, we can
requirements on the radiated field restrict the acceptable y assume D to be closed, which ensures that also G , the
to a subset Yc of the space Y.In the same way, we can graph of a continuous operator, is closed [56].
impose requirements also on the radiating structure and As a consequence of the closeness of the sets G and Ac,
on the sources, and these define the subsets S c and Xc, conditions (6), ( 7 ) are also sufficient to ensure that z E I .
respectively. Hence, the triple (z,s,y), corresponding to a Furthermore, when I is empty, i.e., when the synthesis
synthesized antenna, must also belong to problem does not have solutions in a strict sense, (7) is
useful as well, since it allows to classify the closeness of
(4) the realizable antennas to the set Ac. In other words, a
point z which minimizes (7) under (6) is the best choice
On the basis of these considerations, a solution of the
we can adopt when G nAC = 0, unless some requirement
synthesis problem must belong, at the same time, to both
is relaxed. In this case the solution belongs to G and this
G and Ac, i.e., to the intersection: explains the inclusion in G of all the realizability requisites,
i.e., those that must be satisfied.
I=GnAc (5)
And so, the synthesis problem is equivalent to the varia-
tional problem of finding a minimum point of (7) under the
between the set of the realizable radiating structures and the
condition (6), i.e., a point of G nearest to Ac. This amounts
set of all triples (z,s,y) satisfying the design specifications.
to finding a point (z,s) E D such that the functional
Each requirement on the radiating structure can affect
D, hence G (i.e., can be considered a requisite for the
realizability of the radiating structure) or AC (i.e., can
be considered a constraint), and the choice is a matter of
convenience. To be definite, we will assume that all the attains its infimum. Accordingly, it is relevant to discuss
unrenounceable requirements are included in the definition the conditions under which such a minimum point exists
of G and all the others in Ac. and can be explicitly found.
In the above formulation, the elements of I are considered The existence of a point of G nearest to AC is not
completely equivalent. Of course, we could choose among strictly required in our context. As a matter of fact, in
them by selecting the element of I which maximizes a any practical instance the synthesis requirements are always
suitable quality criterion. As we will see later, our approach affected by a certain imprecision, which makes the “exact”
is able to automatically include a selection method based attainment of the infimum an academic point. What really
on a quality criterion. matters is that we can find points of G whose distance
In many cases, the above general description can be from AC is as near as we want to the infimum. In other
simplified. For instance, if we are not interested in the deter- words, we only need a minimization algorithm providing a
mination of both the radiating structure and its excitation, sequence {zn} of points of G such that the corresponding
since one of them is fixed a priori and the correspond- sequence { d2 ( z , ,A c )} is not increasing and converging
ing set contains only one element, the presented scheme to the infimum of the functional a.
can be applied by considering only the two appropriate The “minimizing” sequence {zn} can fail to converge,
remaining sets: the output set Y and the excitation set X as it happens when the minimum point does not exist, but
or the parameter set S. Typical examples are those of a must be bounded, i.e. the norms of its elements must admit
fixed reflector antennas, or an equispaced array, wherein a an upper bound, in order to ensure that the solution method
further semplification is possible because the operator S is at hand is meaningful.
unitary-see Section IV-A. We are therefore led to the requirement that any minimiz-
In order to give an operative meaning to the identification ing sequence must be bounded.
of the set I with the set of all the solutions of the synthesis The boundedness requirement is also needed to control
problem, note that necessary conditions for a point z to the sensitivity of the solution. Actually, when unbounded
belong to I are minimizing sequences do exist, we have solutions to the
synthesis problem whose norm can be arbitrarily large, and
d ( z , G )= 0 this, as it is well known [46], makes their sensitivity very
d ( z ,Ac) = O high. In other words, the requirements of a small sensitivity

BUCCl et al.: ANTENNA PATTERN SYNTHESIS 361

and of the very possibility of devising a solution method choice depends on what kind of property has physical
tum out to be essentially the same, i.e., that any minimizing relevance to the problem at hand, as well as on the
sequence must be bounded. In a Hilbert space any bounded simplifications it allows in the subsequent treatment of the
sequence contains a weakly convergent subsequence [56], problem.
i.e., a subsequence which converges componentwise. Ac-
cordingly, when the boundedness requirement is satisfied, 111. A GENERALSYNTHESIS
ALGORITHM
we will say that the synthesis problem at hand is weakly In the previous section, it has been shown that any
well-posed. If, moreover, any minimizing sequence contains synthesis problem can be formulated as the search for
a subsequence converging in norm, we will say that the the infimum, over the set D of the allowed inputs and
synthesis problem is strongly well-posed. parameters, of the functional (9), which we rewrite for
According to the above discussion, we must ensure that convenience as
the synthesis problem is at least weakly well-posed.
In the finite dimensional case, i.e., when both X and q x ,s) = d 2 ((z, s, S(Z, 4), AC)
S are finite-dimensional, the problem is strongly well- + +
= d2(2,XC) d2(S,SC) d2(S(Z,S),YC).
posed, provided that the operator S(z,s) is one to one
(10)
and the set Yc is bounded, both conditions being usually
met. From the practical point of view, we could always More precisely, we search for an algorithm which pro-
reduce the problem to a finite-dimensional form by a vides a minimizing sequence for the functional @.
proper discretization. Of course, the number of dimensions A large number of procedures have been developed to this
and the sampling rate should be chosen large enough to end; however, practically all of them deal with particular,
ensure a small discretization error. Unfortunately, if the or even very particular, cases. On the other hand, what is
original problem is ill-posed, the norm of the elements of needed for answering the full synthesis problem stated in
a minimizing sequence can become larger and larger as the Section 11, is an algorithm which is completely general,
number of dimensions of its discretized version increases, i.e., not tied to a particular type of radiating system and
so that the problem becomes practically ill-posed. This design specifications, in order to take into account, as
kind of difficulty can appear also when the problem is much as possible, all specifications and constraints which
intrinsically a finite-dimensional one, as in the array case, if can be practically relevant. A direct minimization of the
we deal with large arrays, or with arrays with a very small full functional (lo), even if possible, is a very difficult
inter-element spacing. The problem is obviously weakly task leading to cumbersome and numerically ineffective
well-posed if G (which contains the minimizing sequence) algorithms.
is bounded. The same is true if AC is bounded, because A very appealing and natural way to alleviate this diffi-
in this case the points of any minimizing sequence have a culty is the decomposition of our problem into a sequence of
nonincreasing distance from a bounded set, hence are all simpler ones. For instance, the design of a shaped reflector
contained in a bounded set. More generally, the problem antenna would certainly be simplified if the synthesis
is weakly well-posed whenever the intersection of every procedure could be reduced to steps involving the synthesis
c-neighb~rhood~ of AC with G is bounded. of the excitation for a fixed reflector and of the shaping
When both G and AC are unbounded (or too large), of the reflector surface for a fixed excitation. Of course,
a “regularization” is required, i.e., we have to modify in the number of subproblems should be as small as possible,
some, possibly harmless, way the problem in order to assure provided that the solution of each subproblem can be
the boundedness of any minimizing sequence. In the light effectively obtained. In the following we will show how
of the discussion conceming the sensitivity, a first natural to perform such a decomposition for whatever synthesis
choice is to require the set D to be a (closed) bounded set problem, thus leading to a general and effective synthesis
with a sufficiently small diameter, which obviously makes method.
G bounded, as the operator S is continuous. In this way, The most immediate decomposition is the straightfor-
we have essentially performed a regularization through the ward generalization of the so-called “alternate projection”
“constrained solution method” [57]. algorithm [58]-[61], which has recently been successfully
On the other hand, we can require the boundedness of applied to synthesis problems [21], [23], [25], [39], [41],
the first two components of Ac. This makes the first [62]-[66]. As implied by its name, the approach relies on
two components of any minimizing sequence bounded; the concept of (metric) projector. Given an arbitrary subset
hence, the boundedness of the full sequence follows again R of a metric space A, the projector Po onto R is the
because of the continuity of S. In this case, the problem (generally nonlinear) operator that associates to each point
+
is regularized by the term d 2 ( x C , X c ) d 2 ( s , S c ) in (9), z E A a point of Sa closest to z , i.e., such that
which acts as a stabilizing functional [57].
Usually, the sets Xc and S c are defined in terms of d 2 ( 2 ,Pnz) 5 d2(2,W), vw ER (11)
some physical property, such as the superdirectivity ratio d2(., .) being the distance in A.
[501, [511, the reflector smoothness [38], and so on. The Of course, it is assumed that at least one nearest point
exists, and that a rule of choice has been given if more
4The e-neighborhood of AC is the set { z E A : d ( r , A ~5) e}. than one of such points does exist. In our case, A is

362 PROCEEDINGS OF THE IEEE, VOL. 82, NO. 3, MARCH 1994

a Hilbert space, and the weak closeness of R, a very In correspondence with Agl(Z), let us introduce the
lose condition, is sufficient to ensure the existence of the "restricted" functional
corresponding projector [67]. Denoting with PA and
C. d J ] ( zs;, Z) = d 2 ( ( x ,s , S ( x ,s)), AE1(Z))
PG the projection operators onto AC and G, respectively,
the altemate projection algorithm consists of the iteration
scheme
zn+1 =PGPA,~~ (12)
+ d2(I/7Yg1(&)) (14)

which provides a minimizing sequence. for the functional which is obtained replacing AC with AE1(Z) in the def-
@ [59], [60]. Accordingly, the scheme (12) can be used to inition of @.
solve our problem. Consider now the iterative scheme whose nth step con-
The procedure is appealing, but requires the explicit sists of the following substeps:
knowledge of PA and PG. Because AC is a Cartesian
C
product, the evaluation of PA is reduced to that of the
C
partial projectors onto its components, i.e., onto the sets
which specify the design requirements. This can be easily
accomplished for a large number of practically relevant
cases (see [25], [62] and Section IV).
On the other hand, the set G has, in general, a much
more complicated structure, so that the evaluation of PG
can be very difficult. Despite the above difficulties, the
altemate projection method has been successfully applied
to a number of cases, such as equispaced arrays [25], [62]
and fixed reflector synthesis [38]. All these cases share a = (&+U, p + l ) ,S(,b+l), p + l )
Z(n+l)
1). (154
common feature, namely, S c contains a single element,
i.e., the antenna structure is fixed. This makes G a subset It is obvious that the scheme (15) enjoys the required
of a linear variety (because the operator S is linear in features: (15b) is a fixed-structure synthesis step, (1%) is
x), and greatly simplifies the determination of PG, which a fixed excitation one, while steps (15a) and (15d) are
becomes equivalent to a standard field synthesis problem. relatively trivial.
This naturally suggests to pursue such a reduction to simpler Now from (15aH15d) and the very definition of projec-
field synthesis problems also in the general case. Of course, tor, we have
both the antenna structure and excitation must now be qx(n), ,(".I) = d 2 ( z ( " ) , P A & ) ) = @lo~0~01(.Jn),J4; 5)
varied, but each step of the procedure should involve either
a fixed structure or a fixed excitation synthesis problem. In 2 @[J1(,(4, s(4;2 ) > - @[J1(Z(n+l)l p);Z)
this way, any algorithm available to solve these particular 2 @[J1(,(..+1),,(n+l); 5)
cases could be used to solve the general one. This approach, >
- @P?lJ1(x(n+l), Jn+l); 2 )
first proposed in [41], [ S I , [66] for particular cases, will
- @(x(n+l) , p + 1 ) ) .
be now developed in its full generality. (16)
To proceed we need a notation to indicate subsets of Accordingly, the scheme (15) provides a monotonous
Ac-see ( 4 t h which some of the components are held sequence for a, converging to a (generally local) minimum.
fixed. The simplest way to specify what components are The following comments are now in order:
held fixed is to give a binary vector in which 1 means that
the component can vary and 0 that it is fixed. Of course, i) The choice of the triplet J is arbitrary as well as the
to completely determine the subset, we need also to set the order in which steps (15b) and (1%) are performed,
value of the fixed components. Therefore, such a subset is and could vary from a step to another. This freedom
identified by a point of Ac, say Z = (5, 9, i j ) , and a binary can be exploited to simplify the solution of the varia-
vector J = (jl,jz,j3),and will be denoted by Agl(Z). It tional problems (15b), (1%). Also, Z could be varied
is defined as between steps (15b) and (1%) by a further projection
onto Ac. This complicates the algorithm but allows
Agl(Z) = Ag1(3,8,ij)
a faster convergence.
ii) In general, the variational problems (15b), (1%)
cannot be solved in closed form and would require
further iterative loops. However, this is unnecessary,
because the attainment of the extrema is not manda-
tory: what really matters is that the functional
decreases at each step. Accordingly, any algorithm
and analogously for S E 1 ( $ )and Ygl(ij).
providing such a decrease, as, for instance, a descent

BUCCI et al.: ANTENNA PATTERN SYNTHESIS 363

 method, can be used in steps (15b), (1%). We can . . . . .
proceed further along this way dividing the varia-
. . . . a
tional problems (15b), (1%) into subproblems. As
an example, in a dual reflector synthesis, we have,
in S , the parameters describing the main reflector
and those describing the subreflector. The reduction
of @ [ J ] ( ~ ( n f l ) , ~ ; Z in
) (1%) can be achieved by
first varying only the components of s corresponding Fig. 3. Array lattice.
the main reflector parameters and then only those
corresponding to the subreflector (or vice versa).
iii) If D is a linear subspace (or also linear variety) steps a shaped reflector. These cases are considered, not only for
(15b), (1%) involve an unconstrained minimization, their practical relevance, but also because they shed light
which is generally much easier to perform than on the various aspects of the antenna synthesis problem and
the constrained one required when these sets are on the generality and flexibility of the developed approach.
bounded. Accordingly, from the computational point
of view, it is convenient to move all boundedness A . Synthesis of a Reconfigurable Array with
(and smoothness) requirements into Ac, leaving in Phase-Only Control
D only the “structural” constraints, as, f.i., antenna The far field of a planar array, whose elements lie on a
size, number of feeds, etc. regular equispaced lattice in the (x,y) plane of a Cartesian
frame -see Fig. 3 --can be expressed as [25]
As a concluding remark, we must stress that (as in many
minimization problems) we are not guaranteed to reach the
absolute minimum of a, since the minimizing sequence (n,m)EI
can converge toward a relative minimum. Such points are (17)
called traps. wherein ~ ( u , Y )and F ( u , v ) are the element and array
In order to avoid traps, a number of strategies are factors, respectively, and U = IC . a, ‘U = . 6, IC
possible. The most natural is a proper choice of the starting the propagation vector, pointing in the direction (e,$).
point of the sequence. This point can be, for example, Accordingly, U and Y are known functions of the spherical
angles e,$. I is the set of the couples (n, m ) specifying
i) an approximate solution of the synthesis problem, e.g., the locations of the array elements.
an optical or, more generally, asymptotic one, Expression (17) shows that any requirement on the far-
ii) a solution of a less constrained problem. field pattem can be translated into an equivalent one for the
array factor. Accordingly, in the following we will focus
In this framework, a more effective solution would be
on the synthesis of the array factor. The set B of all array
the so-called “continuation method,” in which we solve
factors which can be radiated by the considered array is
sequentially a number of synthesis problems in which the
given by
constraints are tighter and tighter. The occurrence of this or

1
another trap, or of the minimum, depends also on the choice
of J and on the order of the steps (15b), (1%). Therefore,
if a certain choice has led to a trap, we can modify it in
order to avoid that trap. It is also important to enforce all
BG
{ F : F=
(n,m)EI
anmej(nu+mv)
,anmeC c L’(s)
(18)
the constraints which are likely to be relevant, in order to wherein S = ( - - T , T ) x ( - T , T ) .
reduce as much as possible the degrees of freedom of the For reconfigurable arrays we have to deal with Q array
problem, whose number is the main source of the secondary factors, radiated by the same array, which must be syn-
minima. The second important source is the nonconvexity thesised at the same time. We therefore have to consider
the ordered Q-tuple of functions ( F l ( u ,U), . . . , FQ(u,U)),

-
of the involved sets, in particular the set Ac. Accordingly,
to make some or all of its components convex, would which belong to the Cartesian product
certainly alleviate the trapping problem. From this point of
view, the choice as output of the squared field amplitude, y = L2 x L2.. . x L2 = (L2)Q. (19)
instead of the field itself, is very appealing, because it tums Q times
Yc into a convex set, at least in the very relevant case The requirement that the array has a finite number of
where the far-field specifications are expressed in terms of

-
elements means that all the Q-tuples of array factors belong
a mask for the power pattem.
to
BQ = B x B x . . . x B (L2)Q. (20)
IV. E~AMPLES
Q times
To point out the steps needed to solve an antenna synthe-
sis problem, we present now two significant cases, namely, The set G of all admissible Q-tuples is a subset of
a planar reconfigurable array with phase-only control, and this set, including only elements satisfying the additional

364 PROCEEDINGS OF THE IEEE, VOL. 82, NO. 3, MARCH 1994

requirement that
Janm,ll = l a n m , ~ I= ... = Ianm,QI, ( n , m )E I (21)
i.e., all array factors satisfying the requirement of a common
amplitude (21). Note explicitly that G is not a Cartesian
product, which prevents decoupling the problem into Q
independent ones.
The requirements on the pattems, which define the set
Y c , can be generally set according to a proper generaliza- --TI '

"
-71 u n
tion of the mask concept. The requirements on the shape
of each pattern define a suitable mask (a)

m q ( u , v )I IFq(U,U)I 5 M q ( u , v )
u , E ~( - T , T ) x (-T,T), Q = l , Q l (22)
and therefore a corresponding set M
have
Y c = M i x M:! x . . . x M Q
q. Accordingly, we

(L')". (23)
Since the masks are defined over ( - T , n)x ( - T , T ) , there
;:r
is no need for a stabilizing functional, therefore we set -71 '
-77
I
UTI
xc = x = (P)Q.
Accordingly, the fundamental functional (9) is reduced (b)
to its last term, the problem can be settled in a single Fig. 4. Prescribed masks for the reconfigurable array.
space, i.e., the space of the array factors, and the solution
method described in the previous section is reduced to the The analytical details are straightforward and are devel-
altemating projection algorithm (see (12)). oped in [64].
The projector onto AC Y c , PA acts separately For a dynamic constraint
C
on each function of the Q-tuple. It maps ( F1, . . . , FQ)
AMINI Icnm,ql I AMAX (26)
onto (PM,FI , . . . , PnfOFQ) wherein, for the pattem re-
quirements defined through the masks (22) some modification are needed. Letting

three cases are in order:

a) AMINI KLm 5 AMAX *
K n m = KLmi
On the other side, the operator PG acts on the whole b) KAm I AMIN *
K n m = AMIN
Q-tuple. Using the Parseval identity, PG is more easily c) AR4AX 5 KAm *
K n m = AMAX.
expressed in terms of the Fourier coefficients of the Fn's. As an example, we report the synthesis of a planar array
Let of 17 x 17 elements arranged in a square grid, whose masks
are shown in Fig. 4. The resulting pattems, obtained by
constraining the excitation dynamic ratio A ~ ~ A X / A MtoI N
n,m=-m 20, are shown in Fig. 5. The synthesized pattems fulfill
the requirements, except for a small fraction of the shaped
be the element of (L2)' to project onto G, and zone of the triangular pattem, wherein the ripple is a little
bit larger (*0.9 dB versus f 0.75 dB). It is worth noting
that the same case has been synthesized also without the
dynamic constraint, resulting in an excitation dynamic ratio
larger than 100 without any significant improvement on the
the projected element. Letting synthesis result.

B . Synthesis of a Shaped Rejector Antenna

be the common amplitudes, we get We examine now the synthesis of a reflector antenna
system. For the sake of simplicity we will refer to a two-
dimensional focussing antenna.
The reflector diameter 2a is given, whereas its geometry
is free, and is specified by the unknown reflector line

BUCCI et al.: ANTENNA PATTERN SYNTHESIS 365

 synthesis procedure is split in two steps [38], [41]. In the
first step, the reflector geometry and the primary radiation
pattem are found. In that step only the overall primary
-5 source extension 2b concems us. In the second step, the
effective structure of the primary source to be used is
-1'1 specified and the feed excitation is synthesized starting
from the primary radiation pattem obtained in the first step.
~ 15
For a feed constituted by a given number of equispaced
elements this second step can be accomplished by using a
symplified version of the method considered in the previous
-20
subsection or by a more standard array synthesis technique.
Accordingly, only the first step will be considered in the
-25
following. Let us denote by &i the y-polarized primary field
and by E the corresponding secondary far-field pattem,
related to the scattered field E by E = E exp ( j / 3 r ) f i .
In the P.O.approximation, we have

wherein U = sinI3; E = px, e = ( p - g)/X, p =

Jx2 + (g + and A = pg (cos8 - 1)

and the other symbols are defined in Fig. 6.

In the case of a focussing antenna cos I3 I
I1 in the whole
angular sector of interest so that we can assume A N 0 in
(28), which accordingly becomes

= 3(PpaJeqe-2j2Te) (30)
wherein 3denotes the Fourier transform operator and Ppa
the characteristic -function of the interval [-pu, pa], i.e.,

(b)
Fig. 5. Synthesized patterns. Because Fourier transforms can be efficiently evaluated
by means of the FFT algorithm, (30) shows that the
synthesis procedure would be greatly simplified if we could
adopt as unknown the equivalent current Jeqinstead of the
primary far field. This would be possible, provided that:
a) the primary far field can be easily recovered from
20 the knowledge of Jeqand the reflector geometry, i.e.,
e(c>;
b) the constraints on the primary source can be translated
4 into equivalent constraints in terms of Jeq.
Both these requirements can be satisfied, as it is shown in
Fig. 6. Reflector geometry. detail in [38], [39]. In particular, the source size constraint
is reflected on the fact that the equivalent current Jeq([),
equation given by (29), is an essentially band-limited function, with
a bandwidth 7 = b/d, d being the distance between the
z = g(x) (27) source center and the reflector vertex, see Fig. 6.
see Fig. 6. Thus although Jeq depends in a complicate way on both
In order to get a general procedure, indipendent from the primary field and the reflector geometry, the require-
the particular type of adopted primary source, the overall ment on the source size is equivalent to the requirement

366 PROCEEDINGS OF THE BEE., VOL. 82, NO. 3, MARCH 1994

that J belongs to W,, a set of the functions band-limited
39
to 7, i.e.

wherein e ( t ) is a function belonging to L2(-7, T).

We can now recast our problem into the general formal-
ism of Section 11. Due to the occurrence of the Fourier
transform in the scattering operator (see (30)), it is con-
venient to extend the range of the variables [ and U
to the full real axis. Accordingly, we-have X = Y =
L2(-oo,co) and S = L2(-pu,pa). The set D is equal to -20 -10 0 10 20
W, x S,because no constraint has been set to the reflector (DA)sin@)
geometry5. Because D is unbounded, the well-posedness of
Fig. 7. Prescribed secondary pattem (continuous line) and related
the problem must be ensured by Ac. mask (dot-dashed line).
A very natural and convenient way to regularize the prob-
lem is to control the ‘‘power’: radiated outside the angular
range of interest (-210, uo) and the “power” associated with However, the results obtained with the approximation
Jeqoutside the reflector coordinate interval (-pa, Pa) thus A = 0 are still useful and can be used as a first ap-
limiting the superdirectivity and the spillover losses. The proximation which can subsequently be ameliorated by
first goal is achieved by choosing Yc as the set of all applying a general minimization algorithm with A # 0.
elements of Y complying with the far-field specifications Since the starting point is already near to the solution,
inside (-uo, U O ) , and equal to zero outside. The second, by only a few iterations of such algorithm will be needed and
choosing XC as the set of all square integrable functions the overall computational effort will not be significantly
equal to zero outside ( - p u , , h ) . Hence the functional to increased.
be minimized becames The reflector synthesis procedure outlined has been ap-
plied by assuming D/X = 40, d / X = 40, and a constraint
for the radiated secondary pattem given by the “mask”
shown in Fig. 7. The pattem is required to be flat, within
wherein E ( u ) is given by (30). f 0.5 dB, in the range (-6, 6); smaller than -30 dB in the
in the range U E ( - U O , U O ) is expressed as ranges ( D / X ) u 5 -8 and ( D / A ) u 28; free in the ranges
PYG
in Section IV-A, whereas Py (e(.)) = 0 when U $!
(D/X)u E (-8, -6) and ( D / X ) u E (6, 8). The starting
C point adopted for the iterative procedure (34) corresponds
( - U O , U O ) . The iteration scheme (15a)-( 15d) can be now to a starting reflector obtained by means of a Geometrical
easily written down, since the solution of the minimization Optics synthesis algorithm and to a starting current Jeq
problem (15b) and (1%) can be explicitly found. The equal to the equivalent current associated with the starting
corresponding equations are reflector and a pointwise-isotropic source located at the
center of the feeding region. In order to show the influence
5 = py ,F ( p 1% J(”),-j2*1(”1
eq ) (344 of the source dimensions and the advantages of a synthesis
J,(;+I) = 3-1p 3e327Jn) Paa3- (6) . (34b)
procedure which provides at the same time both the reflector
geometry and the primary excitation, various values of
2.1rl(”+l) = - a r g { J $ + l ) * P ~ , F - l ( ~ ) ) (34~) the source extensions 2b have been considered. For the
sake of convenience, in the following the geometry of the
wherein P, is the characteristic function of the interval synthesized reflector will be expressed as a deformation
[ - T , 71. Az/A with respect to a “reference” parabolic reflector with
Once Jeq and e, hence g-see (28)-have been deter- a diameter of 40X and f / D = 1.
mined, the primary far field can be evaluated by means of As a first example 2b has been chosen equal to 15X.
the so-called “cancellation technique” introduced in [381. In this case, the focal image of the secondary pattem
It must be stressed that the approximation A N 0 associated with a parabolic reflector with D = f = 40X
exploited in (30) can be avoided if it is unsatisfactory, by would be fully included in the feeding region. Accordingly,
using in (33) for the functional @ the “exact” operator given this reflector, corresponding to Az/X =0, will be a possible
by (28). solution. The deformation corresponding to the actually
The corresponding variational equations cannot be solved synthesised reflector (solid line) is shown in Fig. 8 together
in closed form, so that a general minimization algorithm with the starting one (dashed line). As can be seen we
must be used for the minimization of @. get just the above parabolic reflector although the starting
reflector was quite different. The corresponding secondary
’Note that a continuity constraint does not effectively restrict the set
of allowable reflectors, because the set of the functions continuous on pattem, together with the assigned mask and the primary
( - ? a . , 3 a ] is dense in Lz(:3a,,ja). pattem, are reported in Fig. 9.
.-

BUCCI et U/.: ANTENNA PAITERN SYNTHESIS 361

 2l
AZA k A
2

-1 I -1

-2 -2
-20 -10 0 10 20 -20 -1 0 0 10 20

X A x/h

Fig. 8. Deformation of the synthesized reflector (continuous line) Fig. 10. Deformation of the synthesized reflector. 2b = 1OX.
and starting reflector (dashed line). 26 = 15X.

-20 -10 0 10 20
-20 -10 0 10 20
(DA)sin@)
(DA)sin@)
(a)
(a)

t
4

L 2 1 °0
0

-10

-20 -
-20

-30.
-30

4 0 4
40
-30 -20 -10 0 10 20 30
-30 -20 -10 0 10 20 30
e e
(b)
(b)
Fig. 11. (a) Synthesized secondary pattem (continuous line) and
Fig. 9. (a) Synthesized secondary pattem (continuous line) and mask (dot-dashed line). (b) Primary pattem. 2b = 1OX.
mask (dot-dashed line). (b) Primary pattem. 26 = 1 5 X .

has been considered. The deformation of the synthesized

As a second example 2b has been set equal to 1OX.
reflector is shown in Fig. 12, while the secondary and
Now the source dimensions are smaller than the focal
image of the secondary pattem so that the synthesized the primary pattems are shown in the Fig. 13. Obviously,
reflector is not expected to be parabolic. This is confirmed the deformation is increased with respect to the previous
by Fig. 10, wherein the synthesized reflector is shown. examples although the ‘‘goodness” of the secondary pattem
The corresponding secondary and primary pattems are is essentially the same. This is strongly in contrast with the
reported in Fig. 11. As can be seen, although the source pattem achievable with an undeformed parabolic reflector,
size has been reduced, it is still possible to satisfy the which is completely outside the prescribed mask, as shown
prescribed mask by means of a convenient deformation of in Fig. 14. Hence, the availability of a synthesis procedure
the reflector geometry. Finally, a small source with 2b = 5X which gives both the reflector geometry and the primary

368 PROCEEDINGS OF THE IEEE, VOL. 82. NO. 3, MARCH 1994

 t
2
AZA
1

1
-20 -10 0 10 20
(DA)sin@)
-2
-20 -10 0 10 20 Fig. 14. Secondary pattern synthesized with a prescribed unde-
X A formed parabolic reflector (continuous line) and mask (dot-dashed
line). 26 = 5A.
Fig. 12. Deformation of the synthesized reflector. 2b = 5X.

distance from a required pattem, the superdirectivity ratio,

and so on, we look for an antenna fulfilling the pattem
requirements as well as other constraints, including those
on the excitation or primary source. Since it is possible to
cast those requirements in forms of sets, each one collecting
all elements satisfying that requirement, we are led to
look for an element in the intersection between such sets.
Representing the antenna as a triple (excitation, geometry,
pattem), i.e., as an element of a Cartesian product of sets,
we are able to reduce the problem to the intersection of two
sets, one of which is itself a Cartesian product while the
other is not. The latter carries on the “physical realizability”
-20 -1 0 0 10 20
requirements, i.e., that the solution triple can be actually
(DA)sin@)
realized.
(a)
Next, a solution method has been devised with the
t following advantages:

it allows to solve the full synthesis problem, i.e., to de-

termine both the antenna geometry and the excitation;
it allows to take into account, in a systematic and
flexible way, constraints on both the antenna geometry
and excitation;
it allows to segment the procedure, selecting a good
compromise between speed and complexity;
it allows to use different field synthesis or optimization
-30 -20 -10 0 10 20 30 procedures according to the problem or the availability
0 of specific software.
(b)
Fig. 13. (a) Synthesized secondary pattern (continuous line) and Moreover, the solution method allows to deal with the
mask (dot-dashed line). (b) Primary pattem. 26 = 5X. problem of ill-conditioning and to avoid traps by a number
of useful strategies.
pattem allows to satisfy essentially the same synthesis The synthesis procedure has been exemplified in two
requirements by means of various possible couples of significant cases, usually dealt with, to the best of our
reflectors and sources: a more deformed reflector allows knowledge, by imposing a priori choices strongly affecting
to use a smaller source. the result.
The developed procedure is not, at present, an all-
V. CONCLUSION purposes completely automatic CAD for antennas, nor
A new general approach to the antenna pattem synthesis it pretends to be the final word on this subject. Rather, we
problem has been presented and thoroughly discussed. presented a general and flexible framework which orders
The approach has a number of particular features both and clarifies in a systematic way all the steps needed for
in the formulation and in the solution of the problem. the synthesis, as well as a flexible solution algorithm, while
Instead of optimizing a suitable quality factor, e.g., the leaving the designer freedom to choose how to detail the

BUCCI et al.: ANTENNA PA7TERN SYNTHESIS 369

steps of the procedure in a way more adapt to the particular [27] P.Balling, R. Jorgensen, and K. Pontoppidan, “Study of tech-
problem and/or to the available software. niques for design of high gain antennas with contoured beams,”
Final Rep., ESTEC Contract 3371/NL/AK, Dec. 1978.
[28] V. Galindo-Israel, S. W. Lee, and R. Mittra, “Synthesis of
REFERENCES a laterally displaced cluster feed for a reflector antenna with
application to multiple beams and contoured pattern,” IEEE
[ l ] L. La Paz and G. Millar, “Optimum current distribution on Trans. Antennas Propagat. , vol. AP-26, pp. 220-228, 1978.
vertical antennas,” Proc. IRE, vol. 31, pp. 214-218, 1943. [29] V. Galindo-Israel, R. Mittra, and A. G. Cha, “Amplitude
[2] S. A. Schelkunoff, “A mathematical theory of linear arrays,” aperture and phase control of offset dual reflectors,” IEEE
Bell Syst.Tech. J., vol. 22, pp. 80-107, 1943. Trans. Antennas Propagat., vol. AP-27, pp. 154-164, 1979.
[3] C. L. Dolph, “A current distribution for broad side arrays which [30] B. S . Westcott, F. A. Stevens, and F. Brickell, “GO synthesis
optimize the relationship between beamwidth and side lobe of offset dual reflectors,” Proc. Inst. Elec. Eng. , vol. 128, pt.
level,” Proc. IRE, vol. 34, pp. 335-348, 1946. H, pp. 11-18, 1981.
[4] P. H. Woodward, “A method of calculating the field over a [31] R. Mittra, F. Hyjazie, and V. Galindo-Israel, “Synthesis of offset
plane aperture required to produce a given polar diagram,” Proc. dual reflector antennas transforming a given feed illumination
Inst.Elelc. Eng.,, vol. 93, pt. In, pp. 1554-1558, 1947. pattem into a specified aperture distribution,” IEEE Trans.
[5] T. T. Taylor, “Design of line source antennas for narrow Antennas Propagat. vol. AP-30, pp. 25 1-259, 1982.
beamwidth and low sidelobes,” IRE Trans. Antennas Propagat., [32] D.M. Pozar, “Antenna synthesis and optimization using
vol. AP-3, pp. 16-28, 1955. weighted Inegaki modes,” IEEE Trans. Antennas Propagat.,
[6] J. Ruze, “Circular aperture synthesis,” IEEE Trans. Antennas vol. AP-32, pp. 159-165, 1984.
Propagat., vol. AP-12, pp. 691-694, 19.64. [33] C. A. Klein, “Design of shaped-beam antennas through minimax
[7] Y.T. Lo, S . W. Lee, and Q. H. Lee, “Optimization of directivity gain optimization,” IEEE Trans. Antennas P ropagat, vol. AP-
and signal-to-noise ration of arbitrary antenna array,” Proc. 32, pp. 963-968, 1984.
IEEE, vol. 54, pp. 1033-1045, 1966. [34] P. T. Lam, “Directivity optimization of a reflector antenna with
[SI L. D. Bakhrakh and V. L. Troytskiy, “Mixed problems of cluster feed: A closed-form solution,” IEEE Trans. Antennas
antenna synthesis,” Radio Eng. Electron. Phys., vol. 12, pp. Propagat., vol. AP-33, pp. 1163-1 174, 1985.
404-414, 1967. [35] P. Mehler, S. Tun, and N. Adatia, “Direct far-field GO synthesis
[9] T. S . Fong, “On the problem of optimum aperture distribution,” of shaped beam reflector antennas,” Proc. Inst. Elec. Eng., vol.
J. Franklin Inst., vol. 283, pp. 235-249, 1967. 133, pt. H, pp. 213-220, 1986.
[lo] R. F. Hyneman and R. M. Johnson, “A technique for the [36] V. Galindo-Israel, W.A. Imbriale, and R. Mittra, “On the theory
synthesis of shaped-beam radiation pattem with approximately of the synthesis of single and dual offset shaped reflector
equal percentage ripple,” IRE Trans. Antennas Propagat., vol. antennas,” IEEE Trans. Antennas. Propagat., vol. AP-35, pp.
AP-15, pp. 736744, 1967. 887-895, 1987.
[ l l ] G. V. Borgiotti, “A synthesis method for bidimensional 1371 J. Bergmann, R. P. Brown, P. J. B. Clarricoats, and H. Zhou,
apertures,” IEEE Trans. Antennas Propagat., vol. AP-16, pp. “Synthesis of shaped-beam reflector antenna pattems,” Proc.
188-193, 1968. Inst. Elelc. Eng., vol. 135, pt. H, pp. 48-53, 1988
[12] W. L. Stutzman, “Synthesis of shaped-beam radiation patterns [38] 0. M. Bucci, G. D’Elia, and G. Leone, “Reflector antenna
using the iterative sampling method,” IEEE Trans. Antennas power synthesis, a general and efficient approach,” IEEE Trans.
Propagat., vol. AP-19, pp. 36-41, 1971. Antennas Propagat., vol. 37, pp. 875-883, 1989.
[13] D. K. Cheng, “Optimization techniques for antenna arrays,” [39] 0. M. Bucci and G. D’Elia, “A general and effective approach
Proc. IEEE, vol. 59, pp. 1664-1674, 1971. to the synthesis of shaped reflector antennas,” presented at the
[14] T. S . Fong and R. A. Birgenheier, “Method of conjugate
gradients for antenna pattern synthesis,” Radio Sci., vol. 6, pp. 1989 Antennas Propag. Symp., San Jose, CA, 1989.
[40] A. B. Charrette, S. W. Lee, and A. J. Acosta, “A method for
1123-1 130, 1971‘; producing a shaped contour radiation pattem using a single
[ 151 V. V. Semenov, Two problems in antenna synthesis theory,”
Radio Eng. Electron. Phys., vol. 17, pp. 18-24, 1972. shaped reflector and a single feed,” IEEE Trans. Antennas
[ 161 D. R. Rhodes, Synthesis of Planar Antenna Sources. Oxford, Propagat., vol. 37, pp. 698-706, 1989.
UK: Clarendon, 1974. [41] 0. M. Bucci and G. D’Elia, “A new approach to the synthesis of
[ 171 R. Mautz and R. F. Hanington, “Computational methods for scanning or reconfigurable beam reflector antennas,” presented
antenna pattem synthesis,” IEEE Trans. Antennas Propagat., at the XXIII General Assembly of the URSI Symp., Prague,
vol. AP-23, pp. 507-512, 1975. Czechoslovakia, 1990.
[18] H. Schjaer-Jacobsen and K. Madsen, “Synthesis of nonuni- [42] P. H. Woodward and J. D. Lawson, “The theoretical precision
formly spaced arrays using a general nonlinear minimax op- with an arbitrary radiation pattem may be obtained from a
timization method,” IEEE Trans. Antennas Propagat., vol. AP- source of finite size,” Proc. Inst. Elec.Eng. , vol. 95, pt. m,
24, pp. 501-506, 1976. pp. 363-373, 1948.
[ 191 R. Jorgensen, “Coverage shaping of contoured-beam antennas [43] L. A. Waynshteyn and L. P. Graber, “A contribution to antenna
by aperture field synthesis,” Proc. Inst. Elelc. Eng., vol. 127, synthesis theory, General statement and specific problems,”
pt. H, pp. 201-208, 1980. Radio Eng. Electron. Phys., vol. 16, pp. 2012-2016, 1971.
[20] J. M. Jarem, “Strip and line source synthesis with a quality [44] N. N. Voytovich and P. A. Savenko, “Synthesis of antennas
factor constraint,” IEEE Trans.Antennas Propagat., vol. AP-28, from a specified amplitude pattem and related problems in
pp. 3-1,. 1980. quasi-optics (review),” Radio Eng. Electron. Phys., vol. 20, pp.
[21] H. Elmikati and A. A. Elsohly, “Extension of projection method 1-13, 1980.
to nonunifomly linear antenna arrays,” IEEE Trans. Antennas [45] Y. I. Choni, “Synthesis of an antenna according to a given
Propagat., vol. AP-32, pp. 507-512, 1984. amplitude radiation pattem,” Radio Eng. Electron. Phys., vol.
[22] H. J. Orchard, R. S . Elliott, and G. J. Stem, “Optimizing the 26, pp, 770-778, 1981.
synthesis of shaped beam antenna patterns,” Proc. Inst. Elelc. [46] G. A. Deschamps and H. S . Cabayan, “Antenna synthesis and
Eng., vol. 132, pt. H, pp. 63-68, 1985. solution of inverse problems by regularization methods,” IEEE
[23] G. T. Poulton, “Antenna power pattem synthesis using method Trans. Antennas Propagat., vol. AP-20, pp. 268-274, 1972.
of successive Droiections.” Electron. Lett.. vol. 22. DD. 1042-
L _ . , L .
[47] M. Z. Nashed, “Operator-theoretic and computational ap-
1043, 1986. proaches to ill-posed problems with application to antenna
1241 J. F. DeFord and 0. P. Gandhi, “Minimum peak sidelobe pattem theory,” IEEE Trans. Antennas Propagat., vol. AP-29, pp.
planar arrays,” IEEE Trans. Antennas Propatat.,
. _ vol. 36, pp. -_ 220-231, 1981.
i9i-201, i988. [48] T. S. Angell and R. E. Kleinman, “Generalized exterior bound-
[25] 0.M. Bucci, G. Franceschetti, G. Mazzarella, and G. Panariello, ary value problems and optimization for the Helmholtz equa-
“The intersection approach to array synthesis,” Proc. Inst. Elelc. tion,” J. Opt. Theory Appl., vol. 37, 1982.
Eng., vol. 137, pt. H, pp. 349-357, 1990. [49] T. S. Angell, A. Kirsch, and R. E. Kleinman, “Antenna control
[26] W. Pearson and R. Mittra, “Pattem synthesis for antenna with and optimization,” Proc. IEEE, vol. 79, 1991.
multiple primary beams by minimax optimization,” Electron. [50] M. Uzsoky and L. Solymar, “Theory of super-directive linear
Lett., vol. 12, pp. 100-102, 1975. arrays,” Acta Phys. Hung., vol. 6, pp. 185-205, 1956.

370 PROCEEDINGS OF THE IEEE, VOL. 82, NO. 3, MARCH 1994

 A. Bloch, R. G. Medhurst, and S. D.Pool, “Superdirectivity,” Giuseppe D’Elia was born in Salerno, Italy,
Proc. IRE, vol. 48, pp. 1164-1174, 1960. in 1950. He graduated summa cum Zaude in
A. C. Schell and A. Ishimaru,,“Antenna pattem synthesis,” in electronic engineering from the University of
Antenna Theory, R. E. Collin and F. J. Zucker, Ed. New York: Naples, Naples, Italy.
McGraw-Hill, 1969, ch. 7. He was a C. N. R. (National Research Coun-
G. Franceschetti, G . Mazzarella, and G. Panariello, “Array cil) researcher at the I. R. E. C. E. (Institute
synthesis with excitation constraints,” Proc. Inst. Elec. Eng. , of Research in Electromagnetics and Electronic
vol. 135, pt. H, pp. 400407, 1988. Components) from 1983 to 1987. Since1987
L. D. Bakhrakh and C. D.Kremensky, “ Synthesis of radiating he has been Associate Professor of Antennas,
system,” Sov. Radio (in Russian), Moskow, 1974. now at the Electronic Department, University of
0. M. Bucci, “Reflector antenna synthesis, a general framework Naples “Federico 11, ”Naples. His scientific ac-
and presented at the 2nd Int. On tivity involves the study of the transient behavior ofantennas in dispersive
Electromagnetics in Aerospace Appl., Torino, Italy, 1991. media; the efficient analysis of reflector antennas, the near-fielafar-field
A. E. Taylor, Introduction to Functional Analysis. London, transformation techniques, the synthesis of reflector antennas, the phase
England: Wiley, 1958. retrieval for electromagnetic field, and the phaseless diagnostic technique
A. Tikhonov and V. Arsenine, MCthodes de RCsolution des of scttering
Probltmes Mal Posis. Moskow, USSR: MIR, 1976.
L. G. Gubin, B. T. Polyak, and E. V. Raik, “The method of
projections for finding the common point of convex sets,” USSR
Comput. Math. and Math. Phys., vol. 7, pp. 1-24, 1967.
A. Levi and H. Stark, “Image restoration by the method of
generalized projections with application to restoration from
magnitude,” .I . Soc. Amer., vol. AI, pt. A, pp. 932-943,
Opt.
1984.
R. Barakat and G. Newsam, “Algorithms for reconstruction of
partially known, bandlimited Fourier transform pairs from noisy
data. I. The prototypical linear problem,” .I Integral
. Eq., vol.
9. pp. 49-76, 1985.
-, “Algorithms for reconstruction of partially known, ban-
dlimited Fourier transform pairs from -noisy -data. 11. The
Giuseppe Mazzarella (Member, IEEE) was
non-linear problem of phase retrieval,” .I Integral
. Eq., vol. 9,
born in Naples, Italy, in 1959. He received
pp. 77-125, 1985. the “Laurea” degree summa cum l a d e in
0.M. Bucci, G. Mazzarella, and G. Panariello, “Reconfigurable
arrays by phase-only control,” IEEE Trans. Antennas Propagat., electronic engineering from the University
vol. 39, pp. 919-925. 1991. “Federico 11,” Naples, in 1984 and the
0.M. Bucci, G. Franceschetti, G. Mazzarella, and G. Panariello, Ph.D. degree in electronic engineering in
“A general projection approach to array synthesis,” presented 1989.
at the 1989 Antennas Propag. Symp., San Jose, CA, 1989. In 1984 he joined the Electromagnetic
0. M. Bucci, G. Mazzarella, and G. Panariello, “Reconfigurable Research Group of the University “Federico
arrays by phase-only control,” presented at the 1989 Antennas II,” where he was appointed Research Assistant
ProDae. Svmo.. San Jose. CA. 1989. in 1990. Since 1992, he has been Associate
,V‘g~&ai synthesis with smooth excitation,” presented at Professor Of Electromagnetic Field and Microwaves at the University
the 1990 Antennas Propag. Symp., Dallas, TX, 1990. of Cagliari, Cagliari, Italy. His main research interests are in functional
G. Mazzarella, and G . Panariello, “A projection-based synthesis techniques in electromagnetics, design of slot arrays, antenna pattern
of non-uniform array,” presented at the 1991 Antennas Propag. synthesis, and inverse Profile reconstruction.
Symp., London, Ont., Canada, 1991. Dr. Mazzarella is a member of SIAM.
A. V. Balakrishnan, Applied Functional Analysis. New York:
Springer-Verlag, 1976, sec. 1.4.

Ovidio M. Bucci (Fellow, IEEE) was bom

in Pescare, Italy, on November 18, 1943. He
graduated summa cum Zuude in electronic engi-
neering from the University of Naples, Naples,
Italy, in 1966. He also attende a postgraduate
course in theoretical and nuclear physics at the
Department of Physics, Naples University. Caetano Panariello was born in Herculaneum,
In 1967 he was appointed an Assistant Pro- Italy, in 1956. He graduated summa cum laude
fessor of Theory and Techniques of Electro- in electronic engineering from the University
magnetic Waves at the Istituto Universitario “Federico 11,” Naples, Italy, in 1980 and
Navale in Naples. In 1970 he became Associate received the Ph.D. degree in 1989 from the
Professor of Quantum Electronics in the Engineering Faculty at Naples same university.
University, where in 1976, as a result of a national competition, he was He was staff engineer at the Antenna
appointed Full Professor of Electromagnetic Waves. From 1984 to 1988 Division, Elettronica SpA until1 1984, when
and again since 1992, he has been the Head of the Electronic Department he joined the Electromagnetic Research Group
at Naples University. The main topics of his scientific activity have been of the University “Federico 11,” where he is
scattering from tapered absorbers and surface impedances; antennas in presently Associate Professor in the Department
dispersive media; transient electromagnetics; high-performance antennas; of Electronic Engineering. His main research interests are in’ antenna
antennas analysis and synthesis; near-field/far-field techniques; inverse power pattem synthesis, design of slot arrays, inverse scattering, and
problems; and theoretical electromagnetics. nonlinear electromagnetics.

BUCCI et al.: ANTENNA PATTERN SYNTHESIS 37 1