3262
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
A Simple and Robust Adaptive Parasitic Antenna Marco Donald Migliore, Member, IEEE, Daniele Pinchera, Student Member, IEEE, and Fulvio Schettino, Member, IEEE
Abstract—A novel Uda–Yagi adaptive antenna is numerically and experimentally investigated. The antenna consists of an active element and a relatively large number of parasitic elements closed on two different loads selectable by simple electronic switches. The use of fuzzy-logic based cost function and self-adaptive biological beamforming algorithms allows to obtain quite good performances both in terms of signal to interference plus noise ratio and voltage standing wave ratio. The antenna is simple, low cost, and is robust with respect to mechanical and electrical tolerances and with respect to failures of some passive elements. Experimental results on two different prototypes confirm the good performances of the proposed antenna. Index Terms—Adaptive antennas, evolutionary algorithms, fuzzy logic, genetic algorithms, particle swarm optimization, switched parasitic array.
I. INTRODUCTION
I
N THE LAST years we have witnessed an extraordinary increasing of services based on wireless communications. In fact, the number of subscribers to mobile networks has risen beyond all expectations. At the same time, a quickly increasing number of subscribers requires wireless Internet-based services. Since the radio channel represents a serious bottleneck for the high transmission rate required by these services, the success of the next generation of wireless systems strongly depends on a more efficient use of the spectrum resource. At present, two different philosophies are pursued. The first one is to use spectrum resources that are not licensed adopting sophisticated time-domain processing to reach extremely high bit rate. This way is pursued, e.g., by the ultrawide-band (UWB) systems [1]. However, such systems are limited to short-range communications. A completely different approach is based on the use of spaceprocessing besides of classical time-processing [2]. Two main technologies adopting space-processing have been proposed: adaptive antennas [3] and multiple-input multiple-output systems [1]. Among them, systems based on adaptive antennas have been first investigated, and nowadays some commercial applications are available on the trade. The fast development of adaptive antennas in wireless communications has been helped by the large investigation performed for military applications, e.g., see [4], [5]. However, the architecture proposed in the literature is quite complex, being based on a relatively large array,
Manuscript received December 15, 2004; revised March 24, 2005. This work is supported in part by the Italian Ministry of University (MIUR) under a Program for the Development of Research of National Interest PRIN Grant 2003095273. The authors are with the Microwave Laboratory at the DAEIMI, University of Cassino, 03043 Cassino, Italy (e-mail:
[email protected],
[email protected],
[email protected]). Digital Object Identifier 10.1109/TAP.2005.856361
in which each element has its receiver. This makes the adaptive antenna quite expensive, and sensible with respect to failures. Recently, a new class of low-cost adaptive antennas has been proposed [6], [7]. The architecture of these antennas is based on the Uda–Yagi scheme, and consists of a single active antenna, connected to the receiver/transmitter, and a number of parasitic antennas closed on passive loads whose impedance can be electronically changed. The Uda–Yagi adaptive antennas have been originally proposed for military applications [8], but had a limited success due to the worse performances with respect to full-active adaptive arrays. However, their low cost makes this kind of antennas attractive for the commercial market, and at present a large number of Uda–Yagi adaptive antennas has been proposed in literature. A class of Uda–Yagi adaptive antennas has as objective the reduction of the signal to interference plus noise ratio (SINR). These antennas are usually based on a small number of parasitic elements placed in a circle around the active elements, each of them closed on a passive load whose impedance can be continuously changed in a given range of values [6]. A simpler class of Uda–Yagi adaptive antennas is based on the switched beam philosophy. Also these antennas are usually based on a relatively small number of parasitic elements placed in a circle around the active elements, each of them closed on a passive load. According to the value of the load, each parasitic element can work as reflector, director, or out of resonance [7]. Consequently, in this kind of antenna the loads must assume only three different values, and simple microwave switches can be adopted. Furthermore, the voltage standing wave ratio (VSWR) of the antenna is constant during the adaptive process, and can be kept very low. However, such antennas are not fully adaptive, and the performances reachable by switched-beam Uda–Yagi antennas are modest in presence of interference signals due to the absence of null-forming capacity. In this paper we propose a novel Uda–Yagi adaptive antenna architecture, that is “half way” between the above described two classes. The antenna consists of a relatively large number of parasitic elements placed on circles around the active element. Each element of the rings is terminated in two electronically controllable loads, e.g., a short circuit and an open circuit, by means of electronically controllable switches, i.e., PIN diodes or MEMS. Consequently, each antenna assumes two states, corresponding to different scattered fields. The adaptivity of the antenna is obtained by choosing the state of the elements in order to minimize a proper cost function. Usually, in adaptive antennas the objective is the maximization of the SINR. However, in the proposed antenna the VSWR depends on the state of the switches. Consequently, it is necessary to perform a multiobjective minimization taking
0018-926X/$20.00 © 2005 IEEE
MIGLIORE et al.: A SIMPLE AND ROBUST ADAPTIVE PARASITIC ANTENNA
into account both the SINR and the VSWR. This goal has been achieved by using the fuzzy logic [9] to define the cost function. The minimization of the cost function has been achieved by biological beamforming algorithms. In particular, two algorithms, the binary genetic algorithm (BGA) [10], [11] and the binary particle swarm optimization algorithm (BPSO) [12], [13] have been tested. The BGA is the most adopted biological beamforming algorithm in the adaptive antenna field, and an excellent review can be found in [14]. Instead, the use of the BPSO for adaptive beamforming is not reported in literature at the best knowledge of the authors. In particular, besides the “standard” version of the two algorithms, a self-adaptive version changing its parameters during the minimization process has been also developed and tested. It is worth noting that, while this concept is well known for the GAs [15], [16], and has been applied to adaptive arrays by many groups [17], [18], the development of self-adaptive PSO is a new field of research [19], and no self-adaptive BPSO is discussed in literature. Summarizing, the proposed antenna is based on three key factors: the large number of possible states of the antenna, obtained by using a large number of elements, the use of an efficient (self-adaptive) biological beamforming algorithm, and the use of the fuzzy logic to control both the SINR and the VSWR. The last two points are, at the best knowledge of the authors, new in the framework of null-forming parasitic antennas. Moreover, the use of a large number of elements (up to 25) closed in switched-loads is, at the best knowledge of the authors, new in the framework of null-forming parasitic antennas as well. In fact, though in literature the theoretical advantage of a large number of antennas is generally accepted, null-forming parasitic antennas actually tested have only a small number (less than 8) antennas due to the computational effort of the algorithms required in varactor-controlled loaded antennas. The above mentioned three key factors allow to obtain a number of practical advantages. The antenna is extremely simple, and has good space-filtering performances. Furthermore, the antenna is also robust with respect to the failures of some elements. The use of multiobjective biological beamforming makes the antenna also robust with respect to errors in the position of the elements or in the value of the two loads. Consequently, it is possible to accept relatively large mechanical and electrical tolerances, making the realization of the antenna simpler and cheaper than other adaptive antenna schemes. Furthermore, more complex objectives can be considered, by taking into account other parameters of interest (e.g., broad-band response, etc.). Finally, even if in this example a rotationally symmetric antenna geometry has been chosen, more complex configurations of the parasitic element positions can be adopted to match geometrical constraints of the place wherein the antenna must be placed. The paper is organized in Sections II–IV. In Section II the algorithm to choose the state of the parasitic antennas is presented. The use of a fuzzy-logic approach to determine a cost function giving an acceptable value of SINR and VSWR is discussed. Then, we briefly introduce two evolutionary search algorithms, the genetic one and the particle swarm one; a novel dynamic
3263
version of both types of algorithms is also presented, and compared to nondynamic ones. In Section III a number of numerical simulations are presented. Firstly, classical beamforming algorithms with static parameters are compared to beamforming algorithms with selfadaptive dynamic parameters. Then the dynamic version of the BGA and BPSO (called VD-BGA and VD-BPSO) are compared. Finally, the performances of an antenna consisting of 25 elements are investigated using the VD-BPSO. Section IV is devoted to experimental results obtained with two different prototypes having, respectively, 13 and 25 elements. Conclusions are reported in Section V. II. CONTROL ALGORITHM As described in the Introduction, the antenna that we propose consists of elements: a single active antenna and a relatively of parasitic elements, each terminated on a large number load that can assume two different values, i.e., short circuit and open circuit. Consequently, the number of different states that and the state of the antenna can the antenna can assume is . The goal be represented by a binary vector having length of the control algorithm is to select one of the states giving an acceptable value of SINR and VSWR. This goal is reached by minimizing a proper cost function defined by means of a fuzzy approach. A. The Cost Function The definition of the state of the antenna, i.e., the values of the switches, is a multi target problem, since two different (and concurrent) parameters must be taken into account, the SINR and the VSWR. Furthermore, no unique choice exists for a cost function that globally takes into account both the quantities of interest. Generally speaking, we are interested in obtaining a reasonably high SINR with a fairly low VSWR. The quantification of “reasonably high” and “fairly low” depends on the characteristics of the communication system. Consequently, we must associate quantitative values to a “linguistic knowledge.” This goal can be reached by defining the cost function from a “fuzzy point of view.” Fuzzy logic [9] is a powerful tool widely adopted in the area of control. It consists into three phases: in the first one the variables involved in the calculation are “fuzzyfied,” i.e., a nonlinear by the use of the mapping of these variables in the interval membership functions is considered; then an appropriate fuzzy inference operator is applied, whose role is to map its fuzzy inputs in a proper fuzzy output; at the end this output is “defuzzyfied” to obtain, for example, the desired control action. However, in our work we only use fuzzy concepts for “classification.” Consequently our fuzzy algorithm consists of only the first two steps. In particular first we make a suitable “fuzzyfication” of the variable SINR and VSWR using two membership and , then we combine this two quantifunctions ties by a fuzzy norm. The main advantage of this method is that we are making a sort of block decomposition of the problem:
3264
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
the fuzzyfication of the variables gives us a uniform set of variables, that can be treated by the preferred fuzzy norm. In this particular situation a possible choice for the membership function can be the use of the sigmoid function defined as [20] (1) where and define the position and the gradient of the slope, and represents SINR or VSWR. We also get good results by the use of one of the simplest fuzzy inference rules, the probabilistic OR, defined as (2) B. The Minimization Algorithm Due to the discrete nature of the search space, it is quite natural to use biological beamforming algorithms for the minimization of the cost function. Biological beamforming is a very active research area, and many different algorithms have been proposed. Generally speaking, no minimization algorithm can be considered a priori the “best one” [21]. Consequently, we have tested many different biological beamforming algorithms on our specific problem. In particular, as discussed in the Introduction we tested a simple Monte Carlo approach, the BGA [10], [11] in many versions and the BPSO [13]. The BPSO is an extension of the PSO algorithm [12] to a binary search space in which the “speed” of the particle is substituted by the probability of a bit to assume a value. In particular, besides the standard BPSO, we introduced an improved version of the BPSO, that we will call variable behavior binary particle swarm (VB-BPSO), that, generally, performs better than the other BPSO checked algorithms. Let us now consider the th particle. The probability of the th bit of the vector describing the position of the th particle at to be 1 can be set as step
(3) wherein is the th bit of the vector describing the posiis the th bit of the vector tion of the particle at step , describing the position of the local best found up to step (the ” of the particle, [12]), is the th bit of the vector “ describing the position of the global best found up to step (the ,” [12]), , is a random value chosen uniformly in “ , and is the th bit of a novel vector given by (4) In (4), is the number particles, and in (3) , , , , and are the weights, whose sum must be equal to 1, describing the behavior of the particles. In particular is a sort of “inertial” weight, is a sort of “memory” weight, is a sort of “mode” weight, is used to obtain the desired amount of randomness,
and is a novel parameter that we have introduced to prevent that all particle search around the same local minimum point. Since the behavior of the particles depends on the weights, it is easy to assign to each particle a different vector of the weights so that the particles that are close to a minimum perform a “local” search, while the other particles perform a “wide range” search, obtaining a VB-BPSO. In order to reach this goal the particles are sorted according to their cost function. Then the weight assigned to the th particle is calculated by linear and . In the following, interpolation of two constants we will refer to the complete behavior of the swarm as “dynamic.” Obviously the dynamic of an algorithm has to be chosen to fit the nature of the problem to solve; it would be desirable that an algorithm could adjust its dynamic to automatically fit the problem, changing its dynamic on the base of the actual state of convergence. Firstly we note that in the case of a binary search space it is possible to easily relate the convergence of the al: if is close to 0.5 it gorithms to the values assumed by is clear that there are as many gene vectors with th element equal to 1 than gene vectors with th element equal to 0; even if this is not always true, it is possible to affirm with reasonable are around 0.5 the algorithm is approximation that if many still searching, otherwise the algorithm is almost stalled around a local minimum. So we found that a suitable estimator of the state of convergence of the algorithm is given by the function (5) is the vector defined by (4). In fact we observed where that with a standard implementation of BGA-BPSO is a monotonic decreasing function of the iterations, and we can relate it to the “diversity” of the population: a value of close to 1 indicates that the particles are searching in close to different positions of the space, while a value of 0 indicates that the particles are searching around a localized area. Consequently, it is possible to vary the dynamic according . The above described algorithm allows a to the value of self-adaptive choice of the parameters.1 However, in order to avoid confusion in the use of the term “adaptive” in the text, this kind of algorithm will be referred to as variable dynamic BPSO (VD-BPSO) algorithm in the following. The literature on the BGA is very large, and we do not recall the principles of the algorithm. The interested reader can refer to the excellent paper [11] for an introduction. It is worth recalling that also for the BGAs we can define a “dynamic” as the set of parameters and rules that influence the behavior of the evolutionary process [16]. In particular we obtained a variable dynamic BGA (VD-BGA) by evaluating the vector using (4) on the individuals that have been selected by the chosen selection operator, and by modifying the probability of mutation in function of the estimator defined in (5). This is the simplest form of VD-BGA, since only one of the many parameters is modified
m
1We tried different functions for ( ), such as higher order polynomial functions or simple lambda functions. The Omega function has finally been chosen for its effectiveness and for its relative algebraic simplicity.
MIGLIORE et al.: A SIMPLE AND ROBUST ADAPTIVE PARASITIC ANTENNA
3265
during the evolution, but, as can be seen in the following, the performances obtained are satisfactory. In Sections III and IV some numerical examples of a large investigation performed on both the biological beamforming algorithms are shown, using the best VD-algorithms found after an heuristical search over a large number of possible schemes. III. NUMERICAL EXAMPLES The numerical simulations reported in this Section were obtained by solving the Hallén’s system of integral equations describing the mutual coupling of an array of dipoles [22] using the MoM. The code was built to give the impedance matrix of the entire array and the current distribution over the wires as a linear function of the tension on the gaps. Let us suppose that the first wire is the active one and that wires are loaded on its gap voltage is , while the other . The currents at the gaps impedances whose values are can be derived by the solution of the linear equation
Fig. 1. Geometry of the 24 passive elements antenna (all the distances are in terms of . Circles: passive antennas; square: active antenna; the 12 passive elements antenna consists of the active antenna and the 12 passive antennas placed on the two inner circles.
— (6) and . After evaluating the currents at the gaps, it is possible to obtain the current distribution over the wires. The structure we have chosen for our tests is formed by an , array of 25 monopoles of 0.235 , whose diameter is placed on an infinite plane of P.E.C. as depicted in Fig. 1. The active element is the central one, while the others can be loaded on an impedance whose possible values are and (these values are the mean impedances of the switches used in the experimental investigation); we imagine to feed the structure with a 50 line. This geometry is the one used in the final experimental prototype described in Section IV. In the simulation of this chapter we do not consider the noise, to emphasize the sole interference reduction capability of the parasitic element antenna, so we refer to the SIR instead of the SINR. Moreover, the desired and interfering signals are supposed to have the same power. As a first step, let us investigate the performances of the classical and VD algorithms. Many algorithms of the types described in Section II were tested. In the following we consider the algorithms that gave best mean performances. The genetic algorithms will be labeled in the following as follows. where
—
,
BGA1: this is a standard genetic algorithm, the population size is 30, we have 10 tournaments of 3 individuals, the crossover is made using a random bit mask, the “parents” couples are chosen randomly, and the probability of mutation is 3%; we also used the overlapping of generations (for details see [11]).
—
BGA2: this algorithm is almost identical to the previous, but the probability of mutation is increased to 9%. VD-BGA: even this algorithm is identical to the first one, but the probability of mutation is chosen at each iteration according to the , as . value of
These three algorithms are almost identical, the sole thing changing being the probability of mutation. It could be useful to note that because we have the overlapping of generations only 20 “oracle calls” are to be done every iteration. The particle swarm algorithms are as follows. —
—
—
BPSO1: this is a VB-BPSO made with 20 particles. The constant vector for the best particle is , and for the worst particle is . BPSO2: also this is a VB-BPSO made with 20 particles. In this case the constant vector for the best particle is , and for the worst particle is . VD-BPSO: this is a VD-BPSO obtained switching from BPSO1 to BPSO2 when is below 0.3. the value of
It is useful to point out that it is possible to consider more complex schemes for the BPSO, but we were interested to show the gain in performances obtainable with a simple threshold switch.
3266
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Fig. 3. Fig. 2.
Comparison between BPSO algorithms.
Iso-cost curves.
All the algorithms were launched 300 times for a maximum value of 100 iterations; for each iteration every considered algorithm realizes 20 “oracle calls” [21].2 We considered the mean values obtained for the cost function, the SIR and the VSWR. As a benchmark, we will compare the algorithms presented with a random search function. We also tried different fuzzy cost functions; the best results have been obtained using as membership functions two sigfor the moids whose parameters are and for . In Fig. 2 the iso-cost curves of the global fuzzy cost function considered as function of the SIR and the VSWR is drawn. The first problem we investigate is the maximization of the SIR between a desired signal and an interference signal. In the following we choose the angle of arrival (AOA) of the desired signal equal to 0 , and the AOA of the interference signal equal to 45 . In Fig. 3, a first comparison between different BPSO algorithms is presented; BPSO1 gives optimum performances in the first 35 iterations, whilst BPSO2 gives the best performances after a high number of iterations. As previously discussed, VD-BPSO is obtained simply combining BPSO1 and BPSO2, and it is possible to have good performances for both the first iterations and for the last ones. In Fig. 4 the comparison between different BGA algorithms is shown; also in this case BGA1 is an algorithm that gives very good performances in the first 35 iterations, and BGA2 is an algorithm that gives better performances in the advanced phase: finally VD-BGA outperforms both. The above results confirm the better mean performances of VD biological algorithms with respect to static ones. 2It is worth noting that the algorithms themselves are very fast: an assembler implementation of the proposed GA requires less than 200 s for each iteration on a low-cost 200 MHz DSP (without considering the possible pipelining speedup). Of course, the use of more expensive DSP would significantly decrease the time.
Fig. 4. Comparison between BGA algorithms.
Let us now compare the performances of the VD versions of the BGA and the BPSO. In Fig. 5 it is possible to see the comparison between VD-BPSO and VD-BGA in terms of SIR. The plot shows some small differences between the algorithms. In particular, the VD-BPSO gives slight better mean performances in the first steps, while the VD-BGA slightly outperforms the VD-BPSO after the first 15 iterations. However, the differences are small, so that we can state that in practice the performances of the two algorithms are almost the same. For the sake of comparison, the SIR obtained by a random search is also shown in the same figure, showing that while the beamforming algorithms exceed the mean value of 30 dB within ten iterations the random search do not reach this level within 100 iterations. About the VSWR, in this example its value is always lower than 1.3, and is almost the same for the two biological algorithms.
MIGLIORE et al.: A SIMPLE AND ROBUST ADAPTIVE PARASITIC ANTENNA
3267
Fig. 7. Distribution of the VSWR over 1000 sessions after 100 iterations of VD-BGA in the case of a sole interference with and without VSWR control via the fuzzy cost function (desired signal from 0 , interference from 45 ). Fig. 5. Mean SIR as function of the number of iterations.
Fig. 6. Mean SIR as function of the AOA of the interference signal (desired signal from 0 ).
In Fig. 6 the SIR obtained varying the AOA of the interference is shown, with a fixed 0 AOA of desired signal, after five iterations, and after 100 iterations, considering the two self-adaptive algorithms. As could be expected, a decay of the performances as the angle between the signals gets smaller is observable. It is also interesting to note that VD-BPSO slightly outperforms the VD-BGA for any angle after five iterations, but we have the inverse situation after 100 iterations. Summarizing, numerical simulations show very close performances of the VD-BGA and VD-BPSO. Both of them allow suitably good performances in terms of SIR and VSWR with a relatively small number of iterations. This was true for all the simulations we performed. Consequently, in the following we will show the results obtained by only one of the two biological algorithms. In particular, since the VD-BPSO showed a slightly faster convergence in the first ten steps of the simulation, we will
show the results obtained by using this latter algorithm. However, it must be stressed that this convergence behavior depends on the particular parameters chosen for the two algorithms. Even if the choice has been done with great care, it cannot be excluded that other choices of parameters could give a faster convergence of the VD-BGA in the first iterations. As further general observation, in all the simulations the value of the VSWR has always been kept below 1.5, a value that can be considered suitable low for the practical applications. This is due to the use of the fuzzy cost function: in Fig. 7 the improvement of the distribution of the VSWR after 100 iteration of VD-BGA can be seen; the sole negative effect of the presence of a fuzzy cost function is the reduction of the mean SIR of 1.1 dB, which is an affordable drawback. Consequently, in the next diagrams we will refer to the sole SIR. The above examples showed the performances of the system in case of a single interference. An extensive numerical investigation has been performed considering a large number of different combinations of angles of arrival. As an example in Fig. 8 the mean SIR is shown considering a desired signal arriving from 0 , and an increasing number of interferences arriving from 45 , 120 , and 240 , each of them having the same level. The plot shows that even in presence of three interferences the antenna shows acceptable space filtering properties. A further relevant practical problem is that in real-life application the scenario dynamically changes due to the fact that the AOA of the desired signal and/or of the interferences changes during the communication. The biological algorithms were modified to solve dynamical problems: in BGA we imposed that any individual had to be evaluated every iteration, even if it had been evaluated previously; in BPSO we had instead to evaluate the p-best vector every iteration; to make the right comparison we chose the algorithms’ parameters to make 30 oracle calls every iteration. As a first investigation on the capability of evolutionary algorithms to handle these problems we considered the simple case of an interference whose AOA changes every iteration. In Fig. 9 we show the mean performances of VD-BPSO in the case of
3268
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Fig. 8. Mean SIR (VD-BPSO) as function of the number of iterations in case of multiple interferences (AOA of the desired signal: 0 ): continuous line: 1 interference (AOA = 45 ); dashed line: 2 interferences (AOA = 45 and 120 ); dotted line: 3 interferences (AOA = 45 , 120 and 240 ).
Fig. 10. Mean SIR value obtained with VD-BPSO as function of the number of iterations in presence of failures randomly positioned.
degradation of the mean performances of the antenna as function of the number of failures. Good results have been also obtained varying the values of the impedance of the switches, or the position of the elements to simulate the effect of tolerance in the manufacture process. IV. EXPERIMENTAL VALIDATION
Fig. 9. Mean SIR values obtained with VD-BPSO when the AOA of the interfence signal changes from 25 to 75 (AOA of the desired signal: 0 ); continuous line: 0.25 each iteration; dashed line: 0.5 each iteration; dotted line: 1 each iteration.
an interference moving from 25 to 75 with a velocity ranging from 1/4 degree/iteration to 1 degree/iteration. The plot shows that the algorithm is able to converge toward a suitable solution following the source, even if, as we must expect, the performances decrease in case of “fast moving” source. A further interesting feature of the proposed antenna is its robustness with respect to failures. To show this characteristic, let us suppose that a number of switches of the array, randomly chosen, are failed so that the elements connected to these switches are always in a state, e.g., the OFF state. Of course this causes a degradation of the performances of the antenna. However, this degradation is quite smooth, so that the antenna can operate also in case of failures of some elements. For example, in Fig. 10 the mean SIR in case of 1000 simulations and a number of failures between 0 and 8 is shown. The plot confirms a smooth
In order to validate the numerical results obtained in the previous Section, an experimental investigation has been undertaken. Two different prototypes have been realized, both working at 1 GHz. The first one consists of 13 monopole antennas of copper wire, whose length and diameter was respectively 0.235 and , placed on an aluminum dish having radius 2.08 . The passive antennas were placed on two concentric circles having radius 0.45 and 0.60 , while the active antenna was positioned in the center of the circles (see Fig. 1). The second prototype (Fig. 11) is equal to the first one, but with the presence of further 12 passive antennas, 6 placed on a circle having radius 0.75 and 6 placed on a circle having radius 0.90 , increasing the number of antennas to 25 (Fig. 1). The passive antennas were closed on an electronic switch based on BAP 51–31 PIN diode of Philips Semiconductors. Each switch was accurately characterized in both ON and OFF state by means of a vector network analyzer Anritsu 37 217C with SOLT calibration using the Anritsu 3650 calibration kit. The mean measured impedance values of the switch was and , with a standard defor ON state and for OFF viation of state. The switches were designed to be controlled by means of a voltage ranging in 0–5 V, with a very small current absorption. Consequently, the control of the switches could be obtained by means of a standard DAQ-24 Digital I/O PCMCIA card of the National Instruments. The card was programmed by the Matlab Instrumentation tool, obtaining a strong simplification of the hardware required to control the antenna.
MIGLIORE et al.: A SIMPLE AND ROBUST ADAPTIVE PARASITIC ANTENNA
Fig. 11.
3269
Prototype of the 25 elements adaptive antenna.
Fig. 13. Far-field pattern of the 13 elements antenna in case of AOA of the desired signal from 0 .
Fig. 12.
Set-up measurement scheme.
The antennas were tested in the semi-anechoic chamber of the University of Cassino. The measurement set-up is shown in Fig. 12. The antenna under test (AUT) operated in the receiving mode and was placed on a rotating table at 1.0 m from the floor. Two log periodic antennas Electro-Metrics EM6917C were placed at distance 6.0 m from the AUT, again at 1.0 m from the floor. The two antennas were connected to the output port of the Vector Network Analyzer by means of a switch allowing to select the transmitting antennas. GORE Faseflex cables were used to connect the antennas to the VNA. All the electronic parts of the set-up (electronic switches, and VNA) were controlled by means of a MatLab program running on a Laptop. The same program included also the adaptive algorithms. Consequently, we obtained an integrated code allowing to adapt the antenna and to perform far-field pattern measurements automatically using the rotating plane. The above described set-up allows to perform measurements of the azimuthal complex (amplitude and phase) far-field pattern of the AUT. The semi-anechoic chamber environmental reflections were minimized by carefully positioning anechoic materials on the floor of the chamber. As preliminary step, the performances of the 13-elements antenna were investigated in absence of interfering signal, and only one of the two transmitting antennas was fed. We used the VD-BPSO algorithm with the same parameters adopted in the numerical simulations discussed in Section III. The azimuthal
Fig. 14. Far-field pattern of the 25 elements antenna in case of AOA of the interference signal from 0 .
far-field pattern after ten iterations is shown in Fig. 13 as continuous line, showing that the algorithm tried to maximize the gain of the AUT. For the sake of comparison, in the same figure the far-field pattern obtained by numerical simulation considering the same state of the switches obtained in the experimental result is also shown as dashed line. The good correspondence of the two patterns confirms the validity of the numerical code used to obtain the results shown in Section III. Other preliminary measurements were performed by considering only the presence of one interference signal. As an example, in Fig. 14 the measured far-field (continuous line) pattern of the 25 elements prototype in presence of only an interference signal whose AOA is equal to 0 is shown. The plot shows a deep minimum of the pattern in the direction of the interference. The far-field pattern obtained by numerical simulation considering the same state of the switches obtained in the experimental result is also shown in the same figure as dashed line, confirming again the validity of the numerical code.
3270
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Fig. 15. Measured SINR in case of 0 AOA of desired signal and 45 AOA of the interference signal; circles: 13 elements antenna; triangles: 25 elements antenna.
Fig. 17. Far-field pattern of the 13 elements antenna in case of 0 AOA of desired signal and 45 AOA of the interference signal.
Fig. 16. Measured VSWR in case of 0 AOA of desired signal and 45 AOA of the interference signal; circles: 13 elements antenna; triangles: 25 elements antenna.
Then, the performances of the two AUTs were investigated in the case of interference reduction. The two antennas were positioned at an angle of 45 each other, obtaining two angularly spaced signals. In order to simulate an incorrelated interference the two transmitting antennas were fed alternatively. The signal received by feeding one of the two antennas was considered the desired signal, while the signal received by the other antenna was considered the interference signal. The VD-BPSO algorithm was used to adapt the AUT. In Fig. 15 the SINR as a function of the number of iterations in the case of 13-elements antenna (circles) and 25-element antennas (triangles) is shown. The plot shows good performances of both antennas. In particular, a SINR of almost 20 dB is reached in less than ten iterations in both the AUT configurations, according to the numerical simulations. It is useful to note that the performances of the 25-elements AUT are higher than the 13-elements AUT. In fact, in case of 25-elements AUT the SINR reaches more than 30 dB in less than ten iterations (note that SNR of the system is around 30 dB; consequently the SINR oscillates at about 30 dB due to the presence of random noise). In Fig. 16 the VSWR in case of 13-elements AUT (circles) and 25-elements AUT (triangles) are plotted as function of the number of iterations, showing a VSWR lower than 1.5 during the adaptive processing. The azimuthal far-field patterns of the 13-elements AUT and 25-elements AUT after ten iterations are shown in Figs. 17 and
Fig. 18. Far-field pattern of the 25 elements antenna in case of 0 AOA of desired signal and 45 AOA of the interference signal; continuous line: measured after ten iterations; dashed line: measured after four iterations; dotted line: measured after one iteration.
18, respectively. For the sake of completeness, in Fig. 18 the far-field patterns after one iteration (dotted line) and after four iterations (dashed line) are also plotted. Finally, in Fig. 19 the effect of a number of failures variable between 0 and 4 on the 13 elements prototype is shown, comparing the mean SIR achieved after 20 iteration of BGA3 in the case of desired signal arriving from 0 and interference signal arriving from 40 . The experimental results (stars in Fig. 19) confirm the robustness of the antenna with respect to the failure of the elements. Furthermore, the plot shows a good agreement between the experimental results and the results of the numerical simulation (circles in Fig. 19), confirming again the effectiveness of the numerical results shown in Section III. Globally, the above experimental results are in good agreement with the numerical results and confirm the good performances of the proposed antenna.
MIGLIORE et al.: A SIMPLE AND ROBUST ADAPTIVE PARASITIC ANTENNA
3271
and realization of the switches, and Prof. G. Panariello for his valuable suggestions on many aspects of the antenna design.
REFERENCES
Fig. 19. Comparison of the mean SIR achieved after 20 iteration of VD-BGA, showing the effect of a number of failures variable between 0 and 4 on the 13 elements prototype (signal arriving from 0 —interference signal 40 ).
V. CONCLUSION A new architecture for space-filtering Uda–Yagi adaptive antennas is proposed. The antenna consists of a relatively large number of parasitic elements placed on a number of circles around the active element. Each element of the rings is terminated in two loads by means of electronically controllable switches. The adaptivity of the antenna is obtained by choosing the state of the elements in order to minimize a proper cost function, based on the SINR, by means of “self-adaptive” versions of biological beamforming algorithms whose objective function is chosen using fuzzy logic in order to obtain both a good SINR and a good VSWR. The antenna is simple, low-cost and robust with respect to failures of the passive elements. Experimental results confirm the effectiveness of the proposed antenna. It is worth noting that the convergence of the biological beamforming algorithms investigated is quite fast, requiring less than ten iterations to obtain a fairly good SINR with a VSWR that is kept stably low during the adaptive process. The architecture proposed in this paper is also suitable for further improvements. In particular, the adoption of a neural network in order to further increase the velocity of the adaptive process is worth to be investigated. Furthermore, the use of space-time processing parallelizing the solution adopted in [23] is another interesting field of research. Finally, the results shown in the paper indicate that also a simple switched-loaded antenna can give null-forming performances similar to the more complex varactor-controlled loaded antennas. This is an interesting result, since it is generally accepted that a switched-load antenna would give very poor nullforming performances [24], and re-opens the question on the limits of the performances of simple switched-load null-forming parasitic antennas. ACKNOWLEDGMENT The authors thank Dr. F. Iannuzzo, whose deep experience on electronic circuits was absolutely fundamental in the design
[1] “Special Issue on gigabit wireless,” Proc. IEEE, Feb. 2004. [2] A. Paulraj and R. Nabar, Introduction to Space-Time Wireless Communications. Cambridge, U.K.: Cambridge Univ. Press, 2003. [3] J. Litva and T. K -Y. Lo, Digital Beamforming in Wireless Communications. Boston, MA: Artech House, 1996. [4] R. J. Compton Jr, Adaptive Antennas. Englewood Cliffs, NJ: Prentice Hall, 1988. [5] H. L. Van Trees, Optimum Array Processing. New York: Wiley, 2002. [6] K. Gyoda and T. Ohira, “Design of electronically steerable passive array radiator (ESPAR) antennas,” in Proc. Antennas and Propagation Soc. Int. Symp., vol. 2, Jul. 16–21, 2000, pp. 922–925. [7] N. L. Scott, M. O. Leonard-Taylor, and R. G. Vaughan, “Diversity gain from a single-port adaptive antenna using switched parasitic elements illustrated with a wire and monopole prototype,” IEEE Trans. Antennas Propag., vol. 47, no. 6, pp. 1066–1070, Jun. 1999. [8] R. J. Dinger, “Reactively steered adaptive array using microstrip patch elements at 4 GHz,” IEEE Trans. Antennas Propag., vol. AP-32, no. 8, pp. 848–856, Aug. 1984. [9] J. M. Mendel, “Fuzzy logic systems for engineering: A tutorial,” Proc. IEEE, vol. 83, pp. 345–377, Mar. 1995. [10] D. E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning. New York: Addison-Wesley, 1989. [11] J. Michael Johnson and Y. Rahamat-Samii, “Genetic algorithms in engeneering electromagnetics,” IEEE Antennas Propag. Mag., vol. 39, no. 4, pp. 7–21, Aug. 1997. [12] J. Robinson and Y. Rahamat-Samii, “Particle swarm optimization in electromagnetics,” IEEE Trans. Antennas Propag., vol. 52, no. 2, pp. 39–407, Feb. 2004. [13] J. Kennedy and R. C. Eberhart, “A discrete binary version of the particle swarm algorithm,” in Proc. IEEE Int. Conf. Computational Cybernetics and Simulation, vol. 5, Oct. 1997, pp. 4104–4108. [14] R. H. Haupt, H. L. Southall, and T. H. O’Donnel, “Biological beamforming,” in Frontiers in Electromagnetics. ser. IEEE Press Series on Microwave Technology and RF, D. H. Werner and R. Mittra, Eds. Piscataway, NJ: IEEE Press, 2000. [15] J. D. Shaffer and A. Morishma, “An adaptive crossover mechanism for genetic algorithms,” in Proc. 2d Int. Conf. on Genetic Algorithms, 1987, pp. 36–40. [16] M. Srinivas and L. M. Patnaik, “Adaptive probability of crossover and mutation in genetic algorithms,” IEEE Trans. Syst., Man Cybern., vol. 24, no. 4, pp. 856–667, Apr. 1994. [17] Q. Wu and Z. L. Gong, “On the performance of genetic algorithm based adaptive beamforming,” in Proc. 6th Symp. on Antennas, Propagation and EM Theory, 2003, pp. 339–343. [18] A. Massa, M. Donelli, F. G. B. De Natale, S. Caorsi, and A. Lommi, “Planar array antenna control with genetic algorithms and adaptive array theory,” IEEE Trans. Antennas Propag., vol. AP-52, no. 11, pp. 2919–2924, Nov. 2004. [19] X. Hu and R. C. Eberhart, “Adaptive particle swarm optimization: Detection and response to dynamic systems,” in Proc. Int. Conf. on Evolutionary Computation, 2002, pp. 1666–1670. [20] J. W. Harris and H. Stocker, Handbook of Mathematics in Computational Science. New-York: Springler-Verlag, 1998. [21] D. H. Wolpert and W. G. Macready, “No free lunch theorems for optimization,” IEEE Trans. Evolut. Comput., vol. 1, no. 1, pp. 67–82, Apr. 1997. [22] C. A. Balanis, Antenna Theory. New York: Wiley, 1994. [23] K. Yang and T. Ohira, “Realization of space-time adaptive filtering by employing electronically steerable passive array radiator antennas,” IEEE Trans. Antennas Propag., vol. 51, no. 7, pp. 1476–1485, Jul. 2003. [24] D. V. Thiel, “Switched parasitic antennas and controlled reactance parasitic antennas. A system comparison,” in Proc. AP-Symp., Jun. 2004, pp. 3211–3214.
3272
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Marco Donald Migliore (M’04) received the Laurea degree (honors) in electronic engineering and the Ph.D. degree in electronics and computer science from the University of Napoli “Federico II,” Naples, Italy, in 1990 and 1994, respectively. He was a Researcher at the University of Napoli “Federico II” until 2001. He is currently an Associate Professor at the University of Cassino, Cassino, Italy, where he teaches adaptive antennas, radio propagation in urban area and electromagnetic fields. He teaches microwaves at the University of Napoli “Federico II.” He is also a consultant of industries in the field of advanced antenna measurement systems. His main research interests are antenna measurement techniques, adaptive antennas and medical and industrial applications of microwaves. Dr. Migliore is a Member of the Antenna Measurements Techniques Association (AMTA), the Italian Electromagnetic Society (SIEM), the National Inter-University Consortium for Telecommunication (CNIT) and the Electromagnetics Academy. He is listed in Marquis Who’s Who in the World, Who’s Who in Science and Engineering and in Who’s Who in Electromagnetics.
Daniele Pinchera (S’05) was born in Cassino, Italy, on March 6, 1980. He received the Dr. Eng. degree in telecommunication engineering (summa cum laude) in July 2004. He is currently working toward the Ph.D. degree in the Microwave Laboratory at the University of Cassino. His current researches are in the fields of smart antenna technologies, MIMO systems and the development of efficient evolutionary computation techniques.
Fulvio Schettino (M’99) was born in Naples, Italy, in 1971. He received the Laurea degree (summa cum laude) in electronic engineering in 1997, and the Ph.D. degree in electronics and computer science in 2001, both from the University Federico II, Naples. Since June 2001, he has been a Researcher at the University of Cassino, Cassino, Italy. His main research activities concern analytical and numerical techniques for antenna and circuits analysis and adaptive antennas.
