Dual-Polarized 6–18-GHz Antenna Array With Low-Profile Inverted BoR Elements

This article investigates a dielectric-loaded Vivaldi-type antenna array that is formed inside a plastic block. It utilizes the inverted body-of-revolution (BoR) structure, where cone-shaped cavities are formed in a dielectric material and metalized from the inside. The metalized cavities form an array of inverted BoR elements that are fed by a single printed circuit board. In contrast to previous work, this paper uses a solid, shaped plastic block instead of foam as a dielectric material and exploits its considerably higher permittivity to reduce the height of the elements. The reduced profile provides an excellent polarization purity in the class of tapered slotline (TSL) antennas while a 3:1 bandwidth is maintained with an active reflection coefficient (ARC) $\leq -10$ dB. The shaping of the dielectric block decreases the scan loss of the array and prevents detrimental surface waves. The proposed array can steer the beam up to ± 50° in all planes without exceeding −10 dB ARC. The proposed dielectric loading may be applied also to other antenna types, such as conventional BoR antennas. The excellent electrical performance and manufacturability of the array are confirmed by fabricating and measuring a prototype.


I. INTRODUCTION
T HE VIVALDI element [1] is one of the most common antenna elements used in ultrawideband (UWB) phased arrays. The large bandwidth and well-behaving active reflection coefficient (ARC) over large scan angles are its main advantages. These attractive properties have stimulated numerous studies, where different variations of the basic Vivaldi element have been developed to improve specific performance characteristics of the array. For example, bunny-ear [2], banyan-tree [3], body-of-revolution (BoR) [4], inverted BoR [5], balanced antipodal Vivaldi (BAVA) [6], sliced-notch [7], and all-metal elements [8], [9], [10] are different ways to implement Vivaldi arrays.
One important design target in many studies has been to reduce the height of the Vivaldi elements. Lower-profile Vivaldi arrays are attractive since their cross-polarization levels are smaller and they can be more easily mounted on challenging platforms. In addition, the lower profile often means reduced fabrication costs and weight which may be crucial for many applications. There are multiple possible ways to reduce the element height, such as the optimization of the element shape, the use of low-profile baluns, and the use of dielectric loading. However, in previous studies, mostly the shape of the antenna element has been optimized (e.g., [2], [3], [6], [11], [12], [13]).
In this paper, we continue the research started in [5] by introducing a dielectric-loaded, low-profile inverted BoR array operating at 6-18 GHz. In contrast to our previous work, the inverted BoR structure is now milled inside Rexolite plastic instead of Rohacell 71 HF foam, which provides significant advantages in the electrical performance of the array. Rexolite is a polystyrene-based hard plastic with low permittivity ( r = 2.53) and loss tangent (tan δ = 0.00066 at 10 GHz). Due to its desirable dielectric properties, rigidity, and excellent machinability, Rexolite is particularly suitable for antennas and microwave components with fine details. For example, due to the small wall thickness between the cavities, it would be difficult to manufacture a 6-18 GHz inverted BoR array from Rohacell (as was used in [5] at 2-6 GHz).
However, the larger permittivity of Rexolite (compared to Rohacell 71 HF or air) introduces new opportunities and challenges in antenna design. The opportunity is to exploit the permittivity to reduce the height of the element, which can be obtained by two means: First, the wavelength is smaller inside the Rexolite block than in free space (or in Rohacell). Secondly, the impedance steps between the feeding PCB, Rexolite, and free space can be used to introduce beneficial resonances that increase the bandwidth. That is, the antenna element is partly a stepped impedance-matching layer between the feed and free space. As a result, the antenna element can be considered low-profile in its class. In addition to these fundamental improvements, the radiation efficiency is improved due to the better metalization and surface roughness of the dielectric block than in [5].
The challenges of the proposed structure are also related to the dielectric material between the elements. Due to the relatively large thickness of the dielectric block, surface waves may appear when the beam is steered. If this happens, electromagnetic waves propagate only in the dielectric material and cause scan blindness. This effect, however, can be reduced by milling grooves and holes to the Rexolite block in order to reduce its effective permittivity. The grooves are also used to shape the fields radiating from the antenna elements, which decreases the scan loss of the array and improves its active impedance matching.
In addition to new, shaped dielectric material, this paper proposes a new method to establish a galvanic contact between the feeding PCB and the metalized cavities. The method employs canted coil springs that are simultaneously pressed towards cavity walls and the PCB. In other respects, the feeding structure is similar to that of [5].
The rest of the paper is constructed as follows. In Section II, the antenna structure and its operating principle are presented in detail. The paper proceeds with the infinite-array simulation results in Section III. Section IV presents the measured embedded element characteristics of  the manufactured 7×8 array and compares them to finitearray simulations. Section VI summarizes the main results and concludes the paper.

II. ANTENNA STRUCTURE AND OPERATING PRINCIPLE
This section examines the overall structure of the antenna part by part. The exploded view of a 7×8 array with 8×8 cavities is shown in Fig. 2(a), and the cross-section of two cavities is illustrated in Fig. 2(b). Three functional layers can be distinguished from the structure: 1) dielectric block with metalized cavities; 2) contact springs and inserts to align the springs for proper contact; 3) PCB with integrated balun (back cavity) and SMPS connectors. The rest of the section describes these parts and their operating principles in detail.

A. DIELECTRIC BLOCK AND METALIZATION
The dielectric block, where the antenna elements are formed, is made of Rexolite. The shape of the antenna element cavities follows an exponential curve defined by [14] with c 1 = y 2 − y 1 e Rz 2 − e Rz 1 and c 2 = y 1 e Rz 2 − y 2 e Rz 1 e Rz 2 − e Rz 1 .
Here y 1 , y 2 , z 1 , and z 2 are the coordinate points shown in Fig. 2(b), and R is the curve parameter defining the exponential curve. The sharp top peak of the cavity has been cut and rounded with a 1.5-mm radius in order to make the  milling possible. Holes for M3 screws are placed diagonally between the cavities as shown in Fig. 3(d). Furthermore, the top surface of the Rexolite block is milled. There are 1.3-mm wide and 6.2-mm deep grooves between the cavities, which has a threefold effect: 1) The effective permittivity of the top part is smaller. This helps to prevent surface waves along the dielectric block. Fig. 4 shows the difference between the dielectric block without grooves and with grooves. At 17.8 GHz, the beam-steering in H-plane excites surface waves that propagate along the surface of the array. This would result in scan blindness in the farfield pattern. When the grooves are added, the lowest surface wave frequency is shifted above the intended operational band. In addition to grooves, the dielectric block is perforated to further decrease its effective permittivity. 2) The grooves modify the shape of the wavefront radiating from the antenna element. The dielectric material near the edges of the tapered slot cause phase delay, which makes the wavefront more spherical. As these waves are advantageous in a wide-angle beam-steering scenario, the scan loss of the array becomes smaller. The difference in the curvature of the electric fields can be seen in Fig. 5, where the electric fields are plotted between two cavities. 3) The groove adds an impedance step inside the slot.
Thus, an additional resonance mode is added to the antenna structure, which can be exploited in impedance matching.
The cavities of the dielectric block are silver-plated at Jet Metal, France. Their plating process allows high-quality chemical spray metalization of plastic parts [15]. According to the manufacturer, the thickness of the silver layer is at least 1 µm, corresponding to 1.2 skin depths at 6 GHz. Due to thin plating, the conductivity at microwave frequencies is lower than that of bulk material. However, in the absence of the surface current hot spots, the reduced conductivity does not significantly affect the radiation efficiency. It can be confirmed by comparing the measurement and simulation results, e.g., in Fig. 14.
One advantage of the inverted BoR structure is that the mass of the antenna array can be reduced. In the manufactured prototype, the measured mass of the silver-plated Rexolite block is only 27 g. If the antenna array were equipped with corresponding aluminum cones (64 pcs), the estimated total mass of the cones would be 49 g. However, in reality, the mass would be larger since the cones are longer in the absence of dielectric loading. A similar dielectric block made of Rohacell 71 HF would be only 1.8 g (without silver plating), but it is difficult to manufacture due to the fine details in the antenna structure.
The proposed dielectric material and plating technology are not the only combinations to realize a 6-18 GHz antenna array. Also other low-loss dielectric materials can be used, such as Preperm [16]. Rohacell 71 HF foam used in [5] is not a viable option due to its porous structure, which deteriorates its machinability. Furthermore, the surface roughness of the foam is inherently high, which affects the radiation efficiency, especially at higher frequencies. The photographs of the silver-plated Rexolite cavity and silver-painted Rohacell cavity are illustrated in Fig. 6, which evidently shows the advantage of using solid plastic instead of foam.

B. PRINTED CIRCUIT BOARD
The antenna elements are fed by a 4-layer PCB that includes both the balun structure and the feed network. The PCB is similar to the PCB presented in [5] but scaled to a higher frequency band. In this prototype, the PCB is used only to feed the antenna elements, but the PCB can also be exploited to integrate more complicated feed networks and electronics into the array.
substrate layers (top and bottom) are laminate, but the second substrate layer is a stack of alternating laminate and prepreg sheets.
The slot of the antenna element is fed by an asymmetric stripline that is located on the second PCB layer. The stripline is connected to a via, which leads to the bottom (fourth) layer. This via is fed by a microstrip line, which, in turn, is fed by an SMPS connector. Note that of the standard connectors, only SMPS is small enough if the whole array is equipped with connectors. In a single-polarized scenario, also bigger connectors, such as SMPMs, are a good option.
Similarly as in [5], the dielectric cavity inside the PCB works as a balun. Thus, the thickness of the PCB defines the dimensions of the balun and its bandwidth.

C. CONTACT SPRINGS AND INSERTS
The galvanic contact between the PCB and the cavity walls is established with canted-coil springs whose placement is shown in Figs. 2 and 3(e). The canted-coil spring is set into the triangular groove that is formed between the plastic inserts and the cavity walls. When the PCB is installed on the antenna elements, canted-coil springs apply a force toward the contact pads of the PCB. The triangular groove guarantees that the spring force is applied evenly towards the silver-plated cavity walls.
The canted-coil spring is not the only option to connect the PCB to the antenna elements. Also other flexible contacts, such as circular shield gaskets, can be used. However, solid inserts, as used in [5], are not a viable option because of two reasons: 1) Cavities made of Rexolite are not elastic. Therefore, the manufacturing tolerances of the inserts, cavities, and the PCB would be extremely tight. 2) Solid inserts could be problematic in varying ambient temperature due to possibly different coefficients of thermal expansion (CTE). However, if the inserts can be made of the same material as the dielectric antenna block, the CTE is not a problem.

III. INFINITE ARRAY SIMULATIONS A. ARC, CROSS-COUPLING, AND EFFICIENCY
The antenna element described above is simulated with CST Studio Suite in a unit-cell configuration equivalent to the infinite array condition. Figs. 8(b) (a)-(c) show the ARC in H-plane, E-plane, and diagonal plane, respectively. As can be seen from the figures, the ARC is below −10 dB in all planes when θ ≤ 50 • . Also, larger steering angles are possible if the bandwidth requirement is slightly relaxed.
The cross-coupling of the orthogonal ports is shown in Fig. 8(a) in the diagonal plane. The coupling is negligible at small steering angles and rises up to −20 dB when the beam is steered up to ±50 • . However, the coupling levels are small in comparison with other published works, as shown in Table 2. In the E-and H-planes, the cross-coupling is negligible, ca. −40 dB, independently of the steering angle.
The simulated efficiency of the infinite array is shown in Fig. 9. The resistive losses increase linearly with increasing frequency but they are not dependent on the steering angle. Due to dielectric losses, the proposed antenna array has a slightly smaller efficiency than its all-metal counterparts listed in Table 2. The radiation efficiency may be further improved by using a lower-loss PCB substrate.

B. EMBEDDED ELEMENT PATTERN
The embedded element pattern (EEP) illustrates the beam steering capability of the infinite array. In the absence of grating lobes, the EEP is given by [17] where A phys is the physical area of one antenna element (array lattice cell), λ 0 is the wavelength in free space, and (α, β) is the ARC with progressive phase shifts α and β. According to this equation, the ideal element pattern has a shape of a cosine curve. All discrepancies between the ideal cosine pattern and the realistic pattern are due to the ARC of the antenna element.
The EEPs of the proposed array are shown in Fig. 10 at three different frequencies. Both co-and cross-polarized patterns are given in the principal planes and the diagonal plane (D-plane), following Ludwig's third definition [18]. As can be seen, the scan loss of the co-polarized component is less than 3 dB in E-and H-planes when the beam is steered up to ±50 • . The scan loss is small also in D-plane. However, at 18 GHz, the scan loss in D-plane is a little larger than 3 dB due to the rise of the cross-polarized component. All in all, the performance of the proposed array is excellent in terms of scan loss and polarization purity, as can be confirmed by Table 2.
The embedded element patterns in Fig. 10 show that there is no scan blindness at 6 GHz and 12 GHz. There is a blind spot in H-plane at 18 GHz (Fig. 10(c)) at ±60 • , which is, however, outside the intended scan range. Vivaldi elements typically have a relatively poor crosspolarization ratio (CPR) (the ratio of cross-pol component to co-pol component). Typically high-profile (ca. 3λ high ) Vivaldi arrays may have CPR > 0 dB when the beam is steered above θ = 30 • in the D-plane [19]. However, in this case, the CPR is relatively low due to the small element height.

IV. ANTENNA PROTOTYPE AND FINITE ARRAY MEASUREMENT RESULTS
The manufactured antenna prototype is a dual-polarized 7×8 array in a square lattice. The array is equipped with five SMPS connectors: one for corner element, two for edge elements, and two for center elements of orthogonal polarizations. Other ports are terminated by Vishay CH0402F 50-resistors. The antenna prototype allows measuring reflection coefficients of the aforementioned five antenna elements, cross-coupling between center elements, and embedded element patterns.
The S-parameters were measured by Agilent E8363A PNA which was calibrated using a normal short-open-load-thru (SOLT) procedure. Due to SMPS connectors, extra measurement cables are used to connect the antenna to the VNA. Their effect on the measurement results was corrected afterward using thru-reflect-line (TRL) calibration algorithm based on [20]. The far fields of the array were measured in the anechoic chamber at Aalto University.

A. S-PARAMETERS
The simulated and measured reflection coefficients are shown in Fig. 11(b), and the feed port locations are illustrated in Fig. 3. The coupling of the two orthogonal center elements is presented in Fig. 11(a). Note that these results are obtained in a passive situation when other ports are terminated, i.e., they do not describe the active reflection coefficient in the real-use scenario. The measurement results have small differences from the simulated ones: there is a small frequency shift downwards. Nevertheless, the level of the S-parameters is correct and the results show that the manufactured array works as intended.

B. EMBEDDED ELEMENT PATTERNS
Embedded-element patterns of the antenna are given for the center element only because it best describes the average element in the array. Since the array is not very large, the center element does not exactly mimic the EEP of the infinite array simulations [25]. As the steering-angle-dependent ARC of the center element differs from that of the infinite array, and the edge effects are significant, the measured EEP has a different shape and contains some ripple. The measured patterns should be compared only to the simulated finitearray pattern of the same element.
The embedded element patterns in principal planes and two diagonal planes are shown in Figs. 12 and 13, respectively. They are given at three distinct frequencies, 6 GHz, 12 GHz, and 18 GHz. The measured patterns correspond to the simulated ones, which confirms the intended operation of the antenna. Also, the realized gain is the same at 6 GHz and 18 GHz. However, the realized gain is systematically lower at 12 GHz which is partly caused by the higher reflection coefficient shown in Fig. 11(c). The same effect can be seen in Fig. 14, where the maximum realized gain is plotted as a function of frequency. The measured realized gain has a larger fluctuation than the simulated one but the overall trend is similar. The differences in the measured and simulated realized gains can be explained by a combination of four possible factors: 1) The reflection from the feeding port is higher, i.e., S nn of port n is higher. This explains ca. 0.5 dB of the difference.
2) The coupling of the antenna element to the adjacent elements may be higher, which increases the losses in the 50-terminations. 3) Resistive losses in the PCB and/or in the antenna element are higher. However, resistive losses are not a probable reason since the difference in the gain is not monotonic as a function of frequency. 4) There might be a small error in the measurements.

V. BEAM-STEERING PATTERNS
In general, the embedded element patterns can characterize an antenna array element. However, to better illustrate the beam-steering properties of the proposed 7×8 array, we have computed its beam-steering patterns. As all antenna elements of the array were not measured, we used the average pattern method [26] to generate the far fields. In this method, all antenna elements are approximated to have a similar pattern to the center element. The electric far field pattern E is then computed by [26] where g av is the average element pattern, a q is the complex excitation coefficient of each element, N is the number of elements, k is the free-space wavenumber,r is the unit vector pointing to the observation direction, and r q is the vector pointing to the element q. In Figs. 15 and 16, the approximate beam-steering patterns are shown in the E-and H-planes, respectively. Both simulation and measurement results are computed by the same method in order to make a fair comparison. The results show that the 7×8 array can steer the beam as expected. However, the peak values of the radiation patterns are a little offset when the beam is steered to ±50 • . This is due to the ripple in the EEPs, which increases or decreases the scan loss at some steering angles. Nevertheless, if the size of the array is increased, the EEPs become smoother, and also the beam-steering properties are improved.

VI. CONCLUSION
In this article, a dual-polarized inverted BoR array with low-profile elements is presented. We have shown that the inverted BoR concept can be realized for much higher frequencies and higher-permittivity dielectric material than the previous design [5]. Furthermore, the proposed array exploits Rexolite as a dielectric loading, which provides a significantly lower profile, better polarization performance, and smaller scan loss than the array based on Rohacell foam in [5]. In fact, the polarization purity is excellent in the class of TSL antennas. To the authors' knowledge, only slicednotch arrays (SNA) [7] and planar dipole arrays (e.g., [21]) have a better CPR, but they may have other disadvantages, such as complicated manufacturing process (SNA) or worse scan performance in H-plane (planar dipole arrays). In addition, as Rexolite is hard plastic, it improves the rigidity and other mechanical properties of the array.
Furthermore, as the improvements in the array performance were based on the shaped dielectric block, the same method may also be applied to other antenna types. For example, conventional, all-metal BoR antenna arrays might be improved by introducing dielectric loading. A similar approach may also be applicable for PCB-based Vivaldi arrays, as well as other tapered slot arrays.
The operation of the array was confirmed with the fabricated prototype that was measured. The results show that the manufactured array works as expected: S-parameters, realized gain, and embedded element patterns correspond to simulated ones with small differences. Table 2 shows that the proposed array demonstrates good performance in comparison with other published arrays. The excellent electrical performance, low profile, and presumably lower manufacturing cost (in comparison with conventional BoR array) make the proposed array an attractive option for UWB applications.