Dual-Polarized Highly Folded Bowtie Antenna with Slotted Self-Grounded Structure for Sub-6 GHz 5G Applications

In this paper, a novel dual-polarized highly-folded self-grounded Bowtie antenna that is excited through I-shaped slots is proposed for applications in sub-6GHz 5G multiple-input-multiple-output (MIMO) antenna systems. The antenna consists of two pairs of folded radiation petals whose base is embedded in a double layer of FR-4 substrate with a common ground-plane which is sandwiched between the two substrate layers. The ground-plane is defected with two I-shaped slots located under the radiation elements. Each pair of radiation elements are excited through a microstrip line on the top layer with RF signal that is 180° out of phase with respect to each other. The RF signal is coupled to the pair of feedlines on the top layer through the I-shaped slots from the two microstrip feedlines on the underside of the second substrate. The proposed feed mechanism gets rid of the otherwise bulky balun. The Bowtie antenna is a compact solution with dimensions of 32 32 33.8 mm3. Measured results have verified that the antenna operates over a frequency range of 3.1–5 GHz and exhibits an average gain and antenna efficiency in the vertical and horizontal polarizations of 7.5 dBi and 82.6%, respectively.


I. INTRODUCTION
The proliferation of wireless technologies and the Internet of Things (IoT) has necessitated the development of next generation of technologies, such as 5G, that provide greater channel capacity and significantly higher data transmission rate compared to 4G/LTE [1]. Moreover, this technology acceleration has driven the hardware to be more compact, easier to integrate, and more economic while offering smarter and more multifunctional characteristics. This has spawned the need for wideband antennas to enable high data throughput [2]. Hence, a lot of effort has been invested over the past several years in developing wideband antennas that meet the requirements of 5G systems and support other multiband and multi-standard wireless systems.
In this paper we present the design and implementation of a novel dual-polarized Bowtie antenna that has a small form factor. The proposed antenna is self-grounded with a novel feed mechanism that has the benefit of reducing the overall size of the antenna as well as manufacturing costs. The antenna is intended for applications in sub-6 GHz 5G Massive MIMO wireless communications systems.
Although the structure of the proposed antenna appears relatively simple, its geometry is defined by 39 parameters, which makes the optimization of antenna's performance a big challenge. Hence, an artificial intelligence (AI) driven optimization approach is necessary to realize the required performance specifications from the antenna, which is an extension of our previous work [19], where only a simple parameter sweeping was utilized. The AI method of optimization applied here is based on the parallel surrogate model assisted hybrid differential evolution for antennas (PSADEA) [20], [24]. PSADEA belongs to the SADEA algorithm series [20]- [24] and is applicable for cases with up to 40 optimization parameters. Compared to other popular global optimization methods, e.g., particle swarm optimization and genetic algorithm, PSADEA provides significantly better results at a much faster speed by a factor of 30. This article is an extension of our previous work [25]- [27]. The novelty of this work includes: 1) realization of a small form factor dual-polarized Bowtie antenna design; 2) using a highly compact feeding mechanism; 3) reduction of sidelobes by surrounding the radiation petals with a metal wall; and (iv) realization of a large beamwidth to enable wide-angle scanning in phased array antennas.
The paper is organized as follows: Section II describes the structure of the proposed dual-polarized Bowtie antenna and the optimization method employed. The simulated and measured results are given in Section III. The work is concluded in Section IV.

II. THE HIGHLY FOLDED SELF-GROUNDED BOWTIE
The aim of this work was to design a highly compact self-grounded Bowtie antenna for an application in phased array antennas. In this application the antenna spacing must be limited to 0.5λ o to enable wide scanning angles however this must be achieved without introducing grating lobes and reduction in the gain and bandwidth that result from strong mutual coupling between adjacent radiators. Moreover, as the beam is steered to a wider angle, there arises an issue that the sidelobe level rises to a non-negligible level. To circumvent these issues, we have developed a highly folded self-grounded Bowtie antenna illustrated in Fig. 1. The antenna consists of two pairs of radiation petals whose length is about one wavelength, and the pair of petals are arranged orthogonally to realize dual-polarization performance. The parameters defining the structure of the radiation petals are shown in Fig. 2. The three convex sections in the petals provide the required bandwidth of operating between 3.1-5.0 GHz. The radiation petals are surrounded with a metal wall to reduce the sidelobes from the antenna.
The antenna structure can be divided into two parts: the radiation petal part and the feeding part. Fig. 2 and Fig.3 show the parameters that define the geometries of these two parts. The full list of the parameters is given in Table I and Table II . The mechanism to excite the antenna is implemented on a stack of two PCBs (Printed Circuit Boards) of FR-4 substrate that sandwich the common ground-plane. The substrates have a height of 0.768 mm, a dielectric constant ( of 4.3, and a loss-tangent of 0.025. The two microstrip feedlines created on the upper side of the top PCB connect the two pairs of folded petals, as shown in Fig. 3a. Each of these feedlines is excited by RF signal coupled from the feedlines on the underside of the bottom PCB through the I-shaped slots implemented in the middle ground-plane, as shown in Fig. 3b. The pair of petals are excited with the RF signal of the same magnitude but out-of-phase by virtue of the connecting microstrip feedline lengths. The proposed feeding mechanism avoids the otherwise need for a bulky balun. (a) Petals and feeding mechanism (b) Feeding lines below ground plane The proposed antenna was modelled in CST Microwave Studio, with a mesh density of 20 cells per wavelength, resulting in total about 400,000 mesh cells for the whole structure. Each full-wave simulation costs on average about 7 minutes. All simulations and computations reported in this work have been carried out on a workstation with an Intel 8-core i9-9900K 3.6 GHz CPU and a 64 GB RAM, and the time consumptions are wall clock time. Optimizations have been carried out with target goals of: (1) reflection-coefficients below -dB; (2) isolation between the two ports greater than 20 dB; (3) antenna gain above 6 dBi; and (4) 3 dB beamwidth larger than across GHz. The objective function used for the optimization is stated as follows: ( 1) where , , and are the in-band S-parameters, is the in-band realized gain, and is the in-band 3 dB beamwidth. and are the penalty coefficients set to 1 and 50, respectively, to preferentially ensure that the optimization procedure focuses on satisfying the specifications for and first, by largely penalizing if they are violated. Then, meeting the requirements for the S-parameters becomes the primary focus of the optimization procedure as soon as and are satisfied.  CST Microwave Studio was initially employed for the optimization process. However, the results obtained were not satisfactory. This is because the trust region framework of CST optimizer requires the initial design to be good, however in our case this was not the case. Using our initial design, CST trust region framework was trapped in a  local optimum far from the required specifications. CST global optimization methods, such as genetic algorithm and particle swarm optimization require hundreds of EM simulations for the size of our structure. Using this method, the calculation cost is too high, and the design efficiency is very low. Therefore, we had to employ an optimization scheme described below. We first optimized the radiation petal part by using genetic algorithm optimizer in CST Microwave Studio. Then, using this preliminarily results we performed an AI-driven optimization approach to optimize the whole antenna structure. This was done using an inhouse developed PSADEA algorithm. This was because PSADEA had previously been demonstrated in [20,21] to be suitable for complex antenna designs involving challenging geometric constraints and stringent performance requirements, which other standard global optimization methods are not be able to address [28], [29].  17.76 4 th part length of feeding 1 4.32 5 th part length of feeding 1 14.21 6 th part length of feeding 1 12.48 1 st part length of feeding 2 3.84 2 nd part length of feeding 2 3.46 3 rd part length of feeding 2 11.90 4 th part length of feeding 2 4.80 5 th part length of feeding 2 4.90 6 th part length of feeding 2 4.32 7 th part length of feeding 2 11.52 8 th part length of feeding 2 8.06 9 th part length of feeding 2 3. PSADEA uses a Gaussian process (GP) t o predict new values of geometry parameters based on previous evaluations [30]. PSADEA models a parameter y(x) as a Gaussian distributed stochastic variable with mean μ and variance σ 2 . It uses a Gaussian correlation function to describe the correlation between two variables: (2) where d represents the dimension of x and θ l the correlation parameter which determines how fast the correlation decreases when x i moves in the l direction. The function p l determines the degree of smoothness with respect to x l . To determine the parameters θ l and p l , the likelihood function y = y i at x = x i (i = 1, …, n) is maximized. The function value y(x*) at a new point x* is predicted using: The mean square error value of the prediction uncertainty is: where (8) Several prescreening methods can be used to evaluate the quality of a given design with respect to the predicted value in (3) and the prediction uncertainty in (7). In PSADEA, the lower confidence bound (LCB) method [31] is used. Given the predictive distribution N( , ) for y(x), an LCB prescreening of y(x) can be defined as: (9) where ω is a constant, which is often set to 2 to balance the exploration and exploitation ability [30]. The flow diagram for PSADEA implementation is shown in Fig. 4. More details about how PSADEA works can be found in [20], [21]. The final optimized parameter values are given in Table I and Table II.   assembling. Fig. 6 shows the simulated and measured S-parameters of the proposed antenna. The vertically polarized excitation corresponds to the V-pol feedline, and the horizontally polarized excitation is due to the H-pol feedline on the top PCB, as shown in Fig. 3a. It is observed that the antenna exhibits an excellent impedance matching performance for both V-pol (S 11 ) and H-pol (S 22 ) cases. The simulated and measured and are below -10 dB over 3.08-4.74 GHz and 3.29 -5.0 GHz, which correspond to fractional bandwidths (FBW) of 42% and 41%, respectively. The difference between the simulated and measured |S 11 | and |S 22 | is marginal. The discrepancy in the S-parameters of the two polarizations is due to the asymmetry of the feedlines as well as manufacturing and assembly tolerance. The measured isolation between the ports of different polarizations is larger than 28 dB, which agrees well with the simulation results. These results show the optimization goal of the simulation agrees well with the measurements. The simulated and measured gain and radiation efficiency of the proposed antenna are shown in Fig. 7. The antenna gain was measured by using the substitution method with a calibrated reference antenna. The discrepancy between the simulation and measured results is attributed to ohmic and the mismatching losses. The measured antenna gain varies between 7.4 and 7.8 dBi for vertical polarization, and between 7 and 7.78 dBi for horizontal polarization over 3 GHz to 5.1 GHz. These results satisfy the optimization goals.

III. SIMULATION AND MEASUREMENT RESULTS
Measurements show that the radiation efficiency of the antenna varies between 82% and 86.2% in vertical polarization, and between 78.8% and 83.9% in horizontal polarization over 3 GHz to 5.1 GHz. The average discrepancy in the measured results between the vertical and horizontal polarization of the gain and radiation efficiency are 0.3 dB and 3.7%, respectively. The reason that the antenna gain for H-pol is a bit lower than that for V-pol is that the feedline for H-pol is longer and with more bends, as shown in Fig. 3a. However, there is good coherence between the simulated and the experimental results.  The self-grounded Bowtie antenna is a quasi-BOR 1 antenna and the radiation performance can be described fully by the -plane co-and cross-polar radiation patterns [25]- [27]. Simulated and experimental radiation patterns of the proposed antenna in -plane at three frequency points is shown in Fig. 8. The antenna gain is stable over the frequency band while the beamwidth is broader (3 dB beamwidth is about ). This is quite a unique characteristic which is very good for large scan phased array antennas. The simulated and experimental results are in reasonable agreement. These results demonstrate the proposed antenna is viable for applications in 5G sub-6 GHz wireless systems.  Table III shows a comparison between the proposed highly folded self-grounded antenna and the previously published work on Bowtie antennas in terms of feed approach, bandwidth, antenna gain, efficiency, polarization, footprint size and complexity. Although the self-grounded Bowtie antenna was first introduced in [25] it was a single-polarized antenna that used a PCB microstrip balun to excite the radiation petals. The balun was used to transform a 50-impedance microstrip-line to a 150- twin-line whose length was too large to implement dual-polarization. In [26] a self-grounded dual-polarized Bowtie antenna was designed and excited through four ports connected to two 180° hybrids used as baluns. In [27] the dual-polarized self-grounded Bowtie antenna was realized by using a complex balun circuit to feed its dual-polarized radiation petal pairs. The balun is designed based on the composite right-left handed transmission-line concept.
In this paper, we have replaced the complicated balun designs with a simple feed structure that couples electromagnetic energy from the input feedlines on the underside of the bottom substrate to the feedlines connected to the radiation elements on the top side of the top substrate. The electromagnetic energy is coupled through a pair of I-shaped slots implemented on the common ground-plane which is sandwiched between the two substrates. With the proposed feed mechanism, (i) dual polarization of a Bowtie antenna can be realized in a small form factor, and (ii) the use of a balun is avoided. The proposed technique can be applied to other antennas that require differential feed.

IV. CONCLUSION
A novel configuration of a dual-polarized dual-port Bowtie-antenna is shown to exhibit excellent radiation characteristics over a wideband from 3.1 GHz to 5 GHz making it suitable for sub-6 GHz 5G wireless communications systems. The design has been verified by simulations and measurements. Essentially, the antenna comprises four folded radiation petals embedded on two layers of FR-4 substrate. Each pair of vertical and horizontal petals can be excited separately to create any dual polarization, such as dual linear polarization or dual circular polarization. The new feeding structure is compact and replaces the otherwise bulky balun used in prior designs with no compromise in performance. The isolation between the two excitation ports is greater than 28 dB. The antenna has an average gain of 7.5 dBi and average radiation efficiency of 82.6% for vertical and horizontal polarizations. The antenna's dimensions are 32×32×33.8 mm 3 . The proposed antenna has a small form factor, is straightforward to integrate with RF electronics and is relatively easy of manufacture.