Due to their high speed and low latency characteristics, fifth-generation (5G) mobile communication systems have been employed in various industrial and civilian applications in recent years. Notably, due to the shortage of sub-6 GHz spectrum resources, several current 5G systems work in the millimeter-wave (MMW) band. To overcome the high path loss and strong phase noise in the MMW band [1, 2] and realize high speed, high throughput, and low interference communication, it is crucial to design an antenna that offers a wide operating bandwidth, compact size, high realized gain, and low cross-polarization (XP) radiation.

In this regard, microstrip patch antennas (MPA) have gained significant attention in modern wireless communication systems for their low cost, low-profile configuration, compact size, and light weight [3, 4]. However, one drawback that greatly limits their application is the narrow impedance bandwidth. In the past decades, several methods for broadening the bandwidth of MPAs have been reported, such as reactive slot loading [5, 6], arranging short-circuited pins [7], and integrating parasitic resonators [8]. Among these, a design based on multi-resonant modes emerged as a suitable candidate. Furthermore, the shorting pin-loading technique enables the multi-mode MPA to simultaneously achieve miniaturization and a broad bandwidth [9–12]. However, a shorted MPA usually ends up with high XP radiation in its H-plane [13]. Although some low-XP MPAs have recently been designed and presented [14–16], creating a wide-band design that also ensures reduced XP radiation continues to be a challenge.

Another notable obstacle to the adoption of MPAs in MMW communication systems is the high transmission loss of the feeding structure, which can dramatically suppress the realized gain. Recently, the gap waveguide has attracted extensive attention due to its ability to serve as a low-loss guiding structure in the MMW band [17–20]. Compared to traditional microstrip lines, the gap waveguide can suppress the propagation of unnecessary surface waves [21], making it a good candidate for designing the feeding networks of MPAs and their arrays [22–24]. For instance, a four-element broadband slot antenna array based on the ridge gap waveguide was presented in [22]. In this design, the array attained a realized gain higher than 12.1 dBi over the 12–15 GHz frequency range. However, the all-metal ridge made the array bulky, heavy, and difficult to integrate. Considering these shortcomings, printed ridge gap waveguide (PRGW) can be considered a more attractive option. A two-dimensional (2D) scanning magnetoelectric dipole array with a PRGW-based Butler matrix was introduced in [23]. The proposed 2 × 2 array achieved an impedance bandwidth of 20% and a stable gain higher than 10.3 dBi. Furthermore, to realize high gain, a broadband 1 × 4 magnetoelectric dipole array loaded with split ring resonator (SRR) structures was proposed in [24]. By loading the SRR structure, the impedance bandwidth reached 34%, and the realized gain was higher than 16 dBi. However, the three-layer loading resulted in a high profile and a complex structure.

From the above examples, it can be concluded that integrating multi-mode resonances, the XP-reduction technique, and the PRGW-based feeding structure might potentially help design an MPA with good physical, reflection, and radiation performance for 5G applications. To this end, a 2 × 2 wideband MMW array is proposed in this paper. A compact and shorted MPA comprising the dual resonances of a quarter circular patch and a quarter annular patch was analyzed and designed as the array element. To achieve reduced XP radiation, four-array elements were placed in a rotationally symmetric arrangement to form the array, successfully leading to a drop in the normalized XP levels below −18.7 dB in the xoz and yoz planes. Furthermore, a pair of parasitic structures was loaded to improve the realized gain in the high frequency range. A PRGW-based feeding network was designed and applied to provide differential feeding for the planar array. The array fabrication process and its measured results are also presented for validation. It was found that the designed low-XP array has an impedance bandwidth of 14.2% and an average realized peak gain of 10.2 dBi.

Fig. 1 illustrates the structure of the MPA element. It consists of two parts—the radiation layer and the PRGW feeding structure. As for the radiation layer, a quarter circular patch and a quarter annular patch were printed and then shorted using shorting vias. The quarter circular patch was fed by slot coupling, which adopted a ridge line and a coupling slot etched on the top plate, as plotted in Fig. 1. The ridge line was integrated into the PRGW feeding structure. The geometry of the unit cell of the electromagnetic bandgap (EBG) used for the PRGW is illustrated in Fig. 2. The substrate used for the radiation layer was F4BM-2 (ɛ_r = 2.2, tanδ = 0.0014) with a thickness of 0.508 mm. Meanwhile, the substrates used for the PRGW feeding structure were F4BM-2 (ɛ_r = 2.2, tanδ = 0.0014) with a thickness of 0.254 mm, F4BM-2 (ɛ_r = 3, tanδ = 0.0025) with a thickness of 0.254 mm, and F4BM-2 (ɛ_r = 2.94, tanδ = 0.0023) with a thickness of 0.762 mm.

Fig. 2 also depicts the simulated dispersion diagram of the EBG unit cell, showing that the EBG structure generates a stopband ranging from 17.8 to 37.0 GHz. This indicates that surface waves are not supported within such a broadband. Hence, only quasi-TEM waves can propagate along the ridge line.

The design process of the MPA element is exhibited in Fig. 3. First, a complete circular patch antenna, denoted as Ant 1, was designed. However, it presented the limitation of having a narrow bandwidth. Therefore, to broaden the operation bandwidth, a shorted half annular patch was loaded as a parasitic radiation source to obtain Ant 2. This led to the excitation of another resonant mode, and the dual resonances of Ant 2 were combined. In this context, it should be noted that a shorting pin array was loaded at the center of the circular patch, which had little impact on the reflection and radiation performance of Ant 2. Due to the increases, it reached a size of 0.84λ₀ × 0.56λ₀ (where λ₀ is the free-space wavelength at the centering frequency), which made it difficult to realize the composition of a planar array. Therefore, drawing on the symmetry of the E-field distribution, the annular and circular patches of Ant 2 in the y-axis direction were cut by half, and the circular patches in the x-axis direction were also halved. Finally, the size-reduced design was realized, denoted as Ant 3, which maintained dual resonances and had a suppressed aperture of 0.42λ₀ × 0.42λ₀.

The simulated E-field distributions of Ants 2 and 3 are presented in Fig. 4. In Ant 2, as shown in Fig. 4(a) and 4(b), three equivalent magnetic currents (M₁, M₂, and M₃) along the edges of the circular and half annular patches generate radiation. Moreover, since M₁/M₂ and M₃ have different lengths, dual resonances are supported. In addition, Fig. 4(b) indicates that the E-field distribution on the circular patch is perpendicular to the yoz plane and symmetrical along the y-axis. This means that an electric wall (shorting pin array) can be placed in the yoz plane, and the left part of the circular patch can be removed directly. Moreover, the circular patch can be further cut by half along the y-axis, thus realizing the quarter circular patch of Ant 3. Fig. 4(b) also shows that the E-field distribution on the half annular patch is symmetrical to the x-axis, indicating that this patch can also be cut into a quarter annular patch. Finally, the far-field radiation of Ant 3 is generated by M₁ and M₂, as shown in Fig. 4(c).

Fig. 5 plots the simulated reflection coefficients and realized gains of Ants 1–3. It is observed that the introduction of the parasitic half annular patch enabled Ant 2 to achieve dual resonances. In addition, the peak realized gain of Ant 2 was 0.8 dB higher than that of Ant 1. By adjusting the size and position of the coupling slot, the impedance matching of Ant 3 was further broadened. The simulated results indicate that Ant 3 achieved a −10-dB impedance bandwidth of 15.3% (26.2–30.5 GHz) and a peak realized gain of 6.6 dBi. Notably, it can also be deduced that Ant 3 had high XP radiation in the broadside direction, since the far-field radiation of Ant 3 was generated by both the x-axis and y-axis components of M₁ and M₂, as shown in Fig. 4(c). Therefore, reducing the XP of the proposed antenna in the array design is a critical issue.

In this section, the above-mentioned element is improved upon to generate a 2 × 2 array antenna. The geometry of the proposed array is shown in Fig. 6, which also consists of a radiation layer and a PRGW feeding network. Four MPA elements and a pair of parasitic patches are printed onto the radiation layer. To reduce XP radiation, these array elements are placed in a rotationally symmetric arrangement. The parasitic patches are placed symmetrically along the x-axis to enhance the realized gain, which is analyzed later in this study (Table 1). A PRGW-based differential feeding network is employed to provide phase differences, compensating for the spatial phase differences. Furthermore, the feeding line below the left and right slots is symmetric, leading to the left feeding line having a 180° phase leg with regard to the right one. Therefore, the electrical length of the left feed line was designed to be half-wavelength longer than that of the right feed line to compensate for the 180° phase difference.

To explain the role played by this array layout, the results obtained under the different dual arrangements (Arrangements 1 and 2 depicted in Fig. 7) were compared and investigated. Arrangement 1 involved four elements placed along the same direction, while Arrangement 2 placed the elements in a rotationally symmetric manner. For both arrangements, the distances between the left and right elements and the upper and lower elements (D_x and D_y) were maintained at 3 mm. The far-field radiation of Arrangement 1 was generated by both the x-axis and y-axis components of the magnetic currents. In this case, the XP radiation was quite significant, which will be validated below. For Arrangement 2, the far-field radiation generated by the x-axis components of the magnetic currents canceled each other out, leaving only the y-axis components contributing to the radiation. Therefore, in this case, XP radiation was suppressed.

Fig. 8 shows the simulated radiation patterns of the two arrangements at 28 GHz. Compared to the results of Arrangement 1, the XP radiation of Arrangement 2 is greatly reduced, thus aligning well with the above analysis. It is further observed that the normalized XP levels drop from −4.5 to −18.9 dB and from −4.4 to −18.7 dB in the xoz and yoz planes, respectively. In this context, it is worth noting that compared to Arrangement 1, the relative distance between the quarter-circular patch elements in Arrangement 2 was shorter. This contributed to weakening the realized gain of Arrangement 2, which will be addressed below.

To further improve the gain of the array, two shorted parasitic patches of the same size as the half circular patch of the MPA element were introduced. As depicted in Fig. 6(a), the two parasitic patches were placed symmetrically between the array elements. The simulated E-field distribution of the entire array loaded with parasitic patches is shown in Fig. 9. It is observed that the main far-field radiation at 27 GHz arises from the magnetic currents (M₁, M₂, M₃, M1′,M2′, and M3′) at the edges of the half and quarter annular patches, while the radiation at 30 GHz comes from the symmetrical magnetic currents (M₄, M₅, M₆, M4′,M5′, and M6′) at the edges of the half and quarter circular patches. On the one hand, since the parasitic patches introduced additional magnetic currents (M₅ and M5′) at 30 GHz, the realized gain improved remarkably. On the other hand, since the currents (M₂ and M2′) introduced by the parasitic patches were quite weak at 27 GHz, the realized gain showed only minimal enhancement.

Fig. 10 displays the simulated broadside gains of the arrays with and without the parasitic patches. It is evident that when the parasitic patches are loaded, the simulated realized gain of the array in the high frequency range improves significantly, while little improvement is achieved in the low frequency range. In particular, at the 30 GHz band, the array realized gain increased from 6.7 dBi to 9.6 dBi. Moreover, the simulated results validated the above analysis.

To validate the reflection and radiation performance of the proposed antenna, an array prototype was fabricated and measured. Fig. 11(a) and 11(b) display photographs of all the layers before and after assembly, respectively. These layers were integrated using plastic screws. A PRGW-to-microstrip line transition was utilized for array feeding. Fig. 11(c) shows the experimental environment in a near-field anechoic chamber. It also indicates that only the radiation pattern in the upper half of the space can be measured. Moreover, since the measurement system rotated very slowly, only four frequency points were measured, as shown in Fig. 12.

The simulated and measured reflection coefficients and realized gains are presented in Fig. 12(a). The measured results show that the impedance bandwidth of the proposed array is 14.3% (26–30 GHz). However, due to fabrication inaccuracies, impedance mismatch is observed at high frequencies. This is probably because air gaps may have appeared between the multilayer structures during the assembly process. In addition, impedance mismatch between the SMA connector and the feeding microstrip line may also exist. These circumstances could affect not only the impedance performance of the array antennas, especially at high frequencies, but also the measured radiation pattern. A comparison between the simulated and measured gains is plotted in Fig. 12(a), showing that the measured peak realized gain is 9.8 dBi. Due to limitations posed by the experimental equipment and the assembly error with regard to the multi-layer structures, a slight difference is evident between the measured and simulated results, although the trend is basically consistent with the simulated one. Furthermore, as shown in Fig. 12(b), the total radiation efficiency of the proposed antenna was more than 80% within the operating frequency range.

Fig. 13 presents the simulated and measured normalized radiation patterns of the 2 × 2 array at four different frequencies. The results indicate good agreement between the measured radiation patterns and the simulated results at the four frequencies. In addition, it is observed that the normalized XP radiation is lower than −18 dB in the broadside direction. The measured far-field features validate that the proposed array offers the advantages of both high gain and low XP levels.

To emphasize the strengths of this work, a performance comparison of the proposed antenna and those used in previous studies in terms of the XP-reduction method, broadside XP level reduction, impedance bandwidth, and total dimensions is listed in Table 2. It is evident that using a differential-fed network [14] or a hybrid feeding technique [15] helps the array antenna realize XP reduction within a relatively wide bandwidth. However, these designs usually have complicated feeding structures, which significantly impact the manufacturing cost. Many other methods, such as using defected ground structures [16] and loading shorting pins [25], have also been proposed for XP reduction. Nevertheless, the common drawback of these designs is a narrow bandwidth. In contrast, the −10 dB impedance bandwidth of the proposed MMW antenna array was 26–30 GHz (14.3%), and the XP level attained a reduction of more than 14.0 dB in both planes.

In this paper, a compact wideband 2 × 2 MMW array based on PRGW feeding is presented. First, an MPA element consisting of a quarter circular patch and an annular patch was designed and analyzed. Notably, this element had a compact aperture supporting dual resonance. However, it also sustained high XP radiation. Therefore, to reduce the XP radiation and enhance the realized gain, the element was developed into a four-element array, whose XP level and realize gain were determined and investigated. Furthermore, a PRGW feeding network was designed to provide differential feeding for the final array. Subsequently, a prototype of the final array was fabricated and measured, with the measured results indicating that the array achieved an impedance bandwidth of 14.3% (26.0–30.0 GHz) and a realized peak gain of 9.8 dBi. In addition, the XP level of the final array declined to less than −18 dB at the measured frequencies in the broadside direction. Considering these findings, the proposed antenna can be effectively used as transmitters or receivers in 5G MMW wireless communication systems.