Dual-Band Dual-Polarized Planar Antenna for 5G Millimeter-Wave Antenna-in-Package Applications

This article presents a dual-band, dual-polarized antenna operating from 24 up to 40 GHz at 5G new radio (NR) millimeter-wave (mm-Wave) Frequency Range 2 (FR2) bands. A novel stacking arrangement of ring patches is proposed to achieve wide dual-band operation with stable gain. Two pairs of stacked ring patches operating at lower (24.25–29.5 GHz) and upper (37–40 GHz) FR2 bands are alternately integrated at four metal layers. The antenna offers a sharp roll-off and a filter-like response with radiation nulls between the operating bands due to the stacking configuration. The antenna is aperture-coupled and stripline-fed, and it conforms to the antenna-in-package (AiP) requirements. The proposed single antenna element and a $2\times 2$ array are fabricated utilizing a standard commercial printed circuit board (PCB) fabrication facility. The measurement results show a decent agreement with the simulations in terms of impedance matching, isolation, radiation patterns, and realized gain. The proposed design is a promising candidate for 5G mm-Wave base station antenna arrays.


I. INTRODUCTION
T HE research and commercial activities related to next-generation wireless communication systems are gaining momentum in parallel with the 5G network deployments. Millimeter-wave (mm-Wave) communications is a key enabling technology in 5G networks as it offers wide bandwidth for high data rate applications [1]. From the antenna electrical performance perspective, there are different design challenges at mm-Wave frequencies, such as the requirement of wide operating bandwidth, stable gain, and symmetric radiation pattern suitable for beam scanning applications. Generally, Manuscript  wideband or multiband antennas are designed to cover the multiple-frequency band allocations. The multiband antenna design needs a balance between form factor, manufacturability, and cost structure of the antenna. Due to high path loss and signal attenuation at these frequencies, highly directional antennas or phased arrays are needed to meet the link budget requirements [2], [3]. The antenna design also has to meet the array requirements, such as maximum element dimensions, mutual coupling, and beamwidth. For highly integrated wireless systems, antenna-in-package (AiP) technology is generally adopted. It is proven to be economical and reliable at mmwave frequencies; the antenna design also needs to follow the buildup design rules of the used AiP technology [4].
In addition, dual-polarized antennas are preferred as they offer better performance against multipath fading and an improvement in communication channel capacity [5]. However, it is challenging to achieve high isolation and good cross-polar discrimination (XPD) in a dual-polarized antenna as the input ports are close to each other [6]. Planar antennas have the advantage of being low profile, easy to integrate with the mm-Wavefront end, and costeffective. Recently, different planar antenna solutions have been proposed to meet the mm-Wave AiP requirements. For AiP applications, low-temperature cofired ceramics (LTCC) technology is generally preferred when performance, repeatability, and reliability are essential [7], while a multilayer organic buildup substrate is a cost-effective solution [8]. For LTCC technology, a dual-polarized planar aperture antenna with differential feed, operating at 60 GHz is presented in [9]. The antenna offers wide bandwidth and high gain with a simple geometry. Recently, a dual-polarized antenna array with filtering response based on patch antennas has been demonstrated for 5G mm-Wave applications [10].
Likewise, a Ka-band AiP using a probe-fed stacked patch is presented for multilayer printed circuit board (PCB) technology in [4]. The demonstrated bandwidth was 0.8 GHz with a 3.8 dBi peak gain at 30.8 GHz. Similarly, for wideband mm-Wave applications, a magnetoelectric (ME) dipole is an attractive choice. It provides wide bandwidth and a stable radiation pattern with low cross polarization. A linearly polarized ME dipole covering wide bandwidth from 22 to 33 GHz and flat gain response is presented with a low profile in [11]. Recently, a low-profile, vertically polarized ME monopole antenna covering even wider bandwidth from 23.5 to 44 GHz and a peak gain of 7 dBi has been demonstrated in [12]. For antenna applications with filtering capability, a differentially fed, dual-polarized stacked patch with controllable nulls has been demonstrated for 5G communications [13]. Deckmyn et al. [14] presented a dual-band array based on quarter mode substrate-integrated waveguide (SIW) cavities operating at 28 and 38 GHz frequency bands. The demonstrated antenna is simple to fabricate but only offers linear polarization; also, the available spectrum is only partially utilized. It is observed that many prospective antenna solutions with wide operating bandwidths are being demonstrated in various technologies for the mm-Wave frequency spectrum. However, only a few structures with dual-band or multiband performance are reported at these frequencies in the literature. Their ability to suppress the radiation at other than the required frequency bands can help to ease the load on the filtering circuitry of a transceiver.
For AiP applications, microstrip patch antennas are preferred due to their low profile and ease of integration. However, there are associated challenges such as narrow bandwidth, surface waves, and losses due to the inherent nature of the microstrip antennas [15]. Placing a parasitic patch over the driven patch or antenna stacking is a well-known technique to enhance the bandwidth of patch antennas. Generally, either wideband [16] or dual-band [17] operation is demonstrated by the patch stacking technique. However, the proposed novel antenna configuration combines two pairs of stacked ring patches, as shown in Fig. 1. Four ring patch resonators are merged in an alternating order to achieve a wide dualband operation. Besides, the patch stacking also offers finely tunable radiation nulls, which can be useful for achieving stringent filtering requirements at the band edges that are relatively challenging to attain by the dedicated filters. Increase in complexity and cost due to additional metal layers is a reasonable tradeoff to achieve better performance. Moreover, in the case of an AiP, generally, the interconnection of RF circuitry requires multiple metal layers. To keep the combined stack symmetric and to avoid the violation of the PCB buildup rules, additional metal layers are usually required at the antenna stack [4].
In this article, a dual-band, dual-polarized antenna covering n257, n258, n260, and n261 bands of 5G new radio (NR) Frequency Range 2 (FR2) [18] is presented. The lower band, from 24.25 to 29.5 GHz, covers the n257, n258, and n261 bands with 19.5% of fractional bandwidth, while the higher band from 37 to 40 GHz covers the n260 band with 8% of fractional bandwidth. Earlier, the preliminary simulated results of the proposed design have been presented in [19]. In Section II, the antenna configuration is presented, followed by the feed design and explanation of the antenna operation. Its matching, gain, and filtering characteristics, as well as a parametric study to determine how key parameters impact the antenna performance, are also discussed. The simulated and measured results of the single element and the 2 × 2 array are presented in Section III. Finally, Section IV concludes the work.

A. Antenna Configuration
For a microstrip patch antenna design, a low dielectric constant and a low-loss material are usually preferred to obtain better bandwidth and efficiency. Furthermore, in an AiP, the multilayer fabrication of the selected material needs to be considered. Here, the Panasonic Megtron 7 [20] family of laminates and prepregs is selected for their ultralow loss and multilayer properties. The laminate and prepreg have model numbers R-5785(N) and R-5680(N), respectively, and cloth style 3313. These are selected so that all the used materials have the same dielectric constant of 3.34 and a dissipation factor of 0.003 at the operating frequencies. As the dielectric constant depends on the ply thickness of the selected laminate, multiple plies of the same thickness are stacked to achieve a uniform dielectric constant, as shown in Fig. 2. This also simplifies the simulation model and minimizes the variation during the manufacturing process.
There are nine metal, seven substrate, and six prepreg layers in the PCB stack-up (Fig. 2) of the proposed antenna. Metal layers M1-M4 are etched with the stacked ring patches, and M5 and M8 are the connected ground planes, while the orthogonal feed lines are on M6 and M7 for dual-polarization. Layer M9 with microstrip feeds and via-pads is required for standalone antenna characterization, and it can be excluded for AiP.
Laser vias connect layer M9 to M8, while M9 is connected to M5 with blind vias. Generally, in the PCB fabrication process, a blind via diameter should be at least equal to the required depth, while a through via can be processed with a much lower thickness, depending on the available technology. To minimize the diameter of the blind via, the PCB fabrication was completed in two steps. First, a substack comprising layer M9 to M5 was built with a through via, which was then laminated with the remaining stack (M4 to M1) in the second step.

B. Feed Design
Probe feeding and aperture coupling techniques are successfully adopted to excite the patch antennas for AiP applications [7], [8]. Here, aperture coupling is utilized to feed the antenna as it restricts the spurious feed radiation from distorting the antenna radiation pattern due to a common ground plane [21]. The 3-D simulation model of the proposed antenna is presented in Fig. 3, and a 2-D view of each metal layer (M1-M8) is shown in Fig. 4. Layer M9 only contains a 50 microstrip feed and pads, and there are no tunable dimensions over it. All the optimized dimensions are presented in Table I. Layer M5 is the first (common) ground, while a second ground plane (Layer M8) is placed to isolate the antenna feed from the transceiver circuitry to avoid possible feedback or ringing problems [22]. The second ground plane causes parallel-plate modes to appear in the antenna stripline feed layers. The presence of these modes may reduce the antenna matching and efficiency. Placing shorting vias around the slot aperture is a practical solution to suppress these modes. However, they significantly influence the antenna impedance and must be judiciously positioned. Though it is recommended to have a via spacing around one-tenth of the guided wavelength [23], the minimum spacing cannot go below what is allowed by the PCB fabrication rules. Here, eight shorting vias are placed The antenna is fed by 50 microstrip lines on layer M9 for vertical and horizontal polarizations. The microstrip feed lines are connected to the striplines through feed (center) vias. A vertical microstrip to stripline quasi-coaxial transition is designed to minimize the reflections. The feed via is surrounded by a ground-via fence as increasing the number of ground vias can improve the bandwidth [24]. Here, four ground vias are placed around the feed via. To optimize the matching, the diameter of the ground cut-outs on layers M8 and M9 as well as the distance between the ground-via fence and the feed via are tuned. The coupling aperture on layer M5 is similar to an H-shaped slot with all the edges being chamfered to enhance the coupling [25]. The cross slots are fed by orthogonal dual-offset feed lines, etched on layers M6 and M7 for vertical and horizontal polarizations. These asymmetric stripline structures are sandwiched between the two ground planes, with 0.6 mm of substrate thickness. In order to achieve wide bandwidth, a linearly tapered T-junction is optimized for the reactive power divider of the feed lines. These feed lines are terminated with radial stubs.

C. Antenna Design
To understand the operation of the proposed design, antenna evolution, starting with a single aperture coupled patch (Stage 1) and progressing to four stacked patches (Stage 4), is presented here. Adding one ring patch at a time shows how multiple patches are providing the broadband operation at two separate frequency bands. A single-polarized antenna is studied for the sake of simplicity. Fig. 5 shows the simulation models of the studied stages 1-4, and the corresponding input impedance (rectangular plot of the real part) and impedance loci (Smith charts) are presented in Figs. 6 and 7, respectively. The appearance of the resonance peaks with the addition of ring patches and their mutual coupling is explained with the presented plots. Finally, all the studied stages are compared based on their matching and radiation performance. All the simulations and optimization of the proposed design are carried out by utilizing CST Microwave Studio (MWS).
1) Stage 1: At the first stage, an aperture coupled square-ring patch P 1 embedded in the dielectric material is designed, as shown in Fig. 5(a). For the fundamental TM 11 mode operation, the mean parameter of the ring should be one guided wavelength of the microstrip width. The initial ring dimension can be calculated by [26] where c 0 is the speed of light in free space, f is the fundamental frequency of the square-ring antenna, L o and L i are its outer and inner lengths, respectively, and ε eff is the effective dielectric constant. The initial difference between L o and L i can be set to the minimum manufacturable trace width, and the value can then be increased as required. A practical value for the height of the patch from the ground plane, H 1 , based on available literature and parametric study can be selected at this stage. It can be optimized later in the presence of the remaining patches for optimum performance. Its effect on the antenna coupling to slot and impedance bandwidth will be discussed in parametric studies. The input resistance (real part) shows the four peaks in Fig. 6(a). The peaks at 21.7 and 43.6 GHz are due to the resonant stripline feed, while peaks at 26.8 and 35.3 GHz show the resonance of the ring patch and slot, respectively. The impedance locus of the first stage [ Fig. 7(a)] shows a single loop at the center of the Smith chart. The loop over the Smith chart is due to the coupling of feed and the patch resonance.
2) Stage 2: An aperture-coupled stacked ring patch antenna (P 1 + P 3 ) to cover the lower operating band is designed at the second stage, as shown in Fig. 5(b). The upper patch P 3 is proximity coupled to the lower patch, P 1 . The spacing (H 2 + H 3 ) between the patches and their dimensions are tuned in a manner to yield the maximum bandwidth. The gap basically controls the coupling between the patches. Generally, for any stacked patch design, the lower patch alone is designed to be overcoupled, and the input impedance is compensated by the addition of the top patch [15]. In Fig. 6(b), five resonance peaks can be observed on the input resistance plot. The fifth peak appeared due to the addition of P 3 . It is observed that due to stacking, the resonance of the lower patch is shifted to a lower frequency. The peaks at 25.8 and 30.5 GHz are due to P 1 and P 3 , respectively. Fig. 7(b) shows two tight loops near the center of the Smith chart. The second loop is due to the interaction of two coupled patches.
3) Stage 3: To achieve dual-/high-band operation, a third patch, P 2 , is introduced between the two lower band patches, as shown in Fig. 5(c). The resonance frequency of P 2 is strongly dependent on its spacing from P 1 , as it acts as a ground for P 2 . The effect of dimensions of P 2 and its position between P 1 and P 3 on its operating frequency is observed by the parametric study. On the input resistance plot in Fig. 6(c), the sixth resonance around 37 GHz is due to P 2 . It is observed that P 2 has no influence on the resonance frequency and the input impedance of P 1 and P 3 as the input impedance curves of Stages 2 and 3 completely overlap at the lower operating band. The impedance locus of the configuration in Fig 7(c) shows a third loop, which is due to the coupling of P 2 and the slot resonance. 4) Stage 4: Finally, a fourth patch P 4 is added on top of P 3 to achieve a wide, dual-band operation, as shown in Fig. 5(d). Here, P 3 acts as a ground for P 4 and its operating frequency is dependent on its spacing from P 3 . It is observed in Fig. 6(d) that the addition of P 4 has a minor influence on the lower operating band: the resonance frequency of P 3 is slightly shifted to the lower side with an increase in its input impedance. At the higher frequency band, P 4 has slightly shifted the resonance frequency of P 2 to the lower side and also compensated its high input impedance. In Fig. 7(d), three coupling loops can be observed over the impedance locus. The introduction of P 4 has broadened the second coupling loop at the lower frequency band. At the higher frequency band, P 2 is now coupled to P 4 .

D. Antenna Matching and Gain Performance
The performance of all the presented stages of the antenna design in terms of impedance bandwidth is shown in Fig. 8. The first stage with one resonant patch can partially cover the lower band in terms of impedance bandwidth. The required bandwidth, from 24.25 to 29.5 GHz at the lower band, is achieved with two stacked patches, P 1 and P 3 , at the second stage. By the addition of P 2 , around 1 GHz of impedance bandwidth is achieved at the higher band in the third stage of the design. The desired bandwidth at the two operating bands is achieved with the four patches in the final stage with the addition of P 4 . A minor reduction in impedance bandwidth at the lower band is observed due to the inclusion of P 4 .
The antenna gain at different design stages is compared in Fig. 9. In the first stage, the antenna is mainly radiating at the lower frequencies due to P 1 . The stacking of P 3 at the second stage further improves the antenna gain at the lower frequency band. A radiation null (first null) can be observed here at 33 GHz due to the addition of P 3 . At the third stage, the antenna gain at the higher operating band is improved due to P 2 . It does not affect the lower band radiation performance. In addition, a second null appears around 36 GHz. P 4 in the final stage further improves the radiation performance at the upper band. The third null around 41.5 GHz is also achieved, although it is not very deep due to unwanted radiation at higher frequencies caused by the diffraction of surface waves from the antenna ground edges. In addition, the radiation bandwidth at the lower band is slightly reduced due to the last added patch. In the following, the three nulls or the filtering characteristics of the proposed design are explained in detail.

E. Filtering Characteristics
A stacked patch configuration approaches the upper limit of the bandwidth as the phase difference between the surface currents over the patches becomes 180 • [16]. The proposed design offers three radiation nulls at 31.5, 34.5, and 41.5 GHz (Fig. 9, Stage 4). To explain these nulls, first, the electric field distribution between the patch resonators at operating and null frequencies is compared in Fig. 10. Patches P 1 and P 3 are coupled at 29.0 GHz to operate at the lower frequency band, as shown in Fig. 10(a). The first radiation null, at 31.5 GHz, appears due to the reversing phase of these fields between the lower band pairs [ Fig. 10(d)]. Fig. 10(b) shows the coupling between P 2 and P 3 to cause the unwanted radiation around 33 GHz. The second null around 34.5 GHz occurs as the phase of this coupling field is reversed by 180 • , as shown in Fig. 10(e). Similarly, at the higher band, P 2 and P 4 are coupled at 39.0 GHz as visible in Fig. 10(c). The third null around 41.5 GHz is achieved as the phase of the field is reversed, as marked in Fig. 10(f).
The surface current distribution over the ring patches at different frequencies is presented in Fig. 11 to further clarify the radiation nulls. Figs. 11(a) and (b) show the surface currents over the patches at 29.0 and 31.5 GHz, respectively. It can be seen that P 1 and P 3 are in even mode at 29 GHz while changing to odd mode at 31.5 GHz, which causes the first null. Surface currents on P 2 and P 4 remain out of phase at these frequencies. Similarly, Fig. 11(c) and (d) present the surface currents for 33.0 and 34.5 GHz, respectively, to explain the second null at 34.5 GHz. It can be observed that P 2 and P 3 have the same current direction, which reverses at 34.5 GHz and causes the second null. P 1 and P 4 remain out of phase at both frequencies. Finally, the surface currents at 39.0 and 41.5 GHz are presented in Fig. 11(e) and (f), respectively, to study the third null. The P 2 and P 4 patches are coupled to operate at the higher band, and the null appears as the surface currents over P 4 reverse the phase. Similarly, P 1 and P 3 remain out of phase at both frequencies. It is pertinent to note that with increasing operating frequencies, the proposed antenna becomes electrically thicker. Consequently, the phase difference between the surface currents of the coupled patches increases with frequency. Nevertheless, the phase reversal condition remains valid for a null frequency.

F. Parametric Study
A parametric study is presented to study the influence of different parameters on the performance of the proposed antenna. In the following, all the parameters except for the one being studied are kept constant, as presented in Table I. 1) Effect of the Patch Dimensions: A ring patch geometry offers two parameters, L i and L o (inner and outer length), to tune its resonance frequency, impedance, and bandwidth [27]. Each of the dimensions of the four resonators is varied to observe its influence on the input resistance of the proposed design in Fig. 12. The resonance frequency of P 1 decreases and its input resistance increases with the increase in L op1 , as shown in Fig. 12(a). Due to the frequency shift, it starts to decouple from P 3 . The increase in L op1 also decreases the slot resonance frequency. Similarly, decreasing L ip1 causes an insignificant increase in the resonance frequency of P 1 and a decrease of its input resistance. On the other hand, L ip1 has a considerable impact on the resonance frequency and input impedance of P 3 . Decreasing L ip1 increases the resonance frequency and input resistance of P 3 . This is due to the nature of coupling between P 1 and P 3 , as it changes the effective dimensions of P 3 [28]. Moreover, L ip1 has a pronounced effect on the slot resonance frequency, it increases rapidly with a decrease of L ip1 . It can be concluded here that the dimensions of P 1 are critical for both the operating bands. Likewise, as shown in Fig. 12(c) and (d), in the higher frequency band, an increase of L op2 reduces the resonance frequency of P 2 and it decouples from P 4 . Similar to the effect of L ip1 on the P 3 resonance, decreasing L ip2 significantly increases the resonance frequency and input impedance of P 4 .
It can also be observed that the inner and outer dimensions of P 2 have no impact on the resonances of the lower operating band. Similarly, for P 3 , increasing L op3 will decrease its resonance frequency and increase its input resistance without affecting the upper band, as shown in Fig. 12(e). At the same time, decreasing L ip3 will increase its resonant frequency and decrease its input resistance. It is important to note the effect of L ip3 on the upper band, as its decreasing length will affect the coupling between patches P 2 and P 4 . It means that tuning L ip3 will affect both the operating bands. For P 4 dimensions, increasing L op4 follows the same trend of decreasing resonance frequency with an increase in input impedance, as shown in Fig. 12(g). This also causes a shift in the resonance frequency of P 2 toward the lower frequencies. Patch P 2 will eventually decouple from P 4 due to this frequency shift of P 2 . It also affects the lower frequency band by reducing the resonance frequency and increasing the input resistance of P 3 . Finally, in Fig. 12(h), it is shown that L ip4 has no significant effect on the lower frequency band. At higher bands, a slight shift toward the upper frequencies and a reduction in input resistance are observed with the increase of L ip4 .
From the parametric study, it is observed that the individual response of all the patches to outer and inner length variation is the same, apart from their influence on other resonances. In other words, increasing the outer length results in a lower resonance frequency and higher input impedance, while reducing the inner length increases the resonance frequency and decreases the input impedance. However, there is an interdependence between the individual resonance frequencies, the input impedance of the stacked patches, and the dimensions of the surrounding patches due to the coupling mechanism. In the presented antenna design, optimization tools are utilized to achieve the optimum parameters after the initial tuning of the resonance frequencies.  2) Effect of the Thickness of Antenna Substrate: To observe the influence of the substrate thickness on patch resonance and coupling between the patches, input resistance plots for varying thickness of antenna substrate are presented in Fig. 13. The obvious effect of decreasing the thickness (H 1 ) of the first substrate is an increased coupling to slot (higher input resistance) and lowered resonance frequency of P 1 , as shown in Fig. 13(a). The coupling between P 1 and P 3 and the resonance frequency and input resistance of P 3 are also affected by H 1 . The resonance frequency of the slot shifts toward higher frequencies as H 1 decreases. It is also noted that the resonance frequency of P 2 depends on H 1 as it changes the substrate thickness between P 1 and P 2 . The thickness of the second substrate, H 2 , only influences the higher operating band response, as presented in Fig. 13(b). Reducing H 2 will result in stronger coupling between P 2 and P 4 , while increasing it will eventually decouple the slot and higher band patches. The appropriate choice of the third substrate, H 3 , is required to meet bandwidth criteria at the lower band, as shown in Fig. 13(c). Decreasing H 3 will reduce the coupling between P 1 and P 3 . Moreover, the resonance behavior of P 4 also depends on its spacing from P 3 . The thickness of the last substrate (H 4 ) has a minor influence on the lower operating band performance, as shown in Fig. 13(d). It is also noted that increasing H 4 above the designed value will not yield any significant reward in terms of improvements in the operating bandwidth.
For the proposed antenna design, it is observed that the total antenna substrate thickness is mainly dictated by the bandwidth requirement at the lower frequency band. To achieve the desired 19.5% of fractional bandwidth at lower band, initially, H 1 and the substrate between P 1 and P 3 (H 2 +H 3 ) are selected. The thickness of substrates for the upper frequency band (H 2 and H 4 ) offers more flexibility in antenna design.
3) Effect of the Slot Dimensions: In the proposed aperture coupled design, the slot is mainly coupling the feed line to the stacked patches. For the sake of brevity, only the conclusions drawn from the parametric simulations are presented here. The slot length controls the coupling level along with its resonance frequency. Decreasing it affects both the operating bands as the coupling between the patch resonators reduces significantly. Furthermore, it changes not only the slot resonance frequency but also influences the resonance frequencies of the stripline and patch resonators [29]. It is preferred to keep the slot length as small as possible to avoid back lobe radiation. Compared to its length, the width of the H-slot has a minor effect on coupling and the resonance frequencies, and it is generally designed to be around 10% of the slot length [30].

III. EXPERIMENTAL RESULTS AND DISCUSSION
To validate the proposed design, a single-element and a 2 × 2 array prototype are fabricated using a commercial PCB fabrication facility. The mini-SMP (SMPM) connectors are utilized for the measurements of the prototypes due to their low profile. The S-parameters are measured with a Keysight N5247B PNA-X Network Analyzer.

A. Single Element
The 3-D simulation model with SMPM connectors and the fabricated prototype of the single-element design are shown in Fig. 14. The second ground plane (M8) is extended by 5 mm for placing the connectors. It is observed that this ground extension has a minor effect on the antenna matching. A thru-reflect-line (TRL) [31] calibration kit is designed and fabricated to calibrate out the fixture effects. Furthermore, the time-gating option [32] is enabled in the Network Analyzer to remove the ripples from the measured response. The simulated and measured two-port S-parameters of the proposed design are presented in Fig. 15. There is a minor difference between the reflection coefficient plots of the two ports as the feed lines are asymmetric. Both of the operating bands meet the |S 11 | ≤ -10 dB impedance bandwidth criterion, with minor differences between the simulated and measured results. The lower band is widened by around 1 GHz and the higher band is shifted by 0.8 GHz in the measurements. Simulated and measured isolation between the ports remains better than 20 dB, except simulated |S 21 | around 25 GHz. The extended ground affects the simulated isolation performance as the design without connectors was optimized for at least 20 dB of isolation. The uncertainties in measurements and tolerances in the PCB fabrication mainly account for the differences between simulated and measured results. At the higher frequency band, a possible cause of discrepancy is the plating requirement for the top layer at the fabrication facility, which further increased the manufacturing tolerance of the prototype.
The antenna radiation patterns are measured in an anechoic chamber utilizing an in-house developed far-field measurement setup [33]. The patterns are measured up to ±130 • due to the limitations of the turntable. The normalized simulated and measured patterns for both the operating bands in xz and yz planes are presented in Figs. 16 and 17. In general, there is a good agreement between the simulated and measured patterns. The simulated patterns for the design without connectors are also plotted for comparison to observe the pattern deterioration due to the extended ground. In this case, the XPD at the boresight direction remains better than 20 dB in both the operating bands. However, the larger ground increases the cross-polar component and the back lobe and also degrades the pattern symmetry.
The simulated and measured gains are compared in Fig. 18. The gain is measured by the comparison method, using a standard horn antenna. The frequency shift in the measured gain correlates with the detuning in the |S 11 | measurements. The measured gain is reduced by 1-2 dB across the operating bands most likely due to measurement uncertainties and unaccounted dielectric losses in the antenna substrate. Moreover, the gain stability is also affected by the extended ground and the connectors.

B. 2 × 2 Array
Generally, for a phased array, the element spacing is kept half-wavelength (λ 0 /2) at the highest operating frequency to avoid the grating lobes while steering the beam. However, in the presented case, the two operating bands are wide apart on the frequency spectrum. One possibility is to keep the element spacing λ 0 /2 at 40 GHz and have a maximum scanning range while accepting a reduced array gain and higher mutual coupling at the lower band. Here, a tradeoff is made, and the element spacing is kept at λ 0 /2 at 30 GHz, which equals 5 mm. This offers some flexibility in antenna design, as well as better array gain and mutual coupling performance at the lower operating band with a reduced scan range at the higher operating band. Arranging antenna elements in an image orientation is a well-known technique to suppress the cross polarization and increase the radiation pattern symmetry [34]. In the presented dual-polarized array, antenna elements for both polarizations are arranged in an image configuration, as shown in Fig. 19. This also helps to keep the array symmetric when placing the SMPM connectors. For the mirrored antenna elements, the excitation should be 180 • out-of-phase. This will keep the co-polarized pattern of the elements in-phase, while the cross-polarized patterns will be 180 • out-of-phase, reducing the array cross-polarized pattern significantly.
To demonstrate the performance of the proposed design in an array configuration, a 2 × 2 array is simulated and fabricated, as shown in Fig. 20. As mentioned earlier, the ground is extended by 5 mm on each side to place the SMPM connectors. For the S-parameter measurements, all the ports are terminated by 50 SMPM terminations except for the ones being measured. The simulated and measured two-port S-parameters of an element of the array are presented in Fig. 21. The simulated and measured |S 11 | and |S 21 | curves match well with the single-element results as presented in Fig. 15, with slight deterioration of |S 11 | in the array case. It is to be noted here that the push-on SMPM connectors contributed to the measurement uncertainties. The coupling between one port of the measured antenna element and other elements in the array is presented in Fig. 22. The coupling between ports 1 and 2 is the highest as they are physically the closest ones in the array. For the sake of clarity, only the results of co-polarized ports are presented; coupling between  all the orthogonal ports remains better than 20 dB at both frequency bands.
The normalized simulated and measured radiation patterns of the array in xz and yz planes are presented in Figs. 23 and 24, respectively. Each of the antenna elements of the array is measured individually while keeping the rest of the elements terminated. The patterns are combined in postprocessing with appropriate phase shift (Fig. 19) to achieve the array pattern. The measured results reasonably match with the simulated ones. The possible reasons for minor discrepancies include errors in the antenna placement on the holder and the phase stability of the measurement cables in the sequential measurements of the array elements. It can be noticed that the  XPD is significantly improved due to the mirrored arrangement of the elements, as discussed earlier. The simulated cross-polar component is very low compared to the co-polar component due to perfect symmetry and an ideal environment.
The simulated and measured gains of the fabricated array are presented in Fig. 25. The simulated results of the design without connectors are also presented for comparison to show the deterioration of the gain performance due to the larger ground and connectors. At the lower frequency band, the gain is increasing with the frequency as the antenna dimensions increase in comparison with the wavelength.
As a demonstration of the merits of the proposed design, Table II compares its performance to that of other previously reported 5G mm-Wave antenna elements and arrays.  Accordingly, it can be observed that the proposed design is characterized by the salient features of multiple demonstrated designs, such as wide bandwidth, dual bands, and radiation nulls. Furthermore, its single element size makes it a potential candidate for 5G mm-wave beam scanning applications.

IV. CONCLUSION
In this article, a single-ended, dual-band dual-polarized antenna for 5G mm-Wave applications has been demonstrated. A novel stacking configuration based on two pairs of stacked ring patches operating at two distant frequency bands is proposed. The patches have been printed in an alternating arrangement on four metal layers. Each pair of ring patches offers broad bandwidth in the respective band along with a sharp roll-off and filtering behavior in between the operating bands due to the stacking configuration. The stripline-based aperture coupling feeding technique is utilized to have better isolation and minimal spurious radiation as well as to conform with the AiP requirement of a ground plane in between the antenna and RF circuitry. In addition to a single element, a 2 × 2 array is also demonstrated to show the application of the proposed design in an antenna array. The measured S-parameters show that the antenna is meeting the −10 dB matching criterion at both the lower (24.25-29.5 GHz) and higher (37-40 GHz) operating bands with isolation better than 20 dB. The demonstrated radiation performance shows a stable gain and symmetric radiation pattern also when considering the degradation due to the larger ground required for the connectors. Moreover, the measured frequency response shows a very sharp filtering response, especially at the edges of the stopband. This will significantly relax the filtering requirement at the stopband, which is known to be critical from the perspective of spectrum regulation. The proposed multilayer antenna design is suitable for AiP applications and a prospective candidate for 5G mm-Wave base station arrays.