Flippable and Hermetic E-Band RWG to GCPW Transition With Substrate Embedded Backshort

In this article, a novel millimeter wave (mm-wave), fully hermetic and flippable transitioning structure between grounded coplanar waveguides (GCPWs) and rectangular waveguides (RWGs) is introduced. This transition, realized through standard printed circuit board (PCB) technology, can be employed to feed a waveguide (WG) compliant device on any face of a board given its flexibility in trace routing and component placement. The proposed structure utilizes a mix of transmission line and substrate integrated waveguide (SIW) interconnected with an in-substrate embedded back-short. The transition, in all its configurations, targets a fractional bandwidth (FBW) wider than 20% to cover the commercial $E$ -band spectrum. Moreover, at center frequency the predicted insertion loss (IL) for both normal and flipped transitions is 0.4 and 0.46 dB, respectively. Parametric and yield analyses, together with postmanufacturing inspections have been performed to assess the robustness of the design and to identify the critical manufacturing inaccuracies. The experimental validations confirm the wideband operation with an IL bounded below 2.3 dB for the worst case of a back-to-back configuration.

SECTION I. Introduction

In recent years, researchers’ focus has been shifted away from the crowded sub-40-GHz bands toward the portions of the millimeter wave (mm wave) spectrum, including $E$ -band (70–90 GHz), $W$ -band (90–110 GHz) and $D$ -Band (110–170 GHz). These bands are currently being employed for an increasing number of applications, which embrace terrestrial communications, satellite-to-satellite links, automotive radars, and imaging systems. Among these applications, $E$ - and $D$ -band backhauling point-to-point infrastructure links are receiving particular attention in the deployment of the 5G network and future 6G [1]. For these applications, antenna systems integration aspects are particularly critical as they involve multiple technological domains. Typically, the front end is embedded into a mixed-signal chip, which is packaged and integrated in a multilayered printed circuit board (PCB) [2]. The mixed-signal chip is usually referred to as System-in-Package (SiP) and constitutes a strategic element in modern telecommunication equipment operating in the transport segment of mobile networks. As a consequence of the progressive evolution toward mm-wave bands, technological constraints impact onto the achievable performance, or in turn, demand critical and expensive mm-wave components [3]. To address both aspects, industrial interest focuses on the integration of these sections into packaged devices embedding all radio frequency (RF) functions, namely SiPs. The critical interconnection thus becomes the interface toward the external world as a rectangular waveguide (RWG) port, representing the main goal of the work presented in this article. Additionally, industrial production features must be introduced to answer specific needs of simplification in the manufacturing process, such as the hermeticity of the device, without requiring the application of temporary barriers and seals and so simplifying the production workflow.

More generally, several applications require high-gain radiating structures, which are not easy to be manufactured in a PCB at mm waves. For this reason, most manufacturers package RF transceiver circuitry, which is usually placed on a planar PCB, together with all-metal antennas to minimize losses and reduce costs. Hence, the demand increased for microwave transitional structures capable of easing the assembly of PCBs populated by core chips and passive waveguide (WG) components within the same system [4], [5]. Many of such transitions were proposed in the literature were designed with different topologies, technologies, and substrates. They can be classified into three main groups: 1) requiring a modified WG; 2) requiring an external backshort; and 3) end launchers. The $E$ -plane probe transitions [6], [7], [8], [9], [10] belong to the first group that employs a microstrip-fed monopole probe that is inserted into the WG through an aperture on its broad sidewall. Although this configuration provides wideband coverage, it requires a WG customization, which increases cost, can cause leakage and is prone to misalignment errors. Other similar concepts employ steps or ridges attached to the WG walls as proposed in [10], [11], [12], and [13]. The second class of PCB to RWG transitions makes use of rectangular launchers, which are typically located at a quarter wavelength from a WG backshort termination [14]. When this configuration is employed at mm waves, it is possible to embed the quarter-wavelength WG backshort section into a thick multilayer PCB as it is done, for instance, in [15]. Third, in the past few years, several methods were proposed to simplify the WG to PCB transition employing planar end launchers directly connected to the RWG termination without the use of a backshort. The signal coupling is achieved through slots or other radiating structures integrated into a multilayer PCB and radiates directly into the WG [16], [17], [18], [19]. Although good performance can be achieved with these solutions, backside radiation leakage can arise since no back cavity is employed to shield the transition.

For industrial assembly, circuit hermeticity is another aspect of practical interest. This aspect implies that sealed components are preferred: these can undergo industrial washing without leaking liquids into cavities or surface-mounted device (SMD) modules. One solution to achieve a perfectly sealed transition was recently proposed in [18]. The loss level and achieved fractional bandwidth (FBW) are, however, poor. The use of a wideband patch antenna built on a quartz substrate for $D$ -band applications was reported in [20]. This solution provides about 36% of transmission bandwidth but requires a sophisticated manufacturing technology with very high manufacturing precision (i.e., $10~ \mu \mathrm {m}$ ). Moreover, the proposed design makes use of a custom tapered RWG section to fit the WR6 to the antenna size. A similar approach but based on conventional PCB processes was proposed in [21]. In this example, the transitional element is based on a dual-resonant patch, which provides 22% of transmission bandwidth in the Ka-band. Wideband operation is achieved by setting the resonant frequency of the dual-band antenna at the extremes of the desired band. Nevertheless, this approach has three main limitations. First, the bandwidth cannot be arbitrarily widened because the patch feeder is large and it interacts with the WG walls. Second, in the proposed topology, the dual-band antenna design forces the use of a dual feed. Third, it is complex to implement at higher frequencies as it requires high-capacitance gaps between the patch elements, which cannot be implemented with a standard manufacturing process.

Parallelly, the trend toward phased arrays and distributed antenna systems demands the development of multiport SiPs to simultaneously feed several antenna elements, possibly with individual power and phase control. While several complete RF frontends would allow all such features and present several intermediate frequency or baseband lines to recombine, the cost and energy consumption policies demand that such scenarios are instead handled by a reduced number of SiPs for up- and down-conversion themselves, feeding directly at mm wave via low-loss WG distribution networks, several SiPs implementing only amplification and phase control close to the antenna elements. Therefore, multiple input and output interfaces are required for such devices, often placed on opposite sides of the component. This feature is particularly important in many practical applications such as multipoint backhauling systems where several input–output WGs should be integrated into the same PCB board. In such cases, the ability to have multiple integration solutions is of primary importance.

To answer all the requirements mentioned above, a new wideband transition for $E$ -band applications is presented in this work. The proposed configuration provides a concrete answer to industrial market needs demanding simple WG to PCB transitions. Specifically, the novelty of the design allows the implementation of fully hermetic seamless transitions between standard WR12 RWGs, without requiring alteration in the RWG metallic walls or external backshort elements. Furthermore, these features are achieved with a simple PCB topology, requiring only three metal layers and two metalized via holes. Besides, the proposed architecture is extremely flexible, allowing to employ the same stack-up to implement two types of transitions referred to as 1) normal, when the WG approaches the PCB board from the top face, and the microstrip is also on that side, and 2) flipped, when the WG approaches the board from the bottom face while the feeding signal is still on the top side.

This work is organized as follows. Section II describes the design concept and the detailed structure of the proposed transition. It also explains how the proposed transition functions in every stage and discusses the performance of the final optimized designs. Section III presents the collective results from parametric and yield analyses, leading to the final design. Section IV discusses the design, measurements, and results from the investigation of the back-to-back structures. Finally, a conclusion is carried out in Section V.

SECTION II. Transition Design Rationale and Performance

As explained in the introduction, the design concept behind this work was based on three main industrial requirements:

1. to design a transition structure that delivers a wideband signal to either face of a PCB without the need for protruding backshort;

2. to achieve a hermetic design so it can a) resist water and dirt leakage into the inner transceiver circuitry, and b) eliminate any radiation leakage, which leads to losses and causes electromagnetic interference (EMI) to neighboring components;

3. to have a standard WG interface avoiding any extra metallic structures within the WG walls (e.g., inductive WG irises as reported in [10]).

The signal is guided from a WR12 standard WG to a grounded coplanar waveguide (GCPW), which represents the most practical interface between a given PCB and a monolithic microwave integrated circuit chip (MMIC). The proposed transition is designed on a common substrate using a standard PCB process. The WR12 to GCPW transition is conceived to be implemented in two configurations with structures almost identical to each other, as illustrated in Fig. 1(a). These two configurations are as follows:

1. “normal,” in which the WR12 and GCPW are on the same PCB layer, as shown in Fig. 1(a) (left);

2. “flipped,” where the WR12 and GCPW are on opposite PCB layers, as shown in Fig. 1(a) (right).

Fig. 1.

(a) Topology of the packaging concept of the proposed flippable transition in the “normal” configuration (to the left of the MMIC) and in the “flipped” configuration (to the right of the MMIC). (b) Conceptual functionality stages of the proposed RWG-to-GCPW flippable transition.

Conceptually, both configurations are based on the same blocks, as depicted in Fig. 1(b). The two configuration arrangements stem from the need for interconnecting an RWG to a GCPW printed on either side of a PCB. The transition proposed for this scope makes use of four different sections. A signal from a GCPW is driven into a substrate-integrated waveguide (SIW), then into a stripline (SL). The SL must be implemented using both the top and bottom substrates with all three metal layers, while the SIW and GCPW sections require either the top or bottom substrate, therefore, requiring only two metal layers. From the SL, a launcher is designed to excite an RWG. Thanks to the inherent symmetry of this arrangement, it is possible to implement the RWG feed on both sides of the SL section, while the launcher can be independently rotated to obtain the needed configurations. The configuration at hand can be implemented in both normal and flipped arrangement using a three-metal stack-up with only two different types of metallized via holes, namely a through and a blind via.

The targeted electrical requirements, compliant with $E$ -band back-hauling applications, are as follows.

1. Frequency of Operation: 71–86 GHz.

2. Return Loss (RL): Higher than 20 dB.

3. Insertion Loss (IL): Lower than 0.5 dB.

A. Transition Design

As depicted in Fig. 2(a) and (b), the PCB stackup has three 18-$\mu \text{m}$ -thick copper layers, namely $L_{1}$ , $L_{2}$ and $L_{3}$ , and two 254-$\mu \text{m}$ thick core dielectric substrates, $D_{1}$ and $D_{3}$ , implemented with the material Rogers CLTE-MW (with a dielectric constant of 3.03 and $\text {tan}\delta =0.0015$ ). A bonding layer (Prepreg), $D_{2}$ , based on a 38-$\mu \text{m}$ -thick Rogers Bondply 2929 (with a dielectric constant of 2.94 and $\text {tan}\delta =0.003$ ) was used to infuse $D_{1}$ and $D_{3}$ together. Two types of metalized via holes were employed: 1) through hole vias (THVs) and 2) blind vias between $_{}L_{1}$ and $L_{2}$ . The size of the proposed transitions is $6.8\times5.2$ mm². The standard WR12 RWG fits flush to a rectangular opening. In both configurations, WG access is from $L_{1}$ (flipped config.) or from $L_{3}$ (normal config.), and the SL is always realized on $L_{2}$ as illustrated in Fig. 2. The RWG’s metallic walls sit right on top a THV cage creating a dielectric cavity in the PCB, which houses the actual launcher buried in $L_{2}$ .

Fig. 2.

Exploded 3-D view of the proposed $E$ -band transition. (a) Normal configuration, with the WG received on the top face of the PCB and microstrip lying on the same face. (b) Flipped configuration, with the WG received from the bottom face of the PCB while the microstrip is on the opposite (top) face.

B. Transition Operation

Fig. 3 represents the top view of a semitransparent structure, revealing the labeled layout of the proposed transition (“normal” configuration). Its operation was studied in all its parts using a full-wave simulator [22]. Variable dimensions labeled in Fig. 3 are listed in Table I for both configurations. Three main subtransitions were employed in order to realize this design; these configurations are as follows.

TABLE I Labeled Parameters Dimensions in Micrometers

Fig. 3.

Top view (xy plane) of the proposed $E$ -band transition with labeled dimensions in $\mu \text{m}$ and variables.

1) WR12 to Stripline Subtransition:

The WR12 RWG cavity is sealed by two rows of THVs. A patch, which acts as a launcher, is placed on $L_{2}$ : it is designed and optimized to match the WG’s TE₁₀ impedance and mode pattern across the entire $E$ -band. The total dielectric thickness of the two substrates and the prepreg is 546 $\mu \text{m}$ , where the effective dielectric constant is 3, so the quarter wavelength at the center band frequency, 78.5 GHz, is $548~\mu \text{m}$ . Thus, a fully sealed backshort can be embedded into the PCB dielectric.

2) Stripline to Substrate Integrated Waveguide Subtransition:

Fig. 4 represents the mapping of four cut sections where the electric field vector distribution at 78.5 GHz is simulated. The interface between the monopole launcher and the SL is controlled by a pair of THVs forming an inductive iris, as shown in Fig. 4(a) Cut-I. These inductive irises suppress any higher order mode other than quasi-TEM into the PCB while contributing to the overall matching of overall transition. As illustrated in Fig. 4(b) Cut-I, only a quasi-TEM mode is propagating around the 70-$\Omega $ SL. The signal is guided between the two dielectric substrates in addition to the prepreg layer. Such SL configuration is shown in Fig. 4(a), where it can be noticed that unwanted modes are formed with more tapering, as illustrated in Cut-II and Cut-III, showing that an uneven distribution results from the presence of the Bondply layer. Then, the electrical field at Cut-IV shows that the quasi-TEM SL mode is now gradually converted into two opposing SIW TE₁₀ modes as in [23].

Fig. 4.

Cut sections of the vector $E$ -field distribution at 78.5 GHz showing transmission modes within SL to SIW conversion step. (a) Cut sections mapping. (b) Electric field distribution.

3) SIW to 50-$\Omega $ GCPW Subtransition:

Fig. 5 is a mapping of the electric field vector at 78.5 GHz in the four cut sections that were cut through the SIW and both tapering slots. As shown in the previous part, two opposing SIW TE₁₀ modes exist in Cut-IV. A pair of blind vias, highlighted in Fig. 2, were employed between $L_{1} $ and $L_{2}$ to block any reflected signal from scattering toward the direction of propagation within $D_{1}$ . In Fig. 5 Cut-V, TE₁₀ mode is the only propagating mode inside the SIW ($L_{3}$ ) while in $L_{1}$ the electric field is negligible. The signal from the SIW, implemented between layers $L_{2} $ and $L_{3}$ , is gradually coupled into the 50-$\Omega $ GCPW line in $L_{3}$ , with the assistance of two oppositely tapered slots, which allow only propagation of the fundamental TE₁₀ mode, as in [24]. This transition model, along with a set of THV, was utilized to mitigate the higher order TE modes, which can be generated within the 50-$\Omega $ GCPW. In principle, the bandwidth of the SIW to GCPW transition can be enlarged using tapered or Vivaldi slots as investigated in [25], [26], [27], [28], [29], and [30], slot dipoles as in [31] or capacitive quasi coaxial THV as reported in [15]. As observed in Fig. 5 Cut-VI and -VII, the more we move in the negative $x$ -axis direction, the narrower the slot spacing and the smaller the area where TE₁₀ mode is dominant inside the SIW.

Fig. 5.

Cut sections of the vector $E$ -field distribution at 78.5 GHz showing transmission modes within SIW to GCPW conversion step. (a) Cut sections mapping. (b) Electric field distribution.

Eventually, the quasi-TEM mode of the GCPW remains dominant in Cut-VIII. One Pair of THVs are positioned relatively closer to where Cut-VII is located if compared to rest of the proceeding rows to suppress any unwanted high-order TE mode.

C. Performance

The simulation results (S-parameters) of the normal and flipped versions of the proposed transition are presented in Fig. 6. Such performance is achieved after optimizing the whole design to meet the specifications with some margin at the band edges. In the normal configuration, the RL is higher than 20 dB from 69.5 to 88.1 GHz, which is 23.6% relative bandwidth, and it is better than 10 dB starting from 68.5 to 91 GHz (28.2%). It can also be observed that the IL is 0.39 dB at the center frequency of 78.5 GHz and is worst at 88.1 GHz (0.69 dB).

Fig. 6.

Simulation of transmission and reflection coefficients for proposed normal and flipped transition configurations.

On the other hand, the flipped version has an RL higher than 20 dB starting from 70.2 to 87.1 GHz (21.4%), and it is better than 10 dB starting from 69.2 to 88.4 GHz (24.3%). It can also be observed that the IL is 0.46 dB at the center frequency, and it degrades to 0.67 dB at 87.1 GHz. All the loss mechanisms have been carefully evaluated, from dielectric to conductor losses and radiation leakage.

SECTION III. Parametric and Yield Studies

In this section, the sensitivity studies performed to ensure the design robustness against uncontrolled manufacturing errors are presented. Both normal and flipped configurations were subjected to 1) parametrical analysis and 2) yield study.

A. Parametric Analysis

Six main design parameters were investigated. The chosen ones are those that have the highest foreseen probability to either impact the desired transition performance or be erroneously realized in the fabrication process. The reason for these choices and the impact of each parameter variation are described in each section.

1) Core Substrate Dielectric Tolerance:

According to the supplier of the core substrate, the dielectric constant ($\varepsilon _{r}$ ) is subjected to an error of ±0.04. It is worth mentioning that it is reasonable to suppose that the same variation affected both upper and lower core substrates ($D_{3}$ and $D_{1}$ , respectively, as labeled in Fig. 2) simultaneously. Fig. 7(a) represents the S-parameters within the 67–89-GHz band. It can be observed that, for both normal and flipped configurations, with higher core dielectric values, the curve response tends to shift toward lower frequencies with a rate of approximately −300 MHz/0.04 change in dielectric. Yet, the curve patterns of IL and RL are preserved.

Fig. 7.

S-parameters of normal and flipped configurations’ parametric analyses targeting. (a) Core substrate dielectric constant. (b) Core substrate thickness. (c) Prepreg substrate dielectric constant. (d) Prepreg substrate thickness. (e) RWG alignment. (f) Iris THVs positioning.

2) Core Substrate Thickness:

Unpredictable thickness variations can result from heat/pressure exposure in the final assembly/stacking-up process. In this regard, a ±10% variation was applied to the 254-$\mu \text{m}$ thickness to both upper and lower core substrates simultaneously. Fig. 7(b) shows a significant change in the S-parameters curve pattern, where the curves with reduced thickness were shifted at a greater rate than the ones with larger thickness. The RL, however, remains better than 18 dB, except below 75 GHz for the flipped configuration.

3) Prepreg Dielectric Tolerance:

Given the tolerance variation caused by manufacturing errors and the presence of some possible impurities within the prepreg adhesive layer, the supplier indicated a ±0.05 error in the dielectric constant. Fig. 7(c) shows that the performance of both transitions is only slightly affected by the modification of the Bondply 2929 dielectric constant. These results were somewhat expected, given the small profile of this layer.

4) Prepreg Thickness:

Only one 38-$\mu \text{m}$ -thick layer of Bondply 2929 was used to adhere the stack-up together. Its thickness was varied by ±10%, leading to a variation in the order of a few micrometers. A small shift, compared to previous deviations [see Fig. 7(a)–(c)], in the S-parameters curve pattern is observed at higher frequencies, as presented in Fig. 7(d).

5) WR12 RWG Planar Alignment:

The WR12-sized RWG terminal should be carefully positioned on the PCB since misalignments at such small wavelengths could result in undesired behaviors. In this study, the metallic WR12 opening was placed diagonally on the xy plane at five different positions, as represented in the inset in Fig. 7(e). The incremental step of movement is $50~\mu \text{m}$ in both $x$ - and $y$ -directions. The results reported in Fig. 7(e) indicate that in the normal transition configuration, the curve pattern is not deformed. The matching level has, however, slightly degraded, and the curve has shifted toward higher frequencies in all offset scenarios. On the other hand, for the flipped transition configuration, the RL and operational bandwidth are severely affected at lower frequencies. The degradation is higher for the flipped version due to the asymmetry in thickness introduced by the presence of the prepreg layer.

6) Inductive Irises Positioning:

The location of the inductive THVs (irises) is crucial for controlling the bandwidth of operation and the TEM mode around the SL. In this analysis, five different positions were investigated; these positions coexist diagonally on the xy plane, as illustrated in the inset of Fig. 7(f). For both normal and flipped scenarios, it can be noticed that the RL level is degraded in both ($- 25\,\,\mu \text{m}$ , $-25\,\,\mu \text{m}$ ) and ($- 50\,\,\mu \text{m}$ , $- 50\,\,\mu \text{m}$ ) positions, while a narrower band is reported in cases of ($25~\mu \text{m}$ , $25~\mu \text{m}$ ) and ($50~\mu \text{m}$ , 50 $\mu \text{m}$ ) positions.

B. Yield Analysis

To complement the parametric analysis and to better understand critical and sensitive parameters, a study is presented in this section, where 31 design variables, including all parameters of the parametric study, labeled in Fig. 3, and listed in Table I, were altered with ±15% with respect to their expected tolerances. For both normal and flipped configurations of the proposed transition, 200 simulations were performed with Gaussian parameter variations.

To visualize the resulting data from all iterations in a comprehensible way, two representations were used. The first one is the plot of simulations that passed the desired S-parameters condition at each frequency. In this category, the conditions are an RL greater than 20 dB and an IL lower than 0.5 dB, respectively. The second representation highlights the number of simulations that passed a given level of S-parameters for all frequencies covering the commercial $E$ -band, namely from 71 to 86 GHz.

Fig. 8 is the yield study data representation for the normal transition configuration. Fig. 8(a) shows the first representation. It represents that, across the intended band, more than 150 simulations (75%) pass the $| S_{21} |>-0.5\,\,\mathrm {dB}$ criterion, except at 77 GHz where the number of pass simulations drops to 132 (66%). Furthermore, $| S_{11} | < -20 \,\,\mathrm {dB}$ for the 70–87.2-GHz band in more than 140 simulations (70%) except for 82 GHz, where the number of simulations is 119 (59.5%). Moreover, for the full 71–86-GHz band, Fig. 8(b) indicates that 69% of simulations have $| S_{21} |>-1 \,\,\mathrm {dB}$ while 67% of simulations exhibit $| S_{11} | < -18 \,\,\mathrm {dB}$ .

Fig. 8.

Results of the yield analysis performed on the normal configuration of the proposed transition. (a) Number of simulations versus frequency and (b) S-parameters versus number of simulations, which have full 71–86-GHz coverage.

On the other hand, the data from the flipped transition configuration is displayed in Fig. 9. Fig. 9(a) depicts that for the 72–84-GHz band, $| S_{21} |>-0.5 \,\,\mathrm { dB}$ in more than 127 (63.5%) simulations except for 78.7 GHz where the number of simulations is 117 (58.5%). Furthermore, $| S_{11} | < -20\,\, \mathrm {dB}$ for the 70.5–87-GHz band in more than 66 (33%) simulations. Moreover, considering the intended band, Fig. 9(b), shows that 69.5% of simulations have $| S_{21} |>-0.7\,\, \mathrm {dB}$ while 69% of simulations have $| S_{11} | < -16\,\, \mathrm {dB}$ .

Fig. 9.

Results of the yield analysis performed on the flipped configuration of the proposed transition. (a) Number of simulations versus frequency and (b) S-parameters versus number of simulations, which have full 71–86-GHz coverage.

It can be concluded that the two transitions are reasonably robust to the tolerances, with the normal configuration being slightly more so than the flipped one.

SECTION IV. Experimental Assessment of the Back-to-Back Configurations

To demonstrate its adaptability to all possible WG and PCB arrangements, the transition proposed in Section II was tested in three different back-to-back configurations: 1) normal-to-normal (N2N); 2) flipped-to-flipped (F2F); and 3) normal-to-flipped (N2F). To this end, three different PCBs using all the identical stack-up were manufactured, as shown in Fig. 10.

Fig. 10.

Test vehicles. (a) N2N, (b) F2F, and (c) N2F.

A. Waveguide Assembly Characterization

In order to facilitate the integration of the three test circuits with the R&S ZC90E 60–90-GHz frequency extenders required to implement the test setup, a custom WG assembly was required. The custom component consists of a dual-channel block where two 90° WR$12~E$ -plane bends are engraved (see Fig. 11) so that the transition WG interfaces are separated for easier connection to the measurement system. A jig is present at the interface with the PCB to hold the circuit in place. A single dual-channel $E$ -plane bend block is required for the N2N and F2F experimental assessment as illustrated in Fig. 11(a) and (b), respectively, since the N2N and F2F transitions have both input and output WG ports on the same face.

Fig. 11.

Test set-up. (a) Custom WG assembly used for the N2N, (b) F2F, and (c) N2F test. Active WGs are highlighted in blue.

On the other hand, the use of two identical dual-channel $E$ -plane bend blocks is required to test the N2F, as shown in Fig. 11(c). It is worth mentioning that the pitch distance between both WR12 openings, facing the transition back-to-back transitions, is 14.3 mm [see Fig. 10(a)]. Thus, a 4.35-mm-long GCPW TL was designed in order to connect the realized transitions. In a simulation, the IL of this TL extension is 0.21 dB at 78.5 GHz. The custom WG assembly required for all three test configurations was characterized through the experimental configuration shown in Fig. 12. Two WR12 bends were placed in a back-to-back configuration to evaluate the losses introduced by the custom $E$ -plane WR$12~90^{\circ }$ bends. The measured S-parameters of the two WR$12~E$ -plane bends are reported in Fig. 13. Measurements were carried out from 65 to 90 GHz and compared to simulation results carried out with a full-wave simulator [22]. Measurement results show an average IL across the investigated spectrum of about 1.2 dB. These results are in good agreement with the simulation results. In particular, ripples can be noticed, which are due to nonperfect matching, with worse RL of about 14.8 dB at the lower band; however, such an RL is acceptable to accurately characterize the developed transition, as shown in the following sections.

Fig. 12.

Back-to-Back test set-up used to de-embed the measurement results of the proposed $E$ -band transition.

Fig. 13.

Simulated and measured S-parameters versus frequency of the $E$ -plane WR12 bends in back-to-back configuration.

B. Experimental Validation

The experimental assessment was conducted by measuring the three configurations illustrated in Fig. 11 and by de-embedding the custom $E$ -plane WR12 right angle bend through the data shown in Fig. 13. The achieved results are reported in terms of $| S_{21} |$ and $| S_{11} |$ in Fig. 14 along with the simulations. It is worth noticing that both simulated and measured data refer to the back-to-back arrangement. Although the IL data are satisfactorily flat and in good agreement with the simulations, the measured IL of the N2F case is higher than the simulation at low and high frequencies. This is mainly due to a degree of misalignment between the extra $E$ -plane 90° bend + bottom jig assembly. A noticeable frequency upshift can be observed in the RL curve for all configurations. The values of these upshifts are 3.3, 3.2, and 3.3 GHz for N2N, F2F, and F2N cases, respectively. Despite this shift, the overall behavior remains acceptable and in good agreement with the simulations.

Fig. 14.

Measurement and simulation results for (a) N2N transitions, (b) F2F transitions, and (c) F2N transitions.

C. Experimental Investigation

The reason for the upshift in the frequency response and IL increase appearing in all manufactured prototypes was intensively investigated. Several factors can play a key role in obtaining such results. According to the parametric study conducted in Section III-A, such factors can be: 1) a reduction in the dielectric constant of the core substrate; 2) a reduction in thicknesses of core substrates; 3) extreme misalignment in the positioning of WG on top of the transitions, and 4) misalignment in the positioning of THVs irises.

First, the reduction of the core dielectric constant to the level that causes such a frequency upshift is not compatible with the substrate technical data. In order to investigate a possible change in the dielectric thicknesses, one F2F sample was cut at its middle section to check the actual thickness of each layer, as shown in Fig. 15(a). Yet, reported thicknesses are in accordance with the simulated model. Finally, to check the correct positioning of all internal features, including the inductive THVs, a nondestructive X-ray imaging of the F2F sample was performed, as shown in Fig. 15(b) and (c). With the careful investigation of the circuitry internals, it is noticed that all the realized vias (through and blind included) have a 35% larger diameter than the simulated ones (simulated vias: 300-$\mu \text{m}$ diameter; measured vias: $\approx 400$ -$\mu \text{m}$ diameter). The effect of this change in the via diameter was evaluated through full-wave simulations, which are reported in Fig. 16. As observed, the comparison between measurement and the retro-simulation of the three transitions is now in excellent alignment. It can be thus deduced that the proposed approach could potentially perform as expected in a future manufacturing iteration. Table II lists the values of IL and FBW simulated scenarios and measurements.

TABLE II Comparison Between Back-to-Back Measurements and Simulations With the Nominal (0.3 mm) and Actual (0.4 mm) THV Diameter

Fig. 15.

Optical investigation of (a) microscope image of F2F transition’s cross section after destructive cut and fine polish, (b) nondestructive X-ray imaging of F2F transition (labels from Fig. 2), and (c) X-ray 3-D view of the GCPW section.

Fig. 16.

Comparison between measurement and resimulation with 400-$\mu \text{m}$ diameter THVs for (a) N2N transitions, (b) F2F transitions, and (c) F2N transitions.

Table III represents a comparison between the proposed design and state-of-the-art transitions operating above 18 GHz. All reported values refer to the single transition rather than the back-to-back configuration. It is worth mentioning that reported IL values for both normal and flipped standalone versions include losses associated with the jigs. It is impossible to estimate jigs’ losses because of the air gap shown in Fig. 12 (bottom view). Perfectly sealed and hermetic designs are reported in [18], [32], [33], and [34]. Yet, [33] requires a tapered metallic transition to be compatible with standard WR12. Moreover, the design presented in [32] operates in the lower $K$ -Band, requires huge computational resources and sophisticated manufacturing to be realized. On the other hand, designs reported in [18] and [34] can be considered as fair opponents; however, [18] supports a limited FBW and has a reduced performance in terms of IL despite employing dielectric material with a loss tangent that is less than the one employed in this work in order of magnitude. Furthermore, Li and Luk [34] employ a complex configuration with an increased number of blind vias in a three-layered stackup, which makes the design costly and prone to manufacturing misalignment. As it can be observed, the proposed solution in this work is the only PCB-based one, which is compatible with MMIC integration through GCPW termination, which is not the case for the SIW terminated designs reported in [33] and [34], resulting in an estimated more losses associated with SIW to MMIC transitions.

TABLE III Comparison With State-of-the-Art Circuitry to WG Transitions

SECTION V. Conclusion

A fully sealed transitional structure suitable for $E$ -band mm-wave frequencies was presented. The design, targeting $E$ -band backhauling applications, is based on a two-dielectric layer PCB, and thanks to the utilization of subtransition steps, it can provide access to WGs from the two opposite faces of the stackup referred to as normal and flipped configuration. The proposed design is fully hermetic, cost-efficient, compact, and integrable within close proximity with other PCB components as it neither leaks radiation nor requires external backshort. The termination is standard WR12 without the need for any custom modifications of the WG. A relatively wide bandwidth, larger than 20% in all configurations, was achieved. To understand the origin of the unintended frequency upshift, of about 3.3 GHz, in each of the three configurations, a detailed sensitivity analysis was performed, as well as a postproduction study. As determined by X-ray imaging, up to 35% of errors were identified via hole diameter. Retro simulations carried out with the actual via hole diameter showed that measurements agree well with simulations, provided manufacturing errors are considered. Even with such gross inaccuracies, the standalone normal and flipped transition versions achieved FBW of 21.4% and 22%, respectively. At the center, the frequency measured IL of both standalone normal and flipped transition versions are 0.6 and 0.75 dB, respectively.

ACKNOWLEDGMENT

The authors would like to thank Vincenzo Formoso (Department of Physics, University of Calabria) who has provided the X-ray measurements, and Carmine Maletta (Department of Mechanical Engineering, University of Calabria) for supporting this activity with accurate section-cut measurements.

NOTE

Open Access funding provided by ‘Università della Calabria’ within the CRUI CARE Agreement