In-Line Vector Modulator Integration in Dielectric-Filled Waveguide

This article proposes a scalable substrate-integrated waveguide (SIW) module accommodating an in-line vector modulator monolithic millimeter integrated circuit (MMIC). The SIW module is realized with low-temperature co-fired ceramic (LTCC) technology, and it can be inserted in a dielectric-filled waveguide (DFWG). The module combines <inline-formula> <tex-math notation="LaTeX">$\lambda _{g}/4$ </tex-math></inline-formula>-transformer-based <inline-formula> <tex-math notation="LaTeX">$E$ </tex-math></inline-formula> plane tapering and SIWs on LTCC with the wire-bonded vector modulator. The proposed active LTCC module and two passive test structures (i.e., a constant-height-SIW module and a SIW module with <inline-formula> <tex-math notation="LaTeX">$E$ </tex-math></inline-formula> plane taperings) are manufactured and tested as in-line modules in a DFWG. The passive test structures with the waveguide-to-DFWG and DFWG-to-SIW transitions measure 3.1 and 4.6 dB of insertion loss on average, respectively, at the 71–81 GHz frequency range. The active LTCC module measurements demonstrate a DFWG with phase and amplitude tuning capability and gain up to 17.6 dB within the same frequency range. A four-channel mock-up module with <inline-formula> <tex-math notation="LaTeX">$\lambda _{0}/2$ </tex-math></inline-formula> channel spacing is designed and manufactured to demonstrate the scalability of the design.


I. INTRODUCTION
R APID development in mobile communications demand ever faster networks. Millimeter wave (mm-wave) frequencies are therefore increasingly used in communication network backhauls. For example, E-band (71-76 and 81-86 GHz) provides wide band and low latency for future mobile networks [1], [2]. In addition, a frequency band inside E-band (76-81 GHz) is currently utilized in radar applications [3].
Waveguides and horn antennas are used to create lowloss fixed-beam antenna arrays [4] and printed circuit boards (PCBs) with phase shifters can be implemented in the waveguide environment to create beam steerable antenna arrays [5]. Various in-line transitions from air-filled waveguide (AFWG) to substrate-integrated waveguide (SIW) [6], to microstrip line [7], and to monolithic millimeter integrated circuit (MMIC) have been presented [8], [9] to integrate radio frequency integrated circuits (RFICs) into waveguides. These could also be used to integrate phase shifting capability to the waveguide arrays. When a steerable antenna array is considered, the mentioned proposals share a common limiting factor, namely the AFWG. The interelement spacing in a steerable antenna array should be approximately half a wavelength in free space (λ 0 /2) at the highest usable frequency to avoid grating lobes and the subsequent deterioration of the gain. However, the inner width of an AFWG is λ 0 /2 at the cut-off frequency whereas the lowest usable frequency is approximately 25% above the cut-off frequency. When these two criteria are considered, any amount of bandwidth will increase the interelement spacing far beyond the desired range. This problem increases further because a finite wall thickness between the waveguide elements is needed in any real application. The spacing problem can be mitigated by filling the waveguides with low-loss dielectric material to reduce the absolute width and maintain the same electrical width and cut-off frequency of the waveguide.
Dielectric rod antenna arrays have been previously proposed for mm-wave frequencies [10], [11]. These arrays are based on dielectric-filled waveguide (DFWG) and dielectric rod waveguide (DRW) feed networks and can be constructed in a compact way to suppress the grating lobes. Thus far, these utilize fixed feed and the beam steering is hence not possible.
An in-line phase shifter utilizing liquid crystal (LC) technology has been proposed for the dielectric waveguide antenna array [12], making it possible to achieve λ 0 /2 element spacing. The LC phase shifter provides continuous phase shifting without complex transitions. However, few drawbacks to this approach exist. The maximum applicable phase shift is a function of the phase-shifter length. Full 360 • -phase-shifting range makes phase shifters long and lossy. The LC technology also requires high-voltage electrodes outside the dielectric waveguide and thus introduces further size and voltage isolation requirements. Furthermore, LC technology does not provide signal amplification.
In this work, we propose an in-line IC-integration module (seen in Fig. 1 uses endfire connections to interface a SIW with the existing dielectric-waveguide design [10]. Inside the IC-integration module, the full-height SIW is narrowed down to a thin one using E plane tapering [13], and a SIW to grounded co-planarwaveguide (GCPW) transition is used to accommodate wire bonding for the IC integration. The prototype is designed to be approximately λ 0 /2 wide in free space including realistic waveguide-wall thickness. This enables the grating-lobe-free beam steering if integrated into an antenna array.
Due to the short wavelength at E-band, the manufacturing technology for the SIW and MMIC-integration must be capable of very small details. The Ferro A6M-E-based lowtemperature co-fired ceramic (LTCC) process at VTT produces mm-wave components with high accuracy [14]. The LTCC technology provides multilayer boards and buried microvias for easy manufacturing of complex structures, making it a suitable platform for the proposed IC-integration module.
A single-channel module is designed, manufactured, and measured. The design utilizes previous work based on former research by Kähkönen et al. [10], Haarla et al. [13], Karki et al. [15], and Kantanen et al. [16]. This article presents the measurement results for the single-channel model. A mockup version of the four-channel LTCC module is designed and manufactured to demonstrate that the multichannel in-line integration design can be implemented within the spacing limitations (2.06 mm ≈ λ 0 /2 at 73 GHz).
The article is organized as follows: the design of prototypes is described in Section II, the measurement results are presented in Section III, a discussion of the results and array implementation follows in Section IV, and conclusions are drawn in Section V.

II. PROPOSED STRUCTURES FOR IN-LINE INTEGRATION
OF VECTOR MODULATOR The proposed IC-integration module consists of several interfaces or transitions. First, a phase-shifter module that can be inserted in-line with a DFWG requires a waveguidelike interface. In this design, a SIW of similar size with DFWG [10] is selected to form an endfire type interface. This is a well-suited approach due to the similar dielectric constants of the SIW built in LTCC and the dielectric inside the DFWG. The LTCC stack-up also serves as a platform for other transitions for the MMIC integration [see Fig. 2(b)]. An internal stepped E plane tapering described in [13] is introduced to reduce the SIW height. This allows for an efficient transmission line transition from SIW to a GCPW of the same height. Finally, the vector-modulator MMIC is wire-bonded to the 50-GCPW. The same transition chain in reverse is added between the MMIC output and the DFWG.
The LTCC process allows for the stacking of multiple equally thick layers and blind vias between each layer. The prototypes are constructed from eight 92-µm thick layers, yielding an overall thickness of 0.736 mm. 5-µm thick gold layers and gold vias of 100 µm in diameter are used to construct the conducting surfaces inside the LTCC substrate. The dielectric constant of the LTCC (ϵ r ) is 5.7 and the loss tangent (tan δ) is 0.002. The conductivity of gold (σ ) is 7 × 10 6 S/m. The width of the SIW is set to 1.48 mm. The via spacing in the SIW wall is set by the LTCC design specification [14] as 250 µm (2.5× via diameter). The SIW is tapered in E plane such that the MMIC off-center location is optimal for wire bonding to the GCPW. The E plane transition is implemented with λ g /4 transformers where λ g is a wavelength inside the SIW. The E plane transition is reported in greater detail in [13].
Three separate prototypes are designed, manufactured, and tested. The first two are passive LTCC test substrates that are manufactured to demonstrate test setup operation. The third structure has an integrated vector modulator for the active module demonstration. The first passive structure (⟨I⟩) has a straight LTCC SIW to provide a test case for the  SIW-to-DFWG interface (see Fig. 3(a) for the EM model and Fig. 4(a) for the manufactured prototype). The second passive structure (⟨II⟩) has two E plane taperings in back-to-back configuration. The EM model for ⟨II⟩ is shown in Fig. 3(b) and the manufactured prototype in Fig. 4(b). This prototype has a similar layout as the active prototype but the SIW-to-GCPW transitions and MMIC have been replaced by a single-layer SIW.
The active prototype, the LTCC module with the vector modulator (⟨III⟩), consists of two E plane taperings and SIWto-GCPW transitions in a back-to-back configuration, along with a wire-bonded vector modulator in the middle of the structure. The control signals and power supply are routed below the RFIC for compact integration. The narrow design enables placement of multiple vector modulator MMICs sideby-side, (see Fig. 3(c) for the EM model and Fig. 4(c) for the manufactured active prototype). The larger cavity is to provide protection for the bonding wires from mechanical damage and the internal cavity is to fix the wire-bond ends to the same level.
The MMIC vector modulator operates over a 60-100 GHz range covering a 360 • phase and over 10-dB gain control ranges. A three-stage low-noise amplifier in a cascode configuration is used to drive the active vector modulator core which is based on the Cartesian approach. A two-stage differential poly-phase filter is used to generate the required IQ RF signaling for the variable gain amplifiers. The variable gain amplifiers are realized with two differential Gilbert cells and controlled by analog voltages. The active vector modulator core is followed by a two-stage cascode amplifier. The chip is fabricated using 0.13 µm SiGe BiCMOS technology. The same core circuit with digital control was presented in [16].
The feeding DFWGs are adopted from the previously published design [10]. The DFWG consists of Premix Preperm dielectric enclosed in brass. The dielectric material has the dielectric constant ϵ r = 4.4 and loss tangent tan δ = 0.0029. Brass is modeled with the conductivity σ = 1.59 × 10 7 S/m and surface roughness rms = 0.4 µm. The feeding DFWGs have a transition from a WR-12-size AFWG to a DFWG (see Fig. 3).

III. LTCC VECTOR MODULATOR MODULE MEASUREMENTS
The measurements of the manufactured prototypes are done with a vector network analyzer (VNA) and WR-12 waveguide frequency-extension modules. The presented measurement results accompany the simulation results of equivalent structures obtained with CST Microwave Studio.
The fabricated LTCC modules are placed between DFWGs in the measurement assembly (see Figs. 1 and 3). The frequency range of the AFWG-to-DFWG adapter was optimized for the lower E-band and the emphasis of the measurement result analysis is thus limited to the lower E-band (71-76 GHz) and 76-81 GHz. Due to the fixed, high output power level of the frequency extender, a 50-dB attenuator was used in series with the active LTCC module because of the limited acceptable input power of the vector modulator. Only S 21 measurement is thus available for the active LTCC module. All measurement results include the AFWG-to-DFWG transition and the losses from the dielectric waveguides. Due to symmetry of the LTCC module |S 11 | and |S 22 | in passive protypes are similar and thus only |S 11 | is shown. Fig. 5(a) displays the simulated and measured |S 21 | and |S 11 | of the LTCC module ⟨I⟩ with the constant-height SIW. The simulated |S 21 | is approximately −2.4 dB. The measured response is approximately −3.1 dB with an increased ripple over the frequency range. The simulation shows the nulls that are caused by the AFWG-to-DFWG transition are at approximately 69 and 83 GHz. Both nulls have shifted lower in the measurements. |S 11 | shows −9 dB matching over the whole frequency band. Fig. 5(b) presents the simulated and measured |S 21 | and |S 11 | of the LTCC module ⟨II⟩ with the two E plane taperings of the SIW. Based on the simulations, the |S 21 | is on average −3.7 dB between 71 and 81 GHz. The measurements show the average |S 21 | of −4.6 dB. There is also a small frequency shift between the simulation and the measurement. The measured response was 1.4 GHz higher than the simulated one. Similar frequency shift in LTCC structures have been observed previously [13], [15]. Upward frequency shift could imply a lower effective dielectric constant compared to the design value. |S 11 | shows −5.7 dB matching between 71 and 76 GHz and −9 dB matching between 76 and 81 GHz with a deterioration likely due to a standing wave forming between the two E plane tapers. This section is replaced by the vector modulator in the active LTCC module. The simulated and measured transmission coefficients have maximum values of 18.9 and 17.6 dB, respectively. The measured S 21 shows an approximate 10-dB variation over the 71-81 GHz frequency band. There is an approximate 2 GHz frequency shift upward compared to the simulation results. The shift is similar to the one seen in passive structures. However, the level of the ripple is higher than expected. The measurements also indicate a downward trend in S 21 that was not anticipated based on the passive structure measurements. Section IV-A discusses the reasons for the ripple and downward slope in S 21 .
The group delay and phase stability of the prototype are analyzed using the measured data. In the selected representative data, the vector modulator gain is varied 15 dB while the phase is approximately constant. The group delay of the prototype is shown in Fig. 6(a). The vector modulator gain is varied and

A. LTCC Module Wire Bonding and Transition Analysis
The analysis of the S 21 measurement results divides into two sections: the wire bonding and the material design parameters of the LTCC. The proposed prototype was simulated using optimal wire-bonding locations with 100-µm arc height and the realized wire-bonding locations in the manufactured prototype with 150-µm arc height. In the optimal case, the signal bonding wires are attached in the conductor centerline 60 µm from the end. The ground wires are attached 60 µm from the side and 62 µm from the end of the ground-plane corner closest to the MMIC. In the manufactured case, the signal bonding wire location is moved 30 µm side from the center line and 220 µm toward the SIW. The ground-wire locations are moved 150 and 110 µm, respectively. The wire bonding was done using 25 µm gold wires.
In the manufactured prototype, the bonding wires are longer than expected, which leads to a change in performance due to additional inductance [see Fig. 7(b)]. As the bonding wire attachment point moves toward the SIW, the end of the center conductor starts to work as an additional stub, further deteriorating matching. Based on the wire-bonding simulations, the increase in the wire bond length leads to a reduction in the operation frequency. With longer bonding wires, the ideal operational frequency range is below 71 GHz, and the performance at the designed frequency range gradually decreases with the increase in frequency. Fig. 7(a) displays the simulation results for both bond-wire options. This difference in bonding wires caused a drop of 6.2-8.6 dB over the 71-81 GHz frequency band. The shifted operational frequency of the bonding wires also caused an increase in the variation over the inspected frequency range due to the downward slope.
The AFWG-to-DFWG transition introduces unwanted reflections to the system limits the usable bandwidth of the test setup, and creates ripples in the frequency response. These limitations are not related to the proposed prototype, but rather to the test setup necessary for the prototype measurement. The LTCC module ⟨II⟩ shown in Fig. 3(b) is used to study the effect. The CST-model is simulated with and without the AFWG-to-DFWG transition [see Fig. 8(a)]. Fig. 8(b) demonstrates the effect of the transition. The transmission coefficient without the AFWG-to-DFWG transition is broadband and extends up to 86 GHz without a significant ripple. When the AFWG-to-DFWG transition is introduced, the transmission coefficient is a much narrower band with an increased ripple over the frequency band.
The effect of the uncertainty in the LTCC material design parameters is evident in Fig. 5(b) as an approximately 1 GHz frequency shift in the operational frequency band. The operation frequencies of the E plane transition and the SIW-to-GCPW adapter are based on the λ g /4-transitions that are affected by the wavelength inside the material. The designs manufactured in the same production batch show a similar upward frequency shift in the operation [13], [15].
The combined effect of these factors results in increased ripple in the measurement band. The operational frequency range is shifted to lower frequencies due to the long bonding wires whereas the difference between the design and realized material parameters shifts the operational frequency of the E plane tapering to higher frequencies.

B. Array Implementation
The proposed in-line vector modulator integration is designed for easy scalability. A mock-up four-channel prototype is manufactured to demonstrate the scalability (see Fig. 9). It has four identical channels in parallel with 2.06 mm spacing according to the dielectric antenna array proposed previously in [10]. The mock-up retains the same length of 10.59 mm as the single-channel version. Possible future improvements with the flip-chip attachment of the vector modulator would eliminate bonding wire landing pads and allow tighter integration.
An expansion to 4 × 4 configuration could be a future research topic. The design is not restricted along the normal of the LTCC as the contacts are located on the sides of the LTCC module and the LTCC module height is smaller than the 2.06 mm antenna spacing, and the vector modulator RFIC is embedded below the top surface plane of the LTCC. A 1.324 mm-thick spacers could be used to stack four LTCC modules to produce a 4 × 4 array in case the corresponding antenna array and feeding network can be realized. Table I lists similar proposed designs for the phase shifting in E-band and W -band. As stated previously, these have size limitations that lead to degradation of the antenna array performance. The solution in [12] utilizes an in-line LC structure to introduce continuous phase shift. This approach requires electrodes between each channel to control the phase shift, thus increasing the width of each channel. Maximum phase shift is also a function of length, thus increasing the length of the phase-shifter part. At last, the structure is lossy and does not provide signal gain. The transition in [9] uses a low-loss connection between a waveguide and RFIC. The article uses only an amplifier, but we assume that any vector-modulator IC integration is possible. Such a design, however, utilizes an AFWG, which imposes size constraints for the width of each channel and degrades antenna performance if implemented in an antenna array. In the design in [17], the phase shifters are not in-line with the waveguide. Such a configuration imposes its own restrictions to the antenna spacing of an array as shown in [5] (0.8λ 0 at 75 GHz). The design proposed in this article fulfills the requirements for near λ 0 /2 design width while simultaneously providing amplitude and phase control for future beam-steerable antenna arrays.

D. LTCC Process Variations
There are several process steps which affect the accuracy of the LTCC circuits. Conductors are typically screen-printed where the resolution of conductors is defined by the quality of the screen, paste properties, design layout, and process parameters. The tolerance of linewidth is typically ±10%. It is possible to affect the tolerance in critical areas, such as microstrips or coplanar waveguides, by optimizing the abovementioned parameters.
When the individual sheets are stacked together, misalignment between the layers is over ±10 µm. This is not considered a significant risk for the performance, taking into account that the typical via hole dimensions are in the range of 100 µm. One of the most critical parameters in the processing is shrinkage which occurs during co-firing. Typical variation of shrinkage in the horizontal direction is ±0.15%-0.30% which has a direct effect on the dimensions of the fired substrate. Also, the thickness of substrate can vary. Typically the variation of vertical shrinkage is ±2.0%. There are, however, methods to control the variations if needed.

V. CONCLUSION
This article proposed an active LTCC module with a vector-modulator IC for the in-line integration with a DFWG. The in-line modules allow λ 0 /2 placement of the elements in antenna arrays with DFWG feed networks. The active and passive designs are measured in an application-like setup using two DFWGs. The passive, constant-height-SIW prototype shows a 3.1-dB measured insertion loss on average, whereas the insertion loss of the prototype with E plane taperings measures a 4.6-dB on average over the 71-81 GHz frequency range. The measurements of the active prototype demonstrate the capability for phase and amplitude tuning with a gain of up to 17.6 dB for the same frequency band. The reasons affecting the insertion loss were identified and solutions were presented to improve the performance. The wire-bonding could be improved with shorter bonding wires, or an IC flip-chip integration could be used for improved matching and to further narrow and shorten the design. Additionally, a more robust connection between the DFWG and SIW in LTCC could be implemented. The narrow width and scalability of the proposed design are demonstrated with a four-channel mock-up model. The in-line modules could be implemented in a fixed DFWG array to convert it to a phased array with beam-forming capability. He has been with Aalto University, Espoo, formerly TKK, serving at the Radio Laboratory from 1995 to 2007, the Department of Radio Science and Engineering from 2008 to 2016, and currently with the Department of Electronics and Nanoengineering. He also works as the Staff Scientist. He has been a Researcher and the Project Manager of various millimeter wave technology related projects. His current research interests are the antennas and antenna measurement techniques for millimeter and submillimeter waves, and millimeter wave imaging.
Markku Lahti received the M.Sc. and D.Sc. degrees in electrical engineering from the University of Oulu, Oulu, Finland, in 1993 and 2008, respectively.
Since 2001, he has been with the VTT Technical Research Centre of Finland Ltd., Espoo, Finland, where he is currently the Senior Scientist. He has authored or coauthored more than 20 peer-reviewed publications and more than 80 conference papers. He has been involved in several EU and ESA projects. His current research interests include the manufacturing of low-temperature co-fired ceramic substrates, module-level integration, and packaging of components with ceramic and polymer platforms.  Mr. Holmberg is a member of the Audio Engineering Society, the Acoustical Society of Finland, and the Telecommunication and Electronics Association. He has authored or coauthored more than 90 journal articles and 100 conference papers. He is an inventor in 16 granted patents. His current research interests include antennas for mobile devices and networks, antenna clusters and coupled arrays, RF-powered devices, and antenna measurement techniques.
Dr. Viikari is a Regional Delegate of EurAAP. He was a recipient of the Young Researcher Award of the Year 2014, presented by the Finnish Foundation for Technology Promotion, and the IEEE Sensors Council 2010 Early Career Gold Award.