Bandpass filter with multiple selective absorptive stopbands for 28 GHz transmitters

Abstract

This manuscript presents a novel design for an absorptive bandpass filter for mm-wave applications, specifically the commercial FR2 spectrum. Three bands have been selected to be properly input matched with only one of them being the passband, where the insertion loss is minimized. The proposed approach relies on a multiplexer topology implemented through microstrip lines and on thin-film manufacturing process on alumina to shrink the footprint. Cascades of half-wavelength C-shape open-ended resonators are exploited to create the matched bands and define the filter’s selectivity. The selected passband spans from 26.5 to 28.5 GHz, with a measured maximum insertion loss of 3.05 dB for a −3 dB fractional bandwidth of 7.3%. Two absorptive bands are realized to match signals at 24 and 32.25 GHz. The alumina die footprint is 5500 × 3440 µm², compatible with immediate integration within a mm-wave lineup.

Introduction

The next-generation high-density fixed services will heavily rely on the adoption of mm-waves. Such services comprise the beyond 5G (B5G) radio access network and fixed wireless services where the exploitation of larger bandwidths in the 28 and 39 GHz bands will be the key-enabler for multi-Gbit/s connections [1]. Among the challenges, the signal fading between the base station and the mobile or fixed equipment is one of the major bottlenecks: obstacles like windows, walls, or trees are mostly negligible in the sub-6 GHz spectrum, but across the FR2 band allocated for 5G and B5G (mm-wave chunks beyond 24 GHz) they become causes of signal degradation and connection failures.

Transmitted power and system linearity must be preserved to overcome the challenges and guarantee efficient deployments of mm-wave infrastructure. The main Power Amplifier (PA) biasing has to be finely tuned to maximize its output capabilities but, in a heterodyne architecture (Fig. 1), it is possible that second or third order intermodulation products and spurious signals are generated by the upconverter and amplificated by the following stages. Furthermore, the PA contributes also with its own distortion, which can be countered by extra back-off but penalizing the transmitted power.

Figure 1. Heterodyne transmitter architecture. Spurious can be reinjected into the driver (DR) or back to the upconverter (cross talk 1). If not filtered, intermodulation products will be amplified by the PA (cross talk 2) with the useful signal.

Properly designed filters placed along the transmission lineup are generally enough to prevent the feedthrough of spurious products, traditionally implemented with reflective stopbands. However, at mm-wave reflections given by the out-of-band behavior of the filter generate cross talk between the blocks, even introducing strong distortions or oscillations. To avoid such occurrences, the filter needs absorptive capabilities to suppress and remove the intermodulation products.

Absorptive filters are widely studied within the literature and mainly implemented in the sub-6 GHz spectrum, where a mix of distributed and lumped elements still allows the right balance between resonators Q-factor and compactness. References [2–6] show several examples of these designs. Ladder networks of inductors and capacitors properly designed are generally coupled with a resistive load to provide almost arbitrarily matched band. In the same frequency range, different microwave resonators can be used [7, 8] but loads can be still implemented through Surface Mounted Device (SMD) like resistors. Absorptive filters have been also proved in the lower portion of Q-band [9] implemented through GaAs process: the chosen topology and the target wideband matching however led to a meandered structure with wide unused substrate areas. These topologies are based on bandstop or high-pass filters, then the degrees of freedom to suppress and absorb selectively more than one band is still missing.

The novelty introduced with this work relies on the implementation of a compact absorptive bandpass filter (BPF) centered in the upper portion of Ka band or FR2 spectrum (26.5–28.5 GHz) which exploits the multiplexer approach to maximize the design flexibility. Since the operational frequency prevents the use of lumped components, the matched loads are implemented through distributed resistive elements to guarantee high level of absorption over a wide frequency range. The combination with the selected technology, thin-film on ceramic substrate, leads to a very compact design which is mandatory in modern mm-wave lineup sections.

Since frequencies and bandwidth of the undesired signals are reasonably predictable, absorptive stopbands and matching levels are selected accordingly and the target specifications are given in Table 1. The designed component is intended to be placed between active stages then tight input and output return loss specifications are not requested. In the worst case, a return loss of 10 dB is acceptable since is comparable with matchings given by the involved gain blocks.

Table 1. Filter specifications based on identified use-case

The manuscript first introduces the novel approach to design the multi-stopband absorptive BPF; then the section titled “Filter’s building block implementation” describes the physical dimensioning of the involved resonators, filters, and blocks of the design. Then full-wave simulation results and the experimental validation are shown while the last section concludes the paper highlighting the main outcomes.

Proposed filter topology

The microstrip half-wavelength C-shape open-ended resonator is the main building block of the filter. Thanks to the available planar alumina technology, each resonant node of the filter is compact and shows a decent unloaded quality factor, even beyond 150 from 24 up to 40 GHz. In-line BPFs, made by a single resonator or a linear cascade of them, are branched together in parallel to realize passbands. The topology resembles the “terminated multiplexer,” widely adopted in waveguide assemblies and also proved through planar transmission lines [10, 11], but only partially adopted in reflection-less or absorptive filters. In fact, the balance between the operational frequency, the adopted technology, and the number of branches directly connected to the source makes the proposed structure just similar to the distributed microstrip implementation shown in paper [12].

Figure 2 shows the topology of the proposed absorptive filter. The source S is connected to a binary split junction, marked as a black dot, whose first branch is directly connected to the cascade of three resonators R_PB (resonators defining the passband). It selects the passband of the filter and mainly defines the frequency behavior in terms of in-band insertion loss, return loss, and out of band rejection. This first cascade terminates at the filter output (load, L). The second branch is connected to another binary split junction to feed two paths: one is the cascade of two resonators for the lower absorptive band R_LB and the other one is a single resonator R_HB for the higher absorptive band. These paths are terminated with a properly designed 50 Ω matched load T.

Figure 2. Topology exploited for the proposed design filter. S, L, and T are source, load, and termination respectively. R_PB, R_LB, and R_HB indicate the nodes in the pass, lower, and upper bands.

The exploitation of three branches is needed to address the use-case but also highlights the flexibility of the proposed structure. Additional absorptive stopbands can be introduced by adding more terminated cascades. The effect on the passband insertion loss remains small and given by the sum of the filter Insetion Loss (IL), the losses introduced by transmission line routing and by n-times the loss introduced by each splitter along the main path. An additional insertion loss is potentially introduced by the limited selectivity of the absorptive branches, although this can be minimized by design for instance by increasing the number of resonators of those branches.

Furthermore, the multiplexer-like approach is almost independent of the filter topology chosen along each branch, therefore complex topologies beyond the standard in-line or hybrid technologies can also be adopted, with the only requirements of being compatible with the technological choices. In the proposed design, the usage of single or multiple in-line resonators depends on the desired bandwidth and rejections of the filters.

However, different requirements can be matched with filter topologies like low-pass or high-pass filters. Solutions on alumina have been proved in Ku and Ka band respectively [13] and perfectly suit an integration in a terminated multiplexer filter. Other approaches can be followed with the aim of reducing the number of branches in the overall structure. In fact, the thin-film process on pure alumina tailors the extracted-pole filter topology [14] implemented through microstrip resonators [15]. This accurate synthesis methodology allows the design of high-order dual-band filter, which can be used in the proposed structure to match the lower and higher bands with a single terminated branch. However, the latter design philosophy was discarded here due to its higher design complexity when several stopbands are required and due to its lack of modularity.

Filter’s building block implementation

The manufacturing technology relies on the thin-film process available within SIAE MICROELETTRONICA facilities. Pure alumina is the dielectric material while gold photoetched traces realize the conductive microstrips. The alumina substrate is 5 mils (127 µm) thick from Coorstek Inc. with a nominal dielectric constant (ε _R) of 9.9 and a loss tangent of 1 × 10⁻⁴. The metal stack-up is firstly made by three sputtered layers, each only a few tens of nanometers thick: tantalum (to realize resistors), titanium, and palladium (to permit the right adhesion between the ceramic and the upper layer). The additional final metallization is a 3 µm galvanic-grown pure gold, etched with tight ±5 µm tolerance on the deposition plane. Via holes, only adopted into the Ground-Signal-Ground (GSG) launcher for this design, can be laser drilled and finally plated with gold.

The full-wave simulation carried out to design the filter only takes into account the last gold layer, on top of the metal stack-up, and neglect the sputtered nm-thick layers. The introduction into the model of the first layers (extremely thin with respect to 50 Ω lines, computed to be 120 µm wide, and the guided wavelength at 27.5 GHz, which is 4289 µm) leads to higher computational time without an advantage in terms of performance estimation.

Resonators and filters

Each branch of the structure embeds a single or a cascade of resonators to achieve the requested frequency behavior. The preliminary evaluation of the filter performance has been carried out by the exploitation an ideal Chebyshev synthesis and by take advantage of its circuital equivalence. Then, each filter has been implemented through impedance inverters, RLC parallel resonators, and branched together with phase shifters and circuital nodes. The input and output reference impedance has been set to 50 Ω. The reference circuit is reported in Fig. 3. Resistors have been added into the model to take into account a realistic loaded Q-factor. Investigations regarding the losses are presented in the following paragraphs and shown in Fig. 4.

Figure 3. Ideal circuital representation of the proposed absorptive filter based on RLC resonators, impedance inverters J, and phase shifters T. Values are the following: Z ₀ = 50, R ₁ = 9.7826 Z₀, L ₁ = 0.000564 Z₀, C ₁ = 58.9462/Z₀, R ₂ = 5.39419 Z₀, L ₂ = 0.00035622 Z₀, C ₂ = 122.4268/Z₀, R ₃ = 3.0487 Z₀, L ₃ = 0.00014793 Z₀, C ₃ = 159.1549/Z₀, J ₁ = 0.94525/Z₀, J ₂ = 0.87987/Z₀, J ₃ = 1.0369/Z₀, J ₄ = 1.2868/Z₀, J ₅ = 1/Z₀, T ₁ = 12°, T ₂ = −27°, T ₃ = 11°, T ₄ = 166°. The negative T₂ phase delay is obtained as equivalent reactive effect of the second node microstrip junction.

Figure 4. (a) 3D full-wave model of the C-shape resonator. (b) Top: fundamental and first higher mode frequencies versus the overall resonator length. Bottom: quality factor for the fundamental mode across the 20–40 GHz span.

The frequency response of the circuital implementation of the filter is reported in Section IV where its superposition with the full-wave simulations and the experimental results are reported in Fig. 9.

Once the ideal circuital assessment has been made, the filter has been implemented through realistic half-wavelength microstrip resonators. The choice to fold each resonator as a C-shape follows the paradigm of compactness because the Q-factor penalty introduced by the bends is negligible. The resonators have been simulated through the eigenmode solver of Ansys Electronics Desktop to exactly correlate the resonant frequency, the length of the resonator with the designed chamfered bends, and the unloaded Q-factor (Fig. 4).

It is important to highlight that the main loss mechanism in the alumina-based circuit is given by the finite conductivity of the top layer of gold. The bottom graph of Fig. 4(b) shows a significant drop of quality factor for frequencies below 24 GHz whereas from 26 to 40 GHz the behavior is much flatter. Since the used alumina substrate can be considered very pure, isotropic, with a low dissipation factor and with a negligible surface roughness even in the mm-wave spectrum, other losses can be neglected.

Three resonators have been in-line coupled to realize the main passband of the filter and optimized to reach 7.3% of fractional bandwidth. This configuration also guarantees more than 20 dB of rejection across the selected absorptive bands. Next, two resonators have been coupled together to realize a passband below the central frequency of the filter: from 23.5 to 24.5 GHz spurs are filtered and absorbed once the cascade is terminated on a matched load. One single resonator is then designed to cover the highest band, a 500 MHz bandwidth around 32.25 GHz. The layouts of these three branches are shown in Fig. 5.

Figure 5. Resonator layouts for all the three designed branches. Dimensions in µm: g ₁ = g ₃ = 25, g ₂ = 100, g ₄ = 110, g ₅ = 30, W ₅₀ = 120, W _R = 80, W _IN = 60, L ₁ = 2190, L ₂ = 2200, L ₃ = 2380, L ₄ = 1800. The lengths L _n indicate the overall resonator lengths measured along their centerline.

To verify robustness to manufacturing inaccuracies, the circuit-based analysis with the microstrip models of each stand-alone branch has been exploited to evaluate a large number of geometrical variations. Dimensions like length of traces, coupling gaps between elements and width of microstrips have been randomly varied following a Gaussian function with standard deviation σ equal to 1.667 µm. This value leads to 3σ = 5 µm which is the maximum allowed tolerance considered in the process.

Figure 6 shows the superposition of a hundred of traces per band pass filter involved in the proposed design. Red lines show the specification demanded in terms of input matching. Both lower band and main passband, centered respectively at 24 and 27 GHz, do not significantly suffer from manufacturing inaccuracies, Fig. 6(a) and (b). The higher absorptive band, which covers 500 MHz between 32 and 32.5 GHz, partially suffers because of tolerances and few iterations of the analysis overstep the −10 dB limit-line, as reported in Fig. 6(c).

Figure 6. Superposition of a hundred simulations of |S₁₁| from of each branch of the proposed filter. (a) is related to the lower absorbed band, (b) the main passband, and (c) shows the |S₁₁| given by the single resonator of the upper matched band.

Matched load

The absorptive function of the filter is created by terminating the corresponding branches by a matched load. In the low-frequency domain, lumped resistors can be directly soldered onto the Printed Circuit Board (PCB) to realize the resistive behavior. In the microwave region, high-end resistors can be also wire-bonded to accomplish the task. In the mm-wave spectrum, thin-film technology allows the realization of distributed resistors, proven up to 80 GHz [16].

The sputtered layer of tantalum is everywhere present underneath the metal stack-up but without further process it is still a conductive layer. At the end of the manufacturing process, another step of photolithography etches the metal layers above the designed areas of tantalum. Its exposition to the environment and to the curing process results in the formation tantalum nitride (TaN): the control of its thickness to 500 Å and the curing settings (temperature and time) can provide a 50 Ω square resistance. Its patterning has the same rules of the gold metallization and any 2D shapes can be etched with micrometric accuracy.

The matched load is then simulated and optimized through the full-wave solver where the resistor shape is set as layered impedance. The final is a trapezoidal patch asymmetrically placed with respect to the input 50 Ω feeding microstrip. The simulated return loss value is higher than 20 dB across the intended operational frequency.

Microstrip junction and phase adjustments

Multiple paths are finally branched together with microstrip tee-junctions. The design of each bifurcation is straightforward since no impedance transformer is adopted. The natural matching of each port is enough to branch independent filtering structures. The chosen splitter features compactness, low loss, and wide operational band, up to 40 GHz. By cascading several junctions, it is also possible to join the multiplexer branches, as done in the proposed structure between the source port and the filters for the absorbed stopbands.

The distance between the last junction and the filtering structure, as well as the distance in between two microstrip tees, is an important design parameter. The passband must not be affected, hence the loading effect from the absorptive branches must be close to an open circuit at the center of the passband. Moreover, in order for the stopbands to be effectively absorbing, the loading from the other branches must be also close to an open circuit at the center of the stopbands.

The length of the arms (L₁–L₄ in Fig. 2) should finally be minimized but, for implementation purposes and to reduce the cross talk between the branches, a physical length is actually needed. The delay optimization can be made at circuit level with the following hypotheses:

• • exploiting only straight sections to avoid reactive effects of bends;

• • the physical length is maintained short enough to avoid any resonant effects across the filter operational band.

Optimization has been carried out by using the two-port simulated scattering parameters of the three branches: the simulated 50 Ω load has been used as termination of the absorptive branches whereas transmission line sections and the simulated T junction have been used as interconnection elements. This setup allows to carry out a circuit optimization, much faster than electromagnetic simulations. The feasibility of such approach is allowed by assuming a low coupling between the various branches, achieved by inserting sufficient physical spacing.

The results of transmission line length optimizations lead to values which are then implemented as microstrip sections in the overall filter layout. Again, for routing purposes, chamfered bends will be added and a fine tuning of the model is needed to guarantee better performance. Referring to Fig. 2 symbols: L ₁ = 1244 µm, L ₂ = 729 µm, L ₃ = 1029 µm, L ₄ = 1842 µm. Absolute values are mandatory to be reported for clarity, but to give another view, the same values can be expressed as fraction of wavelength (considering an ε _Reff = 6.65): L ₁ = 0.29 λ _eff27.5GHz, L ₂ = 0.17 λ _eff27.5GHz, L ₃ = 0.21 λ _eff24GHz, L ₄ = 0.51 λ _eff32.25GHz.

With such circuital fast optimization of the junction lines, the achieved in-band return loss is better than 10 dB, while the input return loss in the lower and upper absorptive bands are averagely better than 10 dB and 15 dB respectively. Achieving better input matching in the lower absorptive band is difficult due to the proximity of such a band to the main passband of the filter. Further, an upshift of the lower stopband (for instance due to manufacturing inaccuracies) would cause a strong deterioration of filter performance, specifically in the insertion loss at the lower edge of the passband.

Full-wave design and experimental validation

A full-wave simulation is in fact needed since the branches of the filters are very close to each other and undesired coupling may occur. The filter has been designed to be glued onto the PCB and connected through gold wire-bonding. However, in the mm-wave range the radio frequency connection with thin wires leads to an impedance mismatch which becomes larger when the operational frequency rises. To overcome such effect, a high-impedance meandered line has been designed and placed between the 200 µm pitch GSG pads and the filtering structures.

The full-wave design is then slightly optimized by changing the length of the C-shape resonators only to adjust the passbands and the in-band return loss. The simulated performance is shown in Fig. 7, where the absorptive bands have been highlighted in the top graph while the 2 GHz passband is highlighted in the bottom subplot where the |S₂₁| is reported. The behavior of the filter is characterized by two transmission zeros placed below and above the passband respectively. Their generation is given by the couplings between the filter lines and branches. As counterproof, the simulations of stand-alone branches do not show any transmission zero since the in-line filter topology does not permit to have nulls in the |S₂₁|. Thus, those transmission zeros cannot be precisely controlled at circuit level; their position has hence been optimized through full-wave analyses of the whole structure.

Figure 7. Simulated S-parameters of the proposed filter. Red and black lines respectively correspond to the simulations without terminating loads (hence with an open-circuit termination) and with the designed distributed TaN terminating patches.

The designed layout is excited via full-wave ports to mimic the foreseen measurement setup through 50 Ω-matched probes. Figure 7 also displays the comparison between the structure without and with the final matched loads (red and black lines respectively) which highlights the effectiveness of the designed matches in the passband. In both cases, the insertion loss shows the required transmission zeros. However, when the TaN terminating patches are not present, the return loss in the two stopbands is rather poor (5–10 dB), only due to radiation effects in the open-circuit truncation at the end of each branch but no real absorption takes place. Instead, with TaN terminations, the absorption improves drastically.

When the full-wave optimization has been made, the 3D file is converted into a layout to start the manufacturing process. The realized circuit, 5500 µm wide and 3440 µm tall, is shown in Fig. 8. An absorptive pad has been added close to the GSG launcher to perform DC characterization of the tantalum resistive layer. During the curing process, the resistance is kept under control to precisely match 50 Ω/square value. RF measurements are carried out through 200 µm GSG probes connected to a Rohde & Schwarz ZVA40 VNA, calibrated through a custom TRL cal-kit realized on the same alumina substrate. As anticipated, the filter has been designed to be perfectly matched when connected through wire-bonding. The measurement with 50 Ω-matched probes leads in fact to return loss and insertion loss penalties which does not invalidate the performance of the designed structure. The measured S-matrix is shown overlapped with simulations in Fig. 9. For clarity, input and output matchings have been plotted in separate graphs to show the absorptive bands. |S₂₁| is also reported in a dedicated subplot to show the passband insertion loss and the out-of-band rejection.

Figure 8. Top photograph of the alumina die where the filter is manufactured. Gold traces are the yellow ones in the picture, while dark areas are the TaN-based resistors.

The experimental verification shows a slight passband mismatch in the input return loss since |S₁₁| reaches up to −9.4 dB while the |S₂₂| shows a maximum level of −11.2 dB through the 26.5–28.5 GHz band. Instead, the other measured parameters coincide rather precisely with the predicted behavior. The passband IL spans from a minimum of 2.05 dB to a maximum of 3.05 dB at the lower edge of the operational band.

The absorptive bands are well identified through matching dips at the right frequencies: in the lower absorptive band, the matching is better than 10 dB throughout the 24 GHz band whereas at 32.25 GHz the |S₁₁| notch is below −15 dB from 32 to 32.5 GHz. Across these bands, the filter rejection is compliant with the specifications given in Table 1.

It is finally important to underline why the measurement result fits the target better than expected for what concern the 23.5–24.5 GHz matching level. The final design demanded gaps between the input microstrip and the first resonator down to 25 µm, close to the photolithography resolution. Then, due to manufacturing spread the gaps result narrower: the consequences from filter behavior point of view are related to a higher matching level in the absorptive band and a downshift in the resonant frequency, as highlighted in Fig. 9.

Figure 9. Superposition between the circuital synthesis-based behavior of the filter (marked with crosses), the full-wave simulation (solid lines), and the experimental result achieved (dashed lines). Top subplot in (b) highlights the two selected absorptive bands.

Conclusion

This manuscript presents a novel multi-stopband absorptive band pass filter intended for a mm-wave transmitter lineup. The proposed topology involves microstrip resonators in a terminated-multiplexer topology to realize a band pass filter, and two absorptive stopbands, below and above the central operating frequency. The filter’s passband is centered at 27.5 GHz with a maximum verified insertion loss of 3.05 dB at 26.5 GHz. The selected absorptive bands are centered at 24 GHz and 32.25 GHz respectively, where the measurements show a matching better than 10 dB for both the intended spectra. These passive functionalities have been implemented through a thin-film technology suitable for smart and effective integration of the device in mm-wave systems. The architecture could also accommodate more absorptive or reflective stopbands, thus representing a very flexible solution for modern mm-wave frontends.