Vacuum-Seal Waveguide Feedthrough for Extended W-Band 67–116 GHz

This article describes the design and performance of a vacuum-seal waveguide feedthrough with ultrawide RF band 67–116 GHz. We describe first the initial drivers behind the chosen design, then we present the results of the numerical simulations and optimization and provide thereafter the results of the RF and vacuum tests of the fabricated devices. The demonstrated RF performance is very close to the one expected from the simulation with an insertion loss less than 0.3 dB and a return loss better than 20 dB. Simultaneously, the feedthrough shows excellent vacuum isolation, Helium gas leak rate of < 2x10−8 mbar·L/s is demonstrated, which allows using such a device in various space and ground applications.


I. INTRODUCTION
Different space and ground application at microwave and THz frequencies, involve temperature-or climate factors' sensitive equipment that should be placed into an isolated compartment or vacuum chamber, thus requiring interfacing via feedthroughs. In particular, modern communication technologies and instrumentation actively move to employ short mm-wave bands, creating the demand for wideband, lowloss, low-reflection vacuum-tight feedthroughs suitable for, e.g., moderate-power or low-signal instrumentation applications. This article describes the design and performance of a vacuum-seal waveguide feedthrough with ultrawide RF band 67-116 GHz referred as an extended W-band. In practice, the WR-10 waveguide interface of this feedthrough is used within an RF band significantly exceeding the specified WR-10 RF bandwidth of 75-110 GHz, reaching a fractional bandwidth of 54% of the central frequency.
Commercially available vacuum windows (or feedthroughs) for WR-10 waveguide have limited bandwidth by design, e.g., RF band 92-96 GHz, [1]. In such designs, a piece of solid dielectric, glass, quartz, etc., is typically brazed inside a length of a Kovar waveguide, which yields a helium leak rate of < 10 -8 mbar·L/s. However, the presence of a dielectric sealing element produces unwanted reflections and signal interferences between the two surfaces, i.e., vacuum-dielectric and dielectric-air separated by the length of the sealing dielectric. The interference between the reflected signals could be used to equalize the RF performance over a broader bandwidth [2]. Another common design uses a mica membrane placed between two waveguide flanges [3]. The sizes of such membrane are substantially larger than the waveguide dimensions thus possibly producing resonances within the operational RF band. Various designs have also been suggested for vacuum window or feedthrough [4]- [7]. However, none of these suggested designs combines a low insertion loss and a fractional bandwidth of 54%.
The vacuum-seal feedthrough presented in this article was designed for and intended to be used in a radio-astronomical receiver, ALMA Band 2 [8] with a required bandwidth of 67-116 GHz. The adopted specifications for this device were return loss better than 20 dB at any frequency within the band together with an insertion loss better than 0.3 dB averaged over the entire RF band. The helium leak rate requirement was specified at the level of < 2.5x10 -6 mbar·L/s after 20 minutes continuous exposure to the He gas and a lifetime of over 20 years.

II. FEEDTHROUGH DESIGN
Considering the difference of the pressure between the vacuum on one side of the waveguide window and atmosphere  on the other side, the WR-10 waveguide sealing is exposed to roughly 28 grams of the atmospheric pressure force at sea level. Potentially, this would open up for the usage of a very thin plastic membrane, e.g., thin Mylar film with a thickness of few tens of micrometers for adequate waveguide sealing. While it likely wouldn't produce unwanted resonances within the RF band, the long-term stability and especially the lifetime of such a solution is questionable. Furthermore, assuming an ultimately wide bandwidth of 67-116 GHz, a solution that could ensure single-mode propagation is preferred.
Foam materials, e.g., Rohacell, have lower dielectric constant and thus do not introduce strong discontinuity in the permittivity of material filling the waveguide. However, closecell materials have the known drawback of potentially collapsing or irreversibly changing their shape after vacuum cycling. On the other hand, open-cell versions of such materials do not provide sufficient sealing and therefore cannot ensure long term compliance to the required leakage rate with 20-minutes He-exposure testing. Besides the cell size is quite comparable with the wavelength, which may lead to scattering or resonance effects and enhanced RF loss.
Assessing possible fabrication challenges, the sealing material should advantageously be introduced in a circular shallow structure (waveguide) to ensure tight press-fit tolerances towards all walls' surfaces that provide vacuum sealing. Furthermore, from an electromagnetic point of view the sealing material should be made of a dense material with low dielectric permittivity and RF loss, e.g., Teflon or high-density Polyethylene (HDPE), since their relatively low relative dielectric permittivity (ε r ∼ 2 …2.5) facilitates the RF matching at the sealing interfaces, insert/air and insert/vacuum. These materials are also known for their low RF dielectric loss with tanδ of the order of 2-3x10 -4 [9].
Summarizing the discussion above, the proposed novel waveguide feedthrough design for the ultra-wideband and low insertion loss operation employs the layout illustrated in Fig. 1. The input and output use a WR-10 rectangular waveguide, the sealing section employs a circular single-mode dielectric-filled waveguide. The matching sections connect the input / output WR-10 with the sealing section. The matching sections should not only effectively couple the rectangular WR-10 waveguides with their circular counterpart but also account for the changed propagation conditions in the dielectric of the sealing insert in the circular waveguide. HDPE was chosen as a material for the sealing insert, based on its better mechanical stability and long-term creep-free behavior under load as compared to Teflon.

III. MODELLING AND MECHANICS
An initial model for full 3D electromagnetic simulations using Ansys HFSS TM was built on fundamental assumptions for the waveguide geometry [10], and is shown in Fig. 1. The WR-10 standard interfaces came from the project requirements. In order to provide the required wideband operation, we opted for a two-step transition between the WR-10 waveguide and the circular waveguide sealing section.
The input and output rectangular waveguides made of standard WR-10 waveguide with the dimensions 1.27x2.54 mm 2 , are followed by the roughly quarter-wavelength-long sections with slightly reduced transversal dimensions chamfered corners to 0.3 mm radius, to facilitate the fabrication by CNC milling machine (callout (1) in Fig. 1). In the HFSS TM model, these chamfered waveguide sections were set initially having standard WR-10 dimensions, which were free parameters for the optimization. The custom circular waveguide is filled with HDPE having a relative dielectric permittivity ε r = 2.37, and tanδ = 0.0003 and has a diameter of 1.86 mm (callout (2) in Fig. 1). The dimensions (diameter) of the circular waveguide should allow propagation of the dominating TE 11 mode above F L = 67 GHz, lower edge of the RF band and preclude propagation of the second TM 01 mode up to F U = 116 GHz, the upper RF bandwidth frequency. Using equations from [10], we write the following conditions for the circular waveguide diameter, D: and The diameter of the circular waveguide for the sealing section, Fig. 1 callout (2), was chosen to be 1.86 mm fitting between the two limits given by (1) and (2). The HDPE insert become very flexible and difficult to handle during the pressing inside the metal parts of the feedthrough. For that purpose, we have chosen the diameter towards the upper limit defined by the (2) but with about 0.1 mm margin. A special tool was designed to facilitate pressing he HDPE insert that precludes bending.
The additional second step in the transition from the WR-10 to the circular waveguide, relies on edge modification of the circular waveguide, with small symmetric recesses at the sides, Fig. 1 callout (3b), and a coaxial recess in the HDPE insert, Fig. 1 callout (3a), of about a quarter of a wavelength at both ends. The recesses are also effectively changing the actual dielectric constant in the transition facilitating a reflection-free transition between the circular and rectangular waveguide sections of the feedthrough. The initial dimensions of the recesses were chosen to achieve a "filling factor" of the dielectric in this transition section such that The initial feedthrough geometry model was further parametrized and the electrical performance was optimized in the Ansys HFSS TM .
During the optimization, the overall length of the feedthrough remained constant, while the HDPE sealed circular waveguide dimensions were varied together with the dimensions of the matching elements of the transitions between the WR-10 rectangular waveguide and the circular sealing section. Fig. 2 displays the simulated performance of the proposed feedthrough for the model with the optimized dimensions.
The simulations for the optimized feedthrough were performed for the perfect geometry. For its practical implementation, the mechanical layout of the device should be constructed of two parts, Fig. 3(a). The main dimensions of the optimized feedthrough are given in Fig. 3(b). We used two O-rings in the design of the feedthrough. The internal O-ring (denoted as O2 in Fig. 3(a)) seals the split between the two metal halves and, by this, extends the effective isolating length of the HDPE insert into the air half of the feedthrough. The external O-ring (denoted as O1 in Fig. 3(a)) is used to seal the mounting interface of the feedthrough.

IV. TOLERANCE ANALYSIS
In order to define the manufacturing tolerances, we performed a sensitivity analysis of the design to mechanical fabrication accuracy using ANSYS HFSS TM . In general, the alignment of the two halves of the feedthrough is provided by the dowel pins with the pressing fit giving an estimated accuracy of the two parts alignment better than ±10 µm. Additionally, during the fabrication, we have used a special jig to align the opposite sides' interfaces of the feedthrough metal parts. The simulations of the lateral offset produced quite insignificant effect on the feedthrough performance for the allowed offsets, which is consistent with the results in [11].
The axial alignment of the two halves is naturally made by the pressing fit of the HDPE sealing insert. However, the rotational alignment by dowel pins at the interface between the two halves could allow some angular misalignment resulting into positioning the WR-10 input / output ports at some angle. From the point of view of the EM field propagation, it implies that the wave entering from the input WR-10 port and going through the polarization-neutral circular section hit the output WR-10 waveguide at some angle, leading inevitably to reflection and hence additional losses proportional to cos(α) where α is the misalignment angle.  Such misalignment manifests itself by frequency suck-out in the insertion loss as presented in Fig. 4 for a misalignment angle of α = 0.5 degrees.

V. MEASUREMENTS AND DISCUSSION
The performance of the fabricated feedthrough was measured using a Vector Network Analyzer (VNA), Keysight PNA-X, with frequency extension modules. To cover the extended W-band 67-116 GHz, two VNA waveguide extensions VDI WR-8 (90-140 GHz) and WR-10 (67-110 GHz) with corresponding waveguide adapters were used. The WR-10 extension module exhibits excessive noise that is manifested by small positive picks in the insertion loss plot especially at 97-99 GHz, Fig. 5.
The measurements were performed several times in order to obtain a reliable stitching between the results attained with the two different VNA extension modules. The presence of small amplitude suck-outs with max amplitude of ∼-0.11 dB at the measured insertion loss (Fig. 5) is likely to be ascribed to a small miss-alignment of the rectangular sections causing internal reflections and resonances, while the spread in return loss could be associated with a mixed effect of the small changes in the HDPE insert dielectric permittivity connected to the pressing and deformation during the assembling of the device and other small mechanical fabrication intolerances.
The helium leak test was performed using the Pfeiffer Vacuum Technology GmbH SmartTest HLT560 leak detector. The vacuum-seal feedthrough was exposed to the Helium gas at atmospheric pressure continuously and a leak rate after 20 min exposure of less than 2x10 -8 mbar·L/s has been measured, fully complying with the requirements to be below 2.5·10 -6 mbar·L/s. Table 1 summarizes the performance of several feedthroughs available commercially or reported in the literature, showing that the superiority of our proposed design, when it comes to wideband vacuum-seal waveguide feedthrough operating in the W band.

VI. CONCLUSION
We presented a novel vacuum waveguide feedthrough that is rather tolerant to geometrical inaccuracies achieved by employing, e.g., standard fabrication by the CNC machining. The performance of the feedthrough was verified using RF and leak test measurements. The feedthrough demonstrated a low insertion loss below 0.3 dB and a return loss better than 20 dB over in the entire 67-116 GHz RF band corresponding to a fractional bandwidth of 54 %. In addition, the feedthrough has a leak rate better than 2x10 -8 mbar·L/s. The design is relatively simple mechanically and could be scaled for other frequency band of interest and different applications. The feedthrough presented here was designed to be used in the ALMA Band 2 cold cartridge assembly [8].