3-D Printed Plug and Play Prototyping for Low-Cost Sub-THz Subsystems

Polymer-based additive manufacturing using 3-D printing for upper-millimeter-wave (ca. 100 to 300 GHz) frequency applications is now emerging. Building on our previous work, with metal-pipe rectangular waveguides and free-space quasi-optical components, this paper brings the two media together at G-band (140 to 220 GHz), by demonstrating a compact multi-channel front-end subsystem. Here, the proof-of-concept demonstrator integrates eight different types of 3-D printed components (30 individual components in total). In addition, the housing for two test platforms and the subsystem are all 3-D printed as single pieces, to support plug and play development; offering effortless component assembly and alignment. We introduce bespoke free-space TRM calibration and measurement schemes with our quasi-optical test platforms. Equal power splitting plays a critical role in our multi-channel application. Here, we introduce a broadband 3-D printed quasi-optical beamsplitter for upper-millimeter-wave applications. Our quantitative and/or qualitative performance evaluations for individual components and the complete integrated subsystem, demonstrate the potential for using consumer-level desktop 3-D printing technologies at such high frequencies. This work opens-up new opportunities for low-cost, rapid prototyping and small-batch production of complete millimeter-wave front-end subsystems.


I. INTRODUCTION
Over the past few years, companies have been adopting additive manufacturing techniques as a standard means of rapid prototyping and small-batch production. For example, 3-D printing has inherent advantages with fast turn-around, weight reduction (employing unique topological solutions) and waste reduction, when compared to subtractive manufacturing. These significant advantages already play a key role within the aerospace and high-performance automotive industries.
Polymer-based 3-D printing is now showing great potential for the rapid prototyping of passive components and circuits for microwave-to-terahertz (THz) applications. For example, Timbie et al. demonstrated a 3-D printed metal-pipe The associate editor coordinating the review of this manuscript and approving it for publication was Xi Zhu . rectangular waveguide (MPRWG) and corrugated horn antenna at W-band (75 to 110 GHz) [1]. Three years later, Bieren et al. reported a MPRWG and diagonal horn antenna at WR-3.4 band (220 to 330 GHz) [2]. The following year, we reported our detailed investigation for implementing high performance 3-D printed MPRWGs; covering both X-band (8 to 12 GHz) and W-band, while also demonstrating a high performance E-plane split-block iris-coupled 6 th order band-pass filter (BPF) at W-band [3]. Since then, we have moved 3-D printed MPRWGs up in frequency; covering the WM-380 band (500 to 750 GHz) and WM-250 band (750 GHz to 1.1 THz) [4].
These early works [1]- [4] are non-reconfigurable, in the sense that the reported components are neither mechanically or electrically tunable. In order to expand the functionality for 3-D printing, in addition to metal-plated structures, we have accurately characterized important polymer-based building materials. For example, a relatively low loss acrylonitrile butadiene styrene (ABS) material was characterized over X-band [5] and Ku-band (12 to 18 GHz) [6]. This build material was then employed in mechanically-tunable 3-D printed dielectric-insert phase shifters at both X-band [7] and Ku-band [8]. In addition, carbon-loaded conductive polylactic acid (PLA) was also characterized across both X-and Ku-bands [6]. This build material was then employed in a mechanically-tunable rotary vane attenuator at Ku-band [6].
While mechanically-tunable 3-D printed components are interesting, there are still reliability issues. For this reason, the first passive-active hybrid technology was demonstrated, having 3-D printed waveguide flanges, delay lines and power splitters/combiners and, more significantly, with integrated high-resistivity silicon implants and laser diodes; creating an electronically-tunable I-Q vector modulator at WR-2.2 band (325 to 500 GHz) [9].
Integrating multiple 3-D printed components together offers the important advantage of removing impedance mismatch losses between flanges. For example, both Rohrdantz et al. [10] and later Dimitriadis et al. [11], demonstrated a non-tunable 4-element antenna array at Ka-band (26.5 to 40 GHz). In contrast, our group recently demonstrated a mechanically-tunable 3-D printed beam-steerable 4-element phased-array antenna at Ku-band [8].
In addition to the advances being demonstrated with integrated waveguide-based subsystems, Harris Corp. (USA) recently reported two commercial (conventionally-machined) open-ended waveguides that are coupled via a pair of 3-D printed lens, using an in-line alignment frame at lower millimeter-wave frequencies (ca. 30 to 100 GHz) [12], [13]. In 2021, we reported our detailed investigation for implementing high performance 3-D printed quasi-optical components (rectangular horn antenna, off-axis parabolic mirror (OAPM), convex and Fresnel lenses, polarizer, radiation absorbent material (RAM) and a reconfigurable 'plug and play' housing for undertaking the free-space characterization of these components) at G-band (140 to 220 GHz) [14].
An exhaustive survey of polymer-based 3-D printed microwave-to-THz applications has not been given here, as most of the relevant articles have already been cited in our previous papers. This paper significantly advances 3-D printing by fully integrating both waveguide and quasi-optical components, to create a compact multi-channel subsystem proof-ofconcept demonstrator. Here, the proposed front-end sub-THz subsystem integrates 30 individual components: 1× OAPM, 1× plano-concave lens, 1× reconfigurable flat mirror, 5× flat mirrors, 4× beamsplitters, 6× plano-convex lenses, 6× horn antennas and 6× BPFs. Only the OAPM and horn antennas were previously reported [14], while the plano-convex lens has been redesigned for this new bespoke application. As a result, details of their design, fabrication and characterization can be found in [14]. This work introduces our concave lens and flat mirrors, while concentrating on a novel broadband beamsplitter; requiring custom 3-D printed test platforms.
More importantly, this paper demonstrates the proof-ofconcept for a fully 3-D printed integrated plug and play prototype architecture. This paper will show the effectiveness of 3-D printing for achieving good alignment, without the need for a skilled operator to use costly optical aligners for individual components (normally taking many hours). Potential subsystem applications include low-cost millimeterwave metrology, aerospace radiometer and communications payloads, as well as front-ends for +5G base stations.

II. DESIGN
Our design goal is to implement a compact front-end subsystem architecture (with details given in Subsection D). In this section, we describe the design of the bespoke prototype dielectric lenses, holey beamsplitter and also the test platforms and subsystem housings. From the outset, it was decided that the complete single-piece subsystem housing would fit within the maximum build volume (280 mm × 350 mm × 300 mm) of our fused deposition modelling (FDM) 3-D printer (Raise Pro 2). Multiple housing parts would compromise alignment accuracy and structural integrity.

A. DIELECTRIC LENSES
The compact nature of our architecture dictates a number of design implantation choices. For example, lenses are chosen over mirrors to focus and collimate beams, minimizing signal path complexity; this also allows a significant reduction in both volume and mass, but increases transmission losses (discussed in Section VI).
Two different types of hyperbolic aspherical lenses are designed for the subsystem: a plano-convex lens [14] and a plano-concave lens.
The main input test beam is externally collimated by a commercial antenna-OAPM pair [14]. With the OAPM, the two-sigma (i.e., 1/e 2 normalized power) boundary for the collimated Gaussian beam (containing approximately 95% of the beam power) projected into the subsystem's main input window has an estimated 1/e 2 beamwidth that is less than 50 mm (i.e., below the 76.2 mm diameter of the OAPMs, to minimize diffraction losses with both the external commercial OAPM and its internal 3-D printed OAPM replica [14]).
The internal beamwidth is significantly reduced to approximately 20 mm, to reduce the width of the lenses (helping to make the subsystem more compact) and, therefore, also reduces their thickness (lowering transmission losses with plano-convex lenses). At the output of the internal OAPM, a convex lens could have been chosen. However, in order to significantly reduce the transmission path length (to achieve a compact subsystem), a hyperbolic aspherical planoconcave lens is chosen. In addition, six hyperbolic aspherical plano-convex lenses are employed for our 6-channel architecture, each focusing the internal collimated beam into its corresponding channel output horn antenna.
The building material for the lenses is chosen to be the Elegoo Water Washable Rapid Resin (Ceramic Grey), as its  dielectric properties are already characterized from our previous work [14]: dielectric constant ε r ≈ 2.89 and loss tangent tanδ ≈ 0.032 across G-band. The hyperbolic aspherical curvatures for both plano-concave and plano-convex lenses, use the following design equation [14]: where f pl is the target focal path length, z(r) is the curvature profile and r is the corresponding radial distance. Also, to verify the design, a custom MATLAB code is written for optical ray-tracing, with results shown in Fig. 1(a). It can be seen that the plano-convex lens design can provide a perfect optical focus with the target focal path length of f pl = 50 mm. Figure 1(b) shows that the plano-concave lens is able to collimate a converging beam. In both cases, spherical aberration is not theoretically present. In practice, however, because the input source cannot be perfectly collimated, the lenses will not be able to focus/collimate perfectly. 3-D printed lenses and the key properties of the lenses are shown on Fig. 1 and Table 1, respectively.

B. HOLEY BEAMSPLITTER
To split an incident collimated beam into two separate beams of equal power, a holey beamsplitter is developed for this application. Conventional wideband free-space beam splitting is commonly achieved using a 'polka dot beamsplitter' at optical frequencies (infrared and visible) [15], [16]. The polka dot beamsplitter consists of a 2-D periodic array of metalized patches (typically circular or square) that masks 50% of its glass substrate. Because the size of each patch is sufficiently larger than the maximum in-band optical wavelength, it has an excellent performance over a wide bandwidth [17]. For (sub-)THz frequencies, sub-wavelength metasurface beamsplitters have been reported, based on either a 2-D or 3-D periodic array [18]- [22]. However, as the required minimum feature size for the patch is sub-100 µm, it is not possible to manufacture these beamsplitters for uppermillimeter-wave frequencies (ca. 100 to 300 GHz) using currently available desktop 3-D printers. Also, due to their sub-wavelength periodicity, they exhibit frequency dispersion and, therefore, narrow bandwidth.
Another low-cost approach for achieving free-space power splitting, at THz frequencies, employs an electrically thinfilm metal-coated dielectric substrate. For example, a subskin depth layer of silver paint can be deposited onto a low-density polyethylene substrate [23].
Our holey beamsplitter design is inspired by the concept of the polka dot beamsplitter, by exploiting Babinet's principle to create the complementary structure (i.e., the metal patches are replaced by a metal mesh), which avoids the need for printing the non-metalized areas of the substrate. This solution allows the beamsplitter to be manufactured using conventional FDM 3-D printing and metal plated using gold leaf gilding techniques [14]. FDM 3-D printing has poor dimensional precision and print consistency for sub-millimeter minimum feature sizes. Moreover, a single layer of gold leaf (having a manufacturer's quoted thickness of 0.12 µm [14]) exhibits less than one skin depth across G-band (ranging from 0.167 to 0.210 µm); simulations for a single layer of gold leaf show that this is sufficient to provide a screening effectiveness [24] of approximately 59 dB and return loss of 0.01 dB, across G-band. A conventional approach for achieving equal transmittance and reflectance was not possible with a 50% metallized coverage area. Since we do not have the computational resource to undertake the necessary 3-D electromagnetic modelling, we adopted an iterative 'trial and error' design-manufacturetest cycle; fortunately, our solution lends itself to rapid It is important to note that the beamsplitter's lattice constant p = 2.3 mm λ max (140 GHz)/10 = 0.2 mm, which means it cannot be treated as a homogeneous subwavelength metasurface (which would otherwise require p < λ/10); as such, with p ∼ λ, our beamsplitter design falls under the heading of quasi-optics (which excludes both simple electromagnetic theory and simple geometrical optics ray tracing).
The power transmittance (T = |S 21 | 2 ) and reflectance (R = |S 31 | 2 ) measurement setup is shown in Fig. 3. An ideal, lossless beamsplitter has T = R = 0.5, while our primary objective is just to achieve T = R across G-band.
Absorptance (A), which includes all losses (e.g., scattering, diffraction, ohmic and non-ideal bean collimation), is represented by: With our 2-port measurements, the input return losses are −10log |S 11 | 2 < 20 dB across more than 80% of G-band and so the simplifying approximation in (2) can be justified.

C. TEST PLATFORMS
To test the 3-port holey beamsplitter, at its intended oblique incidence angle of 45 • , two 2-port measurements are needed; with a common input port (Port 1) and two output ports (Port 2 and Port 3), as illustrated in Fig. 3. Both output ports need to face the input port, due to the limited cable lengths between the VNA and frequency extension heads, and also to avoid unnecessary movement of the cables [14]. To accommodate these requirements, a bespoke reconfigurable quasioptical 3-port test platform is required. Our 3-D printed plug and play test platform has pre-aligned slots for the horn antennas (in their 3-D printed adapters) and collimating planoconvex lenses. The beams transmitted through and reflected by the beamsplitter are focused by plano-convex lens into the corresponding commercial horn antennas at Port 2 and Port 3, respectively; with the latter, a flat mirror is needed to  redirect the beam and this should be positioned parallel to the beamsplitter. The 3-D printed 3-port test platform is shown in Fig. 4, without a device under test. Our input and internal beams are all vertically polarized. To evaluate the beamsplitter's polarization response, a bespoke quasi-optical rotating test platform is 3-D printed and fitted with a rotational positioner (having 3-D printed alignment markers at every 5 • intervals of rotation, over a 90 • arc). Here, the intended incidence angle is fixed at 0 • . The 3-D printed quasi-optical rotating test platform is shown in Fig. 5.

D. SUBSYSTEM HOUSING
The complete single-piece subsystem housing is designed to have external dimensions of 279 mm × 265 mm × 92 mm, integrating 30 individual components. Our prototype architecture is designed to support multiple iterations on the development of each component. For this reason, our plug and play methodology employs pre-aligned snap-in brackets for each component.  A block diagram for the compact multi-channel subsystem architecture is illustrated in Fig. 6(a), with associated beam paths for each of the six channels. In addition to the main input, an auxiliary window is included with the option for the subsystem channel receivers to 'see' external hot or cold reference sources. The removable flat mirror has two settings. When inserted, Channels #1 and #2 can 'see' the thermal reference sources (blue beams in Fig. 6(a)), while Channels #3 to #6 'see' the main input (green beams in Fig. 6(a)). In contrast, when removed, Channels #1 and #2 'see' the main input, while Channels #3 to #6 'see' the thermal reference sources. Although we changed the removable flat mirror settings manually (for convenience), our 3-D printed housing accommodates an electrically-actuated switching mechanism; this could then be used to implement a Dicke switch for radiometry applications.
The fully 3-D printed quasi-optical housing with integrated components for our proof-of-concept demonstration is shown in Fig. 6(b). When compared to using a commerciallymachined breadboard and off-the-shelf optical components and their positioners, 3-D printing for prototyping of a complete front-end subsystem offers the important advantage of providing a much more compact solution. In addition, component assembly and alignment are effortless [14]. It can be seen in Fig. 6(b) that a near-optimal configuration is achieved in terms of size and compactness. Table 2 lists the various components, from the test platforms and subsystem housings to individual components, and their methods of implementation [14]. In general, each component has unique methods of implementation; the rationale for this has already been discussed in detail [14].

III. FABRICATION A. 3-D PRINTING
Details will now be given for the new components not previously reported. With the flat mirrors, to avoid the relatively high cost of commercial electroplating, a single sheet of ∼25 µm thick copper foil is simply laminated onto the flat front-facing surface of the mirror. Since the band-pass filters have small and complex internal features, commercial electroplating of ∼50 µm thick copper is chosen [14].
As the holey beamsplitter requires dimensional adjustments with its iterative cycle, the script-based CAD drawing software, OpenSCAD [32], is used. Being only 500 µm thick, this component is too delicate for commercial electroplating; hence, gold leaf gilding is employed [14]. Applying more than one layer of gold leaf causes glue to clump between the holes, causing uneven surfaces. Therefore, only a single layer of gold leaf is gilded onto one side of the holey beamsplitter's 3-D printed skeletal frame.
The 300 µm separation distance between the square holes is large enough to be FDM 3-D printed (having a 200 µm diameter nozzle size), while the extrusion speed is sufficiently reduced to prevent mesh breakage. With an FDM 3-D print layer height of 100 µm, multiple layers were deposited to  give a mesh thickness of approximately 500 µm; this provides sufficient mechanical rigidity for our application. A low viscosity acrylic glue is applied with a very fine brush to one surface of the 3-D printed part; this is allowed to partially-dry before the gold leaf is gilded. The only design variable left is the physical size of the square holes needed to provide an equal power split, as discussed in Section IIB.

IV. COMPONENT MEASUREMENT
All electromagnetic measurements are undertaken at the U.K.'s National Physical Laboratory (NPL), using their Keysight Technologies PNA-X N5247B vector network analyzer (VNA) -with VDI WR-5.1 frequency extension heads [35].
Two different calibration methodologies are needed: Thru-Reflect-Line (TRL) calibration in conventional metal-pipe rectangular waveguide, with reference planes at the VNA frequency extension heads; and Thru-Reflect-Match (TRM) calibration in free space having two reference plane options.
After completing the TRL calibration, commercial Flann Microwave Ltd 20 dBi standard gain rectangular horn antennas are fixed to each port [36]. The antennas, with their interface adapters, are then slotted into the quasi-optical test platform shown in Fig. 4. This set-up still has its calibration reference planes located at the flanges of the frequency extension heads. Therefore, any subsequent measurements will include the adverse effects due to the non-ideal horn antennas, lenses and free-space channel, illustrated in Fig. 7.
To avoid these adverse effects, while isolating the device under test in free-space, TRM calibrations are employed for some of the measurements, which are widely used for free-space material characterization [37]. TRM calibration does not require movement of any components during its calibration process [38], making it ideal for our fixed position test platform. Proprietary software containing an automated TRM algorithm, within the PNA-X, was used to calibrate our 3-D printed test platform using our 3-D printed Thru-Reflect-Match calibration standards.
As shown in Fig. 3, the beamsplitter represents a 3-Port network that requires both transmittance and reflectance measurements. However, since the VNA and its frequency extension heads only support 2-Port testing, two different sets of measurements are required, each having their own calibration. For the transmittance measurements, TRM Calibration #1 is adopted, as illustrated on the left-hand side of Fig. 8. For the reflectance measurements, TRM Calibration #2 is adopted, as illustrated on the right-hand side of Fig. 8. For both calibrations, 3-D printed flat mirrors and commercial pyramidal absorbers (TK Instruments [39]) are used for the Reflect and Match standards, respectively. These calibrations provide an initial common reference plane positioned midway between the lenses.

A. FLAT MIRROR
Using the quasi-optical test platform shown in Fig. 4, with TRM Calibration #1 shown in Fig. 8, the performance of the flat mirror is evaluated. A pair of identical flat mirrors, at an oblique incidence angle of 45 • , creates a thru-line, as illustrated by the insert in Fig. 9. The initial TRM Calibration #1 reference plane is effectively split in two and shifted from the center of the mirrors to their outer edges. As a result, the transmission measurement, in Fig. 9, represents the combined losses from both flat mirrors (e.g., misalignment, ohmic and surface defect-related diffuse reflection scattering) and non-ideal beam collimation exacerbated by excess path lengths between mirrors. It can be seen that a single flat mirror is expected to have an associated loss of 1-2 dB. However, it is important to note that this result only represents a rough pseudo-qualitative estimation of the flat mirror's performance. VOLUME 10, 2022 FIGURE 10. Holey beamsplitter measurements: (a) illustration of the test set-ups using TRM Calibration #1 (with additional de-embedding of the reference planes) and #2 for measuring transmittance and reflectance, respectively; and (b) measured transmittance and reflectance at 45 • incidence (running averaging applied over 3.2 GHz).

B. HOLEY BEAMSPLITTER
Using the same 3-port quasi-optical test platform, shown in Fig. 4, the 3-D printed holey beamsplitter is now evaluated. The measurement set-up employs both TRM Calibrations #1 and #2, illustrated in Fig. 10(a), for the transmittance and reflectance measurements, respectively, with results shown in Fig. 10(b). With the former, the initial TRM Calibration #1 reference plane is effectively split and shifted from the center of the beamsplitter to its outer edges. Figure 10 shows that the holey beamsplitter is able to equally divide the incident power; fulfilling our primary objective. The maximum difference between transmittance and reflectance is less than 9%, across G-band. Ideally, absorptance should be 0%. In practice, due to a number of factors (e.g., diffraction, diffuse scattering, non-ideal beam collimation and ohmic losses), our holey beamsplitter has approximately 30-50% of absorptance. Table 3 shows the performance of ideal, typical commercial optical and our holey beamsplitters. When compared to the commercial optical polka dot beamsplitter, our millimeter-wave proofof principle demonstrator has approximately 20-30% more absorptance.
With our vertically-polarized collimated beams, the S-parameters are measured at different transverse angles of rotation, using the 3-D printed polarization test platform, shown in Fig. 5. For this measurement, conventional TRL calibration at the VNA's frequency extension heads is performed. As shown in Fig. 11(a), the beamsplitter has identical transmission responses at three transverse angles of rotation (0 • , 45 • and 90 • ); confirming polarization independence.
The normalized transmittance of the beamsplitter at normal incidence, with quasi-optical test fixture, T n (dB) ∼ = to the transmittance when the beamsplitter is removed. Here, T n (dB) ∼ 5 dB at 180 GHz, for all three rotation settings; this being comparable to the measured transmittance T = 5.5 dB at 180 GHz, extracted from Fig. 10, at the oblique incidence angle of 45 • . As seen in Fig. 11(b), the reflection responses at all three transverse angles of rotation are also almost identical. It is important to note that, due to the high insertion loss for each lens (>5 dB, discussed in the next sub-section), there is almost no measured difference in reflectance seen in Fig. 11(b), with or without the beamsplitter.
Note that the observed in-band Fabry-Pérot ripples, having frequency period f , correspond to standing-wave reflections between two spatial boundaries that are separated by a theoretical free-space distance l = c f . Without the beamsplitter, the extracted value of l ∼ = 53 mm corresponds to the spatial distance of 55.0 mm between the flat sides of both plano-convex lenses. With the beamsplitter inserted, the extracted value of l ∼ = 26.5 mm corresponds to the spatial distance of 27.8 mm between the beamsplitter and the flat side of the plano-convex lens. This shows that the metalized side of the beamsplitter is positioned exactly midway between the two plano-convex lenses.

C. PLANO-CONVEX LENS
Using the 3-D printed test platform shown in Fig. 4, with conventional TRL calibration at the VNA's frequency extension heads, the performance of a lens pair is evaluated. Similar to the arrangement shown in Fig. 7, transmission between two commercial horn antennas, with and without the lenses, is measured. From Fig. 12, it can be seen that adding the lenses provides between 6 to 17 dB of 'gain enhancement'. This clearly shows that the lens is performing as expected. The commercial antenna and 3-D printed lens pair exhibits, in situ, approximately 5-8 dB of insertion loss across G-band.
Note that the extracted value of l ∼ = 62.5 mm corresponds to the spatial distance of 65.0 mm between the flat sides of both plano-convex lenses.

D. WAVEGUIDE FILTERS
The proposed front-end sub-THz subsystem accommodates up to six channels, each having their own band-pass filter. Measured results for four 5 th order prototype BPFs are shown in Fig. 13. Further details of our 3-D printed G-band filters is beyond the scope of this paper and will be published in a separate article.
With reference to IEEE standards [40], the internal aperture dimensions for WM-1295 (G-band) metal-pipe rectangular waveguides are 1,295 µm × 647.5 µm. All five cavity resonators support the TE 101 mode and are coupled together using transverse offset inductive irises. Photographs of a 180 GHz H-plane split-block filter are shown in Fig. 13(a), with a side-view of the upper-half (above) and plan-view of the lower-half (below). The corresponding measured frequency responses for four G-band prototype BPFs are shown in Fig. 13(b). It should be noted that the top-right   Fig. 13(b) correspond to the filter in Fig. 13(a); this filter is employed to demonstrate the final subsystem performance for Ch. #1 (presented in Section V).

V. INTEGRATED SUBSYSTEM MEASUREMENTS
With conventional TRL calibration at the VNA's frequency extension heads, the performance of the fully assembled 3-D printed integrated subsystem is evaluated. At Port 1, commercial rectangular horn antenna and 90 • OAPM are added to emulate the collimated main input beam [14]; adding an estimated 2 dB of associated losses. Port 2 is connected directly to one of the six output channels, to emulate the channel output receiver. The complete front-end measurement set-up and its results are shown in Fig. 14.
The measured transmission results for the six channels, without their filters, are shown in Fig. 14(b). The lowest transmission loss is observed with Ch. #1, as expected; between 21 and 24 dB across G-band. The highest transmission loss is observed with Ch. #6; between 27 and 40 dB across G-band. Clearly, performance degrades as frequency increases, due to non-ideal beam collimation, which is exacerbated with longer path lengths. With the 180 GHz BPF integrated into Ch. #1, the measured results are shown in Fig. 14(c). With the filter included, despite all possible losses, the subsystem works as expected and exhibits good dynamic range.
Significant computational resources are needed to undertake numerical free-space electromagnetic modeling of individual quasi-optical components. Unfortunately, the lack of such resources prevent us from undertaking rigorous quantitative analysis of the complete integrated system. As a result, it is beyond the scope of this paper to accurately quantify the loss contributions associated with each and every component in situ. However, using the measurements in Fig. 14(b) and with information for individual component (from Section IV), it is possible to make a pseudo-quantitative loss budget analysis for the complete subsystem.
An estimated loss for each component type is assigned and their total loss is then compared with that measured for each channel. From this, we speculate that both the flat mirrors and beamsplitters create an additional loss of ∼0.7 dB at 180 GHz in their reflected paths, because the beams are not perfectly collimated. This is further compounded by both external and internal misalignments of the quasi-optical components. The results of our preliminary analysis are shown in Table 4 and Fig. 6(a). With the exception of Ch. #2, it can be seen that there is good agreement between our estimated and measured total losses.

VI. DISCUSSION AND CONCLUSION
In this paper, we have significantly expanded our previous works on polymer-based 3-D printing for low-cost waveguides and quasi-optical components. Here, we brought together many of the concepts by demonstrating a complete plug and play 6-channel subsystem for operation between 140 to 220 GHz. This sub-THz front-end architecture has successfully integrated 30 individual components: 1× OAPM, 1× plano-concave lens, 1× reconfigurable flat mirror, 5× flat mirrors, 4× beamsplitters, 6× plano-convex lenses, 6× horn antennas and 6× BPFs. It was found that a blend of different 3-D printing technologies and plating methods were needed, depending on the type of component.
In this paper, we introduced bespoke quasi-optical TRM calibrations and measurement schemes with our 3-D printed test platforms. Our 3-D printed flat mirrors demonstrated total losses between 1-2 dB across G-band. The holey beamsplitter was able to divide the input beam into two output beams with equal power. A pair of 3-D printed plano-convex lenses gave up to 17 dB of gain enhancement across G-band. Four BPFs showed good transmission and reflection measured responses. Finally, two test platforms and integrated subsystem housings were all 3-D printed as single pieces, to support plug and play development, while providing sufficient beam alignment accuracy.
Having been able to establish a pseudo-quantitative loss budget for the individual elements, future work on minimizing associated losses can now be investigated. For example, measurements of Ch. #1, having a 180 GHz BPF included, demonstrates a total measured insertion loss of 23 dB for our first proof-of-concept prototype. While this is likely to be too large for most applications, our link budget helps to identify where significant improvements can be made.
Our lenses were made from a relatively lossy 3-D printable dielectric material and so the plano-convex lens exhibited a transmission loss that is >5 dB. Therefore, a 3-D printable dielectric that has a much lower loss tangent will reduce transmission losses; while a greater refractive index will reduce the absorptance due to having a thinner lens, but the impedance-mismatch reflection losses will dominate and increase the net transmission loss. Alternatively, focusing mirrors (e.g., OAPMs) can completely replace the lenses, but this will dramatically increase the size and mass of the assembled subsystem; these and other engineering compromises can be explored in future studies.
Another component worthy of further investigation is the beamsplitter. After considerable investigation, there does not appear to be an obvious 3-D printing alternative to our solution. Nevertheless, with our holey beamsplitter approach, optimization of the internal macroscopic geometrical features (using full-wave electromagnetic simulation software), better material selection and advanced manufacturing techniques, will inevitably reduce its loss characteristics.
One of the major contributors to loss is non-ideal beam collimation and misalignment. Replacing our rectangular horn antenna with a 3-D printed corrugated conical horn antenna will help to preserve the bi-axial symmetry of the beam [14]. Moreover, the manufacturing tolerances associated with our FDM printed housing, plug and play interfaces and adapters can be improved with more advanced, commercial grade 3-D printing technologies.
Diffraction and distortion of beam profiles is not inherent to 3-D printing and so this technology should not have any impact on the design of a future subsystem. With the former, as with any quasi-optical implementation, simply increasing component sizes and/or reducing beamwidths will reduce diffraction losses. With the latter, simply using a better quality 3-D printer will reduce beam distortion.
In conclusion, when compared to using a commerciallymachined breadboard and off-the-shelf optical components and their positioners, 3-D printing for prototyping of a complete front-end subsystem offers the important advantage of providing a much more compact solution. In addition, component assembly and alignment are effortless.
This work opens-up new opportunities for rapid prototyping and small-batch production of complete low-cost quasi-optical upper-millimeter-wave subsystems. Indeed, with the continual advancements in all aspects of affordable 3-D printers, and their materials, better resolution and higher performance can be expected in the near future.
ROSHAN PAYAPULLI was born in Bengaluru, India, in 1995. He received the M.Eng. degree in electrical and electronic engineering from Imperial College London, London, U.K., in 2018, where he is currently pursuing the Ph.D. degree with the Department of Electrical and Electronic Engineering.
His research interests include designing and manufacturing lightweight and low-cost waveguide components and systems for microwave and millimeter-wave applications using additive manufacturing techniques. MANOJ STANLEY (Member, IEEE) received the Ph.D. degree in electrical engineering from the University of Liverpool, in 2019. After the completion of his Ph.D. degree, he joined the National Physical Laboratory, U.K., as a Higher Research Scientist at the Electromagnetic Technologies Group. He has supported the development of next-generation computing and communications technologies and high-frequency electronics applications. He is currently developing high-frequency metrology capabilities to characterize superconducting quantum integrated circuits at milli-kelvin temperatures. His research interests include design and characterization of devices at RF and terahertz frequencies, RF metrology for quantum computing applications, material, and channel characterization for 5G and beyond communication. He is the Operations Officer at European Microwave Week 2021.