Screen-Printed 1 × 4 Quasi-Yagi-Uda Antenna Array on Highly Flexible Transparent Substrate for the Emerging 5G Applications

Abstract

In the Internet of Things (IoT) era, the demand for cost-effective, flexible, wearable antennas and circuits has been growing. Accordingly, screen-printing techniques are becoming more popular due to their lower costs and high-volume manufacturing. This paper presents and investigates a full-screen-printed 1 × 4 Quasi-Yagi-Uda antenna array on a high-transparency flexible Zeonor thin-film substrate for emerging 26 GHz band (24.25–27.55 GHz) 5G applications. As part of this study, screen-printing implementation rules are developed by properly managing ink layer thickness on a transparent flexible Zeonor thin-film dielectric to achieve a decent antenna array performance. In addition, a screen-printing repeatability study has been carried out through a performance comparison of 24 antenna array samples manufactured by our research partner from CEA-Liten Grenoble. Despite the challenging antenna array screen printing at higher frequencies, the measured results show a good antenna performance as anticipated from the traditional subtractive printed circuit board (PCB) manufacturing process using standard substrates. It shows a wide-band matched input impedance from 22–28 GHz (i.e., 23% of relative band-width) and a maximum realized gain of 12.8 dB at 27 GHz.

1. Introduction

The growing applications of the Internet of Things (IoT) and 5G/6G have raised the demand for flexible, wearable, low-cost, and mass-manufacturing microwave and millimeter-wave components. These applications involve billions of connected devices performing several operations such as sensing, data acquisition, processing, and transportation [1,2,3,4]. Modern printed electronic circuits need to be low-cost and mechanically flexible for wearable applications to fulfill the requirements of mass production and ease assembly conditions on different surfaces. Unfortunately, traditional electronic manufacturing techniques such as subtractive fabrication technologies are not technically and economically competitive in mass production [5,6]. They are expensive, resulting in material waste, and are not ideal for mass manufacturing [7]. Recently, various printing techniques have been adapted for electronics manufacturing. For instance, gravure printing, flexography, inkjet printing, and more specifically screen printing [8,9,10,11]. The latter seems to be a promising and cost-effective technique for printable microwave/millimeter-wave/THz circuits such as RFID tags, RFIC packaging, energy-harvesting rectennas, and wireless communication antennas [11,12,13]. Today, screen printing offers repeatable accurate printing of antenna circuits on inexpensive flexible thin-film materials such as glass, laminates, and plastic, providing a large substrate selection for flexible electronics [10,14]. Nevertheless, specific arrangement and parameters such as ink, printing, and screen conditions must be carefully monitored at higher frequencies on highly flexible transparent substrates. Moreover, depending on ink composition and ink/substrate interaction, post-processing through heat treatment at 135° might be required [11].

Given what has been achieved in the screen-printing research area, high-frequency circuits involving the screen-printing process on transparent thin-film substrates are still missing from the literature. Moreover, the implementation challenges associated with those types of substrate materials have not been previously revealed. For these reasons, we believe that clarifying screen-printing issues on highly transparent substrates at higher frequencies will provide practical performance optimization guidelines to microwave/millimeter-wave circuits and antenna designers. As a proof of concept, a full-screen-printed 1 × 4 Quasi-Yagi-Uda antenna array on a high-transparency flexible Zeonor thin-film substrate is presented and analyzed for emerging 24.25–27.5 GHz band 5G applications. Thanks to the suggested screen-printing improvement, the implementation of the 1 × 4 Quasi-Yagi-Uda antenna array prototype has been highly optimized. Moreover, a repeatability test has been performed by manufacturing 24 antenna array samples equally distributed over 4 substrate sheets for performance comparison. The achieved comparison performances show a good concordance over the considered frequency range from 20 GHz to 30 GHz. They exhibit a large bandwidth and high gain as expected from the traditional subtractive printed circuit board (PCB) manufacturing process using standard substrates.

2. Screen-Printing Technique

Figure 1 shows a screen-printing process description. Ink is spread on a stencil by a squeegee. Thus, when the stencil is transparent, the ink is deposited on the substrate. The process requires a precise management of stencil mesh and squeegee properties, as well as production techniques. It may also require tightly controlled parameters during printing. Furthermore, ink transfer is mainly influenced by the screen mesh, squeegee speed, and pressure on the screen. The thickness and accuracy of the printed layer are affected by the ink’s rheological properties. It should be noted that thicker ink deposition involves lower resolution [15].

The parameter selection involves some rules consideration for antenna design, including minimum line width, spacing between adjacent lines, and feeding techniques. The screen-printing prototyping uses a commercial screen-printing machine. However, to improve cost-effectiveness, only textile stencils were utilized to print all the designed antennas. The substrate is the first element to be selected appropriately, especially at the millimeter-wave frequency range, where it plays a key role in antenna performance. It must have low-loss characteristics and be compatible with the physicochemical properties of the selected ink. For our screen-printing process, a Zeonor [16] substrate with a thickness of $100\mathrm{\mu }\mathrm{m}$ (see

Table 1. Properties of the silver ink and substrate.

Ink Name	Conductivity	$\mathit{\delta }$	Thickness
ECI1011	$3.3\times {10}^6\mathrm{S}/\mathrm{m}$	$2.5\mathrm{\mu }\mathrm{m}$	$5\mathrm{\mu }\mathrm{m}$
Substrate name	thickness	${\epsilon }_r$	$tan\left(\delta \right)$
Zeonor	$100\mathrm{\mu }\mathrm{m}$	2.3	0.0009

for characteristics) has been chosen. In order to ensure a proper printing and good adherence of ink with the substrate, it must have an oxygen-free plasma treatment. Substrate selection is a key factor in ensuring good compatibility with inks, particularly with regard to the substrate’s physical and chemical properties. In this section, we expose the essential properties that promote this compatibility. For instance, in the case of inkjet or screen printing with metallic inks, such as silver ink, the substrate must have a microporous receiving layer, either by plasma etching to create nanometric roughness, or by depositing a porous low dielectric such as a fluoropolymer or polystyrene, which enables better adhesion and avoids excessive ink diffusion. The characteristics of silver inkjet printing were intensively investigated with control of surface energy, which is why the choice of substrate can greatly affect the quality of the print and consequently the electrical performance of the deposited silver layer. Plasma introduces high-energy chemical species (such as ions and radicals) that activate the surface, increasing its surface energy. This improves wettability, allowing inks or adhesives to distribute evenly. It removes organic contaminants and fragile boundary layers, creating a cleaner surface and promoting stronger adhesion. Oxygen plasma introduces reactive oxygen species capable of oxidizing silver, which degrades its conductivity and appearance. Even in oxygen-free plasmas (e.g., nitrogen, argon), the treatment generates polar or reactive sites that promote adhesion, although to a lesser extent than with oxygen plasma. This results in better ink adhesion, sharper prints, and longer-lasting bonds in applications such as flexible electronics [17]. The last parameters to choose are the speed of the squeegee movement and the pressure applied to it. These two parameters ensure that the ink is properly spread on the stencil. In our case, a squeegee speed of 150 mm/s and a pressure of 100 N allow us to achieve the desired resolution.

To minimize losses in metallic parts of the antenna, ink selection is a key point in our study. The selected ink must be sufficient for screen printing and have high conductivity with relatively low cost. Thus a trade off between cost and conductivity should be found. As initial value, the thickness of ink is fixed to two penetration depth ($\delta$). Taking into account those constraints, the ECI011 nano-particles silver ink from Henkel has been selected with a conductivity of $\sigma =3.3\times {10}^6\mathrm{S}/\mathrm{m}$. According to Equation (1), the penetration depth can be computed at $24.5 \mathrm{GHz}$ and the needed ink thickness deduced. Table 1 summarizes the properties of this ink.

$$\delta ={\displaystyle \frac{1}{\sqrt{\sigma {\mu }_0\pi f}}}$$

(1)

With such ink thickness, the expected resolution should be around $150\mathrm{\mu }\mathrm{m}$ for ink strips and $200\mathrm{\mu }\mathrm{m}$ for slots formed on ink strips. This resolution should be suitable for printing antennas in the 5G ETSI millimeter-waves frequency range. As the substrate and the ink properties are set, a feeding method is essential. Various feeding solutions on flexible substrates can be employed for this purpose. For a comprehensive characterization of the manufactured antenna prototypes, a 50 $\mathrm{\Omega }$ microstrip line along with a $1.85\mathrm{m}\mathrm{m}$ Southwest Microwave End launch connector are implemented. The designed micro-strip lines consist of a signal strip printed on the top face of the substrate, and a ground plane printed on the bottom face. To assess the thickness of selected ink layer, the lineic attenuation of a 50 $\mathrm{\Omega }$ micro-strip line is extracted from full-wave simulations. The results plotted in Figure 2 shows that an increase in the thickness of printed layers up to $5\mathrm{\mu }\mathrm{m}$ allows reduction of metallic losses. Indeed, increasing the thickness does not significantly reduce metallic losses and does not improve the resolution of the printed patterns [15].

3. Quasi-Yagi-Uda Antenna

In this section, the design, the optimization, and the measurements of the single element antenna of the proposed Quasi-Yagi-Uda antenna array are presented.

3.1. Design Optimization

The single-element Quasi-Yagi-Uda antenna configuration is shown in Figure 3.

Its geometrical parameters have been optimized according to the design rules described in [18,19,20,21]. The initial Quasi-Yagi-Uda antenna configuration is designed using a full ground plane as shown in Figure 3a. Typically, a Quasi-Yagi-Uda antennas antenna consists of two driven elements fed by a 50 $\mathrm{\Omega }$ microstrip line and several directors (parasitic elements), as shown in Figure 3. For design optimization, 3D HFSS 2023-R2 electromagnetic (EM) simulation software has been used while considering the southwest end-launch connector effect. It was demonstrated from simulation results that the distance between the ground plane and driven element (d’) impact significantly matching bandwidth, however, the distance between different directors (d) mainly impacts the gain. The length $L_{rad}$ sets the antenna operating frequency, whereas $L_{dir1}$ and $L_{dir2}$ set the antenna’s lower and higher cut-off frequencies. Several director configurations have been tested. A trade-off between directivity, directors’ losses, and the overall antenna size has enabled us to limit the directors number to 4 directors placed on the top layer. It was reported that a double row of four directors placed respectively on the top and bottom layers may increase the antenna gain [21].

In the context of our study, the expected gain has not been achieved due to high metallic losses and the thickness of the selected substrate. The computed values from [19] have been modified by a scale factor while considering the dielectric constant parameter. We must reiterate that, in our case, the antenna is designed on a substrate contrary to [19], where the antenna was implemented in air. Thus, the dimensions of the antenna are recalculated according to the guided wavelength in the substrate, using the approximation given in Equation (2), where ${\epsilon }_r$ is the permittivity of the substrate.

$${\lambda }_g={\displaystyle \frac{C_0}{\sqrt{({\epsilon }_r+1)/2}{f}_0}}$$

(2)

Then, the antenna dimensions were optimized to achieve good impedance matching and the flattest realized gain over the 24.5 GHz to 27.5 GHz frequency band. For instance, simulation results show that an optimized 4 directors configuration exhibits a realized gain of 9.8 dB (11.1 dB in an 8 directors configuration). This represents an increase in antenna size from $450\mathrm{m}{\mathrm{m}}^2$ to $600\mathrm{m}{\mathrm{m}}^2$ (about a 33% size increase) and a radiation efficiency variation from 65% (8 directors) to 69.5% (4 directors).

Now, we optimize the ground plane to reduce the ink consumption while increasing the flatness band of the gain and improving the side lobe rejection. The proposed ground plane configuration is shown in Figure 3b. $W_{gnd2}$ and $L_{gnd2}$, have been optimized through a rigorous parametric study. Several effects can be observed on the variation in the gain versus frequency as shown in Figure 4a and Figure 5a. However, simulation results demonstrate that if $L_{gnd2}$ is too short, the resulting ground does not affect antenna performance. An optimum value of $L_{gnd2}$ has been set to $15\mathrm{m}\mathrm{m}$ in the proposed antenna configuration; nevertheless, higher $L_{gnd2}$ length causes a reduction in antenna bandwidth (see Figure 4b). Regarding the impact of $W_{gnd2}$ variations, the parametric study results are plotted in Figure 5a. They show that the optimum value is approximately equal to the length of a reflector given in [19]. For lower values of $W_{gnd2}$, the edge of the ground plane does not act as a reflector and higher values than optimal decrease the bandwidth and the maximal gain. Careful attention to input impedance matching (Figure 4b and Figure 5b) show that those parameters can optimize the antenna’s bandwidth. Setting the proper values of $W_{gnd2}$ and $L_{gnd2}$ enables the flattest peak gain in the desired bandwidth.

A further effect concerning the out-of-band gain reduction is also observed. As can be seen, the gain at 20 GHz has been reduced by more than 8 dB (from 1.2 dB to −7.5 dB). Under the same design modifications, the in-band gain variation is reduced to less than 0.5 dB (from 1.3 dB to 0.3 dB).

Table 2. Geometrical parameters of the elementary Yagi antenna.

Parameter	Value	Parameter	Value
$L_{power}$	$5.1\mathrm{m}$$\mathrm{m}$	$L_1$	$2.38\mathrm{m}$$\mathrm{m}$
d	$1.95\mathrm{m}\mathrm{m}$	$L_{rad/2}$	$2.93\mathrm{m}$$\mathrm{m}$
$W_{power}$	$300\mathrm{\mu }$$\mathrm{m}$	$W_{rad}$	$400\mathrm{\mu }$$\mathrm{m}$
$l_{dir1}$	$4.07\mathrm{m}$$\mathrm{m}$	$l_{dir2}$	$3.93\mathrm{m}$$\mathrm{m}$
$W_{gnd2}$	$4.75\mathrm{m}$$\mathrm{m}$	$L_{gnd2}$	$5\mathrm{m}$$\mathrm{m}$

summarizes the optimized geometrical parameters of the antenna, including the 4 directors configuration and modified ground plane.

Figure 6 shows the simulation results of the optimized Quasi-Yagi-Uda antenna. Those results show a maximum gain of 9.8 dB at 27 GHz with a variation of only 0.3 dB within the considered frequency band. In conclusion, the proposed technique allows ground plane surface reduction from $453\mathrm{m}{\mathrm{m}}^2$ to $163\mathrm{m}{\mathrm{m}}^2$, representing a decrease of 64%.

The analysis of the current distribution in Figure 7 shows that the surface current density is intense on the edge of the truncated ground plane while it is much lower when the ground plane is not truncated. This demonstrates that the edge of the truncated ground plane contributes to the radiation more than that of the non-truncated ground plane.

With regards to the radiation pattern performance shown in Figure 8, the newly optimized antenna parameters have an influence only on the H plane radiation pattern with no major changes on the E plane. In the proposed antenna configuration, side lobes at $\pm 75°$ can be reduced from −6 dB to −12.5 dB.

3.2. Measurements

3.2.1. Ink and Printed Layer Characterization

The conductivity of the ink and the thickness of the printed layers were characterized for each ink. For each ink, the resistance value of a given pattern and the metallization thickness have been measured using, respectively, the four-terminal sensing (4 T sensing) technique and a mechanical profilometer. Then, the conductivity of the ink has been deduced. A value of $5\times {10}^6\mathrm{S}/\mathrm{m}$ has been measured. The obtained printing resolution was controlled by measuring the dimensions of slits and lines of 150 µm width under a microscope. A deviation of less than 5% was observed. Similarly, the average thickness of the patterns was measured using a mechanical profilometer. The thickness of the sheets is $5\mathrm{\mu }\mathrm{m}$ except for sheet 2, for which the thickness is $6\mathrm{\mu }\mathrm{m}$.

3.2.2. Antenna Measurements

This section deals with the experimental characterization of the single Quasi-Yagi-Uda antenna. For that purpose, 24 antenna prototypes, printed on 4 different sheets, have been printed by CEA-Liten “Ares” facilities. Measurements for performance investigation have been performed in the im2np lab. Figure 9 shows a photograph of the test bench with a 4 × 1 Quasi-Yagi-Uda antenna under test. Measurements have been performed using a DRH 67 GHz RF-SPIN antenna and an MS4647B 4-port 70 kHz-70 GHz (Anritzu VNA). The antenna under test and reference antenna are set on a rotating mast with a rotating joint to avoid the cables twisting. All measurements are automated thanks to a Python-3.5 script, which records data from the VNA and drives all step-by-step motors used in the test bench. The script used allows for performing radiation pattern measurements in E and H planes, realizing gain in a given direction, and matching as a function of frequency up to $67 \mathrm{GHz}$. The height of the mast and the distance between the antenna under test and the reference antenna have been set to satisfy the free space criteria.

The size of the printed antenna and the end-launch connector compared to a 50-cent coin is given in Figure 10.

The measurement results are plotted in Figure 11 and Figure 12. For each sheet of printed antennas, the mean (see Figure 11a) and standard deviation have been computed from individual measurement of the realized gain. Standard deviation for each sheet plotted in Figure 11b gives further information on the repeatability of the screen-printing process. The standard deviation is lower than $0.4\mathrm{d}\mathrm{B}$ over the $24 \mathrm{GHz}$ to $27.5 \mathrm{GHz}$ frequency band. It involves a very low dispersion for 24 measured antennas. The maximum measured gain is $9.5\mathrm{d}\mathrm{B}$ in the line of sight direction. A small shift of approximately $200 \mathrm{MHz}$, (less than $1\%$) can be observed on measurements. Geometrical measurements of the printed antennas and retro-simulations does not allow to affirm that geometrical issues introduced during the printing cause this mismatch. Retro-simulation of the antenna with ${\epsilon }_r=2.1$ instead of ${\epsilon }_r=2.3$ allows to explain this slight frequency shift between simulations and measurements (see Figure 11a). By adopting the same approach $S_{11}$ measurements results are plotted in Figure 12. The same discrepancy between simulations and measurements can be observed and same reasons can be invoked to explain the observed frequency shift. Higher dispersion on this parameter can also be observed. Still, in this case, a little frequency shit between measurements of different antennas causes higher variations in $S_{11}$ than for gain versus frequency values. The measured $-10\mathrm{d}\mathrm{B}$ matching bandwidth ranges from $24.5\mathrm{to}27.5 \mathrm{GHz}$.

Radiation patterns in the E and H planes have been measured at $24.5 \mathrm{GHz}$, $25 \mathrm{GHz}$, $26 \mathrm{GHz}$, and $27 \mathrm{GHz}$. The results of the simulated and measured normalized gain are plotted in Figure 13. As can be seen, a good agreement between measurements and simulations is achieved. For the H plane, at all measured frequencies, side lobes are below $-15\mathrm{d}\mathrm{B}$ and front to back ratio is lower than $-12.5\mathrm{d}\mathrm{B}$. For the E plane, the side lobe levels are below $-10\mathrm{d}\mathrm{B}$ except at $25\mathrm{G}\mathrm{Hz}$. Measurement shows a maximum $-3\mathrm{d}\mathrm{B}$ beam-width of $56°$ in the H plane and $48°$ in the E plane.

Realized gain performances are compared for Quasi-Yagi-Uda antennas manufactured using different manufacturing processes in

Table 3. Comparison with other published 5G 26 GHz Quasi-Yagi-Uda antennas.

Ref	${\mathit{f}}_{\mathit{o}}$	Bandwidth	Gain	Substrate	Metallization	Radiators
This work	$25.8 \mathrm{GHz}$	$13.7\%$	$9.5\mathrm{d}\mathrm{B}$	Zeonor	Silver ink	4
[22]	$29 \mathrm{GHz}$	$10\%$	$4.9\mathrm{d}\mathrm{B}$	Screen-printable ink	AgNW ink	1
[23]	$26 \mathrm{GHz}$	$14\%$	$8.9\mathrm{d}\mathrm{B}$	Megtron6 *	Copper	4
[24]	$28 \mathrm{GHz}$	$18\%$	$8.1\mathrm{d}\mathrm{B}$	Neltec NX9240 *	Copper	5
[18]	$26 \mathrm{GHz}$	$18\%$	$8.9\mathrm{d}\mathrm{B}$	RO4350B *	Copper	3
[21]	$26 \mathrm{GHz}$	$28\%$	$6.2\mathrm{d}\mathrm{B}$	MFLEX *	-	3 × 2
[25]	$26 \mathrm{GHz}$	$31\%$	$7\mathrm{d}\mathrm{B}$	Liquid crystal	Copper	3
[26]	$26 \mathrm{GHz}$	8.2%	$8\mathrm{d}\mathrm{B}$	Ro3010 *	Copper	2
[27]	$37 \mathrm{GHz}$	15%	$12\mathrm{d}\mathrm{B}$	ASTRA MT40 *	Copper	3 × 3

* Standard PCB technology.

. According to comparison results, antennas printed using silver ink have lower performance than antennas implemented through standard PCB processes. However, our study shows that it is possible to obtain comparable performances by optimizing the manufacturing screen-printing process (thickness of screen-printing inks, selected substrate, and design optimization).

4. Quasi-Yagi-Uda Antenna Array

In this section, the design, optimization and, measurements of the proposed 1 × 4 Quasi-Yagi-Uda antenna array are presented.

4.1. Design Method

A corporate feed network using three rounded-shape Wilkinson dividers is adopted for the 1 × 4 Quasi-Yagi-Uda antenna array design. The selected feed network configuration allows more control of feeding each antenna element (amplitude and phase); hence, they are ideal for several antenna array applications such as multi-beam arrays, scanning phased arrays, or shaped-beam arrays [19]. Typically, two power divider types may be used for the power distribution in the antenna arrays, reactive T-junctions and Wilkinson power dividers. The T-junctions are lossless and simpler to design but do not provide isolation between the output ports. The Wilkinson dividers involve higher design and fabrication complexities, and they are not entirely lossless since reflected power is dissipated in the resistor. However, they provide isolation between output ports, which prevents mismatches and reflected signals from propagating to other parts of the array. To reduce the undesirable coupling and improve isolation between the Wilkinson output ports at higher frequencies, two $\lambda /2$ microstrip lines have been added between the two output ports [28]. It should be noted that the 100 $\mathrm{\Omega }$ isolation resistor between the output ports (ports 2 and 3) has been implemented using a 50 $\mathrm{\Omega }$ per square thin layer carbon past.

The designed Wilkinson power divider and its simulated S-parameters are shown in Figure 14a and Figure 14b, respectively. The simulation results show a maximum insertion loss of $0.8 \mathrm{d}\mathrm{B}$ and isolation between the output ports (ports 2 and 3) better than $25 \mathrm{d}\mathrm{B}$ in the frequency range from $22 \mathrm{GHz}$ to $28 \mathrm{GHz}$.

Electromagnetic simulations have been performed without the feeding network (see Figure 15a) to evaluate the mutual coupling between the antenna elements, impacting the overall performances of the antenna array. Theoretically, the optimal distance required to avoid side lobes is $\lambda /2$ [19] with the proposed network design. The adopted arrangement of 1 × 4 antennas is expected to add $6 \mathrm{d}\mathrm{B}$. Thus, if there is no coupling between the antennas, the gain should reach $15.5 \mathrm{d}\mathrm{B}$ in the main radiation direction.

The obtained results show that the added gain is much lower and that the matching is modified (see Figure 15b). By observing the electric field at the substrate surface for a 6.25 mm spacing between the antennas, it can be observed that the electric field is very intense between the radiators and the directors of two adjacent antennas. This generates a high coupling between the radiating elements and reduces the gain of the antenna array.

Given the constraints of the screen-printing process we use, we chose to move the antennas further apart. Observation of the electric field for a pitch of $8\mathrm{m}\mathrm{m}$ shows that the coupling has decreased. The electrical parameters plotted in Figure 16 confirm this trend. The gain obtained is very close to the theoretical value of $15.5 \mathrm{d}\mathrm{B}$.

As a final step of the design, the whole antenna array is simulated including feeding network. The overall dimension of the designed 1 × 4 Quasi-Yagi-Uda antenna array is around $40 \mathrm{m}\mathrm{m}$ by $40 \mathrm{m}\mathrm{m}$, as shown in Figure 17a.

Figure 17b shows input impedance matching and gain simulation results including the end launch $1.85 \mathrm{m}\mathrm{m}$ connector effect, over the frequency range of $20 \mathrm{GHz}$ to $30 \mathrm{GHz}$. As can be seen, a maximum gain of $12.5 \mathrm{d}\mathrm{B}$ in the main radiation direction is achieved. The antenna array maintains a good input impedance matching over the $22 \mathrm{GHz}$ to $28 \mathrm{GHz}$ frequency range. The simulated gain for the the antenna array with the three Wilkinson power dividers is 2.5 dB lower than the gain obtained with the configuration in Figure 15a. This is due to the insertion losses (0.8 dB per divider) of the three dividers and the metallic losses caused by the microstrip transmission lines to interconnect them to the antenna elements.

As expected, the minimum side lobe rejection level is $15 \mathrm{d}\mathrm{B}$. The simulated radiation efficiency spans from 50 to 55% in the frequency range of interest. Such values are lower than typical PCB-based designs. This is mainly due to losses in the interconnects, which are printed with silver ink that has lower conductivity than the copper used in PCB technologies.

4.2. Measurements

For comprehensive characterization, 24 antenna arrays have been prototyped over 4 sheets by the CEA-LITEN laboratory. Figure 18 shows the real size of an antenna array with its connector beside a 50-cent coin. 24 antenna arrays have been printed by the CEA and equally distributed over 4 sheets.

The gain in the main direction of radiation and the matching of each array has been measured with the same test bench as for single Quasi-Yagi-Uda antenna elements. The results plotted in Figure 19a,b show the mean of measured gain and its standard deviation for each of the 4 sheets. The measurements results compared with simulations show a good agreement for the four printed sheets.The maximum deviation between measured and simulated gain does not exceed $1 \mathrm{d}\mathrm{B}$ between 24 and $25 \mathrm{GHz}$. In addition, the standard deviation of each leaf is less than $0.25 \mathrm{d}\mathrm{B}$ (see Figure 19b).

By adopting the same approach, $S_{11}$ measurements results were plotted in Figure 20. The results plotted in Figure 20a,b show the mean of measured $S_{11}$ and its standard deviation for each of the 4 sheets. As for the Quasi-Yagi-Uda antenna element higher dispersion on this parameter can be observed. Still, in this case, a little frequency shift between measurements of different antennas causes higher variations in $S_{11}$ than for gain versus frequency values. The measured matched bandwidth is very close to the simulated one and it ranges from $23.5 \mathrm{to} 27.5 \mathrm{GHz}$. This results shows high reproducibility of the screen-printing process. Sheets 1, 3, and 4 show almost the same results; however, only sheet 2 shows a slightly higher average gain than others. Measurements of thickness of the layer with a mechanical profilometer and ink conductivity measurements have shown that a slight increase in the thickness of the printed patterns may cause this shift. Indeed, thickness measurements have shown that for sheet 2, the printed patterns have a thickness of 6 µm compared to 5 µm for the other sheets. Retro simulations confirm this trend.

Radiation pattern measurements in both the E and H planes have also been performed. The results plotted in Figure 21 show very good agreements with simulations results. The obtained results exhibit an antenna beam-width of 28° and side lobes rejection in the E plane of, respectively, $15 \mathrm{d}\mathrm{B}$ at $25 \mathrm{GHz}$ and $20 \mathrm{d}\mathrm{B}$ at $27 \mathrm{GHz}$. The maximum front-to-back ratio is $-21 \mathrm{d}\mathrm{B}$. A deviation of $5 \mathrm{d}\mathrm{B}$ between simulated and measured results may be explained by the employed measurement setup and alignment errors. In addition, for alignment accuracy investigation, geometrical measurements have also been performed. The most important disagreement between simulated model and printed prototypes show a misalignment between top and bottom layers up to $150 \mathrm{\mu }\mathrm{m}$. This issue is due to alignment imperfections that appear when the sheet has to be turned over to screen print the second side. Comparison of measured performances and simulation results allows us to conclude that such mismatch does not impact the proposed antenna and antenna array behavior.

5. Conclusions

In this paper, a 1 × 4 Quasi-Yagi-Uda antenna array has been designed and fabricated on a low-loss transparent substrate using a cost-effective screen-printing process. The screen-printing process has been optimized to achieve a resolution of $150 \mathrm{\mu }\mathrm{m}$ with a $5 \mathrm{\mu }\mathrm{m}$ printed layer thickness. The feeding network of the proposed 1 × 4 Quasi-Yagi-Uda antenna array uses three Wilkinson power dividers whose 100 $\mathrm{\Omega }$ isolation resistors are implemented using carbon paste. A ground plane optimization technique has also been performed to reduce gain variations across the bandwidth and limit the amplitude of the side-lobes in the E plane. A maximum realized gain of $9.8 \mathrm{d}\mathrm{B}$ with a variation of less than 1.3 dB in the frequency band, from 24.25 to 27.5 GHz, has been achieved for a single Quasi-Yagi-Uda antenna element. However, the 1 × 4 Quasi-Yagi-Uda antenna array exhibits a maximum measured gain of 12.8 dB and good impedance matching over the 22 to 28 GHz frequency band (5G band). The measurement of the radiation patterns shows that the adopted topology allows for maintaining the secondary lobes of the radiation pattern at −15 dB in the H plane and below −10 dB in the E plane. The designed screen-printed Quasi-Yagi-Uda antenna can be employed in highly complex communication systems, such as massive MIMO architectures, commonly used in 5G wireless systems. This is thanks to its high gain, wide bandwidth, planar structure, and low-cost manufacturing. Moreover, the antenna array feeding network uses high-performance Wilkinson power divider circuits, which confirm that the adopted manufacturing process allows for printing high-performance passive circuits, paving the way toward designing Butler matrix-based beamforming networks, including hybrid couplers and crossovers for two-dimensional (2D) beam steering operations commonly used in 5G technologies.