Phased antenna arrays (PAA) are essential for mm-wave communications [1] due to their high directivity and beam steering capability. Low-loss beamforming networks with high power handling capability become critical for applications such as backhauling [2], high-speed train-ground communication [3], and satellite communication systems [4]. Feed networks of PAAs typically employ phase shifters (PSs) that are often based on CMOS [5], GaAs [6], and SiGe [7] technologies. However, the losses in these technologies are high [8] with typical power handling being on the order of 10 dBm. Different approaches such as MEMS-based PSs have been reported with lower insertion losses (ILs) (1-5 dB) [9], [10], [11] with power handling capability ∼30 dBm. More recently, beam steering has also been demonstrated with beamforming ICs [12], [13], [14], which typically support multiple antennas with active amplification and phase shifting capabilities. The output power is typically around 10 to 15 dBm. Aside from power handling and losses, PSs are also challenging to employ due to the significant increase in hardware complexity associated with their control bits and bias lines.
Therefore, alternative beam steering approaches have also been proposed to reduce complexity. For example, beamforming networks such as butler matrices can perform beam steering with only switches, but at the expense of a large device footprint and by sacrificing the precision in the scan direction. Combinations of such switched beamforming concepts (butler matrices [15], [16] and lens antenna subarrays [17], [18]) with PSs have also been considered to reduce complexity, but usually at the expense of side lobe performance.
Power handling capability, loss, and control complexity concerns motivated investigation of microfluidics-based device reconfiguration as an alternative technique. In microfluidics based reconfiguration, liquid metals [19], [20], [21] and dielectrics [22] are typically introduced to the vicinity of the device to reconfigure the RF response. However, utilization of liquid metals has challenges associated with actuation [23] and liquid metal oxidization [22]. In [24], the lossy solution of NaOH alleviates the oxidation problem but degrades the RF performance. Refs. [25], [26] have utilized an integrated actuation mechanism with piezoelectric disks and replacing the liquid metal functionality with selectively metallized plates (SMP). SMP was used in [27] to demonstrate a reflection type X-band PS; but, actuation was implemented with external and costly pumps, while reconfiguration distance/time was large (9.5 mm/1.95 s).
The use of microfluidics-based reconfiguration for beam steering has been limited to the SMP technique and solely based on switched beams. [23], [28], [29] demonstrated mmwave beam steering by repositioning an SMP based antenna element at the focal plane of a dielectric lens and achieved a beam scanning range of ±30 • . Later on, [25] significantly increased the speed by using SMP as a multi-throw switch.
More recently, in [30], we have introduced a microfluidically reconfigurable slow-wave PS (MRPS) that is much more compact and operates at 28 GHz mm-wave band with integrated actuation. As compared to the reported dielectric liquid-based phase shifters [31], [32], MRPS has been shown to provide a significant phase shift with small IL (a figure of merit of 170 • /dB as opposed to ∼ 10 • /dB), high power handling (up to 5 W), and improved reconfiguration speed (50 ms).
MRPS has never been employed in the operation of mm-wave beam steering antenna arrays. For practical operation, it is important to control multiple MRPS devices through a single piezoelectric actuator. The phase shift provided by the MRPS is analog, providing continuously variable phase shifts. Since single actuator can displace SMP of all MRPS devices same amount, the MRPS devices should be non-identical and customized to provide a continuous progressive phase shift distribution across the antenna elements. Hence, it is necessary to develop and employ accurate circuit models in the design to account for the motion dependent RF parasitics of the MRPS unit cells. Consequently, this manuscript advances the MRPS concept introduced in [30] in following ways: 1) a new parametric circuit model is introduced which is capable of representing the RF response of the MRPS unit cells by taking account the RF parasitics that alter with unit cell shape variations; 2) a beamforming network design consisting of customized MRPS devices is introduced that can provide progressive phase shift across the devices while using a single actuator; 3) a mm-wave microfluidically reconfigurable phased antenna array (MRPAA) is designed, constructed, and characterized for the first time. Section II details the operational principle and phase shift requirements for the proposed MRPAA. Section III introduces a new circuit model (which is not available in [30]) and subsequently employs it in the design of the beamforming network that will consist of multiple MRPS devices. Specifically, a N = 4 element 28.5 GHz MRPAA is designed to perform continuous beam steering in the range of ±30 • with an SMP actuation range of ±100 μm. Section IV presents fabrication details. Experimental verification provided in Section V shows that the MRPAA achieves >5.6 dBi realized gain and reconfiguration time of 75 ms with an expected power handling capability of 10 W.
Fig. 1 shows the top view of the MRPAA which consists of three main design sections: 4-to-1 power divider, microfluidically reconfigurable beamforming network, and antenna elements. The antenna element spacing is taken as d = 5.4 mm which corresponds to half wavelength of free space (λ 0 /2) at 28 GHz. Fig. 2(a) depicts the exploded 3D view of the MRPAA. The substrate stack-up is shown in Fig. 2(b) and it is identical to the one presented in detail in [30]. The SMP is encapsulated inside a photolithographically manufactured microfluidic channel made of SU8 side walls and sealed with 1 mm thick fused silica) to capacitively and periodically load the microstrip lines patterned over the PCB. As in [30], the separation between SMP and PCB is kept constant by coating the PCB traces with a 5 μm thick Benzocyclobutene. This coating is spun and cured on top of the PCB structure and enforces a minimum constant separation between the SMP and PCB metallizations and potentially lowers friction. The idea is for SMP to touch the spun BCB on PCB at a constant separation, however, fabrication tolerances cause an additional gap between the BCB and the SMP metal traces. Microfluidic channel walls are 242 μm in total height to accommodate 203 μm thick SMP, 17 μm thick SMP metallizations, 5 μm BCB bonded to PCB metallizations, 17 μm thick PCB metallization, and 5 μm FC-40. FC-40 thickness is due to channel height non-uniformity in the fabrication process as previously reported by our group in [17]. SMP and PCB metallization overlap areas are reconfigured by repositioning the SMP using liquid FC40 ( r = 1.9, tan δ = 0.0005), which has also been used in numerous prior works due to its low viscosity and dielectric loss [32], [33]. Compression and decompression of the fluid reservoirs under the device ground plane by a piezoelectric disk are used to actuate FC40 and reconfigure the SMP. The microfluidic channel shape constraints SMP displacement to 200 μm, while different piezoelectric disk actuation voltages can be used to generate a different SMP position within this range. The PCB also contains grounding pads parallel to the microstrip lines so that SMP traces can create a capacitive loading from the microstrip lines to the ground as in [30].
SMP motion along the x-axis changes the overlap area between the SMP metallizations and the microstrip lines. Therefore, a variable capacitive loading is realized to implement a slow-wave wave phase shifter [30]. If the motion of the SMP along x-axis relative to its reference position is denoted by s, then the MRPSs within the beamforming network should provide progressive phase shift of β. Based on the coordinate system shown in Fig. 1, the progressive phase shift of β will imply a beam steering angle of θ s , φ s that is determined from β = -kd sin θ s cos φ s (1) where k=2π/λ 0 and φ s is 0 or π for steering in the x-z plane, which is the H-plane of the antennas. The limit of the SMP motion range s = s max then determines the maximum scan angle θ s = θ max . Different than [30] that developed a stand-alone MRPS, the progressive phase relationship must be established across multiple MRPSs to utilize a single SMP and actuator to perform the beam steering. This will be achieved in below by customizing the SMP traces of each MRPS within the beamforming network through the utilization of a new circuit model.
Keysight Advanced Design System (ADS) Momentum suite is employed for the design of the layout. Ansys EDT HFSS is used to verify the performance under a realistic case of finite size ground plane, substrate, and microfluidic channel walls.
Wilkinson power dividers are cascaded to perform the power splitting [34], [35], [36], [37]. It is implemented on the PCB with no other substrate materials on top. The divider is well matched at 28 GHz with a return loss (RL) better than 15 dB and a worst-case IL of 0.5 dB above theoretical. At center frequency (28.5 GHz), the divider exhibits an RL better than 20 dB and a worst-case IL of 0.4 dB above theoretical. The physical length of quarter-wave transformers is 1.91 mm, and 100 SMD resistors are from CH02016-100RGFT from Vishay.
The design process starts with selection of desired scan range, SMP actuation range, and unloaded microstrip line impedance. The design example presented in this manuscript sets θ max as 35 • which implies a maximum progressive phase shift β max of 103 • . Denoting the phase shifts of the MRPSs within the beamforming feed network at SMP reference state zero
where s denotes the SMP displacement with respect to its reference state as shown in Figs. 1 and 3(a).
Controlled impedance line tool of ADS is employed to determine the width of the microstrip line within the substrate stack-up as w = 0.12 mm. This corresponds to 78 characteristic impedance, which will be lowered with capacitive loading of the SMP during phase shifting. SMP actuation range ±s max is constrained to be about 80% of the microstrip line width as in [30] resulting in s max = 100 μm. 78 implies a worst-case reflection of ≈ -10 dB for 50 terminations. From the electrical length and impedance of the line, its L (unit length inductance) and C (unit length capacitance) are calculated as 486.72 nH/m and 79.21 pF/m, respectively.
The layout of the unit cell metallization is shown in Fig. 3(a). The SMP layout consists of two main sections that overlap with different traces of the PCB, as shown in Fig. 3(b). The trapezoidal shape and varying base (w b ) widths across different MRPSs within the beamforming network is a major difference with respect to the design in [30]. The RF parasitics that is tied to this shape must be carefully accounted for in modeling to realize the desired progressive phase shifting. Overlapping area between the SMP metallization and the ground pad on the PCB (A gnd ) forms capacitor C gnd . The ground pad is 0.5 mm wide and carries a 0.3 mm diameter via. Ground pads are placed on both sides of each microstrip line. The distance between the pads and the microstrip lines is 0.35 mm to avoid changes in |S 11 | of the stand-alone line at 28 GHz. The section of the SMP metallization that does not overlap with the PCB line and ground pad exhibits a capacitance to the PCB ground ranging from 0.006 pF to 0.013 pF, which is relatively small in comparison to C gnd and therefore negligible to the RF performance. Repositioning the SMP over the microstrip line generates a capacitive loading on the microstrip line that can be expressed as 1/C added = (1/(C ov + C par )) + (1/C gnd ) with its equivalent circuit model shown in Fig. 4(a). The value is therefore dependent on the overlapping area between the SMP metallization and the microstrip line on the PCB (A ov ) forms the capacitor C ov , the RF parasitic capacitance C par , and C gnd . A gnd does not change with the motion of the SMP due to SMP metallization being 0.1 mm larger than the ground pad in SMP rest position s = 0. A gnd is made sufficiently large (0.32 mm × 0.5 mm) to generate C gnd = 0.6 pF. Since C gnd is significantly larger than C ov (0.0007-0.036 pF) and C par (0.0021-0.012 pF), C added is approximated as
The trapezoidal shape of A ov allows customization across the different phase shifters of the beamforming network. At a specific SMP position, A ov can be parameterized as a function of the top (w t ) and base (w b ) widths of the trapezoid and s. Modification to w b will not only alter the value of C ov , but the rate at which it grows as function of s. w t is set as 20 μm. Selecting w t as small as possible minimizes the parasitic capacitiance at s = 0.
The unit cell length is set as 0.3813 mm which implies a λ g /15 size (λ g is the wavelength for the stack-up shown in Fig. 2(b)). Per unit cell inductance L ML UC and capacitance C ML UC of the microstrip line can therefore be determined as 0.1856 nH and 0.0302 pF, respectively. C ov is related to A ov as
where o is the vacuum permittivity, r is the relative permittivity of the BCB, and d is the distance between the PCB traces and the metallization on the SMP (i.e. BCB thickness + gap).
A ov can be expressed as
where w e is the bottom edge of the trapezoidal area that overlaps with the microstrip line and depends on the s and w b . At s = 0 A ov is not zero due to the l min = 10 μm overlap as shown in Fig. 3(a). l min is useful to maintain the SMP parallel to xy plane at any position including s = 0. If no overlap of traces occurs, SMP may slightly tilt inside the channel, and this may prevent or slow-down the SMP motion. From the geometry, w e can be expressed as
Substituting ( 5) and ( 4) into (3) results in C ov (s, w b ) = P 01 + P 02 + P 03 + P 04 + P 05 + P 06 (6) where Our recent work on the stand-alone MRPS [30] showed the importance of C par in modeling. However, in [30], C par was constant for any SMP position. On the other hand, the trapezoidal shape leads to a C par that is dependent on the SMP position and trapezoidal area shape, which is a direct consequence of w e being a function of s and w b . C par can be obtained from ADS or full-wave simulations by fitting the phase response of the unit cell to the lumped element circuit model for many s and w b combinations. Typically, the circuit model extraction is performed with the simulation of multiple unit cells to account for mutual coupling effects, which is selected as 15 in our work corresponding to a full wavelength. This makes simulation of all w b and s combinations time consuming to pursue a design. C par can be obtained from
where C added represents the value of overall shunt capacitance added to the microstrip line section of the unit cell which is series L ML UC and shunt C ML UC . We extract C par from simulations of few states of s ∈ {s 1 , s 2 , . . . s Q } and w b ∈ {w b1 , w b2 , . . . w bR }. We employ a nonlinear least-squares curve fitting approach with s ∈ {0, 0.025, 0.050, 0.075, 0.100}, and w b ∈ {0, 0.05, 0.10, 0.15, 0.20} obtaining a 25 element discretized set of C par (s q , w br ) from (7). We assume that C par can be approximated with a second order quadratic equation F ( c, s, w b ) as
where c = [c 0 , c 1 , . . . , c 7 ]. The nonlinear least-squares curve fitting solves for c from the minimization problem min c (q,r) F c, s q , w br -C par s q , w br
The minimization problem in ( 9) is solved with MATLAB function lsqcurvefit. Randomly selected 80% (i.e. 20) of C par (s q , w br ) are used in solution of (9) and remaining 20% (i.e. 5) is used to validate (8). This process is repeated 10 times and phase error of 15 unit cells associated with the difference of actual C par (s q , w br ) and its prediction F ( c, s, w b ) is plotted as in Fig. 4(b). It is shown that selected second order form of F ( c, s, w b ) is satisfactory with phase errors being constrained within {-2.5} • and 5 • . If in this step the errors are quite large, the presented procedure can be repeated with a higher order polynomial for better fit. The c values that gives the best fit are [0.0008, 103.65, 18.95, 0.0054, 0.001, 0.001, 0.0068, -0.0048], with a contributions vector c[%] = [11, 75, 14, 0, 0, 0, 0, 0]. The contribution vector is calculated as the percentage of the output caused by each parameter at s max . Therefore, the system can be further expressed with a simpler expression where contributions below 0.1% are rounded to zero.
C par ≈ 0.0016 + 103.65s + 18.95w b (10) The design continues with estimating [w b1 , w b2 , w b3 , w b4 ] to obtain the desired progressive phase shift (s) between 4 phase shifters as [φ 1 , φ 2 , φ 3 , φ 4 ] = N UC [φ 0 , φ 0 + , φ 0 + 2 , φ 0 + 3 ], where φ 0 (s)=φ 1 (s)/N UC represents the phase delay of the unit cell of the first phase shifter and φ 1 (0) = φ 2 (0) = φ 3 (0) = φ 4 (0) is approximately satisfied due to the identical usage of w t in design of all phase shifters. In our design process, we first set w b1 as 0.02 mm, which is same as w t resulting in φ 0 (s max ) = -32.01 • . Next we set w b4 as 0.16 mm which provides a C added = 0.0472 pF at s max and maintains the Bragg frequency above 80 GHz as in [30]. This results in φ 0(smax) + 3 φ(s max) = -42.09 • . φ(s max) is therefore calculated as -3.36
• . The desired [φ 1 , φ 2 , φ 3 , φ 4 ]/N UC at s max is then estimated as [-32.01 • , -35.37 • , -38.73 • , -42.09 • ].
To facilitate the selection of w b2 and w b3 that will result in φ 2(s max) = -35.37 • and φ 3(s max) = -38.73 • , the circuit model is utilized to generate a phase surface plot of the unit cell using (7) with s and w b sampled in 0.01 mm steps (see Fig. 5(a)). From this surface plot, w b2 and w b3 are found as 0.065 mm and 0.110 mm, respectively.
The number of unit cells N UC required to get a maximum progressive phase shift of 103 • is calculated as 103 • / φ(s max) = 31. A safety margin of 2-unit cells is also introduced into the design to accommodate for the maximum phase error observed for the model (±5 • ≈ 2 φ ) in Fig. 4(b). The simulated phase from the circuit model of 33 cascaded unit cells for different s using [w b1 , w b2 , w b3 , w b4 ] = [0.02,0.065,0.11,0.16] mm is presented in Fig. 5(b) (solid lines). The simulated phase values [φ 1 , φ 2 , φ 3 , φ 4 ] are then compared against ideal response (dotted lines) and show small errors of (2-5) • in PS3 and PS4. The ideal response is calculated by applying identical phase difference φ = φ 2 -φ 1 to all the remaining phase shifters as [φ 1 , φ 1 + φ , φ 1 + 2 φ , φ 1 + 3 φ ]. The figure shows that progressive phase delay is not a linear function of s. This is due to the trapezoidal shape of A ov . Selecting a different shape such as a trapezoid with non-linear edges can be investigated in future to achieve linearity in this response. The layout implemented in the comparison is shown in Fig. 5(c), showcasing the unit cells of four phase shifters each using the w b values determined in the previous section. Additionally, For scanning in both φ s = 0 and π elevation cuts, the SMP metallizations are mirrored and placed into the other side of microstrip lines. Therefore, a single actuator becomes satisfactory due to simultaneous actuation of all unit cells s is varied from -0.1 to 0.1 mm. This results in the final SMP metallization layout shown in Fig. 6(a). mirrored metallizations increases the simulation (solid lines) vs ideal phase response (dotted lines) error previously shown in Fig. 5(b) by 25 • at s max . This discrepancy is related to the loading/unloading of the metallized plates for s = ±0.01 range. For instance, as s reconfigures from 0 to 0.01 mm, one side of the metallization loads the line while the other side of the metallizations unloads the line.
The patch antenna element of the array has a footprint of 2.68 mm × 3.5 mm with insets that are 1 mm long and 0.1 mm wide. The patch is designed to resonate at 28 GHz with a 700 MHz of |S 11 | < -10 dB bandwidth, 4.05 dB realized broadside gain with 80 • half power beam width (HPBW) in the H-plane. The inter-element separation of the array is λ 0 /2 = 5.4 mm. The layout of the antennas is shown in Fig. 7(a). The broadside realized gain of the stand-alone array is 11.55 dBi with an antenna efficiency of 87%. This simulation only accounts for the patches and do not include the feed lines, beamforming network, power splitter or connector losses. The stand-alone array exhibits a half-power beamwidth of 22 • . To compare beamformer simulations against ideal response, the antenna array is fed with both simulated and ideal phase shift from the beamformer section at s = 0.1 mm (which is the worst-case error in Fig. 6(d)). The comparison is presented in Fig. 7(b), where the error incurred by mirroring degrades the maximum scan (at s max ) by 5 • when compared against the same antenna array feed ideally. The final layout of the MRPAA (which integrates the power splitter section, the beamforming network section, feed lines, and the antenna elements) is presented in Fig. 1. The total footprint area is 40 × 45 mm 2 , including the RF edge connector. The entire layout is simulated in Ansys HFSS, and results are used to compare against measurements in the following sections.
The losses of the entire beam steering antenna array system can be analyzed through simulation of individual components. The edge launch connector is expected to introduce a 0.5 dB insertion loss at center frequency of 28 GHz, which translates in a power loss of 5.6%. As indicated in Section III-A, the power divider exhibits a worst-case IL of 0.4 dB above theoretical, marking a power loss of 8.8% between port 1 and reference plane 1. The power loss between reference plane 1 and reference plane 2 is 1.5 dB per branch, which represents a loss of 30%. The loss introduced by the microstrip lines connecting the reference plane 2 and the antennas introduce 0.35 dB loss at 28 GHz (7.8%). Combined, all the losses account for a drop of 46.6% or 2.7298 dB. Since simulated realized gain of the antennas at maximum reconfiguration is 9.8 dBi, the expected realized gain of the antenna (at max reconfiguration position) with cascaded components (including connector loss, power divider, Beamforming network, and antenna feed lines) is therefore 7.08 dBi.
The PCB and SMP are patterned using standard photolithography techniques; hence, these details are omitted for brevity. The microfluidic channel sidewalls are constructed with photoresist SU8-2075. The 0.203 mm thick PCB is first bonded with a 2 mm thick structural substrate to eliminate the process variations due to RO4003 C flexibility. Subsequently, it is spin-coated following the procedure established in [30]. This step is followed by the preparation of microfluidic channel walls using SU-8 developer and rinsing of the PCB substrate.
The channel sidewalls are bonded with the fused silica using the adhesive bonding technique described in [17]. Once the channel is sealed, a fluid circulator structure that is 3D printed is attached to the bottom side of the PCB + structural substrate. The actuation is similar to [25] using vertically stacked reservoirs for achieving compact size. However, the reservoirs in this investigation are placed at the backside of the PCB in a stacked configuration and constructed from ANYCUBIC 3D printing UV sensitive resin using a Stereolithography (SLA) printer. The final prototype is shown in Fig. 8(a), where the structural top fixture (used to reduce device deformation during handling) is separated to facilitate the visualization of its components. Fig. 8(b) shows the setup utilized to characterize the realized gain of the prototype.
As shown in Fig. 2(b), the reservoir structure is 18 mm thick. The larger size of the 3D printed holder is to achieve stability during testing, which may be compromised due to cable weights on connectors. This allows for a conservative design for reduced deformation risk. Further thickness reduction of the structure requires additional tests. The piezoelectric disk compresses/decompresses the bottom reservoir and circulates the dielectric liquid. Reversing the polarity of the actuation voltage flips the state of the reservoirs and, therefore, the position of the SMP between -100 μm and 100 μm. The device also exhibits an internal cavity represented as the absence of resin in Fig. 2(a) that is 4 mm tall to allow for free movement of the inner membrane. Both membranes are 0.2 mm thick and made of cured PDMS for increased flexibility. The inner membrane is bonded between the two sections of the reservoirs structure using liquid PDMS as a reversible bonding agent. The most external membrane is reversibly bonded to the bottom section of the 3D printed reservoirs. The reservoirs structure is then cured at room temperature screws. A T216-A4NO-05 piezo disk from Piezo.com with a thickness of 0.41 mm and a rated drive voltage of ±180 V and a 4 nF capacitance is bonded to the external membrane. The reconfigurable footprint of the MRPAA is 8.38 mm × 9.64 mm, while the total footprint, including connectors and piezo actuation, is 40 mm × 45 mm.
The reconfiguration velocity of the prototype is assessed utilizing an AmScope Microscope Digital Camera Model MU300, which operates with a constrained temporal resolution of 25 milliseconds between successive frames. Applying a peak voltage of 160 V pk to the piezo disk enforces a System Maximum Performance (SMP) velocity, resulting in a cumulative displacement of 0.2 mm over three frames. This cumulative displacement equates to a reconfiguration time of under 75 milliseconds. This achievement signifies a substantial advancement by a factor of 40 compared to earlier efforts documented in [27]. The experimental configuration utilized to activate the mechanism involves a square wave signal generator operating at approximately 7 Hz. This frequency choice ensures a period of 150 milliseconds and, consequently, enables a complete reconfiguration every half period, corresponding to 75 milliseconds. In alignment with these circumstances and utilizing the piezo model T216-A4NO-05, which possesses a capacitance of 4 nF and a Dielectric Loss Factor of 0.02, the power delivered to the piezo remains below 3 mW. of the realized gain at three states (s max , s 0 , -s max ). The data included in the set corresponds to H-plane beam scanning only and has been compensated for connector loss. The measured realized gain is 6.8 dBi for the extreme case of beam scanning angle = 30 • . The compensated realized gain value is between 0.2 dB from simulations (7.03 dB) as shown in Fig. 9(d). It becomes 5.6 dBi when the beam is scanned towards 0 • . Similar to the previous work in [17], the large ground plane size in E-plane causes pattern ripples and tilting, causing the broadside gain in H-plane to appear lower in comparison to when antennas are on a smaller ground plane. The loss in broadside gain is also increased by distortions in φ near s = 0 mm. Fabrication errors in the SMP metallizations and alignment of the plate might be the main cause for increasing this distortion in the array factor by altering φ among the PS. As s changes toward the edges, the φ exits the transition zone and better resemble ideal array factor, showing higher gain. Fabricated prototype has a measured scan range between -20 • to 30 • . The gap between expected -30 • to measured -20 • beam at max actuation can be related to fabrication errors associated with over etching, photolithography process misalignment and/or SMP dicing resolutions. The missing 10 • scan range corresponds to a state of s = -70 μm instead of -100 μm. Nevertheless, the measurements prove the proposed beamforming network concept and the design challenging due to SMP dragging issues within the microfluidic channel. A frequent reset of the SMP position was necessary to relocate the SMP to its desired location. The longest time a position was sustained in the lab was 70 seconds, after which, an air bubble inside the reservoirs started moving and the shifting the SMP. The air bubble could not be removed in several liquid loading attempts, which may be due to the cylindrical shape of the reservoirs. Although 70 seconds is
TABLE 1. Performance Comparison With Other Radiation Pattern Reconfigurable Antenna Arrays
quite a large time frame for mm-wave beam-steering arrays, our pattern measurements in anechoic chamber take a long time (>2 hrs) to obtain. The prototype was also tested for reliability, remaining operative after continuous operation of 10 days with actuation of 2 × 0.2 mm per cycle at a frequency of 10 Hz, for a total of 9 million cycles. The device has been characterized for power handling capabilities up to 1 W (1 W represents the maximum RF power achievable in our lab at the time). Measurements at 1 W show an average surface temperature of 25.3 C • vs 27.01 C • in simulations (using a film coefficient of 20 W/m2K) with a simulated internal temperature of 28.6 C • . Simulations predict a maximum power handling capability of 15 W where the internal temperature reaches the boiling point for the FC-40 liquid.
Finally, Table 1 compares the MRPAA presented in this manuscript against recent state-of-the-art radiation pattern reconfigurable antenna arrays utilizing beamformers. The table shows that the presented MRPAA has good performance with the lowest hardware components complexity between equivalent architectures, better power handling capabilities in addition to being the first microfluidically actuated phased antenna array at mm-wave bands.
This article demonstrated a microfluidically reconfigurable phased antenna array (MRPAA) operating at 28 GHz. It is based on a beamforming network consisting of multiple microfluidically reconfigurable phase shifters (MRPS) that can be all controlled with a single actuator thanks to their customized layouts. The MRPAA operates at 28 GHz and achieves beam steering as is actuated within a 200 μm reconfiguration range. MRPAA exhibits a measured realized gain between 5.6 dBi and 6.8 dBi for the scanning range [-20 • , 30 • ] in the H-plane as the SMP is repositioned inside the microfluidic channel. Therefore MRPAA offers the possibility for a fully microfluidically actuated phased antenna array (1D) operating at 28 GHz or 38 GHz and with spatial adaptation capability for an improved signal-to-interference ratio. However, this technology is better suited for high-power applications, such as backhauling, due to the reconfiguration times inherent to mechanical actuation. The MRPAA can be extended to large planar arrays (1D) with appropriate design considerations, such as different overlapping profiles and different numbers of overlapping elements per branch to overcome the maximum capacitive coupling limit due to overlapping area constraints.This concept can also be utilized in a 2D antenna array 4 × 4 by designing a secondary beamforming network to enforce phase shift on the secondary axis of the phase profile. Future work may consider improvements in design and fabrication processes to extend the scan range and frequency range by reducing the metal thickness using electrodeposition and implementing higher dielectric constant materials.
VOLUME 3, NO. 4, OCTOBER 2023
VOLUME 3, NO. 4, OCTOBER 2023