1. Introduction
Antenna arrays are a popular choice for point-to-point wireless communications due to their ability to concentrate radiated power enabling high directivity [
1]. To increase their functionality, they can also be made reconfigurable. Reconfigurability is a broad term which can refer to several concepts, e.g., electrical beam steering, polarization control, pattern diversity, beamforming, beamsplitting, and frequency reconfigurability [
2,
3,
4,
5]. Examples of typical reconfigurable antenna array systems are phased arrays, transmitarrays/reflectarrays, metasurfaces, and frequency selective surfaces (FSS). In recent years, reconfigurable intelligent surfaces (RIS) have garnered more interest and offer reconfigurable properties similar to those in the other aforementioned systems as well as this work [
6,
7,
8]. RIS can be considered as a reflectarray and metasurface hybrid [
9]. The main difference with RIS is its distance to the transmitting source and receiver as it is typically located between the two to create a smart radio environment which enhances the signal propagation between the two points [
10,
11]. For future antenna array systems, hotspot-mediated control [
12] shows promise in being the next possible step in antenna array development as it provides reconfigurability through plasmonic resonances instead of semiconductor components. This removes the frequency limitation imposed by the said components and also enables the nanoscale antenna designs.
Among the above reconfigurable properties, this paper focuses on the electrical beam steering and also presents some results for polarization control. The former is a highly desired property as it removes or reduces the need to move a system mechanically. Furthermore, precise beam steering is critical in long-distance communications as a few degrees of variation in the steering angle can lead to the transmitted signal missing the receiver and causing transmission failure in highly directive systems, i.e., narrow main beam. Practically, beam steering can be implemented by embedding active components, e.g., PIN diodes, varactors, microelectromechanical systems (MEMS), and monolithic microwave integrated circuits (MMIC), to patterns created on printed circuit boards (PCB). Another option is materials whose material properties are altered with a voltage bias, such as a liquid crystal and graphene.
As mentioned in [
13], PIN and varactor diodes are preferred to radio frequency (RF) and MEMS switches due to their lower cost and higher reliability. Therefore, the design of this paper also employs PIN and varactor diodes to achieve reconfigurability. The PIN diodes are used to implement polarization control and the varactors are utilized in beam steering.
Even though this paper and other works utilized varactors [
14,
15,
16,
17] for implementing beam steering, it was also accomplished with PIN diodes, as was demonstrated in [
18,
19,
20,
21,
22]. Interestingly, a combination of both components was utilized in [
23,
24,
25]. With regard to beam steering, the main difference between these two components is the phase profile they provide. PIN diodes are discrete devices meaning that they are used in ON or OFF states, i.e., two phase states. Varactors on the other hand can be considered as tunable capacitances and provide continuous phase control. However, references [
22,
24] demonstrated that the continuous phase can also be discretized. Nevertheless, these are important factors to consider since, in order to perform the aforementioned beam steering, it is required to generate a progressive phase difference between the antenna elements as was discussed, for example, in [
13]. If the phase is not precisely generated, it introduces a phase error. With PIN diodes, this phase error is commonly called a quantization error due to its discrete nature. The effect of the quantization error was studied in [
26] where the authors reported that the phase error generated strong side lobes and a grating lobe. Additionally, an increase in side lobe levels (SLL) was also observed. More importantly, the beam steering became less accurate and the main lobe gain degraded especially at large steering angles, i.e., high scan loss. Therefore, implementations with PIN diodes have an inherent disadvantage compared to varactors. However, the strength of the PIN diode is in its simple characterization as it only has the two states. For the varactor, the characterization is not only difficult due to its continuous phase profile but it is also strongly nonlinear. In the other works using varactors for beam steering [
14,
15,
16,
17,
23,
24,
25], there was little discussion about the characterization. Instead, references [
14,
23,
25] only mentioned the capacitance range of the varactor which is not sufficient to create an accurate phase profile because inductive components also affect the phase, as mentioned in [
17]. These works successfully performed the beam steering but there was a lack of discussion about its accuracy. Furthermore, looking at Figure 15 in [
14], it was visually confirmed that some steering angles do not have their peaks exactly at the claimed angles. This appears to have been rectified, but not discussed in the following work [
23], using the same varactor.
Reis et al. [
17] characterized the phase of a single FSS unit cell for beam steering. However, in the far-field measurements, a maximum steering error of 16
was reported at the steering angle of 45
in the azimuth direction. The authors mentioned the need to compensate for angle mismatches in angles above 15
with lookup tables. Nevertheless, looking at the gains and SLLs, there is strong variation in these values between the different steering angles hinting towards a strong phase error, as per earlier discussion.
Improved results can be achieved by characterizing each unit cell individually as was performed in Lau et al. [
15]. However, the characterization method was not mentioned. Additionally, the steering accuracy was not discussed and it was visually confirmed that the steering angles of >
in Figures 13–16 are not exactly at the claimed angles. There were also gain variations between different steering angles.
Similarly to the above, Nicholls et al. [
16] characterized each unit cell individually. This was performed with a near-field probe and the information of each unit cell’s phase characteristics was added to a lookup table. Moreover, they provided exhaustive data about the accuracy of each steering angle. The main beam was scanned in the E- and H-planes as well as diagonally. For example, in the E-plane, the maximum steering error was 1.9
and 6.6
in theta and phi directions, respectively. From this, it is evident that even though each unit cell was accurately characterized, accurate beam steering was not a trivial task and there is bound to be some error, even with careful characterization.
Considering the above points, this work proposes a characterization setup which only contains the phase shifters and feeding network of the antenna array design of this paper. The input and 16 outputs are terminated with SubMiniature version A (SMA) connectors in order to measure each unit cell individually with a vector network analyzer (VNA). With these measurements, the operation of the phase shifters and feeding network is confirmed. Moreover, the total phase shift of each unit cell is measured and the results are used to form an average model of the phase shift profile. This is not expected to provide an optimal performance in the beam steering as the objective is to create a single fairly precise model used as a baseline and tolerant to component and fabrication variations. The authors want to emphasize that even though each unit cell is measured individually, these results cannot be used as-is in the antenna array. The reason for this is that the characterization board is not the full system ( antenna array) used in far-field measurements. Therefore, there will be small variations in the etched board line widths and components between the two boards which causes some discrepancy. Hence, the average phase shift model is used instead.
The model is used in far-field measurements to steer the main radiated beam. If there is a discrepancy between the target and measured angle, the beam direction is adjusted by over or under steering the beam in the beamforming calculations until the target angle is reached within an acceptable angle error. The benefit of this approach is that the full antenna array can be pre-calibrated offsite before performing any far-field measurements and the final adjustments are performed later during the radiation pattern measurements. This can be a very beneficial approach in situations where the limited measurement time in an anechoic chamber is available or there is some fault in the board requiring readjustment. This kind of adjustability is not commonly discussed in related works to the best of the authors’ knowledge. As demonstrated later, the approach of this paper is simple and easy-to-use as it requires only changing the steering angle value in the beamforming calculations.
In order to demonstrate the feasibility of the above characterization method, the design and measurement results of a antenna array with electrical beam steering are presented. In addition, the array is also capable of performing polarization control, which will also be shown. The results are analyzed as well as compared to other similar works to highlight the strengths of the approach we employ.
Then, we discuss the contributions of this work. First, we provide a method to analytically calculate the phase shift of each unit cell for beam steering. This does not require the individual characterization of each unit cell as in [
15,
16]. Instead, we create an average model that offers robust beam steering performance. Due to this approach, the beam steering is also adjustable by modifying the steering angle in beamforming calculations, making it simple to re-adjust. Additionally, the method presented herein only requires a VNA instead of a near-field probe. Therefore, this characterization does not require radiation measurement equipment for calibration. Hence, the system can be pre-calibrated and re-adjusted for far-field measurements meaning that this method requires less measurement time in an anechoic chamber. Finally, we also provide insight into different loss sources such as the surface finish and surface roughness through a comparison between simulations and measurements which is not commonly discussed in related works.
This paper is organized as follows. First, the complete structure and its general details are introduced. After this, the structure is broken down into smaller sections followed by the introduction of the power divider and phase shifter network. Then, a more detailed analysis of the phase shifter and the phase shifting is provided from where we move to the antenna. In this section, the polarization control is presented. This concludes the theory part and we proceed to the measurement results. The first measurement setup is the phase shift characterization with a board containing only the power divider and phase shifter network. From this, the full structure far-field measurement results and an analysis are presented. The results are compared to other works after this. Lastly, a conclusion is provided.
3. Phase Shift Characterization
In order to determine the overall phase shift of the
structure, a board containing only the power divider and phase shifter was fabricated, shown in
Figure 9. An SMA connector was soldered to each output of the 16 unit cells as is displayed in
Figure 9b. The unit cells were measured one by one with an Anritsu MS46122B VNA by adjusting the respective unit cell’s bias voltage for varactors from 0 V to 12 V in 1 V steps. The other unit cells were terminated with a 50 Ω load. The voltage control was performed with a third-party control board that was custom made for this work based on specifications provided by us. The measurement was repeated with an identical board to collect sufficient data for an average model of the phase shift at several bias points. The measurement setup is shown in
Figure 10.
The measured total average total phase shift was
with a
difference between the highest and lowest total phase shift. The average phase shift is
higher than in the simulation. This is believed to be due to the additional inductance from the solder that was not considered before. The differences in phase shift between the unit cells is assumed to be due to variations in varactors. For example, in the varactor datasheet [
30], the minimum and maximum capacitances are 0.30 pF and 0.40 pF at 4 V, respectively.
and
results are shown in
Figure 11. In the legend, codes representing each unit cell are shown. The position of each unit cell is mapped in
Figure 11c which represents the phase shifter and power divider layer illustrated in
Figure 2. The average
and
were −10.9 dB and −18.9 dB, respectively. Ideally, the
would be −12 dB, however, due to the losses in the power dividers, phase shifters (imperfect reflection), and non-ideal conductors, the electroless nickel immersion (ENIG) surface finish and surface roughness, there was 6.9 dB of additional loss on average. These surface finish and surface roughness losses are discussed in greater detail in
Section 4. The variation in loss between different bias voltages is due to the changing reflection coefficient of the RTPS. The highest loss is at 1 V and, in this case, the series inductance and capacitance are in resonance, i.e., (
4) is close to zero. In this case, the reflection coefficient is purely based on the internal resistance of the varactor (
in
Figure 4). Additionally, imbalanced power division was observed for unit cells near the feed of the structure, i.e., B21, B22, B31, and B32. This was also observed in the simulation results of
Figure 3, however, the difference between the highest and lowest loss was only 0.2 dB. The higher losses caused by the non-ideal conductors, ENIG, and surface roughness are believed to have increased this difference even more.
To obtain an updated varactor model with the measurement data, a curve fit was performed with MATLAB’s built-in curve fit toolbox by tuning the parameters of (
2) with the minimum square error (MMSE) technique. While tuning, the phase shift at different bias points was calculated with (
3) and compared to the measurement data. This was iterated until the error between the calculations and measurement results was minimized. The result of the curve fit is in
Figure 12, showing good agreement with the measured data. The varactor model parameters obtained from the curve fit are presented in
Table 1. Although these values emulate the phase profile well, they are not expected to be accurate in terms of inductance and capacitance. Therefore, a frequency shift in simulation results is expected when compared to measurements. In future work, the goal is to improve this method to accurately capture the inductance and capacitance values as well. Nevertheless, it is sufficient for beam steering purposes as will be demonstrated in
Section 4.
3.1. Error Analysis
To understand the effect of the average phase shift model to phase error, four different measurement scenarios were performed. Each scenario represents a different beam steering angle of 10
, 20
, 30
, and 40
, respectively. Based on calculations, a different voltage was applied to each column of unit cells in the board. More details about these calculations are provided in
Section 4.1.
The results are shown for 10
and 40
cases in
Table 2 and
Table 3, respectively. The bottom two rows of the board, marked as row 3 and row 4, were measured and compared to each other. The phases are normalized in relation to the first unit cell of each row. Both rows have identical voltages applied to their respective unit cell’s phase shifters. The voltages were confirmed to be correct with a multimeter with a maximum error of 0.05 V. Ideally, this kind of control would generate equal phase shifts at the output of each unit cell of a column. However, there is a discrepancy in the phase shifts due to variations in varactors, as discussed earlier. For the 10
case in
Table 2, the maximum phase error is −4.6
at the unit cell B42 of row 4 which is greater than the maximum error of Row 3, 1.7
.
For the 40
case in
Table 3, the amount of error increases since a wider voltage range is utilized. The maximum error is observed at B31 of Row 4 equaling to 16.3
. The maximum error between the rows is 6.1
, which is a fairly good result considering the use of the average model instead of individual characterization.
Overall, the worst error observed compared to calculations was for the 30 case at B33 which was 22.7. As for the error between rows, the worst case was at a 20 steering angle with an error of 6.5. Since the phase error between the rows is small, the radiated wave fronts of the full antenna array will not cancel each other destructively. Therefore, it is expected that the use of the average model will not significantly degrade the gain of the full antenna array in beam steering. However, the steering angle will not likely be exactly at the expected angle due to the error between the calculated and measured phase value which will require some adjustment in the beamforming calculations.