Metamaterial Arrays with Improved Antenna Isolation and Phase Balance for More
Accurate Bluetooth Direction Finding
- AoA/AoD Basics
- Critical AoA/AoD Array Parameters
- Techniques to reduce the antenna size
- AoA/AoD Accuracy Improvements
- Technology Spreading
- Advantages of Dual-Fed Antennas
Direction Finding Basics
Anetenna arrays are used for angle of arrival (AoA) or angle of departure (AoD) applications. These direction-finding technologies are standardized in the Bluetooth specification 5.1.
With AoA, the angle estimation is based on the received signal phase differences between the array antennas as shown in Figure 1. Here, the spatial phase change of the incident plane wave is projected to a lower spatial phase change in the plane of the array. The incident angle (ϴ) can be calculated from the phase change difference.
With AoD, the signal is transmitted by switching through the antenna array. A receiver with a single antenna determines the direction of the transmitter by observing the phase differences of the incoming signal.
From an RF hardware array perspective, the AoA and AoD RF operation differs only in the RX and TX operation; the antennas and the RF feeding circuitry operate in the same way. Thus, only the AoA array operation is described in this paper. For more information about AoA and AoD operation and angle estimation, see the Silicon Labs white paper, “Bluetooth Angle Estimation for Real-Time Locationing” .
Figure 1. AoA Basic Operation
Figure 2. Two-dimensional (4x4) AoA Array Example with Dual-Fed Patch Antennas
Silicon Labs’ AoA arrays operate by receiving incoming signals and making incident angle calculations separately in two orthogonal polarizations. This approach provides a more robust operation, especially in indoor environments where multipath propagation is typical due to the reflected waves. With this capability, the applied radiators can receive both horizontal and vertical polarizations. Figure 2 shows a 4x4 AoA array embodiment with dual polarized patch antennas.
An AoA array can be one- or two-dimensional. A one-dimensional array can calculate the incident angles in one plane only (e.g., in an azimuth or an elevation plane). A two-dimensional array like the one shown in Figure 2 can calculate the incident angles in two orthogonal dimensions (e.g., elevation and azimuth) and thus, makes a full 3D coverage.
Critical AoA and AoD Array Parameters and Design Challenges
Aside from the general antenna requirements such as optimal in-band matching, radiation efficiency, and gain, in an ideal AoA array, the incident signal-induced phase differences between the antennas should depend only on the incident angle and separation distances between the antennas. Unfortunately, in real-life situations, several unwanted effects can deteriorate the phase differences and thus, cause AoA angle accuracy degradation.
These effects include the following:
- Limited isolation between the antennas
- Non-uniform S11 and resonant frequencies
- Antenna-to-antenna radiation phase pattern variations and their dependence on incident signal direction
- Limited cross-polarization attenuation for dual-fed antennas
- Feeding circuitry phase unbalances
- Feeding circuitry cross-talks
Additional requirements to reduce the array size make the effects mentioned above even more critical. Also, the need for low-cost PCB technology can cause unwanted frequency shifts due to the technological spreading that occurs during PCB manufacturing. Mitigating these issues is essential for low-cost, high-performance direction finding array designs.
Achieving Antenna Size Reduction by Applying Metamaterial Techniques
As mentioned above, the AoA arrays designed by Silicon Labs apply dual-fed (or dual-polarized) patch antennas. In conventional designs, the patch antenna size is primarily determined by the frequency and the PCB laminate used. Theoretically, the patch size should be half wavelength, which is approximately 30 mm on a typical 1.6 mm thick FR4 PCB stack-up at 2.4 GHz. To reduce antenna size, the metamaterial technique is used. The metamaterial surface is a periodic structure that forms a reactive impedance surface (RIS) and produces a parallel resonant, frequency-dependent characteristic impedance for the incoming plane wave [2,3]. Below the surface’s resonant frequency, the metamaterial has inductive characteristic impedance, while above the resonant frequency, it has capacitive characteristic impedance.
The metamaterial is located in an intermediate layer beneath the patch antenna, as shown in Figure 3. Here the metamaterial resonators are square-shaped rings with tuned size and interstage gap and width, all of them serving as tuning knobs. The figure below denotes the RIS ring width parameter by “IRW” in the figure. With a resonance frequency higher than the operation frequency, the inductive RIS characteristic impedance can also resonate with a smaller-than-half lambda patch antenna. For example, in applications where low-cost 1.6 mm thick FR4 PCB technology is used, the typical 29 mm patch size can be reduced to ~22 mm with an RIS surface.
Figure 3. Patch Antenna above a Reactive Impedance Surface (RIS) Layer
Figure 4. Antenna Ports 1 and 2 Reflections (S11 and S22)
Figure 4 illustrates the simulated reflections of the dual-fed patch shown in Figure 3. S11 and S22 are the vertical and horizontal port reflections, respectively. The IRW parameter (width of the RIS rings) is denoted in mm. In the figure, the matching is quite good as it results in a -10 dB bandwidth of 90 MHz. The figure also illustrates that the ring width parameter (IRW) is an effective resonant frequency tuning knob.
Isolation and S11 Uniformity Improvement
To improve the patch antenna uniformity within an array, the relative positions of the antennas to the RIS surface are fixed. Figure 5 shows a 3x3 array example and Figure 6 shows the corresponding magnitude and phase of the patch antenna reflections. Despite the fixed uniform patch positions, the S11 spreading is unacceptably large due to the couplings between the neighboring antennas.
Figure 5. 3x3 AoA Array with Conventional Metamaterial RIS Layer (i.e., without guard rings)
Figure 6. Antenna Port Reflections (S11 curves) of the Array in Figure 5, Showing Pure S11 Univormity
To reduce coupling and improve the reflection uniformity, GND guard rings are applied around each antenna. The RIS surface is also divided into 2x2 ring blocks, which are beneath the patch antennas and also surrounded by GND guard rings on the RIS layer. The GND guard rings on the different layers are axially aligned and connected to the bottom GND layer by numerous stitching vias. Figure 7 shows the building block consisting of a patch antenna, a 2x2 RIS block, and the GND guard rings. Figure 8 shows a 3x3 array example built from these RIS antenna blocks. Figure 9 shows the port reflections (magnitude and phase), where the S11 uniformity is improved significantly compared to the antenna port reflections illustrated in Figure 6 without GND guard rings.
Figure 7. Metamaterial Patch Antenna Block with 2x2 RIS Cell Block and GND Guard Ring
Figure 8. 3x3 AoA Array Built from the RIS Antenna Blocks Shown in Figure 7 with GND Guard Rings
Figure 9. Antenna Port Reflections (S11 curves) of the Array with Guard Rings in Figure 8 Showing Good S11 Uniformity
Radiation Phase Pattern Balancing Problems
The uniformity of the array antennas’ radiation phase patterns is critical because any antenna-to-antenna phase variation causes unwanted phase differences and thus, direction finding errors.
Figure 10 shows the 4x4 linear array example built from the 16 basic dual-fed patch antenna building block cells presented in Figure 7. In this 4x4 design, the building block structure is slightly different as GND guard rings are applied in the non-used intermediate metal layers as well. With this layout, the metal distribution within the PCB stack-up is more balanced and unwanted PCB tensions are avoided.
Despite the applied guard rings, any linear AoA or AoD array construction that optimizes space constraints and positions the antenna blocks near each other will result in significant residual internal couplings (i.e., in the -15 to -20 dB range), especially through the bottom GND layer. In addition, the radiated signal of a patch antenna interacts with the neighboring cells as well. These residual couplings cause position-dependent deterioration in the radiation patterns within the array.
As the perimetric antenna cells have asymmetric neighbor GND structures, their radiation characteristics are deteriorated more asymmetrically. This effect is strongest with the corner antenna cells and weakest with the center cells.
Figure 11 shows the applied far-field coordinate system for the arrays' far-field radiation investigations. Odd ports (Ports 1, 9 and 11) are the vertical (V) and even ports (Ports 2, 10, 12) are the horizontal (H) polarized patch feeding ports of the investigated antenna cells. All ports are situated at the bottom layer, and the patches at the top layer are fed by through-hole vias. Additionally, all ports receive both Theta and the Phi polarized signals, so the total number of radiation phase characteristics is double the number of ports. For example, in a 4x4 dual feed array, there are 32 ports and 64 radiation phase characteristics.
Figure 10. 4x4 AoA Array Built from the RIS Antenna Blocks in Figure 7 with GND Guard Rings
Figure 11. Applied Far-Field Coordinate System for the Far-Field Radiation Investigations of the 4x4 AoA Array
Figure 12a and Figure 12b show the Theta and Phi polarized radiation phase characteristics of vertical port1 of the right top corner patch antenna. Figure 12c and Figure 12d show the same characteristics for the horizontal port2 of the same corner patch antenna. For Theta polarized characteristics (Figure 12a and Figure 12c), the main radiation direction falls into the port axis (i.e., to the port polarization), and there are nulls in the orthogonal directions. For Phi polarized signals, the main radiation direction is orthogonal to the port axis.
Based on the illustrations in Figure 12, it is evident that the Phi polarized radiation phase characteristics (right column of Figure 12) become more asymmetric at low PCB elevations, especially where the signal polarization is nearly parallel to the PCB plane.
Figure 12. 4x4 AoA Array of Figure 11, Radiation Phase Characteristics of the Right Top Corner Patch (Figure 12a and Figure 12b are the Theta and Phi signal polarized signal reception cases for V polarized port1, respectively. Figure 12c and Figure 12d are the same for the H polarized port2.)
Figure 13 compares the most sensitive H port, Phi polarized patterns of a corner, center, and edge patch antenna. Figure 13a, 13b, and 13c shows the 3D phase radiation patterns. Figure 14 shows the Theta = 60-degree cuts of these patterns. Here the azimuth (Phi) is swept, and there are significant phase pattern differences between the corner, center, and edge patch radiators. The uniformity is degraded, which in turn degrades the angle estimation accuracy. The sudden jumps shown in the Figure 14 phase curves are due to the shown phase values being limited to the 0–360 degrees region. Also note that there are phase transition regions close to the pattern nulls (e.g., around 0 and 180 deg azimuth (Phi) angles). At these pattern null azimuth angles, the reception mainly happens in the orthogonal (i.e., Theta) signal polarizations, where there are pattern maximums at these azimuth directions. Therefore, phase uniformity is not important here.
Figure 13. 4x4 Linear Array of Figure 12, Most critical H Polarized Port, Phi Polarized Signal Phase Radiation Characteristics for a Corner (Figure 14a), Center (Figure 14b), and Edge Patch (Figure 14c)
Figure 14. Theta = 60-degree Elevation Cuts (Azimuth Runs) of these Patterns
Metamaterial BRD4191A Array with GND Skirt along the Perimeter for Better Radiation Pattern Balance
A GND skirt around the array makes the GND environment more balanced around the edge and corner radiators as shown in Figure 15. Due to this balance, the antennas’ Phi polarized phase radiation characteristics become less asymmetric and thus, closer to that of a center patch. This behavior is demonstrated in Figure 16, where the “ground skirt” array (Figure 16b and Figure 16d) has less phase variations than their “non-ground skirt” counterparts (Figure 16a and Figure 16c). The azimuth phase regions where uniformity improvement is observed between the left (without skirt) and right (with skirt) column figures are highlighted by dashed ellipses. Again, these are fixed elevation cuts (Theta = 60 degrees), and the azimuth (Phi) is swept. Note: the sudden jumps in the phase curves are due to the shown phase values being limited to the 0-360 degrees region. Also note that the wider the skirt, the better the phase pattern uniformity. However, a wider skirt increases the array size as well, and therefore, is always a compromise with the direction finding performance.
Figure 15. 4x4 Linear Metamaterial Array with 10 mm GND Skirt along the Array Perimeter (the skirt is black in the figure)
Figure 16. Theta = 60-degree Elevation Cuts of the 3D Phase Radiation Patterns of a Corner, Edge, and Center Patch Antenna with Azimuth (Phi) Sweep (Figure 16a and 16b show the H port Phi polarized pattern cuts without and with 10 mm ground skirt, respectively. The same characteristics are shown for the orthogonal V ports in Figure 16c and Figure 16d. With a ground skirt, the variations decreased.)
Even in the presence of a ground skirt, there are some limited phase variations between the antennas. A software manifold compensation method during the data process can accommodate these residual phase variations. During the software manifold compensation method, all radiation phase patterns (64 characteristics for the 32 ports) are measured, stored, and processed.
Low-cost PCB Technology Spreading Problem Mitigation
In Figure 9, the -10 dB S11 bandwidth of the metamaterial patch antenna design is around ~80-100 MHz (i.e., only slightly wider than the 2.4 GHz band). Thus, mistuning due to technological spreading can degrade the antenna realized gain and thus, the range, especially for channels which fall to the band edges. As a result, the design cannot exceed mistuning of 15-25 MHz.
Accuracy is less sensitive to frequency detuning as all radiators shift with the same amount of frequency and thus, the radiation phase pattern differences practically do not change. The metamaterial antenna resonant frequency is basically determined by the laminate dielectric constant and thickness. The dielectric constant of the FR4 laminates can vary in a wide range, typically between 3.9 and 5 at 2.4 GHz. The real value depends on many laminate and PCB manufacturing parameters. Thickness variation is typically less significant.
In a typical FR4 PCB, the woven glass fiber yarns provide the mechanical integrity of the PCB, and the resin is the glue which fills and fixes the PCB. The dielectric constant (epsilon) is primarily determined by the resin to fiberglass ratio (i.e., the resin content). The higher the resin content, the lower the epsilon. Most important is the laminate style. Figure 17 shows two different laminate styles . With a dense fiberglass net (shown on the right side of Figure 17 and which is typical for 2116 and 7628 styles), the resin content is lower, and thus the dielectric constant is higher. On the other hand, in a rare fiberglass net (shown on the left side of Figure 17 and which is typical for 106 or 1080 styles), the resin content is higher, which lowers the epsilon.
Figure 17. FR4 PCB Fiberglass Net Structures
Aside from the laminate style, there are several other secondary effects which can cause unwanted epsilon variations: surface roughness of the copper foil, baking technology and parameters during PCB manufacturing, solder mask thickness, and epsilon.
Surface roughness has an effect on the copper foil to dielectric transition surfaces . The typical rms variation is 0.5-2 um, which is in the order of the skin depth at the GHz frequency range. Due to the surface roughness, the propagating signal runs a significantly longer route, which causes an effective epsilon increase. The detuning caused by this effect can be up to 30-50 MHz at the 2.4 GHz frequency band, so an accurate value of the roughness must be acquired from the laminate manufacturer and factored into the RF simulations.
The prepreg PCB layers, which are made from woven glass fiber yarns pre-impregnated with resin and only partially cured, are baked together with the fully pre-cured core layers during the PCB manufacturing process. Here the temperature, humidity, applied pressing force variations, etc., can cause slight (1 -2 %) epsilon and thickness variations, which again can cause an additional 10-20 MHz detuning.
Solder mask layer variations cause only moderate (5-10 MHz) detuning.
The effects described above can cause an overall worst-case mistuning of 70-80 MHz, which is unacceptably high for the array design. In addition to enhancing the stack-up (e.g., 1.6 mm, 6-layer, and fixed laminate styles), Silicon Labs proposes to mitigate the risks created by mistuning by improving the laminate type and, if possible, the PCB vendor as well. In the application note, AN1195: Antenna Array Design Guidelines for Direction Finding , designs tuned for three different low-cost and widely popular laminates are detailed. With these restrictions, both the sample-to-sample and build-to-build variations can be minimized as shown in Table 1. The PCB vendor variation is limited as well.
In the following table, sample-to-sample variation is the antenna resonant frequency variation among the samples from the same build from the same PCB vendor. Build-to-build variation is the difference between the average antenna resonant frequencies across a number of samples from different builds from the same PCB vendor. Finally, PCB vendor-to-vendor variation is the difference between the average antenna resonant frequencies across a number of samples from different PCB vendors.
Table 1. Sample-to-Sample, Build-to-Build and Vendor-to-Vendor Variations of the Investigated, Low-Cost 1.6 mm, 6-Layer FR4 Stack-Ups with Three Different Laminate Types.
Switched Dual-Polarized or Circular-Polarized Reception
A practical problem with the dual-polarized AoA array operation is that the received signal level can be very different in the two polarizations. The receiver chain amplification is adjusted at the beginning of the Bluetooth packet and usually done only on one polarization to save time and energy. If there is significant signal level difference, the gain may be adjusted to the weaker polarization and thus, strong distortion and even phase errors appear during the reception of the stronger polarization. To avoid these problems, 2 of the 16 dual-fed array antennas are switched to circular polarized (CP) mode, and one of them is used at the preamble phase to adjust the receiver gain.
The block diagram for the switched RF feeding network, which makes this operation possible, is illustrated in Figure 18. Here, signals corresponding to the horizontal and vertical ports of the 16 antenna patches are routed to two 16-pin switches, as shown on the right side of the figure. The signal routes’ spatial division is optimized to minimize and equalize the length of the feeding lines to the antennas (denoted by Section 4 in the figure).
However, there are two patch antennas where the signals’ ports are cross-connected so that the V- and H-polarized signals appear at the outputs of the two 16-pin switches at the same time. These signals can be directed either to a 90-degree quarter-wave hybrid (also shown in the figure) to generate the CP signal or directly to the RF input through the CP/DP switches. Here, the phase balance of the Section 2/H traces (CP mode) and the Section 2/D traces (dual-polarized mode) is critical.
Figure 18. BRD4191A AoA 4x4 Array RF Feeding Circuit Blocking Diagram
RF Feeding Circuitry with Phase Balancing and Good Isolation
Phase balancing and effective isolation is also essential in the RF feeding network. As shown in Figure 18, all routes in the switched RF feeding network are phase balanced. For example, in Section 4, the 2x16 routes from the dual-feed antenna ports to the 16-pin switches, or the routes in Section 2/D and Section 2/H. Figure 19 shows the layout of the RF feeding circuitry situated at the bottom layer of the BRD4191A array board. Here, the Section 4 co-planar antenna feeding lines are highlighted. They are folded to have equal length and thus, phase rotation. The round-arched routing close to the 16-pin switches minimizes the coupling as well. Stitching vias along both GND sides of the co-planar lines improve the isolation. The stitching vias are connected to a massive GND layer, which is the next intermediate layer above.
The phase balancing and isolation of the additional feeding line sections (shown in Figure 18 but not highlighted in Figure 19) are a much simpler task. The 90 deg hybrid, additional switches, and the Silicon Labs BG22 Bluetooth transceiver IC are situated at the right side of the PCB, partly at the ground skirt area. This is also the case for the antenna feeding lines on the other PCB sides. This bottom layer circuitry is well isolated from the top patch antenna array structure and the ground skirt by a massive, intermediate GND layer between them.
Figure 19. BRD4191A AoA 4x4 Array RF Feeding Circuit Layout Configuration with highlighted Phase-Balanced Patch Feeding Lines
Bluetooth AoA and AoD are relatively new direction-finding technologies that can be used for asset tracking, indoor positioning, and wayfinding. These phase-based, direction-finding systems require a switched antenna array (or a multi-channel ADC) to feed the input data for the direction estimation algorithms. Phase uniformity of the array radiators, low internal coupling between the antennas, and a phase-balanced RF feeding network are all essential factors in achieving direction finding accuracy. Novel techniques, such as the use of GND guard rings and a ground skirt along the array perimeter, were shown to improve these parameters.
Small size and low cost are also important when deploying location devices. Metamaterial-based patch design techniques help reduce antenna size, and using a 1.6 mm, 6-layer FR4 PCB stack-up can reduce costs. Finally, the technological variations that result from using these techniques can be mitigated by thorough laminate specification and selection.
-  https://www.silabs.com/whitepapers/bluetooth-angle-estimation-for-real-time-locationing
-  Antenna Miniaturization and Bandwidth Enhancement Using a Reactive Impedance Substrate, Hossein Mosallaei, Kamal Sarabandi, IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 52, NO. 9, SEPTEMBER 2004, p2403
-  Minimized Low-Profile Wideband Antennas Using High Impedance Surface, Yizhu Shen, Hindawi International Journal of Antennas and Propagation Volume 2017, Article ID 2563927, 12 pages, https://doi.org/10.1155/2017/2563927
-  https://blog.lamsimenterprises.com/2010/12/14/
-  https://www.isola-group.com/wp-content/uploads/PCB-Material-Selection-for-High-speed-Digital-Designs-1.pdf
-  Silicon Labs, AN1195: Antenna Array Design Guidelines for Direction Finding