2024 International Conference on Telecommunications and Intelligent Systems (ICTIS)

AI-Driven Beam-Steering Metamaterial Lens for 5G

and Autonomous Vehicles
2024 International Conference on Telecommunications and Intelligent Systems (ICTIS) | 979-8-3315-2739-6/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICTIS62692.2024.10894320

1st Noureddine Melouki 2nd Tayeb Ahmed Denidni

Centre-Energie Matériaux et Télécommunications (EMT) Centre-Energie Matériaux et Télécommunications (EMT)
Institut National de la Recherche Scientifique (INRS) Institut National de la Recherche Scientifique (INRS)
Montréal, QC H5A1K6, Canada Montréal, QC H5A1K6, Canada
[email protected] [email protected]

Abstract—This paper presents an AI-driven metamaterial- time beam reconfiguration with minimal power consumption
based transmissive lens for 5G and autonomous vehicle ap- [5].
plications. A genetic algorithm (GA) model is used to design These transmitarrays offer several advantages over phased
and optimize unit cells, achieving precise phase control through
the integration of two PIN diodes. The unit cell design is arrays:
optimized to ensure high transmission coefficients and a 180° • Cost Efficiency: Metamaterial-based transmitarrays
phase shift difference between the diode states. Furthermore, an eliminate the need for costly phase shifters and RF chains
extended AI-based framework leverages real-time sensor data
for each antenna element, drastically reducing the overall
from autonomous vehicles to dynamically control beam steering
using machine learning (ML) models. Full-wave simulations system cost.
validate the performance of this system, demonstrating beam • Reduced Complexity: The integration of binary phase-
steering at different angles from 0 to 45 degrees. This capability shifting mechanisms, such as PIN diodes, allows for
can be extended to enable wider 2D beam scanning, showcasing simple yet effective control over the transmitted wave-
enhanced functionality for real-world autonomous applications.
front, reducing the complexity associated with traditional
Index Terms—Genetic algorithm (GA), AI model, beam steer-
ing, metamaterials, 5G, autonomous vehicles, machine learning phased array architectures.
(ML) • Scalability: Unlike phased arrays, transmitarrays can be
easily scaled to larger apertures, making them suitable
I. I NTRODUCTION for mm-Wave applications like 5G communications and
autonomous vehicle radar systems.
The exponential growth of 5G networks and the increas-
In addition, reconfigurable transmitarrays have shown sig-
ing demand for high-speed, low-latency communication have
nificant potential in autonomous vehicle applications, where
driven the need for advanced antenna systems. Traditional
real-time beam steering is essential for functions such as
phased arrays, while capable of precise beam-steering, come
radar-based object detection, vehicle-to-vehicle (V2V) com-
with significant drawbacks in terms of cost, complexity, and
munication, and adaptive navigation in dynamic environments
power consumption, especially when scaled for millimeter-
. By integrating machine learning models to control beam
wave (mm-Wave) frequencies. Phased arrays rely on expensive
steering, these systems can adapt to changes in the surrounding
phase shifters for each antenna element, which increases
environment based on real-time data from onboard sensors,
the overall system complexity and limits their scalability for
such as cameras, LiDAR, and radar [6], [7].
applications like 5G communications and autonomous vehicles
This paper introduces a novel AI-driven metamaterial-based
[1], [2].
Reconfigurable transmitarray (MTM-RTA) that utilizes a hy-
Reconfigurable metamaterials-based transmitarrays have
brid approach: genetic algorithms (GA) are employed for
emerged as a promising alternative to traditional phased arrays.
the design and optimization of individual unit cells, while
Metamaterials offer unique electromagnetic properties that
an extended AI model controls the beam-steering system in
enable dynamic control of wavefronts using sub-wavelength
real-time using sensor data. This approach addresses the cost
elements. A key feature of these materials is the ability
and complexity challenges of phased arrays, while offering
to engineer their effective refractive index, allowing for the
high-performance beam steering for both 5G and autonomous
manipulation of electromagnetic waves, such as beam focusing
vehicle applications.
and steering, without the need for bulky or expensive phase-
shifting components [3], [4]. II. T UNABLE UNIT CELL DESIGN AND ANALYSIS
Transmitarrays, composed of reconfigurable metamateri-
als, can achieve beam-steering functionality by dynamically A. Unit cell design
adjusting the transmission phase of each unit cell. This is The proposed unit cell structure is designed on a rogers
typically done using passive elements or integrating active RT/duroid® 6002 dielectric slab, with a 0.762 mm per layer,
components like PIN diodes or varactors, allowing for real- to operate in the range of 26 to 32 GHz and is equipped

979-8-3315-2739-6/24/$31.00 © 2024 IEEE

Authorized licensed use limited to: Instituto Federal de Santa Catarina (IFSC). Downloaded on March 24,2025 at 20:51:54 UTC from IEEE Xplore. Restrictions apply.
 cell’s geometry and material properties, such as the dielectric
thickness, radiating element size, and PIN diode placement.
The GA iteratively evolves the population of designs,
evaluating each one according to a fitness function. The
optimization process consists of three key steps [9]:
• Selection: Designs are selected based on their fitness
scores, with higher-performing designs more likely to be
chosen for reproduction.
• Crossover: Pairs of selected designs are combined to
produce offspring, inheriting characteristics from both
parent designs.
• Mutation: Random variations are introduced to some
Fig. 1: Vehicular communication by MTM-RTA antennas, in designs, allowing the algorithm to explore new areas of
different scenarios for self-driving vehicles. the design space.
C. Fitness Function Formulation
with two PIN diodes for binary phase control (see Fig.2). The fitness function evaluates the performance of each unit
This design concept employs the current reversal technique, cell design based on two criteria:
where the incident wave on the receiving layer of the unit • Transmission Coefficient (T):
cell, changes its current direction, when coupled through the The design should maximize the transmission coefficient,
via to the transmitting layer. Meaning, when the combination ensuring that electromagnetic waves pass through the unit
of the On/Off and Off/on of the two pin diodes, is changed, a cell with minimal loss.
180 phase shift is achieved, as seen in Fig. 3.b.
The initial unit cell exhibited a -3dB transmission (70%) |S21 |2
bandwidth spanning from 26.25 to 31.4 GHz, and two distinct T = (1)
|S11 |2 + |S21 |2
narrow bands with transmission coefficient of more than 90%
in the ranges of 26.8 to 27.51 GHz, and 29.45 to 30.2 GHz. • Phase Shift (∆ϕ): The design should achieve a phase
So far from the required bandwidth of 26 to 32 GHz, with shift of 180° between the two diode states (on/off and
more than 90% (-1dB) in transmission amplitude. off/on). The difference between the phase shift of each
A GA-based framework will then optimize the unit cell’s state should be as close as possible to 180°.
geometric and material properties, to achieve a high transmis-
sion coefficient and stable 180 degree phase shift between the ∆ϕ = |ϕon − ϕoff | (2)
”on/off” and ”off/on” states of the diodes. This ensures the The overall fitness function is a weighted sum of these
system can reliably switch between two distinct phase states, objectives, where wT and wϕ are weights for the transmission
forming the basis of a 1-bit phase-shifting system [8]. coefficient and phase shift difference, respectively:
F = wT · T + wϕ · (180 − ∆ϕ) (3)
This formulation ensures that designs with both high transmis-
sion efficiency and precise phase control are prioritized during
the optimization process. The final transmission amplitude and

Fig. 2: Geometry of the proposed unit cell.

B. Genetic Algorithm Framework

(a) (b)
Genetic algorithms offer an efficient method for exploring
the large design space of metamaterial unit cells. Each poten- Fig. 3: (a) Unit cell simulation setup, and (b) its initial
tial design is encoded as a set of parameters that define the unit simulated transmission results.

Authorized licensed use limited to: Instituto Federal de Santa Catarina (IFSC). Downloaded on March 24,2025 at 20:51:54 UTC from IEEE Xplore. Restrictions apply.
phase optimized by the genetic algorithm are attained after the can predict the optimal beam configuration and adjust the PIN
10th iteration when the termination criterion is met [10]. diodes accordingly.
Fig. 4 depicts the final optimized transmission performances
C. Application in Autonomous Vehicles
of the proposed unit cell, where it is clearly seen that both
transmission amplitude and phase meet the requirements of a Autonomous vehicles rely on real-time data to navigate
highly efficient transmissive 1-bit unit cell. safely and efficiently. The proposed beam-steering system
can be integrated into the vehicle’s radar and communication
systems, ensuring that the beam dynamically adapts to the
vehicle’s surroundings. For instance, if the AI model detects
an obstacle or a nearby vehicle, the beam can be redirected to
improve radar accuracy or establish communication with the
other vehicle.
IV. S IMULATION R ESULTS
The AI-based beam-steering framework is currently in its
initial stage of development, with the simulation results being
derived from data emulating real-world conditions. The data
used in these simulations were captured through a camera’s
(a) (b)
Region of Interest (ROI), providing the necessary coordinates
to determine the beam’s desired direction. This approach
simulates the information that would be obtained from an au-
tonomous vehicle’s onboard sensors, such as cameras, LiDAR,
and radar.
The AI-based beam-steering system was evaluated by con-
figuring the optimized unit cells into a 20 × 20 transmitarray.
The beam-steering capability was tested at various angles, and
the results were compared in terms of peak gain, side lobe
levels, and overall steering accuracy.

(c)

Fig. 4: Optimized unit cell design: (a) Transmitting layer, (b)

Receiving layer, and (c) its final simulated transmission results.

III. AI M ODEL FOR B EAM -S TEERING C ONTROL

A. AI-Driven Beam Steering Framework
In addition to unit cell optimization, the beam-steering
system leverages an extended AI-based model that processes
real-time data from onboard vehicle sensors (e.g., cameras)
to adapt the steering direction. This extended framework uses
machine learning models trained on sensor data to adjust the
configuration of the unit cells dynamically, ensuring the beam Fig. 5: Proposed MTM-RTA setup.
is optimally steered in response to environmental changes.
The system demonstrated accurate beam-steering across
B. Machine Learning Integration various angles (0°, 15°, 30°, 45°), guided by the coordinates
The AI model is designed to interpret data such as obstacle derived from the ROI data. This emulation of sensor data
positions, vehicle speed, and road conditions. By feeding this mirrors the real-time input from an autonomous vehicle’s
data into a pre-trained machine learning model, the system environment, such as detecting obstacles or other vehicles. The

Authorized licensed use limited to: Instituto Federal de Santa Catarina (IFSC). Downloaded on March 24,2025 at 20:51:54 UTC from IEEE Xplore. Restrictions apply.
AI framework used the input data to dynamically reconfigure
the unit cells, ensuring that the beam was directed at the target
angles with minimal error.
At 0°, the system achieved a peak gain of 25.1 dBi, with
only a slight decrease as the steering angle increased, reaching
19.7 dBi at 45°. This behaviour is expected, as steering
larger angles introduces more complexity in maintaining gain.
Nevertheless, the AI-driven beam control performed well,
maintaining high gain across the tested range.
The side lobe levels were kept below -10.3 dB for all
steering angles, with a minimum of -18.5 dB at 0°. This
low side lobe radiation demonstrates the system’s ability
to focus energy in the intended direction while minimizing
interference, an important aspect for both 5G communications
and autonomous vehicle sensing.
The AI-based system demonstrated efficient real-time adapt- (a)
ability, adjusting the beam’s direction based on the simulated
sensor data. The overall transmission efficiency remained high,
exceeding 90% across all configurations, ensuring reliable
communication and radar functionality in autonomous vehicle
applications.

V. C ONCLUSION
The AI-driven beam-steering framework, while in its initial
stage of development, has shown promising results based on
data emulated from a camera’s ROI. This simulates the real-
time input from an autonomous vehicle’s onboard sensors, ef-
fectively guiding beam steering in dynamic environments. The
combination of genetic algorithms for unit cell optimization
and AI for real-time beam control demonstrates the system’s
potential for both 5G communications and autonomous driving
applications.
(b)
R EFERENCES Fig. 6: Simulated beam-steering performance at 28 GHz, for
[1] N. Lu et al., “Connected vehicles: Solutions and challenges,” IEEE (a) H-plane and (b) E-plane.
Internet of Things Journal, vol. 1, no. 4, pp. 289–299, Aug. 2014.
[2] T. Zhou, M. Yang, K. Jiang, H. Wong, and D. Yang, “MMW radar-based
technologies in autonomous driving: A review,” Sensors, vol. 20, no. 24,
[8] Melouki, N., Hocini, A., Fegriche, F. Z., PourMohammadi, P., Naseri, H.,
Dec. 2020.
Iqbal A. and Denidni, T. A. ”High-Gain Wideband Circularly Polarised
[3] L. Di Palma, A. Clemente, L. Dussopt, R. Sauleau, P. Potier, and Ph.
Fabry–Perot Resonator Array Antenna Using a Single-Layered Pixelated
Pouliguen, “Circularly-polarized transmitarray with sequential rotation in
PRS for Millimetre-Wave Applications,” Micromachines, vol. 13, no. 10,
Ka-band,” IEEE Trans. Antennas Propag., vol. 63, no. 11, pp. 5118-5124,
pp. 1658, 2022.
Nov. 2015.
[9] Melouki, N., Ahmed, F., PourMohammadi, P., Naseri, H., Bizan, M.
[4] H. Kaouach, L. Dussopt, J. Lanteri, R. Sauleau, and T. Koleck, “Wideband
S., Iqbal, A., & Denidni, T. A. (2024). 3D-Printed Conformal Meta-
low-loss linear and circular polarization transmitarray in V-band,” IEEE
Lens with Multiple Beam-Shaping Functionalities for Mm-Wave Sensing
Trans. Antennas Propag., vol. 59, no. 7, pp. 2513-2523, Jul. 2011.
Applications. Sensors, 24(9), 2826.
[5] N. Melouki, F. Ahmed, H. Nasseri, P. Pourmohammadi, A. Iqbal and T. A.
[10] Melouki, N., Hocini, A. and Denidni, T.A. ”High gain and Wideband
Denidni, ”Multi-Bit Wideband Transmitarray Aperture with Independent
Fabry-Perot Resonator Antenna based on a compact single PRS layer,”
Phase and Amplitude Control for High Gain with Low Sidelobe Mm-
IEEE Access, vol. 10, pp.96526-96537, 2022.
Wave Applications,” 2024 18th European Conference on Antennas and
Propagation (EuCAP), Glasgow, United Kingdom, 2024, pp. 01-04, doi:
10.23919/EuCAP60739.2024.10501189.
[6] B. J. Xiang, X. Dai and K. -M. Luk, ”A Wideband 2-Bit Transmitar-
ray Antenna for Millimeter-Wave Vehicular Communication,” in IEEE
Transactions on Vehicular Technology, vol. 71, no. 9, pp. 9202-9211,
Sept. 2022, doi: 10.1109/TVT.2022.3178843.
[7] Melouki, N., Ahmed, F., Pourmohammadi, P., Naseri, H., Iqbal, A.,
and Denidni, T. A., “3D-Printed Conformal Metamaterial Lens with
Multiple Beam Steering Functionalities,” in 2023 IEEE International
Symposium on Antennas and Propagation and USNC-URSI Radio Sci-
ence Meeting (USNC-URSI), Portland, OR, USA, 2023, pp. 791-792,
doi: 10.1109/USNC-URSI52151.2023.10237856.

Authorized licensed use limited to: Instituto Federal de Santa Catarina (IFSC). Downloaded on March 24,2025 at 20:51:54 UTC from IEEE Xplore. Restrictions apply.