2022 PhotonIcs & Electromagnetics Research Symposium (PIERS), Hangzhou, China, 25–27 April

Machine Learning Based MIMO Antenna Arrays Optimization for

5G/6G
Maxim A. Dubovitskiy
National Research University “Moscow Power Engineering Institute”, Russia
2022 Photonics & Electromagnetics Research Symposium (PIERS) | 978-1-6654-6023-1/22/$31.00 ©2022 IEEE | DOI: 10.1109/PIERS55526.2022.9793294

Abstract— Utilization of multiple-input multiple-output (MIMO) systems as a means of in-

creasing channel capacity has been an area of increasing consideration in radio communications.
This research is important because high-frequency communication using MIMO allows for in-
ternational communication at long distances using lower power consumption than many other
approaches. The objective of this research is to develop and implement software algorithms for
the synthesis of MIMO-type arrays, which entail an increase in the efficiency of their operation,
including the suppression of side lobes by optimizing their structures, taking into account the in-
terference of electromagnetic waves between neighboring elements, increasing the signal-to-noise
ratio (SNR) at the receiver input, increasing the bandwidth of the receiving and transmitting
modules of LTE/5G communication systems. Since it is assumed that 6G communication net-
works will use the terahertz and sub-terahertz frequency ranges and provide a significantly lower
level of delay in data transmission than 5G/IMT-2020 networks, the proposed Machine Learning
(ML) algorithms should be universal and capable of providing computer-aided design of aperiodic
multi-element antenna arrays not only in existing LTE/5G communication systems, but also in
the terahertz frequency range.

1. INTRODUCTION
This article will present the results of a theoretical analysis of the main dependencies of antenna
characteristics on the amplitude and phase distributions in the near radiation zone of an aperture
antenna, which can consist of a large-sized antenna array applicable to consumer equipment or a
base station (BZ) of a new generation of 5G and 6G communication systems. Modern technical ob-
jects, in relation to which it is necessary to solve the problems of modeling electrodynamic systems,
are quite often characterized by a compact placement of emitting, re-emitting and third-party bod-
ies of significantly different configurations and geometric dimensions. Under these conditions, the
characteristics of each radiating element significantly depend on other elements, as well as on the
surrounding scatterers, so their totality should be considered and modeled as a single complex elec-
trodynamic system. The constantly increasing requirements for the quality, reliability and accuracy
of the results of applied research and technical solutions assume the most complete and accurate ac-
count of the geometric and electrophysical characteristics of all the components of the corresponding
complex electrodynamic system, which is an antenna array of densely arranged aperture elements.
This, in turn, imposes increased requirements on the adequacy of electrodynamic models and the
accuracy of calculation methods, as well as on their resource intensity and efficiency. The relevance
of studying the near field is confirmed by the need for effective estimation of parameters and signal
resolution, which is possible due to the accurate determination of the amplitude-phase distribution
in the opening of an aperture antenna or antenna array. When calculating antenna arrays with an
arbitrary contour boundary, it is important to understand how the spatial characteristics change
when changing parameters such as the total diameter of the effective opening of the antenna, phase
distortions at the edges of the aperture, inhomogeneous amplitude distribution, etc. For receiving
equipment, the formation of the antenna array radiation pattern is a process of linear weighing and
combining signals from the grid elements in order to form the desired spatial dependence of the
signal response. Digital diagram formation implies a digital synthesis of the radiation pattern in
the receiving mode, as well as the formation of a given distribution of the electromagnetic field in
the opening of the antenna array — in the transmission mode. In this case, the diagram formation
can be considered as a spatial function, where discrete samples in space form a spherical wavefront
— the near field of each of the elements of the phased antenna array. As a rule, the antenna array
is designed in such a way as to adjust the weight coefficients of the elements in order to achieve
a higher resolution, directivity and sensitivity, as well as a signal-to-noise ratio, much better than
is possible for each element individually. In fact, the main share of energy is concentrated in the
sector of angles forming the main beam of the radiation pattern, and lower response levels are
spaced apart and are undesirable (side lobes).

690
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on September 05,2022 at 09:55:47 UTC from IEEE Xplore. Restrictions apply.
2022 PhotonIcs & Electromagnetics Research Symposium (PIERS), Hangzhou, China, 25–27 April

2. VISION AND REQUIREMENTS

The relevance of the use of MIMO terahertz antenna arrays in wireless communication technology
lies in the increase in the volume of data and the need for their rapid transmission; the spread
of Internet of Things systems, including the industrial Internet; the growth of e-commerce; the
development of unmanned transport; the demand from users for qualitatively new content (for
example, VR broadcasts); the need to increase the bandwidth of networks due to the increase in
traffic volume (Fig. 1). The tasks of the project can be called system-forming, due to the fact that
today the speed of technological development of states and their transition to the digital economy
is becoming a determining factor of leadership in the international market, both in the field of
information technologies and in classical sectors of the economy. The relevance of the development
of new technologies for designing MIMO antenna arrays for use in 6G communications is also the
lack of a full-fledged technical base for antenna components in the terahertz wavelength range.

(a) (b)

Figure 1: (a) MIMO Digital Phased antenna array: interface design and operation of large-dimensional and
autonomous networks, (b) hybrid analog-digital diagramming.

In 5G networks, beam, formed by a BS with large-scale antenna array, is much narrower. It pro-
vides extremely high directional selectivity and array gain. Furthermore, BS operates at mmWave
frequencies can pack more antennas within a small size of an antenna array. Built with high spa-
tial resolution, an antenna array can considerably increase the data rate while reducing power
consumption. However, the acquisition of channel state information (CSI) becomes an essential
issue. In FDD systems, CSI can be obtained through CSI estimation and feedback. As the number
of BS antennas is increased, the size of reference signal symbols and feedback cost overheads are
unfeasible high. Alternatively, channel reciprocity in time-division duplex (TDD) systems can be
used to conquer such problems. Thus, TDD systems with beamspace multiplexing in terms of
massive beamforming will be the mainstay in development of multiple antenna technology for the
upcoming 5G. Moreover, beam-domain signal processing, especially beam-space multiplexing will
play an important role in 6G and beyond, which is viewed as an intelligent network enabled by ma-
chine learning (ML). In this article, we outline several potential research directions and challenges:
massive beamforming for extremely large-scale MIMO (XL-MIMO), low earth orbit (LEO) satel-
lite communication, and data-driven intelligent massive MIMO. Besides, multi-target spatial signal
processing for joint communication and sensing with massive MIMO needs further investigations.
In large-dimensional networks, the same areas can be covered by multiple access network tiers,
which will result in severe interference among tiers. This interference can be suppressed by collab-
orative scheduling, which shares information (such as CSI) to optimize user scheduling. However,
unlike links between terrestrial BSs, the intersatellite, interaerial, intersatellite-aerial, intersatellite-
terrestrial, and interaerial-terrestrial links are nonideal, so they will suffer from large latency,
which is a challenge for collaborative scheduling in large-dimensional networks. Furthermore, long-
distance and high-mobility moving BSs will generate severe channel estimation errors, another chal-
lenge for designing robust and high-performance scheduling algorithms. AI-Enabled Autonomous
Networks Softwarization, cloudization, virtualization, and slicing are still important characteristics
of autonomous networks; thus, software-defined networking (SDN), network function virtualization
(NFV), and network slicing (NS), first introduced to design the 5G network architecture, are still
an important technique set for designing 6G. However, intelligence is the key characteristic of 6G
autonomous networks. AI techniques can provide intelligence for wireless networks through learn-
ing and big data training; therefore, AI will be the most innovative technique for designing 6G
autonomous networks. The combination of AI and SDN/NFV/NS can achieve dynamic and zero-

691
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on September 05,2022 at 09:55:47 UTC from IEEE Xplore. Restrictions apply.
 2022 PhotonIcs & Electromagnetics Research Symposium (PIERS), Hangzhou, China, 25–27 April

touch network orchestration, optimization, and management, which promotes the evolution from
5G to autonomous 6G networks. AI-enabled network orchestration can dynamically orchestrate
network architecture and slices and self-aggregate different radio-access technologies to achieve
liquidized networks and satisfy the demands of constantly changing services and applications. AI-
enabled network optimization can monitor real-time network key performance indicators (KPIs)
and quickly adjust network parameters to continuously provide extreme QoE. AI-enabled network
management can monitor real-time network status and maintain network health. To promote the
development of AI for wireless networks, the ITU-T established a focus group on.
3. OPTIMIZATION TASK
The task of the project is to develop and program implementation of algorithms for the synthe-
sis of MIMO-type digital phased antenna array, which entail an increase in the efficiency of their
functioning, including the suppression of side lobes by optimizing their structures, taking into ac-
count the interference of electromagnetic waves between neighboring elements, an increase in the
signal-to-noise ratio at the receiver input, an increase in the bandwidth of the receiving and trans-
mitting modules of LTE/5G communication systems. Since it is assumed that 6G communication
networks will use terahertz and sub-terahertz frequency ranges and provide a significantly lower
level of data transmission delay than 5G/IMT-2020 networks, the proposed algorithms should be
universal and capable of providing automated design of aperiodic multi-element antenna arrays not
only in existing LTE/5G communication systems, but also in the terahertz frequency range. One
of the technologies that can be implemented in the sixth generation of cellular communications is
the use of photonic digital antenna arrays at base stations in combination with Massive MIMO
technology.
Practically, massive MIMO system chooses the antenna spacing as half-wavelength to maintain
an implementable array aperture, which is quite different from the convention where the antennas
are often placed further away from each other to achieve the spatial diversity. With half-wavelength
arrays, nevertheless, the array signal processing techniques for phased array in Radar and Sonar
applications can possibly be applied to enhance the quality of wireless data communications. Ar-
ray signal processing has long been used in military applications to extract the angles of Radar
targets or to formulate narrow beams for jamming/anti-jamming. The difference from wireless
communications is that the information contained in signals is not cared but rather the AoA that
represents the target’s position. In fact, the famous term beamforming was originated from array
signal processing that means physically formulating an electromagnetic beam towards the target.
Later on, the term beamforming was replanted in wireless communications to represent the weights
of multiple antennas that can maximize the signal-to-noise ratio (SNR) of a user while does not
necessarily formulate a physical beam over the space.
The genetic algorithm based cognitive engine uses the MC simulation technique to estimate the
performance of the first proposed genetic algorithm based CE. MC is a commonly used technique
for measuring the performance of a digital communication system. At each iteration, the system
generates an A × B matrix of independent and identically distributed (i.i.d.) random sequences.
Then, the sequences are passed through modulation and fading blocks to simulate the channel
before estimation and equalization blocks attempt to reduce the fading effects in conjunction with
the CE (Fig. 2). The BER obtained by each equalizer is observed by the CE. The equalization
algorithm with the lowest BER is selected. The CE then selects the nearly optimum equalization
parameters using a genetic algorithm. Finally, the graphing block visually illustrates the BER
versus SNR for the given channel conditions.
Thus, it is possible to evaluate the various CAD systems performance, for example, for the task
of synthesizing the MIMO antenna array for 5G communication with the number of elements 64 by
64, the number of iterations with an increase in the number of variables for intelligent CAD with
Machine Learning will differ tenfold.
4. COMPUTING FEATURES
Scientific computing has traditionally required the highest performance, yet domain experts have
largely moved to slower dynamic languages for daily work. We believe there are many good reasons
to prefer dynamic languages for these applications, and we do not expect their use to diminish.
Fortunately, modern language design and compiler techniques make it possible to mostly eliminate
the performance trade-off and provide a single environment productive enough for prototyping
and efficient enough for deploying performance-intensive applications. The Julia programming

692
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on September 05,2022 at 09:55:47 UTC from IEEE Xplore. Restrictions apply.
2022 PhotonIcs & Electromagnetics Research Symposium (PIERS), Hangzhou, China, 25–27 April

Figure 2: Neural network for simulation MIMO design flow chart.

language fills this role: it is a flexible dynamic language, appropriate for scientific and numerical
computing, with performance comparable to traditional statically-typed languages. Because Julia’s
compiler is different from the interpreters used for languages like Python or R, you may find that
Julia’s performance is unintuitive at first. If you find that something is slow, we highly recommend
reading through the Performance Tips section before trying anything else. Once you understand
how Julia works, it’s easy to write code that’s nearly as fast as C. Julia features optional typing,
multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT)
compilation, implemented using LLVM. It is multi-paradigm, combining features of imperative,
functional, and object-oriented programming. Julia provides ease and expressiveness for high-level
numerical computing, in the same way as languages such as R, MATLAB, and Python, but also
supports general programming. To achieve this, Julia builds upon the lineage of mathematical
programming languages, but also borrows much from popular dynamic languages, including Lisp,
Perl, Python, Lua, and Ruby.

5. MIMO DESIGN
As in any other multi-element antennas, the MIMO antenna design of the base station interacts
with elements operating in different frequency ranges. This interaction affects the directivity and
alignment of the antennas, but this interaction can be eliminated by the methods used in antenna
arrays. For example, one of the most effective methods is to change the design parameters of the
elements by shifting the exciting device, as well as changing the size of the irradiator itself and
the thickness of the separating dielectric layer. An important point is that all modern wireless
technologies are broadband, and the bandwidth of the operating frequencies is at least 0.2 GHz.
Antennas based on complementary structures have a wide operating frequency band, a typical
example of which is the “bow-tie” (butterfly) type antennas. The matching of such an antenna
with the transmission line is carried out by selecting the excitation point and optimizing its con-
figuration. Modeling and calculation of such antennas are carried out in specialized CAD software
packages. Modern programs allow you to simulate an antenna in a translucent case in the pres-
ence of the influence of various structural elements of the antenna system and thus allow you to
make a fairly accurate engineering analysis. The design of a multi-band antenna is carried out in
stages. First, a microstrip printed antenna with a wide bandwidth is calculated and designed for
each operating frequency range separately. Next, printed antennas of different bands are combined

693
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on September 05,2022 at 09:55:47 UTC from IEEE Xplore. Restrictions apply.
 2022 PhotonIcs & Electromagnetics Research Symposium (PIERS), Hangzhou, China, 25–27 April

(Fig. 3) and their joint work is considered, eliminating, if possible, the causes of mutual influence.
However, such modeling for multi-element antenna arrays is quite resource-intensive and requires
either large computing power or long sequential calculations. One of the traditional solutions is
parallel optimization with the definition of the objective function, and various evolutionary and
hybrid algorithms are used for this, for example, to optimize the inter-element distance in the
antenna array.

Figure 3: Layouts of microstrip antenna array.

Figure 4: Layouts of microstrip antenna array.

An artificial neural network, also simply called a neural network (NN), is an interconnected group
of artificial neurons (simply called neuron) that uses a mathematical or computational model for
information processing based on a connectionist approach to computation.
The neuron receives input from some other neurons, or perhaps from an external source. Each
input has an associated weight w, which can be modified so as to model synaptic learning. A NN
may consist of an input layer, one or several hidden layers and an output layer. A typical NN
structure is depicted in Fig. 4. NNs are commonly trained so that a particular input leads to a
specific target output by adjusting the values of the weights between neurons.
The network is adjusted, based on a comparison of the output and the target, until the network
output matches the target. To use neural networks as surrogates for full-wave analyses in microwave
structure optimization, one must first build a collection of input-output data with the structure
of interest (e.g., geometrical parameters and S-parameters of filter). The data are then usually
divided into two groups; the first group is used to train the neural networks whereas the second
group is used to validate the trained networks.
6. CONCLUSION
In the early days of high-level numerical computing languages, the thinking was that the perfor-
mance of the high-level language did not matter so long as most of the time was spent inside the
numerical libraries. These libraries consisted of blockbuster algorithms that would be highly tuned,
making efficient use of computer memory, cache, and low-level instructions. What the world learned

694
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on September 05,2022 at 09:55:47 UTC from IEEE Xplore. Restrictions apply.
2022 PhotonIcs & Electromagnetics Research Symposium (PIERS), Hangzhou, China, 25–27 April

was that only a few codes spent a majority of their time in the blockbusters. Most codes were being
hampered by interpreter overheads, stemming from processing more aspects of a program at run
time than are strictly necessary.
The project expects the results of analytical calculation and their comparison with the char-
acteristics obtained by electrodynamic modeling in the software packages CST Microwave Studio,
Altair FEKO, as well as mathematical programming devices MATLAB, FAZAR, Python and Julia.
Among the expected results, we can distinguish:
- creation of a mathematical model of antenna arrays, developed on the basis of a system
analysis of the technical features of radiating structures and allowing to describe antenna
array composed of aperiodic subarrays;
- patent research and verification of the patent purity of the developed products and software
planned for patenting has been provided;
- development of a method for optimizing the placement of subarrays with a number of elements
up to 256, based on modernized evolutionary genetic algorithms (registration of a patent for
a utility model);
- software for solving the problem of optimizing the structure of antenna arrays, based on the
proposed algorithm and allowing you to design antenna arrays in the terahertz frequency
range (registration of a patent for a computer program);
- simulation electrodynamic model - simulation results showing the effectiveness of the pro-
posed algorithm and the structures of antenna arrays obtained with its help, allowing for a
comparative analysis with achievements in the field of digital phased antenna array design;
- determination of the main dependencies of the spatial characteristics of MIMO-type antenna
arrays on the design parameters of their elements. If a number of hypotheses are proved,
these dependencies can form the basis for creating a universal mathematical apparatus for
calculating non-equidistant inhomogeneous antenna arrays of arbitrary shape.

ACKNOWLEDGMENT
The reported study was funded by Russian Foundation for Basic Research (RFBR) according to
the research project No. 20-37-90072.
REFERENCES
1. Bozzi, M., et al., “Future architectures for European space agency deep space ground stations,”
IEEE Antennas Propag. Mag., Vol. 54, No. 1, 254–263, Feb. 2012.
2. Jha, R. M., S. A. Bokhari, V. Sudhakar, and P. R. Mahapatra, “Applicability of Artificial
Intelligence languages to solving the scattering and diffraction problems using a personal com-
puter,” Digest on Antennas and Propagation Society International Symposium, 1989.
3. Weng, K. and S. Poynting, “Artificial intelligence in CAD/CAM,” Proc. of 1st National CAD/-
CAM Conference, 1995; New and Emerging Technologies for Industrial Competitiveness, 1995.
4. Gladyshev, A. I. and A. O. Zhukov, “Advantages and disadvantages of simulation using neu-
ral networks,” Series: Complex Systems: Models, Analysis and Management, No. 4. 53–55,
Bulletin of the Russian new University, 2013 (in Russian).
5. Zhukov, A. O., A. I. Zakharov, M. E. Prokhorov, N. I. Shakhov, V. S. Gedzyun, A. I. Glady-
shev, M. S. Tuchin, V. G. Moshkalev, and N. L. Krusanova, “Modeling of the background
environment of near-earth space,” MINTSEVSKIE READINGS Proceedings of the Third All-
Russian Scientific and Technical Conference of Young Designers and Engineers, Dedicated
to the 70th Anniversary of the Radio Engineering Institute Named After Academician A. L.,
160–169, 2016 (in Russian).
6. Chen, X. H., X. X. Guo, J. M. Pei, and Y. W. Man, “A hybrid algorithm of differential evolution
and Machine Learning for electromagnetic structure optimization,” Proc. 32nd Youth Acad.
Annu. Conf. Chin. Assoc. Autom., 755–759, 2017.
7. Yao, H. M., L. J. Jiang, and Y. W. Qin, “Machine learning based method of moments (ML-
MoM),” IEEE International Symposium on APS/URSI, Jul. 2017.
8. Yao, H. M. and L. J. Jiang, “Machine Learning based neural network solving methods for
the FDTD Method,” 2018 IEEE International Symposium on Antennas and Propagation &
USNC/URSI National Radio Science Meeting, 2321–2322, 2018.

695
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on September 05,2022 at 09:55:47 UTC from IEEE Xplore. Restrictions apply.
 2022 PhotonIcs & Electromagnetics Research Symposium (PIERS), Hangzhou, China, 25–27 April

9. Jiang, L. J., H. M. Yao, H. H. Zhang, and Y. W. Qin, “Machine Learning based computa-
tional electromagnetic analysis for electromagnetic compatibility,” 2018 IEEE International
Conference on Computational Electromagnetics (ICCEM), 1–2, 2018.
10. Chu, H.-S., E.-P. Li, E. Liu, and J.-L. Dubard, “Coupled computational intelligence and time-
domain method for design of the microwave devices,” Proceedings of Asia-Pacific Microwave
Conference, 2006.
11. Robinson, J. and Y. Rahmat-Samii, “Particle swarm optimization in electromagnetics,” IEEE
Trans. Antennas Propag., Vol. 52, No. 2, 397–407, Feb. 2004.
12. Jin, N. and Y. Rahmat-Samii, “Parallel particle swarm optimization and finite-difference time-
domain (PSO/FDTD) algorithm for multiband and wide-band patch antenna designs,” IEEE
Trans. Antennas Propag., Vol. 53, No. 11, 3459–3468, Nov. 2004.
13. Jain, S. K., A. Patnaik, and S. N. Sinha, “Neural network based particle swarm optimizer
for design of dual resonance X/Ku band stacked patch antenna,” 2011 IEEE International
Symposium on Antennas and Propagation (APSURSI), 2932–2935, Spokane, WA, 2011.

696
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on September 05,2022 at 09:55:47 UTC from IEEE Xplore. Restrictions apply.