ArticlePDF AvailableLiterature Review

Application of Reinforcement Learning and Deep Learning in Multiple-Input and Multiple-Output (MIMO) Systems

December 2021
Sensors 22(1):309

December 2021
22(1):309

DOI:10.3390/s22010309

License
CC BY 4.0

Authors:

Antonio Coronato

Italian National Research Council

Muddasar Naeem

Italian National Research Council

Giuseppe De Pietro

Italian National Research Council

The current wireless communication infrastructure has to face exponential development in mobile traffic size, which demands high data rate, reliability, and low latency. MIMO systems and their variants (i.e., Multi-User MIMO and Massive MIMO) are the most promising 5G wireless communication systems technology due to their high system throughput and data rate. However, the most significant challenges in MIMO communication are substantial problems in exploiting the multiple-antenna and computational complexity. The recent success of RL and DL introduces novel and powerful tools that mitigate issues in MIMO communication systems. This article focuses on RL and DL techniques for MIMO systems by presenting a comprehensive review on the integration between the two areas. We first briefly provide the necessary background to RL, DL, and MIMO. Second, potential RL and DL applications for different MIMO issues, such as detection, classification, and compression; channel estimation; positioning, sensing, and localization; CSI acquisition and feedback, security, and robustness; mmWave communication and resource allocation, are presented.

The reinforcement learning problem.

…

MIMO communication.

…

Number of papers from 2010 to 2021. Next, we have considered different categories. Therefore, Figure 4 reports the total number of papers concerning different issues of MIMO communication and exploiting RL and/or DL technologies.

…

Number of papers surveyed by category.

…

shows how the surveyed papers are distributed with respect to the analyzed categories. Note that such papers are more or less equally distributed. From Figures 4 and 5, it is possible to deduce that most of the article concerns three categories (Sections 4.1, 4.2 and 4.4). In these three categories, we have considered more topics and there are more papers related to them. Similarly, the number of papers related to the other three categories (Sections 4.3, 4.5 and 4.6) is almost the same, but definitively lower. Moreover, there is also a significant amount of work presented in Section 4.7 that we have not listed among the previous six categories. Most of the works presented in this section are related to channel encoding-decoding and adaptive control in uncertain MIMO nonlinear systems.

…

Figures - uploaded by Antonio Coronato

Content may be subject to copyright.

Content uploaded by Antonio Coronato

Content may be subject to copyright.





Citation: Naeem, M.; De Pietro, G.;

Coronato, A. Application of

Reinforcement Learning and Deep

Learning in Multiple-Input and

Multiple-Output (MIMO) Systems.

Sensors 2022,22, 309. https://

doi.org/10.3390/s22010309

Academic Editor: Biswanath

Samanta

Received: 6 December 2021

Accepted: 24 December 2021

Published: 31 December 2021

Publisher’s Note: MDPI stays neutral

with regard to jurisdictional claims in

published maps and institutional afﬁl-

iations.

Licensee MDPI, Basel, Switzerland.

This article is an open access article

distributed under the terms and

conditions of the Creative Commons

Attribution (CC BY) license (https://

creativecommons.org/licenses/by/

4.0/).

sensors

Review

Application of Reinforcement Learning and Deep Learning in

Multiple-Input and Multiple-Output (MIMO) Systems

Muddasar Naeem * , Giuseppe De Pietro and Antonio Coronato

Institute of High Performance Computing and Networking, National Research Council of Italy,

80131 Napoli, Italy; giuseppe.depietro@icar.cnr.it (G.D.P.); antonio.coronato@icar.cnr.it (A.C.)

*Correspondence: muddasar.naeem@icar.cnr.it

Abstract:

The current wireless communication infrastructure has to face exponential development

in mobile trafﬁc size, which demands high data rate, reliability, and low latency. MIMO systems

and their variants (i.e., Multi-User MIMO and Massive MIMO) are the most promising 5G wireless

communication systems technology due to their high system throughput and data rate. However,

the most signiﬁcant challenges in MIMO communication are substantial problems in exploiting the

multiple-antenna and computational complexity. The recent success of RL and DL introduces novel

and powerful tools that mitigate issues in MIMO communication systems. This article focuses on

RL and DL techniques for MIMO systems by presenting a comprehensive review on the integration

between the two areas. We ﬁrst brieﬂy provide the necessary background to RL, DL, and MIMO.

Second, potential RL and DL applications for different MIMO issues, such as detection, classiﬁcation,

and compression; channel estimation; positioning, sensing, and localization; CSI acquisition and

feedback, security, and robustness; mmWave communication and resource allocation, are presented.

Keywords:

reinforcement learning; deep learning; MIMO systems; signal; channel estimation; detec-

tion communication; BS; positioning; localization; CSI; resource allocation; mmWave communication

1. Introduction

The main problem with the current wireless communication infrastructure is its depen-

dence on either increasing the spectrum or densifying the cells to obtain the targeted area

throughput. Unfortunately, such resources are scarce and are approaching their saturation

level in near future. Moreover, increasing the spectrum or densifying the cells increases the

hardware price and latency. The spectral efﬁciency, which can enhance the area throughput,

has remained essentially unchanged during the fast development in the wireless systems.

Therefore, a wireless access technology must improve the wireless area throughput without

increasing the spectrum or densifying the cell to fulﬁll the essentials requirements of the

wireless carriers.

MIMO is the most promising and fascinating wireless access technology that is able to

deliver the requirements of 5G and beyond networks. The performance of communication

systems has been enhanced thanks to the use of MIMO schemes. MIMO utilizes many

dimensions that account for multiple antennas, multiple users, and time and frequency

resources. Multi-User MIMO (MU-MIMO) is a MIMO system with one BS equipped

with many antennas and provides service to more than one downlink user in a one-time

slot. Massive MIMO (Ma-MIMO) further extends MIMO systems by using hundreds

and thousands of antennas at BS to enhance throughput and spectral efﬁciency. MIMO

technology is integrating bandwidth, radios, and antennas to achieve higher speed as well

as capacity for the incoming 5G [

]. The ability of Ma-MIMO, in particular, to improve

spectral efﬁciency and throughput has turned it a powerful technology for emerging

wireless standards [2,3].

Recent advances in AI and ML, speciﬁcally the success of RL [

] and DL, have brought

many signiﬁcant applications and advancement in different research areas including

Sensors 2022,22, 309. https://doi.org/10.3390/s22010309 https://www.mdpi.com/journal/sensors

Sensors 2022,22, 309 2 of 41

robotics and autonomous control [

–

], communication and networking [

–

], natu-

ral language processing [

–

], games and self-organized systems [

–

], autonomous

IoT [

–

], scheduling management and conﬁguration of resources [

–

], computer

vision [

–

], and healthcare [

–

]. Embedding AI technologies like DRL into the 5G

mobile, MIMO communication, and wireless communication systems is well justiﬁed [

More precisely, data produced by mobile environments are increasingly heterogeneous

due to their collection from many sources with different formats, and they have complex

correlations.

In the areas of MIMO systems, recently RL and DL have been applied as an emerging

solution to address many challenges efﬁciently. These technologies have introduced many

solutions to different aspects of MIMO communication such as signal detection, classiﬁ-

cation, and compression; positioning, sensing, and localization; security and robustness;

mmWave communications; and resource allocation. AI-enabled MIMO communication is

an optimal tool that can give wireless systems the ﬂexibility, intelligence, and efﬁciency

needed to handle the scarce radio resource efﬁciently and enable top quality of service

to the customers. In our work, we are making a research contribution by providing a

comprehensive overview of the potential applications of RL, DL, and the mixture of both

in different areas of MIMO communication.

The remaining of the paper is organized as follows. Section 2presents a brief intro-

duction to RL, DL, and MIMO technology. Section 4presents a comprehensive application

of RL and DL in different areas of MIMO communication. Next, we present statistics and

impacts in Section 5, followed by a discussion in Section 6. Section 3overviews some recent

survey papers. This survey is concluded in Section 7.

2. Technical Background

This section provides a brief introduction to RL, DL, and MIMO systems.

2.1. Reinforcement Learning

RL is a sub-ﬁeld of Machine Learning where an agent interacts with an environment to

achieve a goal, and learning takes place interaction-after-interaction. This section introduces

some basic mechanisms and terminology. A detailed presentation of RL can be found in [

In RL, an

Agent

is an entity (algorithm/robot/player, etc.) that interacts with a given

environment (problem/smart space/game, etc.) by performing actions, and receives a

feedback (penalty/reward) from the environment after any action selected as described in

Figure 1. The reward is the mechanism that enables the agent to understand whether the

action selected has produced a positive or a negative effect concerning the ﬁnal goal.

Policy

is a strategy that indicates to the agent which action to select in every state of

the environment. The agent has to learn the optimal policy, that is, the one that maximizes

the cumulative reward over the long run.

A RL problem is deﬁned as a

MDP

. A MDP is a tuple

γ)

, where

is a set

of states;

is a set of actions;

Pa=Pr(st+1=s0|st=s

at=a)

is the transition probability

(i.e., the probability of achieving

at time

1, after having selected

at time

Ra(s

s0)

is the expected reward or immediate reward obtained when transitioning from

state sto state s0when action awas taken, respectively; and γis a discount factor.

RL schemes are normally categorized into two major types: model-free and model-

based algorithms. Model-based RL algorithms need a precise description of the dynamics

of the environment in terms of the state-transition probability distribution. These methods

(e.g., DP) compute the optimal policy by solving systems of equations. Whereas, model-free

RL techniques are adopted when there is not a precise description of the model or its

solution would be too complicated. This class of algorithms interacts directly with the

environment (or with an emulator of it) using Trial&Error schemes to learn the optimal

policy. In inverse RL [

–

], we study an agent’s objectives, values, or rewards with the

help of employing insights of its behavior. Several methods are available (e.g., MMC, TD,

Sensors 2022,22, 309 3 of 41

etc.). An overview of such methods is reported in [

], whereas a guideline useful to help

to choose the algorithm depending on the kind of problem is deﬁned in [35].

Figure 1. The reinforcement learning problem.

2.2. Deep Learning

DL has revolutionized many research areas with its ability to learn better models from

huge volumes of data [

]. Such technology relies on a new generation of Artiﬁcial Neural

Networks (ANNs) called Deep Neural Networks (DNNs). This subsection presents a brief

overview of DL techniques before approaching the next section.

The premises is that the performance of a DNN is generally superior to the one of a

classic ANN at the cost of greater training time that, however, can be reduced by using

advanced hardware (e.g., GPU) and/or special techniques (e.g., Transfer Learning) [

The design of a DNN is crucial for success. We start this subsection by discussing some of

the most used DL architectures.

•Convolutional Neural Networks

Convolutional Neural Networks (CNNs) are ANNs with a much higher number of

layers and nodes. They are typically adopted for images classiﬁcation. An CNN needs

less preprocessing as compared to other classiﬁcation schemes. Relevant ﬁlters are

used in CNNs to capture the temporal and spatial dependencies in the image [

The most common CNN architectures are ZFNet, ResNet, GoogleNet, VG-GNet,

AlexNet, and LeNet [48].

•Recurrent Neural Networks

RNN is another important architecture for DL. Differently from CNN, where the

layers are sequentially connected, in a RNN there are some nodes whose output is

reported back in the input of a previous node. In this way, the network is capable of

remembering some information time-related. Recurrent Neural Networks (RNNs),

indeed, are massively applied for time-series analysis and prediction in a conﬁguration

called LSTM [49].

•Generative Adversarial Networks

A Generative Adversarial Network (GAN) consists of two sub-networks—the dis-

criminator and the generator, where the later produces the content and the former

validates it. GAN adopts feed-forward and relies typically on CNNs [50].

•Deep Belief Networks

Deep Belief Networks (DBNs) are generative neural networks with undirected con-

nections between some layers called Restricted Boltzmann Machines. These layers

can be trained using a very fast unsupervised learning algorithm called Contrastive

Sensors 2022,22, 309 4 of 41

Divergence. In DBNs, hidden patterns are learned globally, while in every layer of

other deep nets complex patterns are learned progressively [51].

•Autoencoders

Autoencoders are applied to reduce the dimension of data and to detect problems. The

ﬁrst layer in an autoencoder is an encoding layer, whereas the transpose of it is used

as a decoder. Training is unsupervised and the Regression/Classiﬁcation problems

may be addressed and optimized using Stochastic gradient descent. Input data are

translated to a latent space denoted by the encoder, as given below:

h=f(x)(1)

Input data are reconstructed from the latent space denoted by the decoder as described

below.

r=g(h)(2)

Autoencoders can be represented essentially by the below equation.

is the decoded

output and it will be identical to input x

g(f(x)) = r(3)

•Radial Basis Function Neural Networks

RBFNN is a type of ANN that utilizes radial basis functions (RBF) as activation

functions. The output of the RBFNN is a linear combination of RBF of the neuron

parameters and inputs. RBNN has only one hidden layer which is known as a feature

vector. Training in RBNN is faster than in MLP but classiﬁcation in RBNN takes more

time than MLP.

2.3. MIMO Communication

MIMO is a wireless technology that employs multiple antennas at transmitters and

receivers sides to communicate more data simultaneously as can be seen in Figure 2. MIMO

are supported by all wireless devices compliant to 802.11n standard. This subsection brieﬂy

reviews MIMO technology and its different schemes.

MIMO communication uses a multi-path, which is a natural radio-wave phenomenon.

Multi-path-transmitted signals may bounce off objects, like ceilings, walls, etc., and arrive

at the receiver multiple times at different times and angles. Before the inception of MIMO,

interference occur due to multi-path, which can slow down the communication. MIMO

technology with multi-path, however, uses smart transmitters and receivers with the

addition of spatial dimension that grants enhanced performance and range.

MIMO improves signal-capturing of the receiver by empowering antennas to combine

signals coming from multiple paths at different times. Smart antennas get the beneﬁt of

spatial diversity technique that makes surplus antennas useful. The antennas can increase

the range by adding receiver diversity when outnumbered spatial streams.

•Single User MIMO

Single user MIMO, or multi-antenna MIMO, has more than one antenna both at the

transmitter side and receiver side. There are some special variants of MIMO such

as “Multiple-input single-output” (MISO) (only one antenna at the receiver side),

“Single-input multiple-output” (SIMO) (single antenna at the transmitter side), and

a special scenario when transmitter as well as receiver have one antenna is called

SISO [52].

•Multi-user MIMO

Multi-User MIMO MU-MIMO has been considered in recent WiMAX and 3GPP

standards as a candidate technology by various companies such as Freescale, Nokia,

Philips, Huawei, TI, Ericsson, Qualcomm, Intel, and Samsung. MU-MIMO systems are

more suitable for low-complexity mobile phones with a few receiving antennas, while

single-user MIMO’s are more suitable for complex devices having many antennas due

Sensors 2022,22, 309 5 of 41

to their higher per-user throughput. Moreover, enhanced MU-MIMO uses advanced

precoding and decoding techniques.

•Cooperative MIMO (CO-MIMO)

Cooperative MIMO (CO-MIMO) employs multiple surrounding BS to jointly trans-

mit and receive signals to and from users. This prevents inter-cell interference in

neighboring BS as may be experienced with traditional MIMO systems.

•Macrodiversity MIMO

Macrodiversity MIMO is a type of space diversity approach that applies many trans-

mit/receive BS for coherent communication with single/multiple users. It is possible

that users are distributed in a coverage area that has the same resources of time and

frequency [

–

]. The transmitters, as well as the users in multi-user microdiver-

sity MIMO, are far apart as compared to that of conventional microdiversity MIMO

approaches (e.g., SU-MIMO). As a result, each constituent connection in the virtual

MIMO connection has a unique average link SNR. Macrodiversity MIMO techniques

face some theoretical and practical challenges. One of the most fundamental issues is to

get knowledge about how aggregated system capacity is affected by different average

link SNRs and the performance of users individually in fading environments [56].

•Massive MIMO

Massive MIMO Ma-MIMO is a scheme where the number of terminals is inferior to of

BS antennas [

]. The maximum beneﬁts of the massive MIMO in a rich scattering

environment can be obtained by applying simple beamforming schemes such as zero-

forcing (ZF), maximum ratio combining (MRC) [

], or maximum ratio transmission

(MRT) [

]. However, it is difﬁcult to achieve these advantages without the availability

of accurate CSI.

Figure 2. MIMO communication.

3. Related Survey Papers

This section will review some recent related surveys, their contribution, and limitations

as well as the contribution of our work.

The list of related survey papers is shown in Table 1. The work in [

] considers the

area of wireless networks and compares application of three DRL sub-techniques: DDPG,

NEC, and VBC for optimization. More precisely, the comparison of three DRL approaches

is carried out on experiments being performed on a real-world network operation dataset.

Although authors have performed extensive experiments, they limited their analysis to

three methods only and very little knowledge can be extracted on the application of DRL

in MIMO systems.

Sensors 2022,22, 309 6 of 41

Another survey paper [

] focused on the radio-resource allocation in multi-cell

networks via DL. Authors have also compared methods qualitatively, in terms of their

data training and techniques, inputs/outputs, and objectives. A supervised DL design is

presented as a solution to power allocation and sub-band problems in a multi-cell network.

The mmWave communication is discussed in [

] using channel estimation and

signal processing methods, respectively. The former work comprehensively reviews the

state-of-the-art channel estimation methods linked to different mmWave system frame-

works. Ma-MIMO mmWave was also considered, but without the use of RL and DL

methods. The Ma-MIMO wave aspect is also considered in [

], but it mainly reviews

challenges of signal processing in mmWave wireless systems with a particular concern for

higher carrier frequencies MIMO communication.

An encyclopedic overview of the application of DL in wireless and mobile networking

is presented in [

]. The authors have bridged the research gap between DL and wireless

and mobile networking by highlighting the crossovers between these two areas. However,

the area of the MIMO system (i.e., the focus of our current work) was not appropriately

discussed.

The work in [

] presents a comprehensive state-of-the-art on ML based link quality es-

timators generated from empirical data. The ML-based link quality estimation architectures

are analyzed and existing open-source datasets are also reviewed.

The Ma-MIMO systems are highlighted in [

] while presenting the enabling tech-

niques needed for 5G and 6G architecture. The fundamental challenges associated with

signal detection, energy efﬁciency, user scheduling, precoding, channel estimation, and

pilot contamination in a Ma-MIMO communication are discussed. The authors have also

outlined visible light communication, ultra Ma-MIMO, terahertz communication, as well

as ML and DL for Ma-MIMO systems, but they did not consider other areas of MIMO

communication.

A survey paper on DRL techniques that was proposed to solve emerging problems in

communications and networking is presented in [

]. The authors have addressed issues

such as connectivity preservation, network security, data ofﬂoading, data rate control, wire-

less caching, and dynamic network access. Moreover, DRL applications for data collection,

resource sharing, and trafﬁc routing are discussed. However, MIMO communication was

partially discussed in few subsections. Similarly, the work in [68] presents an overview of

array signal processing techniques for Ma-MIMO communication.

An overview of the DL-based cybersecurity applications for mobile and wireless

networks is presented in [

]. The authors have addressed cybersecurity features like

privacy preservation, software attacks, attacks, and infrastructure threads.

Different cases of 5G wireless communication including cybersecurity, energy efﬁ-

ciency, caching, Ma-MIMO, channel coding, and modulation classiﬁcation based on AI

techniques are discussed in [

]. The authors were interested in more general AI-based

applications for 5G wireless communication.

Although there are few review papers related to applications of ML, RL, and DL

reported in Table 1, they do not discuss the applications of RL and DL for MIMO communi-

cations. There is a need for a review that particularly outline useful applications of RL and

DL for different aspects of MIMO communication.

In this paper, we have presented a comprehensive state-of-the-art on the application

of RL and DL in different aspects of MIMO communication such as detection, classiﬁcation,

and compression; channel estimation; positioning, sensing, and localization; CSI acquisition

and feedback, security and robustness; mmWave communication; and resource allocation.

We have also listed the contribution of some survey papers, their limitations for MIMO

communication areas, and our contribution in Table 1.

Sensors 2022,22, 309 7 of 41

Table 1. List of related surveys.

Paper Technology Year Area Contribution Limitation

[60] DRL 2020 Wireless Network

Optimization

Only three DRL techniques: DDPG,

NEC, and VBC, are considered for

wireless network optimization.

Their performances are compared

in terms of rate and convergence

speed improvement.

Only three DRL

methods are taken into

account without

concerning about

MIMO aspects.

[62]

Channel

Estimation

Techniques

2020 mmWave

communication

Review of the channel estimation

methods associated with the

different mmWave system

architectures

Only one area of MIMO

communication (i.e.,

mmWave) is discussed,

as well as DL and RL

techniques are not

considered.

[63]Signal

Processing 2016

mmWave Ma-MIMO

communication

Survey of signal processing

challenges in mmWave systems,

especially focusing on issues due to

utilizing MIMO communication at

higher carrier frequencies.

Only mmWave

communication with

signal processing

techniques are

discussed, as well as DL

and RL techniques are

not considered.

[61] DL 2019 Multi-cell networks

Review the application of DL for

the radio resource allocation in

multi-cell networks.

Focused only on

resource allocation.

[64] DL 2019 Mobile and Wireless

Networking

Application of DL in mobile and

wireless networking

MIMO systems are not

considered.

[65] ML 2021 Link Quality

Estimation

Review ML-based link quality

estimation models. It addresses

quality requirements and standard

design steps perspectives using

performance data.

General ML techniques

are concerned.

[66] — 2020 Ma-MIMO

Presents fundamental challenges

related to signal detection, energy

efﬁciency, user scheduling,

precoding, channel estimation and

pilot contamination in a Ma-MIMO

system, and solutions to these

challenges.

General aspects of

Ma-MIMO are

considered without

application of DL

and RL.

[67] DRL 2019 Communications

and Networking

Connectivity preservation, network

security, data ofﬂoading, data rate

control, wireless caching, and

dynamic network access issues

are addressed.

MIMO systems are not

discussed in detail.

[69] DL 2021 Cybersecurity in

Mobile Networks

Cybersecurity aspects: privacy

preservation, software attacks,

attacks and infrastructure threads

are discussed.

No MIMO application.

Sensors 2022,22, 309 8 of 41

Table 1. Cont.

Paper Technology Year Area Contribution Limitation

[70] AI 2020 5G Wireless Systems

An in-depth review of AI for 5G

wireless communication systems

including cyber-security, network

management, and radio resource

allocation

Only Ma-MIMO were

discussed in one

subsection using general

AI approaches.

[68]

Array Signal

Processing

Techniques

2019 Enhanced Massive

MIMO

A review of array signal processing

in Ma-MIMO communications.

Only Ma-MIMO systems

are considered with

array signal processing

techniques. No

application of DL in

MIMO.

Our

work RL and DL 2022 MIMO

communication

Comprehensive overview of the

application of RL and DL in

different aspects of MIMO

communication.

The tutorial aspect of

our survey only presents

a brief introduction to

RL and DL.

4. RL and DL Application in MIMO

This section presents a comprehensive review of the applications of DL and RL in

different areas of MIMO.

4.1. Detection, Classiﬁcation, and Compression

A growing interest has been developed in recent years to apply DRL techniques

to optimize operations of wireless network [

]. Incorporating RL and DL into MIMO

detection has evolved as a promising method for future wireless communications [71].

An RL-based cognitive BF scheme for co-located MIMO radars is proposed in [

The RL-empowered optimization algorithm enables the MIMO radar to iteratively sense

the radar scene involving an unknown number of targets where the angular positions of

targets are unknown. Therefore, the protocol synthesizes a set of transmitted waveforms

whose relevant beam pattern is tailored to the learning. A BF algorithm based on online

model-free RL approach SARSA is introduced in [

] to study the case of multi-target

detection for a Ma-MIMO cognitive radar when the disturbance is unknown. The study

has shown that RL enabled method able to detect the targets in a dynamic environment

with very low SNR. The work may be improved further by reﬁning the DoA estimate of

the detected targets having a disturbance with unknown distribution.

A DNN architecture is presented in [

] in the context of MIMO detection. The authors

have considered both cases, i.e., constant MIMO channel and multiple varying channels.

A DNN model is also used for MIMO detection in [

]. Two architectures have been

proposed, i.e., “a standard fully connected multi-layer network and a Detection Network

(DetNet)”. The model of DetNet is designed by unfolding the iterations of a projected

gradient descent approach into a network. The authors of [

] propose a model-driven DL

model for MIMO detection. The model is particularly designed by unfolding the iterative

algorithm. Deep unfolding is employed in [

] for MIMO detection for QPSK and BPSK

constellation cases. A new DL-based detector using BP algorithm belief propagation is

discussed in [

], which combines the belief propagation method with the DL techniques.

The MIMO factor graph is used for the detection of signals from the transmitter by pressing

likelihood ratios messages.

A quasi-static ﬂat channel with many antennas is considered for detection of multilevel

modulation symbols using DNN in [

] and partial learning using NN detection method for

Ma-MIMO is given in [

]. A CSI sensing and recovery framework is proposed in [

]. The

method learns to use channel structure effectively from training samples and authors have

shown through experimental results that their method can recover CSI with reasonably

better reconstruction quality. This work was further extended in [

] by developing a

Sensors 2022,22, 309 9 of 41

real-time CSI feedback framework known as “CsiNet-LSTM”. CsiNet-LSTM signiﬁcantly

improves recovery quality and enhances the trade-off between complexity and compression

ratio by learning directly spatial structures that are combined with time correlation from

training samples of Ma-MIMO with time-varying channels.

The authors of [

] have studied the feasibility of using DL techniques for auto-

matic classiﬁcation of modulation types of received signals. For example, an end-to-end

CNN-based automatic modulation classiﬁcation is proposed in [

] that extracts features

automatically from the long symbol-rate observation sequence along with the estimated

SNR. A “deep complex-valued convolutional network” that does not rely explicitly on

the Fourier transform is designed in [

] to recover bits from time-domain orthogonal

frequency-division multiplexing signals. A novel feedback network CRNet is presented

in [

] using advanced training techniques to get superior performance by extracting CSI

features on multiple resolutions.

A likelihood function learning technique for MIMO systems having a one-bit analog-to-

digital converter is proposed in [

] using an RL method. The main idea of the work is to use

input–output samples that are obtained by detecting the data and conduct compensation

in the likelihood function for any mismatch. In another work [

], a robust MLD technique

is considered for an uplink Ma-MIMO communication system with low precision ADCs

under non-perfect CSI at a receiver. The same problem is addressed in [

] for a wideband

SIMO system with one-bit ADCs.

A letter in [

] considers the case of uplink Ma-MIMO network with 1-bit ADCs to

develop a DL-based channel estimation mechanism where the prior channel estimation

record and DNN are able to learn the non-trivial mapping from quantized received mea-

surements to channels. A DL framework for channel estimation and detection has been

investigated in [

], and the results suggested that the proposed DL schemes lead to better

performance in the large SNR regime. However, the architecture provides good results only

for relatively small MIMO dimensions. Another detection method for MIMO optimized by

back-propagation NN is given in [92] to uplink the Ma-MIMO systems.

Two DNN based detectors, i.e., damped belief propagation (BP) and max-sum by

unfolding damped BP [

] and max-sum BP [

] methods, respectively, are designed

in [

]. However, the framework may be further improved for optimization of the DNN

architecture and efﬁcient training schemes. Joint signal detection and channel estimation of

a MIMO system using model-driven DL architecture are performed in [

]. The network for

signal detection is designed by unfolding the Orthogonal AMP detection method. Due to

the requirement of a few adjustable parameters for optimization, the proposed framework

can be easily and efﬁciently trained.

A MPD using DNN is introduced in [

] by modifying the message passing detector

technique to adjust the approximation error. The modiﬁcation was done to achieve good

performance and accelerate the convergence. The proposed architecture is then designed

by unfolding the modiﬁed MPD, and the DL method is used for the optimization of the

modiﬁcation factors. The scaling of DNN-enabled MIMO detectors [

] in terms of system-

atic complexity was performed in [

]. The framework applies a part of the DNN inputs by

scaling their values via weights that follow monotonically non-increasing functions. The

architecture is further improved by employing a sparsity-inducing regularization constraint

along with trainable weight-scaling functions. This improvement enables the model to keep

a balance between detection accuracy and complexity and at the same time, robustness to

variation in the activation patterns increases.

The performance of a large-scale MIMO receiver is investigated in [

100

]. The de-

ployment of the MIMO receiver is done using DNN and a low-density parity check code

to detect and decode disturbed signals. The model experimented with different large

scale MIMO conﬁgurations to get a trade-off between the performance and complexity.

An RL-based detection method for time-varying MIMO systems having one-bit ADCs is

developed in [

101

]. Input–output samples that are being received from data detection are

exploited to perform tracking of the temporal variations of likelihood functions. An MDP

Sensors 2022,22, 309 10 of 41

is modeled to handle the uncertainty of the information due to a data detection error and it

enhances the accuracy of the likelihood function. In the end, an RL algorithm is used to

solve the modeled MDP with less computational complexity.

Implementation methodologies for conventional MIMO transmitters of DL-based

signal detection are presented in [

102

]. The authors have used a DNN architecture of a

one-tap MIMO channel for signal detection while CNN and RNN models are applied with

a multipath fading channel. A DL framework for the detection of MIMO signal for high-

speed railway case is given in [

103

]. The proposed architecture is divided into two steps:

ofﬂine training process and online detection. Another detection scheme for large-scale

overloaded MIMO systems by employing DL is given in [

104

], where the optimization of

trainable internal parameters can be performed by using standard DL schemes, i.e., SGD

and back-propagation methods.

Two scenarios of channel information at the receiver using the DL model are consid-

ered in [

105

] for uplink pilot-assisted MU-MIMO systems. In the ﬁrst case, the channel

matrix is available at BS and the DNN is used as a detector and the channel matrix in

the second case is not known at BS. Signal detection for Ma-MIMO systems is performed

in [

106

] using a DL-based trainable AMP scheme. The trainable AMP model includes a

preprocessing layer and a few detection layers. Moreover, the network adds trainable

parameters to control prior mean-variance of MMSE denoiser. Another method for MIMO

detection known as MMNet is designed in [

107

]. MMNet’s architecture is based on the

idea of iterative soft-thresholding methods and applies a new training scheme that takes

advantage of spectral and temporal correlation in real channels to speed up the training.

The BP-MMSE-based algorithm is proposed in [

108

] that initiates from the MMSE

solution and updates the prior in every iteration with the loopy BP belief. The graph NNs

are used to prevent the complexity of computing MMSE, we use Graph Neural Networks

(GNNs). The graph NN model learns a message-passing scheme to solve the inference

problem on the same graph. A simpliﬁcation with three improvements is introduced in [

109

]

to simplify the detection network. The ﬁrst improvement is the reduction of the number

of inputs, and the second is changing the network from full to sparse connectivity and

decreasing the number of network layers by 50% to simplify network connection structure.

While the ﬁnal improvement is to optimize the loss function to prevent irreversible issues

with the matrix. With these improvements, the network complexity may be reduced from

O(64n2)to O(3n).

A DL empowered detector is designed in [

110

] that after an off-line training able to

detect signals communicated in a channel with impulsive noise. The proposed detector has

less complexity than the average sphere decoder complexity and shows better performance.

A Multisegment mapping model based on DNN for Ma-MIMO detection is proposed

in [

111

]. The proposed network is developed by optimizing the prior detection networks

(ScNet and DetNet) and minimizes network complexity. The authors have also introduced

an activation function to enhance the performance of Ma-MIMO detection for high-order

modulation cases. Tabu search detection in Ma-MIMO systems is considered in [

112

]. First,

A DNN model for symbol detection is proposed by optimizing the DetNet and ScNet

networks. Then, a tabu search algorithm based on DL is presented, where the starting

solution is approximated by the ﬁrst DNN model.

A likelihood ascent search detection approach based on CNN is given in [

113

] by using

a graphical detection type for uplink multiuser Ma-MIMO systems. The proposed method

needs lower average received SNRs to achieve better BER performance. Signal detection in

MIMO-OFDM system has been addressed in [

114

] by applying extreme learning machine

and autoencoder network. This combined architecture does not require the channel matrix

for the signal detection process. A novel NN architecture, radial basis function networks

are proposed in [

115

]. The model is optimized by a quantum genetic algorithm and is used

to solve signal detection issues in MIMO-OFDM.

Sensors 2022,22, 309 11 of 41

4.2. Channel Estimation

RL techniques, in particular, the DL method, have proven important tools to improve

the accuracy of channel estimation to enhance the performance of Ma-MIMO [

116

] utilized

in different applications such as Virtual reality, 5G systems, Internet of Things, and au-

tonomous driving [

117

]. These applications demand the design of wireless systems [

118

] to

provide ultra-low latency, large numbers of connections, and ultra-high data rates [

119

MIMO systems especially large-scale MIMO are one of the potential candidates to meet

these requirements. This subsection will present the application of RL and DL for MIMO

channel estimation.

Direction-of-arrival and channel estimation using RL techniques for Ma-MIMO sys-

tems are discussed in [

120

]. Authors have employed a DNN to realize end-to-end per-

formance and conduct an online–ofﬂine learning mechanism. This is an efﬁcient way of

learning wireless channel statistics and the spatial structures in the angle domain, while

only direction-of-arrival estimation using DL was considered in [

121

]. A DL method is

used in [

122

] for channel estimation in ultra Ma-MIMO systems. The methodology can

address the issues of high processing complexities and computation time with estimation

efﬁciency and accuracy.

Channel estimation for double directional channels with limited feedback for mmWave

Ma-MIMO is done in [

123

]. The BS estimates the downlink channel by recovering a low-

rank matrix, using samples of the compressed channel matrix and feedback from the

mobiles. This results in the prevention of doing resource-consuming tasks for users. The

letter in [

124

] applied DL for estimation of the uplink channels for mixed ADCs Ma-

MIMO systems. Authors have considered some antennas equipped with high-resolution

ADCs and some use low-resolution ADCs at the BS. First, a direct-input DNN is used to

estimate channels by employing the received signals of all antennas. Then, a selective-input

prediction DNN is applied for elimination of the adverse impact of the coarsely quantized

signals. Similar work was also done in [

125

] where low-resolution, ADC-quantized received

pilot signals are used with one of three different methods: (1) “High-resolution quantized

pilot + All low-resolution quantized pilot (High + All)”, (2) “High-resolution quantized

pilot + Argument of low-resolution quantized pilot (High + Arg)”, and (3) “High-resolution

quantized pilot + Modulus of low-resolution quantized pilot (High + Mod)”.

Channel estimation with few-bit ADCs in MIMO systems is the focus of the work

in [

126

]. A DNN is considered ﬁrst and then trained as a nonlinear MMSE channel estimator.

Next, the authors used a DNN for concurrent optimization of the MMSE channel estimator

and the training signal. The work in [

127

] also considers MIMO systems with low-resolution

ADCs and proposes DL based channel estimation method. Several deep multimodal

learning-enabled frameworks in Ma-MIMO systems for channel prediction are developed

in [

128

] by taking advantage of fusion levels and modality combinations. This work on

channel prediction in massive MIMO provides a signiﬁcant example that may be followed

in designing different deep multimodal learning-based communication approaches.

A new concept of channel mapping in space and frequency is introduced in [

129

]. The

mapping has been done between (1). Channels at one group of antennas and one frequency

band and (2) channels at another group of antennas and frequency band. Authors have ﬁrst

proved the existence of such channel-to-channel mapping under certain conditions and

then use DNN for learning this non-trivial channel mapping function efﬁciently. The joint

impact of nonlinear hardware impairments at UEs and the BS on the uplink performance

of single-cell Ma-MIMO in practical Rician fading channel is considered in [

130

] using DL

approach. The quality of estimation is improved as compared to distortion-aware and

distortion-unaware Bayesian linear MMSE.

A new channel estimation framework is proposed in [

131

] with the assistance of DL

technology to improve the channel estimation that is received by the least-squares method.

Authors have used a MIMO system with a multi-path channel case for simulations in

5G-and-beyond networks for the scenario of mobility expressed by the Doppler effects.

Although, the architecture is developed for an arbitrary number of transmitter and receiver

Sensors 2022,22, 309 12 of 41

antennas, but it can be generalized. In another work [

132

], the potential application of NN

to optimize a particular physical layer block in a communication system by considering

some salient characteristics of the emerging radio methods based on 5G standards such as

beamforming, MIMO, and mmWave are investigated.

A DL-enabled method for channel estimation to improve the performance of the

Ma-MIMO system is given in [

133

]. The used approach improves the recovery quality and

enhances the trade-off between the complexity and compression ratio of the Ma-MIMO

system. The method based upon using the CSI network combined with the gated recurrent

unit and dropout technique scheme was used to minimize overﬁtting during the learning

process. However, the use of the gated recurrent unit layers can increase the complexity

due to the subsequent expected increase in the run time. The authors of [

134

] propose two

channel estimation schemes using DL in TDD Ma-MIMO systems under the presence of

pilot contamination and claim to have better performance than the least-squares and the

covariance estimation methods in terms of the channel normalized mean square error for

imperfect timing synchronization and perfect one.

The theoretical analysis on the use of DL in MIMO system channel estimation is dis-

cussed in [

135

]. Authors have ﬁrst interpreted DL-based channel estimation by considering

multiple antennas system under linear, nonlinear, and inaccurate channel statistics and

have shown that DL estimator equipped with a rectiﬁed linear unit DNN is equal to that of

a piecewise linear function mathematically. The Bayesian learning method is used for the

estimation of channel parameters only of the interesting links in the desired cell only for the

interference connections from adjacent cells [

136

]. The authors have claimed the possibility

of obtaining an accurate estimation of the channel parameters with the exploitation of the

propagation properties of Ma-MIMO systems.

Channel estimation when both low-resolution ADCs and hybrid analog–digital pro-

cessing in Ma-MIMO systems are utilized is considered in [

137

]. The authors have formu-

lated the quantized sparse channel estimation into a sparse Bayesian learning architecture

and provide the solution with variational Bayesian technique. A Bayesian channel estima-

tion method based on the AMP algorithm in Ma-MIMO systems is introduced in [

138

], and

this framework requires statistical CSI. The derivation of the covariance is done for CSI

acquisition by analyzing the channel model in the beam domain.

A DL-enabled architecture for channel estimation and joint pilot design of MU-MIMO

channels is given in [

139

]. The pilot is designed using two-layer NNs and the channel

estimator is modeled using DNNs and they reduce MSE of channel estimation after join

training. Authors have also applied successive interference cancellations to minimize

the interference that may be present among the multiple users. The pilot design for the

MU-MIMO system using DL technology is also considered in [

140

] to reduce the sum of

MSE of channel estimation where the pilot signal of every user and the power assigned

to every pilot sequence is represented as a weighted superposition of orthonormal pilot

sequence basis and corresponding weight respectively. Moreover, DNN is used for the

optimization of power allocation to every pilot pattern to reduce the sum MSE where the

input is channel large-scale fading coefﬁcients and the pilot power allocation vector is

the output.

A decision-directed method for channel estimation using DNN is developed in [

141

]

for the MIMO system. The work was done for STBC MIMO systems by considering the

highly dynamic vehicular scenario. DNN was used for k-step channel prediction for STBC

while the DL empowered decision-directed channel estimator removes the requirement

for Doppler rate estimation where channels are time-varying quasi-stationary. Hybrid

precoding and channel estimation for multi-user mmWave MIMO system are considered

in [

142

] using DL compressed sensing method. The prediction of beamspace channel

amplitude is performed by training ofﬂine the channel estimation NN using simulated

environments and the reconstruction of the channel is done using the acquired indices

of entries of the dominant beamspace channel. Then, after the channel estimation, the

Sensors 2022,22, 309 13 of 41

quantized phase hybrid precoder is developed using DL and its training is performed

ofﬂine with approximate phase quantization.

Channel estimation with received SNR feedback is investigated in [

143

] to estimate the

MIMO channel coefﬁcients using the received SNR feedback from a receiver at a transmitter

based to reduce the MSE. Authors have considered time-varying fading and quasi-static

block fading cases for their experiments. In another study [

144

], fast and ﬂexible denoising

CNN has been used for channel estimation. The framework is suitable for a wide variety

of SNR levels with a variable noise level map in the shape of input. Channel estimation

for Terahertz Ultra-Ma-MIMO Systems with array-of-subarrays is considered in [

145

]

using a ﬁfteen-layer deep CNN spherical-wave scheme. The training labels are phase

shift matrix, the amplitude of the channel gain, and spherical-wave channel parameters

including elevation and azimuth angles. In the end, supervised learning is employed with

a self-deﬁned loss function using the labeled data and the training of the deep CNN is

performed ofﬂine and implemented online to conduct channel estimation.

Channel estimation using DL for MIMO systems for the case of multi-cell and

interference-limited is taken into account in [

146

]. The MIMO system considered in this

study is equipped with BSs and each BS serves many single antenna UE with a large

number of antennas. DNN is used on the deep image prior system to denoise the received

signal ﬁrst and conventional least-squares estimation is done. A blind wireless channel

and bandwidth-efﬁcient estimator for the uncoded space-time labeling diversity system is

designed in [

147

]. An NN-ML channel estimator with transmitting power-sharing is used

to perform blind channel estimation for the given system and to reduce the bandwidth

utilization. A wideband low-complexity DOA estimation scheme for Ma-MIMO systems is

given in [

148

] using principal component analysis NN. The framework prevents complex

angle pre-estimation by designing a focusing matrix and minimizes the complexity of

the eigenvalue decomposition by using the signal subspace estimation method. Another

DL-based channel estimator is designed in [

149

] by considering the impact of hardware

impairments in a multiple-antenna BS and UEs on the uplink performance.

Effective interference cancellation and reliable channel estimation are necessary for

improving the performance of MIMO UAC systems. A single carrier MIMO UAC has been

considered in [

150

] to study a robust receiver framework using Bayesian learning for itera-

tive channel estimation that is embedded in Turbo equalization. The proposed architecture

updates the joint estimates of a channel covariance matrix, residual noise, and channel

impulse response. Authors have also designed a “low complexity space-time soft decision

feedback equalizer” using Bayesian learning with successive soft interference cancellations.

Bayesian learning is also used in [

151

] as sparse learning via iterative minimization for the

MIMO UAC system. The authors have also implemented a linear MMSE enabled symbol

detection method by using conjugate gradient scheme and diagonalization characteristics

of circulant matrices.

A MIMO receiver with inadequate pilots in a fast fading channel is considered in [

152

]

as well as a DL-empowered Turbo-MIMO receiver that contains channel decoding, signal

detection, and modules. A short pilot sequence is used to generate an estimate of the

channel matrix by applying the linear MMSE method. Then, a re-estimate is done with

the support of symbols that are reliably estimated. Data symbols are re-detected using the

channel decoder’s soft statistics. Signal detection is performed at the receiver by applying

the expectation propagation technique as multi-layer deep feed-forward networks. A blind

channel estimation scheme based on DL technology is designed in [

153

] for OFDM-based

large-scale MIMO systems. A denoising CNN has been deployed to mitigate the remaining

channel and noise effect. This will help to detect the transmitted data symbols accurately at

the channel sounding step. The CSI of all used was then detected as virtual pilots at each

BS antennas.

A task-oriented quantization model with scalar ADCs based on DL for MIMO channel

estimation is designed in [

154

]. The proposed quantization system does not require explicit

recovery of the system model and proper quantization rule. Two receiver designs—pilot-

Sensors 2022,22, 309 14 of 41

assisted and model-drive—using DL for uplink MIMO systems are proposed in [

155

]. The

formal receiver is developed using a data-driven full connected NNs and the transmitted

signal can be recovered directly in this scheme in an end-to-end manner without the need

for channel estimation. The later receiver combines communication knowledge with DL

and divides the MIMO receiver into signal detection subnet and channel estimation subnet.

Moreover, the application of CNN estimators has been studied in [

156

] for MIMO-OFDM

channel estimation. The channel values of the reference signal are interpolated to estimate

the channel of the full OFDM resource element matrix. Authors have developed a 2D CNN

model using U-net, and a 3D CNN structure to tackle spatial correlation.

A joint design of channel estimator and pilot signal based on data-driven DL method-

ology for wideband Ma-MIMO systems are designed in [

157

]. High-dimensional channels

are reconstructed using DL from underdetermined measurements by exploiting the MIMO

channel’s angular-domain compressibility. The DNN architecture is employed for downlink

multiuser precoding, feedback, quantization, and distributed channel estimation for a FDD

Ma-MIMO system in [

158

], where a BS serves many mobile users. However, the feedback

from the users to the BS is rate-limited. A feedback and channel estimation mechanism

is developed in [

159

] using DL. The framework can estimate, compress and reconstruct

downlink channels for FDD Ma-MIMO systems.

4.3. Positioning, Sensing, and Localization

MIMO systems are meeting the growing need for reliable and faster communications

in wireless systems having a large number of terminals. MIMO systems can also be used to

estimate the position of a terminal utilizing multipath propagation in multiple antennas.

CSI-based user positioning systems using DL technology achieving a high accuracy without

any overhead have shown great potential in the MIMO system. These systems are capable of

positioning indoor users in both LoS and non-LoS environments with reasonable accuracy.

Moreover, with the availability of smart devices and recent development in AI-based

wireless systems [

160

], localization is enabling many location-based applications [

161

162

Demand for localization has increased because of more application of high accuracy, high

bandwidth, and location-based services [

163

]. These applications need precise localization

of users to enhance BF and resource management [

164

]. In this subsection, we will review

DL-based architectures for exploiting MIMO-based CSI to improve indoor localization and

positioning.

Fingerprinting has been an emerging research area for indoor positioning due to its

location-related characteristics and ubiquity [

165

]. A novel DL method for Ma-MIMO

ﬁngerprint-based positioning has been investigated in [

166

]. They have used CNNs with a

feedforward structure and measured channel snapshots. CNNs can compactly summarize

and generalize relevant positioning information for channels with large data sets, e.g.,

in highly clustered propagation cases with Line of Sight (LOS) or without LoS. Similarly,

an “angle-delay channel amplitude matrix” method for extraction of ﬁngerprint and a

deep convolutional NN-based localization scheme for Ma-MIMO-OFDM systems are

proposed in [

167

168

]. The latter method can overcome the error of modeling for ﬁngerprint

similarity calculation.

A deterministic “uplink-to-downlink mapping function” is revealed in [

169

] and a

sparse complex-valued neural network is proposed for the approximation of this function,

where the position-to-channel mapping is bijective. The authors have demonstrated that the

proposed architecture performs well as compared to the other network in terms of predic-

tion accuracy with remarkable robustness. Later on, some of these authors have formulated

the downlink channel prediction as a deep transfer learning problem and introduced the

direct transfer method [170] using the fully-connected NN framework, when the network

is trained in the form of classical DL and ﬁne-tuned for new environments afterward.

The achievable rate of the mmWave Ma-MIMO system can be improved by minimiz-

ing training pilots using the DL model for sensing joint channel and hybrid precoding

framework [

171

]. According to this study, the channel encoder ﬁrst considers the NN to

Sensors 2022,22, 309 15 of 41

improve the channel sensation vectors to strengthen the sensing ability on its successful

attempts, and then the precoder predicts the hybrid framework RF BF/combining vectors

directly from the received sensing vector. Similar work on joint channel sensing and hybrid

BF for mmWave Ma-MIMO systems using DL techniques is considered in [

172

]. The en-

coder learns to optimize the channel sensing vectors to concentrate the sensing power on

the potential directions, while the precoder learns to predict the RF BF/combining vectors

of the hybrid framework from the obtained sensing vector directly.

A channel sounder framework that can measure multi-subcarrier and multiantenna

CSI different propagation, antenna geometries, and frequency bands are introduced in [

173

The architecture can acquire a superior accuracy of more than 75 cm for LoS and is com-

parable to the conventional positioning schemes and achieve the same precision for the

challenging scenario of non-LoS. The feasibility of an indoor positioning framework based

on NNs and CSI of a large-scale MIMO system is investigated in [

174

]. The proposed tai-

lored NN architecture has a feature extractor in the shape of an additional phase branch that

minimized the number of trainable parameters, resulting in a reduction in the amount of

target training data. The measurements were performed for indoor environments covering

a big area of 80 m2with up to 64 antennas.

MIMO user positioning using DL techniques that are based on only the OFDM complex

channel coefﬁcients is examined in [

175

]. The proposed architecture is employed on the top

of available OFDM-MIMO system and does not need any extra piloting overhead. Training

of the model is done in two phases: in the ﬁrst step, training on simulated LoS data, and then

in the second phase, ﬁne-tuning on measured NLoS positions. This results in minimization

of the necessary measured training locations and consequently decreases the attempt for

data acquisition. CNN, multilayer perceptron NN and K-nearest neighbors techniques

are used in [176] for localization of a MIMO transmitter in indoor–outdoor scenarios. The

proposed work has won the ﬁrst position among eight teams worldwide in the indoor

positioning competition organized by “IEEE’s Communication Theory Workshop” by

achieving an MSE of 2.3 cm2.

Accuracy of localization using CSI is improved by combing multi-layer perceptron NN

and K-nearest neighbors techniques in [

177

]. Both schemes then tested for generalization

aspect in different scenarios by dividing the training and validation data in a sense that

the intersection is minimized as compared to the uniform random splitting. In another

work of [

178

], deep NN was used in the development of a robust and accurate localization

scheme for a distributed massive MIMO system. CSI-based positioning is discussed in [

179

]

using CNNs a black box and experimented on the opening of the black box using. The

authors have also discussed the advantages and disadvantages of the use of an open dataset

collected in a real scenario 64-antenna Ma-MIMO system. The position of a user using

the CSI is inferred through CNN and then evaluated on a dataset that consists of indoor

Ma-MIMO CSI measurements of three different antenna conﬁgurations, i.e., covering a

(2.5 m

2.5 m) indoor area [

180

]. The CNN model can be trained for the estimation of the

user position inside (2.5 m ×2.5 m) with an average error of less than half a wavelength.

Indoor localization based on DL and CSI for 28 MIMO antenna is considered in [

181

The input to the multi-layer perceptron NN is the change in the magnitude component of

the CSI and the learning process is improved using data augmentation. To enhance the esti-

mation of the position, an ensemble NN scheme is applied to process the predictions of the

MLPs. An improvement in indoor positioning is done in [

182

] by exploiting the MIMO-CSI

using the proposed CNN architecture. The performance of the proposed three CNN vari-

ants is then compared with ﬁve state-of-the-art NN schemes in terms of accurate estimation

of position. User positioning in OFDM Ma-MIMO systems based on 3D CNN is considered

in [

183

184

] when the BS has a uniform planar array and traditional ﬁngerprint type is

replaced with the “angle-delay channel power matrix”. This methodology is beneﬁcial to

positioning as it embeds angles, with power in the horizontal and vertical directions.

Dynamic localization using predictive RNN for Ma-MIMO systems is investigated

in [

185

]. The authors have designed dynamic localization structures in time-varying en-

Sensors 2022,22, 309 16 of 41

vironments to perform localization and demonstrated the performance of the proposed

architecture in indoor and outdoor scenarios with reasonable localization accuracy. Accu-

rate mmWave positioning is important, and recently few works have been done in this

area using DL techniques [

186

]. Positioning in mmWave Ma-MIMO using DL is consid-

ered in [

187

], and different NN models are applied over beamformed ﬁngerprints such

as CNN, Deep Convolution GP, LSTM, and GP LSTM to reduce the location error in the

outdoor environment near to 1 m. An actor–critic RL scheme is proposed in [

188

] using

NN approximator afﬁne MIMO nonlinear discrete-time type systems, where disturbances

and functions are not known. One NN is used as an action network to produce the best

control signal while the second NN is used as a critic network cost function approximation.

4.4. CSI Acquisition and Feedback, Security, and Robustness

MIMO communication systems are a major enabler of the excessive throughput re-

quirements NGN, e.g., 5G due to its ability to serve many users at a time with high energy

and spectral efﬁciency. However, the MIMO system requires timely and accurate CSI,

which is obtained by a training process including the transmission of a pilot, estimation of

CSI, and feedback [

189

]. The training process experiences a training overhead, that scales

with variation in the number of subcarriers, users, and antennas. Therefore, minimizing

the training overhead in MIMO systems has been an important area of research over the

last few years. Recently, DL-enabled methods have been used to reduce the overhead

in feedback and CSI acquisition and have shown signiﬁcant improvement compared to

traditional schemes [

190

191

]. Here, we will present state-of-the-art DL frameworks used

for CSI acquisition and feedback. This subsection will also review literature work on MIMO

security and robustness aspects.

A DL approach for secure MIMO communications has been employed in [

192

] by

exploiting the advantage of CNN learning network to generate more accurate CSI and to

reduce the BER of the receiver. Both the ideal CSI and imperfect CSI are included in the

training set that then may be used in different scenarios. An RNN-based DL approach is

proposed in [

193

] to learn temporal correlation. The architecture uses depthwise separable

convolution to shrink the network. Results have shown a reasonable performance in terms

of recovery accuracy and quality and obtain considerable robustness at low CR.

Time-varying features have been exploited in [

194

] using two modules: recurrent

compression/uncompression to provide an approach to minimize the parameter size. The

work is extended to MU-MIMO by separately assigning a decoder network for every user

at the BS. A DL-based scheme for channel calibration in Ma-MIMO systems in nonlinear

settings is proposed in [

195

]. The framework is able to exhibit robustness in generic

nonlinear scenarios even with the limited number of training sequences.

A CSI feedback mechanism based on bi-directional reciprocal channel properties and

limited feedback is introduced in [

196

]. The Ma-MIMO BS uses the uplink CSI to recover

the unknown downlink CSI from low-rate UE feedback. The DL enables architecture to

minimize the CSI feedback payload signiﬁcantly based on the multipath reciprocity. A DL-

based denoise network is designed in [

197

] to enhance the channel feedback performance,

and it has shown good performance at low SNR.

A CS and DL empowered CSI feedback framework for FDD Ma-MIMO communication

system is propose in [

198

] where the CSI is compressed ﬁrst at the UE using on CS scheme

and then at BS CSI is reconstructed using a DL enabled signal recovery solver. Deep

autoencoder is used in [

199

] to study CSI feedback in the FDD-MIMO system by considering

the feedback delay and errors. The autoencoder is constructed by using the CSI feedback

process, which contains feedback transmission delays and errors. The proposed architecture

claims to minimize the effect of the feedback delay and errors.

The work in [

200

] presents a multiple-rate CS-NN model for compression and quanti-

zation of the CSI. The authors have adopted two network design principles and develop a

novel quantization mechanism and training scheme. The proposed model improves the

network feasibility by reducing the storage space at UE and enhancing the reconstruction

Sensors 2022,22, 309 17 of 41

accuracy. Similarly, a quantization method and training framework for CSI feedback using

DL are given in [

201

]. The model uses the current CSI feedback in a real communication

network and but reduces the introduced quantization distortion to enhance the quality of

reconstruction.

A Bayesian CS-based feedback scheme for MIMO systems is considered in [

202

]. The

wireless channel used in the study is time-varying temporally and spatially correlated

vector autoregression. The relationship between downlink capacity and the feedback rate

is obtained in closed form in statistics to perform rate-adaptive feedback. A CNN-enabled

analog feedback method that maps the downlink CSI to uplink channel input directly

is given in [

203

204

]. The DL channel estimate is reconstructed by another CNN-based

corresponding noisy channel output. The framework gives a low-latency solution for

rapidly changing MIMO channels because the model does not need explicit modulation,

coding, and quantization.

A compression technique for channel state matrix using DL that is consists of convolu-

tional layers and quantization and entropy coding blocks come after is proposed in [

205

The model enhances the quality of CSI reconstruction even at signiﬁcantly low feedback

rates. The distributed version of this work for an MU-MIMO environment is proposed

in [

206

], where each user compresses its CSI matrices in a distributed form and recon-

struction is done jointly at the BS. The Distributed version not only uses the inherent CSI

pattern of a single MIMO user, but also supports the channel correlations among neighbor

MIMO users.

CSI reporting which is important for MIMO system transceivers to acquire energy

efﬁciency and high capacity in FDD form is considered in [

207

] using DL technology.

The proposed DL-based compression architecture jointly handles recovery, codeword

quantization, and CSI compression under the bandwidth constraint to enhance the encoding

performance of CSI feedback. The correlation between nearby UE has been exploited

in [

208

] by designing a DL-based CSI feedback and cooperative recovery mechanism

to minimize the overhead of the feedback. Authors have also proposed a baseline NN

framework with LSTM for a UE equipped with multiple antennas to extract the correlation

of surrounding antennas and two magnitude-dependent phase feedback schemes that

present instant CSI and statistical magnitude information.

Spatial correlation-based CSI compression feedback for FDD Ma-MIMO systems is

considered in [

209

] and a DL-based CSI compression feedback scheme is used in single-

user as well as multi-user environments. The framework takes into account the spatial

correlation of Ma-MIMO system uniform linear antenna arrays and takes advantage of full

of the channel information during the training. Single-cell and multi-cell scenarios are also

discussed in [

210

] in terms of CSI feedback for BF to optimize the BF performance gain

instead of the feedback accuracy. The encoder at the user in a single-cell does compression

of the CSI and the BF vector is generated at the decoder, while in the multicell system, two

kinds of CSI feedback has to be sent, i.e., the targeted and the interfering CSI. A binarization

assisted feedback NN is proposed in [211] to improve the performance under customized

training and inference approaches.

Implicit feedback framework based on DL is applied in [

212

] to inherit the low-

overhead features. The given architecture uses NNs to interchange the precoding matrix

indicator encoding and decoding components. Moreover, a correlation between sub-bands

is also employed for more improvement in feedback efﬁciency. A compressive sampled CSI

feedback scheme for Ma-MIMO system using DL is proposed in [

213

], where the channel

matrix is sampled in frequency/time dimension uniformly before feeding into NNs. This

will minimize the computational time/resource at UE and improve the accuracy of the

CSI recovery at the BS. A CSI feedback network using DL for the FDD-MIMO system is

studied in [

214

], but its application to the mobile terminal is not effective due to the large

numbers of parameters. Thus using the developed network, authors have designed a new

lightweight CSI feedback framework. Similar work on the development of lightweight

NN for MIMO CSI feedback was also done in [

215

]. An FDD Ma-MIMO communication

Sensors 2022,22, 309 18 of 41

system that prevents signaling overhead by applying a DL-enabled channel extrapolation

is demonstrated in [216].

A scheme to protect the DL-based CSI feedback process from white-box adversarial is

presented in [

217

]. The authors have also shown that jamming attacks may be crafted with

some precautions.

A CNN-based network known as aggregated channel reconstruction model is con-

structed in [

218

] to enhance feedback performance with parametric ReLU activation and

network aggregation. In particular, the elastic feedback method is introduced to ﬂexibly

adjust the network to address various resource limitations. Moreover, the network bi-

narization scheme is integrated with the feature quantization for practical deployment.

Long-range dependencies are captured efﬁciently in [

219

] by using DL-based CSI feedback

method and taking beneﬁts of non-local blocks. Additionally, the feature of the reﬁne-

ment part is strengthened by adopting dense connectivity. AnciNet, a DNN empowered

framework is designed in [

220

] to manage CSI feedback with limited bandwidth. The

proposed architecture extracts noise-free patterns from the noisy samples of CSI to obtain

CSI compression for the feedback effectively.

An uplink-assisted downlink channel acquisition architecture using DL is presented

in [

221

] to minimize feedback and high training overheads. The proposed framework takes

into account the full downlink CSI acquisition process such as channel estimation, downlink

pilot design, and feedback. A CNN model is developed on the Markovian model in [

222

]

to encode forward CSI differentially in time to enhance reconstruction accuracy effectively.

Authors have also explored convolutional layers for the compression of feedback and

spherical normalization of input data. A fully convolutional NN is presented in [

223

]

for the compressing and decompressing the downlink CSI. The proposed model enhance

the reconstruction accuracy of downlink CSI and minimize the training parameters and

computational parameters.

Superimposed coding and DL techniques are combined in [

224

] for CSI feedback. The

proposed methodology ﬁrst spread downlink CSI and then superimposed it on uplink user

data patterns toward the BS. A NN framework is then designed for BS for the recovery

of downlink CSI and uplink user data sequences. A deep transfer learning scheme is

developed in [

225

] to addressed the high training cost of the NN used for 5G MIMO

downlink CSI feedback. The proposed architecture uses a comparatively less number of

samples for ﬁne-tuning of a pre-trained model and provides the possibility to achieve a

new model with reduced training cost.

A DL-based scheme for the prediction of downlink CSI in Ma-MIMO FDD systems

is presented in [

226

]. The given architecture utilizes a complex-valued NN in a complex

domain to tackle complex CSI matrices and adjusts 3D convolution operations for the ex-

traction of features. Similarly, a model-driven DL-enabled downlink channel reconstruction

design is proposed in [227] for FDD massive MIMO systems.

4.5. mmWave Communications

Deep learning techniques have been utilized recently for interesting and important ap-

plications in mmWave and Ma-MIMO systems. DL provides solutions to hard optimization

problems due to its powerful capabilities of learning unknown models.

A DL-based compressed sensing channel estimation and quantized phase hybrid

precoder design scheme is proposed in [

142

] for the MU mmWave Ma-MIMO communica-

tion systems. The proposed work claims to have better performance in terms of spectral

efﬁciency as compared to other techniques having low phase shifter resolution. A novel

DL-based method is developed in [

228

] to estimate the channel for beamspace mmWave

Ma-MIMO communication systems. This framework uses a learned denoising-based

approximate message passing network by taking advantage of iterative signal recovery

methods and DL techniques. A DL-based analog and digital beamforming method for

mmWave point-to-point Ma-MIMO system are given in [

229

230

] for reduction of system

bit error rate and improvement in the spectral efﬁciency.

Sensors 2022,22, 309 19 of 41

An integration of ML and coordinated BF scheme is employed in [

231

] to enable

highly mobile applications for large antenna array mmWave MIMO systems. Authors

have taken the beneﬁt of DL that learns the mapping between beam training results and

Omni-received uplink pilots. Similarly, another RL-based solution for mmWave Ma-MIMO

system is proposed in [

232

] for effective hybrid precoding, where every choice of the

precoders for achieving the optimal decoder is considered as a mapping mechanism in

DNN. In particular, the hybrid precoder is chosen via DNN-based training to optimize the

process of precoding process in mmWave Ma-MIMO.

A generic dataset for mmWave/Ma-MIMO channels known as the DeepMIMO dataset

was introduced in [

233

] with two important features: (1) The construction of the DeepMIMO

channels is based on accurately obtained ray-tracing data from the “Remcom Wireless

InSite”. That means it captures the dependence on the locations of transmitter/receiver and

environment geometry/materials and this is essential for many ML applications. (2) The

DeepMIMO dataset is parameterized/generic so that one can adjust a set of channel and

system parameters to tailor the generated DeepMIMO dataset for the target ML application.

A ray-tracing [

234

] and vehicle trafﬁc simulator are combined in [

235

] to produce channel

realizations that represent 5G situations with mobility of both objects and transceivers.

Authors have used a particular dataset to investigate beam selection method on vehicle-to-

infrastructure using mmWave MIMO.

A hybrid processing model for the mmWave Ma-MIMO system is normally employed

to minimize cost and complexity. However, channel estimation may be very challenging

through this method. The work in [

236

] presents a deep CNN that can exploit both the

frequency and spatial correlation, while the input into the CNN has corrupted channel

matrices at adjacent subcarriers simultaneously. The same research group used Deep CNN

to carry out estimation for the wideband channel of mmWave Ma-MIMO systems in [

237

The proposed scheme exploits the frequency correlation in addition to the exploitation of

spatial correlation and here the input into CNN are channel matrices estimated tentatively

at multiple adjacent subcarriers.

Beamforming gains are achievable and high path loss is preventable in mmWave

systems by deploying a large number of antennas. However, with a large number of

antennas, implementation of digital precoders is difﬁcult due to hardware constraints and,

at the same time, analog precoders have performance limitations. Hybrid precoding is an

important task in mmWave MIMO systems to reduce the cost and complexity as well as to

obtain a sufﬁcient sum rate. Greedy methods or optimization techniques have been used

in literature for hybrid precoding. However, these schemes depend on the channel data

quality and achieve sub-optimum performance, and also give higher complexity. Therefore,

in the next few paragraphs, we will discuss few proposals on the use of DL-based hybrid

precoding. Some alternating algorithms for hybrid precoding for mmWave may be seen

in [238].

Two schemes, i.e., CNN-based and equivalent channel precoding, are designed in [

239

]

for mmWave Ma-MIMO systems. The complexity is decreased signiﬁcantly with equivalent

channel precoding but the performance is a little less than full digital precoding while

the CNN precoding method shows much better robustness to imperfect CSI. In another

related work [

240

], authors have considered the case of multi-user and proposed a DNN

based hybrid BF system. They have simultaneously inferred users and sub-optimal beam

codewords of the BS by applying the received signals only on the target RF beamformers

and hence reducing the complexity of beam training.

A DL empowered hybrid precoding architecture is proposed in [

241

] that uses large-

scale information for the prediction of decoder and hybrid precoders parameters. The

statistics of the channel covariance matrix are applied to design the hybrid precoders and

decoders. The architecture is able to optimize the hybrid precoder and decoder in terms

of maximum spectral efﬁciency after training. Moreover, a CNN architecture for the joint

design of precoder and combiners is designed in [

242

] that takes the input of the channel

matrix and returns the output of analog and baseband beamformers. The underlying CNN

Sensors 2022,22, 309 20 of 41

scheme does not need knowledge of steering vectors of array responses and it achieves

higher capacity performance.

Hybrid precoding for MU-MIMO system is done in [

243

] using CNN. The framework

accepts an imperfect channel matrix as input and at the output gives the combiner and

analog precoder. An exhaustive search algorithm is developed ﬁrst, to chose combiners

and the analog precoder from a predeﬁned codebook. In the second step, combiners and

precoder are employed as output labels during the training network.

Fully convolutional denoising (FCD) AMP scheme is introduced in [

244

] by combining

FCD networks with learned AMP networks in mmWave Ma-MIMO system by considering

NN framework able to learn channel patterns and extract noise features. A beamspace

channel estimation method using prior-aided Gaussian mixture DNN empowered learned

AMP is given in [

245

]. A prior-aided GA beamspace channel estimation method using

prior-aided Gaussian mixture learned AMP network is designed by replacing the original

shrinkage function with that of the derived Gaussian mixture for accurate estimation of the

beamspace channel.

A DL-CS channel estimation method consisting of channel reconstruction and beamspace

channel amplitude estimation is given in [

246

]. The NN is trained ofﬂine based on simu-

lated environments using the mmWave channel model and the correlation between the

measurement matrix, and the received signal vectors are applied as input to the trained NN

which is used for the prediction of the beamspace channel amplitude. Then, reconstruction

of the channel is done using the acquired indices of entries of the dominant beamspace

channel. A mmWave OFDM-MIMO receiver is considered in [

247

] with a generalized

hybrid structure where RF chains and low-resolution ADCs are deployed simultaneously.

The authors have developed a computationally efﬁcient Bayesian data detection scheme

that gives an MMSE estimate on data symbols. The authors have also designed a low-

complexity realization where only matrix-vector multiplications and fast Fourier transform

are needed.

Beamforming for mmWave MIMO systems is considered in [

248

] using the multi-

agent distributed double deep Q-learning approach, where many BSs can dynamically

and automatically adaptable their beams to serve many highly mobile UEs. Authors have

assumed the largest received power mapping criterion for UEs with a realistic channel

model. A frequency-selective wideband mmWave network is considered in [

249

] with two

DL compressive sensing assisted schemes. The proposed methodology learns important

a priori information from training data to give the most accurate channel estimates with

reduced training overhead. Estimation of a channel for mmWave MIMO system is dis-

cussed in [

250

]. A modiﬁed convolutional blind denoising model is developed to boost

the robustness in the noisy channel by adjusting asymmetric joint loss functions, on-blind

denoising subnetwork, and noise level estimation subnetwork for the blind channel estima-

tion. Moreover, the proposed network can minimize the noise interactively by adopting

the estimated noise level map in the channel matrix.

Uplink mmWave Ma-MIMO systems are considered in [

251

] using Bayesian learning.

In this work, an angle domain off-grid channel estimation method is developed by using the

spatial sparse pattern in mmWave channels. Similarly, sparse Bayesian learning is also used

in [

252

] in hybrid mmWave systems for channel estimation. The proposed model exploits

spatial sparsity in the wireless channels that exist due to a highly directional pattern of

propagation. The large intelligent surface-assisted mmWave Ma-MIMO systems are the

focus of work in [

253

] and DL architecture is proposed for channel estimation. A twin

CNN framework is developed for estimating both the direct and the cascaded channels by

feeding CNN with the received pilot signals. Hybrid precoding and channel estimation

mmWave MIMO systems using DL has been studied in [

254

]. Authors have used the

hierarchical codebook based method for channel estimation. An adaptable DNN based

low-rank channel recovery methodology is presented in [

255

] for a hybrid array based

massive MIMO system. The proposed framework includes a common feature extraction

element and the adaptable recovery module.

Sensors 2022,22, 309 21 of 41

4.6. Resource Management and Scheduling

Radio Resource management [

256

] and user scheduling [

257

258

] in MU-MIMO and

Ma-MIMO is very crucial for achieving good performance and often solve using techniques

from optimization theory. The heterogeneity and increased complexity of MIMO systems

like Ma-MIMO demands a paradigm shift from conventional resource management mecha-

nisms. RL and DL are powerful techniques wherein DL a multi-layer NN may be trained

to model a resource allocation algorithm using available data. Therefore, there is no need

for intensive online computations for resource management decisions which would be

required otherwise for the solution of resource allocation problems. This subsection focuses

on the applications of RL and DL on solving the problem of radio resource allocation for

different types of MIMO systems.

Deep learning has been used in [

259

] for the prediction of the power allocation proﬁles

for a new group of users’ positions. A DNN was used that learns the mapping between

optimal power allocation policies and the positions of UE after training. A Q-learning

method is proposed in [

260

] to maximize the overall capacity of the network when BS is

densely and randomly distributed with reasonable improvement in stability and conver-

gence speed. The authors of [

261

] have introduced a DL technique for “heterogeneous

ﬁfth-generation new radio networks” to improve the performance of the downlink coordi-

nated multipoint. The proposed methodology is based on the construction of a surrogate

coordinated multipoint trigger function where the cooperating set is a single-tier of sub-6

GHz heterogeneous BS operating in the FDD mode.

A universal DRL-based framework is given in [

262

] for access control and resource

management. The framework adopts both CNN and RNN for automatically modeling the

sequential features and potential spatial features from the raw wireless signal. An RL-based

power control method is presented in [

263

] for the downlink NOMA transmission without

the knowledge of radio channel parameters and jamming. They formulated the power

allocation of a multiple antennas BS in a NOMA system contending with a smart jammer

as a zero-sum game where the ﬁrst BS selects the transmit power on multiple antennas and

then the jammer chooses the jamming power to cause an interruption in the transmission

of users.

A DQN-based technique is used in [

264

] for resource allocation in Ma-MIMO-NOMA

systems. The RL method is employed for the development of an iterative optimization

structure for beamforming, power allocation, and user clustering. In particular, a DQN is

modeled to group the users in accordance with the reward that has been calculated after

beamforming and power allocation. Moreover, as a part of the study, the use of DL for

the optimization of power control in Ma-MIMO systems has been investigated in [

265

The article [

266

] considers deep spatial learning methods for scheduling that have the

possibility to bypass the channel estimation step. The authors have applied a DNN to

produce a near-optimal schedule only based on the geographic locations of the receiver

and transmitters in the system.

The optimization of sum spectral efﬁciency for multi-cell Ma-MIMO communication

systems is considered in [

267

] for a different number of active users. The proposed method-

ology employs the information of large-scale fading for the prediction of both data power

and the pilot. Authors have used the problem structure to model a single NN able to handle

a different number of active users that are varying dynamically. A similar optimization

algorithm is presented in [

268

], inspired by the weighted MMSE method, to get a stationary

point in polynomial time. Authors then use DL to train a CNN for performing the pilot

power control and joint data in sub-millisecond runtime.

The optimization of downlink beamforming via DL techniques is for the MISO system

is done in [

269

270

]. The method is based on CNN and exploitation of the downlink-uplink

duality and the known pattern of optimal solutions. A novel DRL-based capacity and

coverage optimization method is proposed in [

271

]. The architecture also contained a DRL-

enabled user scheduling method and a novel intercell interference coordination technique

to address capacity and coverage in Ma-MIMO networks. Similarly, the work in [

272

]

Sensors 2022,22, 309 22 of 41

discusses the use of the DL approach for load balancing and user association for sum rate

optimization.

A pilot assignment technique using DL for a Ma-MIMO system equipped with a large

number of antennas is given in [

273

] to improve the performance of cellular networks with

severe pilot contamination by learning the mapping between users’ location pattern and

pilot assignment. The proposed architecture was implemented through a commercially

available deep multilayer perceptron model. A combination of RL and radio service maps

is used in [

274

] to switches off BSs effectively and evaluated by utilizing a 3D ray-tracing

model on computer simulations. An inter-cell interference management method for MIMO

systems is presented in [

275

]. The framework contains interference cancellation on the

receiver and NNs power control on the transmitter end. The authors have evaluated

networks of MIMO systems with the power optimization using a few intercell interference

coordination techniques: the belief propagation, the greedy search, and NN, combined

with IC on the receiver side. Another work in [

276

] considers the suppression of intercell

interference for OFDM-MIMO systems. The authors have employed a complex-valued NN

architecture using the traditional interference rejection combining.

An intelligent algorithm to optimize the performance of the Ma-MIMO beamforming

is introduced in [

277

]. The proposed framework combines three NNs to implement the deep

adversarial RL workﬂow cooperatively. One NN is trained to produce realistic patterns of

user mobility, being used by the second NN to generate a corresponding antenna diagram.

The third NN does the estimation of the efﬁciency of the generated antenna diagram and

returns respective reward to two networks.

4.7. Miscellaneous Applications

The results of some recent works indicate that deep learning models can learn a form of

decoding algorithm, instead of only a classiﬁer. These studies provide that DL architectures

can be applied for improving a standard belief propagation decoder, although having large

example space [

278

]. Moreover, identical improvements are achieved for the min-sum

algorithm.

Metric normalized validation error is introduced in [

279

] to investigate the applications

and limitations of DL-based decoding for different performance metrics, e.g., complexity.

The authors of [

280

] present an iterative belief propagation CNN model for channel decod-

ing under correlated noise. The framework concatenates a trained CNN with a standard

belief propagation decoder. A recurrent neural decoder model using the technique of

successive relaxation is introduced in [

281

]. The authors have observed better performance

over standard belief propagation are on sparser Tanner graph representations of the codes.

A practical issue of imperfect successive interference cancellation decoding for real-

world NOMA system is considered in [

282

]. A novel DL-based scheme is proposed by

authors for the downlink of the MIMO-NOMA system where both successive interference

cancellation decoding and precoding are jointly optimized. Another problem of dynamic

multichannel access is discussed in [

283

] in which multiple correlated channels observe an

unknown joint Markov model and UEs choose the channel for the transmission of data.

The goal of the study is to ﬁnd a policy that optimizes the aggregated future successful

transmissions. The scenario is formulated as a POMDP without known system dynamics.

A novel physical layer DL scheme for MIMO system using an autoencoder is devel-

oped in [

284

] by using a transmitter and receiver which is an extension to the work on joint

optimization of physical layer representation as well as encoding-decoding processes from

to the multi-antenna case. An unsupervised DL-based autoencoder is also used in [

285

]

for single-user MIMO communications to introduce a novel physical layer approach. The

research contribution is an extension of joint optimization [

286

] of physical layer represen-

tation as well as the encoding-decoding processes as a single end-to-end task. The work is

extended for multi-antenna cases by expanding transmitter and receivers.

A DL-based channel prediction scheme is developed in [

287

] to enable FDD large-scale

MIMO for deployment. Authors have removed large signaling overhead using DL based

Sensors 2022,22, 309 23 of 41

channel prediction method and used a NN at the BS to infer the DL CSI that is centered

around a frequency

fDL

by only observing uplink CSI on a different but nearby frequency

region around

fUL

. Then, there is no requirement of reporting/pilot overhead with a

genuine TDD-based system. A DL-based autonomous channel measurement framework

that can accurately predict channel information consisting of a few multi-path effects is de-

veloped in [

288

]. The architecture attains channel magnitude measurements autonomously

using eight antennas through a mobile robot containing a transmitter that receives wireless

commands from a central computer.

Channel characteristics are predicted in [289] using the ML method and CNN for 3D

mmWave Ma-MIMO system indoor channels. Elevation angle of arrival, azimuth angle

of arrival, the elevation angle of departure, azimuth angle of departure, amplitude, and

delay are produced by ray-tracing software. While channel statistical characteristics can be

obtained after data preprocessing to train the CNN. A DNN-enabled decoding framework

for screen-camera communications and a unity 3D-based evaluation scheme is introduced

in [

290

] to boost the obtainable throughput and to synthetically learn the DNN structure for

being robust against multiple different screen-camera scenarios respectively. Jointly sparse

support and jointly sparse signal recoveries have been investigated in [

291

] in multiple

measurement vector schemes for complex signals that may appear in various applications

in signal processing and communications.

An RL actor–critic enabled fault-tolerant control problem is discussed in [

292

] for

MIMO nonlinear discrete-time communication systems. The authors have considered

both abrupt faults and incipient faults. An action NN is designed to produce the optimal

control signal while the cost function is approximated with the critic network. MIMO

uncertain nonlinear dynamic networks having unknown varying control direction matrix

and external disturbance are considered in [

293

] and a continuous tracking control law

is introduced. The proposed framework includes a robust term, an online approximator

(represented by a two-layer NN), Nussbaum gain matrix selector, and high-gain feedback.

A robust adaptive NN control is discussed in [

294

] for a general type of uncertain

MIMO nonlinear systems having input nonlinearities and control coefﬁcient matrices are

not known. The proposed framework combines Lyapunov synthesis and backstepping

with variable structure control for nonsymmetric input nonlinearities of deadzone and

saturation. In another work on MIMO uncertain nonlinear systems with actuator satura-

tion and extern disturbances [

295

], an adaptive controller by taking into account a priori

actuator saturation effects is presented and gives the guarantee of performance tracking.

Authors have used adaptive radial basis function NNs for the approximation of unknown

nonlinearities. Moreover, an auxiliary system is designed for the compensation of actuator

saturation effects.

ANC for uncertain MIMO nonlinear systems is introduced in [

296

] when input satura-

tion and external disturbances are present. The uncertainties of the system are handled by

NN approximation and unknown disturbances are tackled by the Nussbaum disturbance

observer. A dynamic adaptive output feedback NN controller for MIMO afﬁne in the control

uncertain nonlinear systems is developed in [

297

]. The controller can guarantee prescribed

performance limits on the system’s output and the boundedness of closed-loop signals.

An adaptive backstepping control approach is proposed in [

298

] uncertain MIMO

incommensurate fractional-order nonlinear systems. Approximation of unknown nonlinear

uncertainties is done by the radial basis function NN in every step of the backstepping

process. An ANC scheme for MIMO nonlinear systems with different constraints is devel-

oped in this [

299

]. The ANC architecture is combined with disturbance observer, barrier

Lyapunov function, radial basis function NN, backstepping method to tackle the con-

strained states, and nonsymmetric input nonlinearity. Similar works on ANC for uncertain

MIMO nonlinear systems using NN are done in [

300

301

]. An adaptive DL empowered

unmanned aerial vehicle receiver is designed in [

302

] for coded MIMO systems. Authors

have employed the linear convolutional code at the transmitter. The proposed iterative

unmanned aerial vehicle receiver consists of three parts such as ZF or MMSE detector, the

Sensors 2022,22, 309 24 of 41

deep CNN that can suppress the noise by capturing the correlation characteristics among

noise, and the Viterbi decoding decoder.

5. Statistics and Impact

This section presents statistics about the surveyed papers and an analysis of their

impact.

First, we have grouped the literature in four different periods as can be seen in

Table 2

It is possible to quantify the remarkable improvement in the adoption of DL and RL over

the last ﬁve years in MIMO systems. Details about the number of papers published by

years, from 2010 up to 2021, are reported in Figure 3.

Table 2. Number of papers from 2010 to 2021.

Period Number of Papers

2010 to 2012 4

2013 to 2015 7

2016 to 2018 51

2019 to August 2021 148

Figure 3. Number of papers from 2010 to 2021.

Next, we have considered different categories. Therefore, Figure 4reports the total

number of papers concerning different issues of MIMO communication and exploiting RL

and/or DL technologies.

Figure 4. Number of papers surveyed by category.

Sensors 2022,22, 309 25 of 41

Figure 5shows how the surveyed papers are distributed with respect to the analyzed

categories. Note that such papers are more or less equally distributed. From

Figures 4and 5

it is possible to deduce that most of the article concerns three categories (Sections 4.1,4.2

and 4.4). In these three categories, we have considered more topics and there are more

papers related to them. Similarly, the number of papers related to the other three categories

(Sections 4.3,4.5 and 4.6) is almost the same, but deﬁnitively lower. Moreover, there is also

a signiﬁcant amount of work presented in Section 4.7 that we have not listed among the

previous six categories. Most of the works presented in this section are related to channel

encoding–decoding and adaptive control in uncertain MIMO nonlinear systems.

Figure 5. Distribution of the surveyed papers with respect to the different categories.

Figures 6and 7show the distribution of surveyed papers according to DL and RL

architectures respectively. Figure 6concerns DL architectures. Most of the applications

rely on DNN architectures. While a signiﬁcant amount of papers also take advantage of

the CNN framework. We can also see the contribution of other DL architectures such as

RNN, RBFNN and Autoencoder. Similarly, Figure 7presents the exploitation of different

RL schemes. Data indicates that most of the works rely on Bayesian and Q-learning.

Figure 6. Percentage of papers surveyed by DL architecture.

Sensors 2022,22, 309 26 of 41

Figure 7. Percentage of papers surveyed by RL algorithm.

Finally, as a further detail of data reported in Figure 8, we have also listed the top ﬁve

most cited papers for each category in Table 3. This can provide information about the

papers with a higher impact on the research community.

Table 3. Top most cited papers from each category (left to right in descending order).

Category Paper-1 Paper-2 Paper-3 Paper-4 Paper-5

Detection, Classiﬁcation, and Compression [81] [103] [75] [82] [76]

Channel Estimation [120] [285] [144] [129] [124]

mmWave Communication [228] [232] [231] [233] [236]

Positioning, Sensing, and Localization [188] [166] [169] [172] [173]

Resource allocation [266] [262] [268] [273] [277]

CSI acquisition, Security and Robustness [200] [194] [196] [199] [193]

Miscellaneous [294] [279] [281] [136] [283]

Figure 8. Number of citations by category.

Sensors 2022,22, 309 27 of 41

6. Discussion

This section discusses the surveyed papers by highlighting the strengths and limita-

tions, along with future directions.

It is noted from results of channel estimation that DL frameworks show better results

in the large SNR environments, but are outperformed by standard iterative message passing

methods. Moreover, adopted DL schemes are suitable for high MIMO dimensions, but

converged for comparatively small MIMO dimensions for decoding [91]. Therefore, there

is the need for efﬁcient DL architectures to handle the convergence issues that appear in

time-varying channels and 1-bit quantization.

Many problems are associated with 5G Ma-MIMO technology, which can be mitigated

through the use of DL. For example, it is difﬁcult to estimate the accuracy of the channel

by employing conventional estimation schemes and with a suitable number of pilots. The

performance of the low complexity least-squares estimator is not satisfactory. On the other

hand, MMSE channel estimation is relatively complex [

]. Therefore, DL can be applied to

bypass these issues as done in [81,117,124,228,303–305].

In addition, within linear systems, the DL estimator is near to the linear MMSE

estimator, but it outperforms this last one signiﬁcantly when there is a nonlinear signal

model. However, it is sensitive to the training data quality and estimation performance

may degrade appreciably when the data in real regimes distribute wider than that of the

training data [

135

]. Both the advantages and cost of the DL estimator should be considered

when applying it in real wireless communication systems for channel estimation. There is a

need to keep a balance between state-of-the-art channel estimation and DL-empowered

channel estimation.

The computing capabilities and limited memory of wireless devices may not suite

for complex DL algorithms. A considerable amount of time is required for collecting

a sufﬁcient number of samples and for training DL models. This can become a critical

impediment to implement such algorithms on wireless devices with limited storage and

power. Moreover, some MIMO applications need on-ﬂy sampling as well as real-time

processing, which makes training difﬁcult. Obtaining more samples and long-time model

training results in slow feedback. Therefore, DL models should be designed to acquire

optimal accuracy with fewer samples and shorter periods.

User privacy is the most important concern of service providers. While deploying DL

in wireless systems, one challenge is how the training is enabled on a dataset associated with

users without sharing the input data and exposing personal data to risks. It is important to

have a security scheme to speed up the integration of DL in MIMO communications.

Security of DL networks is another challenge, as NNs are prone to adversarial attacks.

These attackers may affect the process of training by inserting fake training data, which can

reduce the accuracy of the DL models. This may lead to a wrong design, which can affect

the overall performance of the network. Research in the security of RL and DL techniques

remains shallow.

Multiple antennas are required to mitigate high path loss and to achieve BF gains in

mmWave systems. However, it is difﬁcult to employ digital precoders in presence of many

antennas due to hardware constraints. Similarly, the performance of analog precoders

is limited [

241

]. Therefore, hybrid precoding using DL architecture is a feasible solution

as the DL-based precoder takes advantage of the large-scale information for parameters

prediction of hybrid precoders, as well as of decoders.

Despite the remarkable progress of DL in communication, still, research efforts are

required in many directions to ease the integration of RL and RL. The acceleration of DNN

alongside distributed RL systems, cloud computing, faster algorithm, and advanced parallel

computing provides an opportunity for 5G to develop the intelligence in its communication

systems to provide ultra-low latency and high throughput. Recently, some efforts have

been done in DNN acceleration [

306

]. The acceleration of DNN can be at architecture,

computation, and implementation levels.

Sensors 2022,22, 309 28 of 41

Techniques such as knowledge distillation [

307

], projection [

308

], pruning [

309

] and

layer decomposition [

310

] can be used at architecture level. Many characteristics may be

investigated for the implementation level, including FPGA designs [

311

] and advanced

GPU [

312

]. With the use of DL acceleration schemes, we can lower the complexity of DL

with reduced loss in the accuracy of this architecture. A combination of these approaches

may decrease the amount of parameters by more than 50%.

Furthermore, more exploration of the acceleration of these models may have a signiﬁ-

cant impact on the adoption of DL to develop intelligence in MIMO systems. Integration

of DL and RL in MIMO communication systems can speed up by data collection and

subsequent cleansing. With the availability of datasets, researchers can build and test their

architectures. Therefore, efforts are required to build systems that can produce datasets.

7. Conclusions

We presented a comprehensive review of the applications of RL and DL to different

issues of MIMO communication systems. First, we presented an introduction of both

classes of AI methods (i.e., RL and DL) and MIMO systems. Afterward, we have presented

various applications of such AI technologies in MIMO systems. Then, we have analyzed

the impact of research papers in the ﬁeld. Finally, we have outlined open issues, some

limitations, and future research directions.

Author Contributions:

Conceptualization, M.N. and A.C.; methodology, M.N. and A.C.; investiga-

tion, M.N. and G.D.P.; writing—original draft preparation, M.N.; writing—review and editing, G.D.P.

and A.C.; supervision, G.D.P.; funding acquisition, G.D.P. and A.C. All authors have read and agreed

to the published version of the manuscript.

Funding:

This research has been partly supported by the project ASMARA—Applicazioni pilota

post Direttiva 2010/65 in realtà portuali italiane della Suite MIELE a supporto delle Authority per

ottimizzazione della inteRoperabilità nell’intermodalitA’ dei ﬂussi città porto—SCN_00529—MIUR

P.O.N. Smart Cities and Communities and Social Innovation.

Data Availability Statement: No database has been used during this work.

Conﬂicts of Interest: Authors declares no conﬂict of interest.

List of Acronyms

RL Reinforcement Learning

DRL Deep Reinforcement Learning

AI Artiﬁcial Intelligence

MDP Markov Decision Process

MC Monte Carlo

TD Temporal Difference

DP Dynamic Programming

VIA Value Iteration Algorithm

PIA Value Iteration Algorithm

AC Actor–Critic

A2C Advantage Actor–Critic

A3C Asynchronous Advantage Actor–Critic

DQN Deep Q-Network

NN Neural Network

TS Thompson Sampling

CNN Convolutional Neural Network

FC Fully Conntected

UCB Upper Conﬁdence Bound

RMSE Root Mean Square Error

SARSA State-Action-Reward-State-Action

ANN Artiﬁcial Neural Network

DBN Deep Belief Network

Sensors 2022,22, 309 29 of 41

NEC Neural Episodic Network

DRL Deep Reinforcement Learning

IRL Inverse Reinforcement Learning

DNN Deep Neural Network

MIMO Multiple-Input and Multiple-Output

RNN Recurrent Neural Network

VBC Variance-Based Control

ML Machine Learning

BS Base Station

TD Temporal Difference

UE User Equipment

LTE Long-Term Evolution

DS Delivery System

DL Deep Learning

RMS Real-time multimedia streaming

ITS Intelligent transportation systems

MC Monte Carlo

DP Dynamic Programming

FDD Frequency Division Duplex

MU-MIMO Multi-User MIMO

NOMA Non-Orthogonal Multiple Access

POMDP Partially Observable Markov Decision

Process

ADC Analogue-to-Digital Converter

MLD Maximum-Likelihood Detection

CSI Channel State Information

SISO Single-Input Multiple- Output

BF Beamforming

LSTM Long Short-Term Memory

MMSE Minimum Mean Square Error

mmWave millimeter wave

BER Bit Error Rate

SNR Signal-to-Noise Ratio

TDD Time Division Duplex

CR Compression Ratio

NGN Next-Generation Network

MSE Mean Square Error

STBC Space Time Block Coded

AMP Approximate Message Passing

MPD Message Passing Detector

DoA Direction of Arrival

UAC Underwater Acoustic Communication

CS Compressive Sensing

SGD Stochastic Gradient Descent

OFDM Orthogonal Frequency Division

Multiplexing

MLP Multi-Layer Perceptron

3D Three Dimensional

GP Gaussian Process

5G Fifth Generation

3GPP 3rd Generation Partnership Project

DDPG Deep Deterministic Policy Gradient

NEC Neural Episodic Control

Ma-MIMO Massive MIMO

ReLU Rectiﬁed Linear Unit

ANC Adaptive Neural Control

RBFNN Radial Basis Function Neural network

Sensors 2022,22, 309 30 of 41

References

Rusek, F.; Persson, D.; Lau, B.K.; Larsson, E.G.; Marzetta, T.L.; Edfors, O.; Tufvesson, F. Scaling up MIMO: Opportunities and

challenges with very large arrays. IEEE Signal Process. Mag. 2012,30, 40–60. [CrossRef]

Larsson, E.G.; Edfors, O.; Tufvesson, F.; Marzetta, T.L. Massive MIMO for next generation wireless systems. IEEE Commun. Mag.

2014,52, 186–195. [CrossRef]

3. Marzetta, T.L. Massive MIMO: An introduction. Bell Labs Tech. J. 2015,20, 11–22. [CrossRef]

4. Sutton, R.S.; Barto, A.G. Introduction to Reinforcement Learning; MIT Press: Cambridge, MA, USA, 1998; Volume 135.

Kormushev, P.; Calinon, S.; Caldwell, D. Reinforcement Learning in Robotics: Applications and Real-World Challenges. Robotics

2013,2, 122–148. [CrossRef]

Argall, B.; Chernova, S.; Veloso, M.; Browning, B. A survey of robot learning from demonstration. Robot. Auton. Syst.

2009

,57,

469–483. [CrossRef]

Guenter, F.; Hersch, M.; Calinon, S.; Billard, A. Reinforcement learning for imitating constrained reaching movements. Adv. Robot.

2007,21, 1521–1544. [CrossRef]

8. Kober, J.; Bagnell, J.A.; Peters, J. Reinforcement learning in robotics: A survey. Int. J. Robot. Res. 2013,32, 1238–1274.

Chen, M.; Challita, U.; Saad, W.; Yin, C.; Debbah, M. Machine learning for wireless networks with artiﬁcial intelligence: A tutorial

on neural networks. arXiv 2017, arXiv:1710.02913.

10.

Mao, Q.; Hu, F.; Hao, Q. Deep learning for intelligent wireless networks: A comprehensive survey. IEEE Commun. Surv. Tutor.

2018,20, 2595–2621. [CrossRef]

11.

Fadlullah, Z.M. State-of-the-art deep learning: Evolving machine intelligence toward tomorrow’s intelligent network trafﬁc

control systems. IEEE Commun. Surv. Tutor. 2017,19, 2432–2455. [CrossRef]

12. Xin, Y. Machine learning and deep learning methods for cybersecurity. IEEE Access 2018,6, 35365–35381. [CrossRef]

13.

He, J.; Chen, J.; He, X.; Gao, J.; Li, L.; Deng, L.; Ostendorf, M. Deep reinforcement learning with a natural language action space.

In Proceedings of the Association for Computational Linguistics Annual Meeting (ACL), Berlin, Germany, 7–12 August 2016.

14.

Guu, K.; Pasupat, P.; Liu, E.Z.; Liang, P. From language to programs: Bridging reinforcement learning and maximum marginal

likelihood. In Proceedings of the Association for Computational Linguistics Annual Meeting (ACL), Vancouver, BC, Canada, 30

July–4 August 2017.

15.

Narasimhan, K.; Yala, A.; Barzilay, R. Improving information extraction by acquiring external evidence with reinforcement

learning. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), Austin, TX, USA,

1–4 November 2016.

16.

Radford, A.; Jozefowicz, R.; Sutskever, I. Learning to Generate Reviews and Discovering Sentiment. arXiv

2017

, arXiv:1704.01444.

17.

Jeerige, A.; Bein, D.; Verma, A. Comparison of Deep Reinforcement Learning Approaches for Intelligent Game Playing. In

Proceedings of the 9th Annual Computing and Communication Workshop and Conference (CCWC), Las Vegas, NV, USA, 7–9

January 2019; pp. 0366–0371.

18.

Shao, K.; Tang, Z.; Zhu, Y.; Li, N.; Zhao, D. A Survey of Deep Reinforcement Learning in Video Games. arXiv

2019

arXiv:1912.10944.

19.

Aytar, Y. Playing hard exploration games by watching youtube. In Proceedings of the Annual Conference on Neural Information

Processing Systems 2018, NeurIPS 2018, Montreal, QC, Canada, 3–8 December 2018.

20. Christopher, B.; Greg, B.; Brooke, C. Dota 2 with large scale deep reinforcement learning. arXiv 2019, arXiv:1912.06680.

21.

Saravanan, M.; Kumar, P.S.; Sharma, A. IoT Enabled Indoor Autonomous Mobile Robot using CNN and Q-Learning. In

Proceedings of the 2019 IEEE International Conference on Industry 4.0, Artiﬁcial Intelligence, and Communications Technology

(IAICT), Bali, Indonesia, 1–3 July 2019.

22.

Künzel, G.; Cainelli, G.P.; Müller, I.; Pereira, C.E. Weight adjustments in a routing algorithm for wireless sensor and actuator

networks using Q-learning. IFAC-PapersOnLine 2018,51, 58–63. [CrossRef]

23.

Liu, X.; Qin, Z.; Gao, Y. Resource allocation for edge computing in IoT networks via reinforcement learning. In Proceedings of

the ICC 2019—2019 IEEE International Conference on Communications (ICC), Shanghai, China, 20–24 May 2019; pp. 1–6.

24.

Testa, A.; Cinque, M.; Coronato, A.; Pietro, G.; Augusto, J. Heuristic strategies for assessing wireless sensor network resiliency:

An event-based formal approach. J. Heuristics 2015,21, 145–175. [CrossRef]

25.

Bakhouya, M.; Campbell, R.; Coronato, A.; Pietro, G.d.; Ranganathan, A. Introduction to special section on formal methods in

pervasive computing. ACM Trans. Auton. Adapt. Syst. 2012,7, 1–9. [CrossRef]

26.

Qiu, X.; Nguyen, T.A.; Crow, M.L. Heterogeneous energy storage optimization for microgrids. IEEE Trans. Smart Grid

2016

,7,

1453–1461. [CrossRef]

27.

Zhang, W.; Dietterich, T.G. A reinforcement learning approach to job-shop scheduling. In Proceedings of the Fourteenth

International Joint Conference on Artiﬁcial Intelligence, IJCAI 95, Montreal, QC, Canada, 20–25 August 1995; Morgan Kaufmann

Publishers Inc.: San Francisco, CA, USA, 1995; Volume 2, pp. 1114–1120.

28.

Mao, H.; Alizadeh, M.; Menache, I.; Kandula, S. Resource Management with Deep Reinforcement Learning. In Proceedings of the

15th ACM Workshop on Hot Topics in Networks (HotNets ’16), Atlanta, GA, USA, 9–10 November 2016; ACM: New York, NY,

USA, 2016; pp. 50–56.

29.

Tesauro, G. Online Resource Allocation Using Decompositional Reinforcement Learning. In Proceedings of the AAAI, Pittsburgh,

PA, USA, 9–13 July 2005; ISBN 978-1-57735-236-5.

Sensors 2022,22, 309 31 of 41

30.

Boyan, J.A.; Littman, M.L. Packet routing in dynamically changing networks: A reinforcement learning approach. In Proceedings

of the 6th International Conference on Neural Information Processing Systems (NIPS’93), Denver, CO, USA, 29 November–2

December 1993; pp. 671–678.

31.

Brunner, G.; Richter, O.; Wang, Y.; Wattenhofer, R. Teaching a machine to read maps with deep reinforcement learning. In

Proceedings of the AAAI Conference on Artiﬁcial Intelligence (AAAI), New Orleans, LA, USA, 2–7 February 2018,

32.

Cao, Q.; Lin, L.; Shi, Y.; Liang, X.; Li, G. Attention-aware face hallucination via deep reinforcement learning. In Proceedings of

the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21–26 July 2017.

33.

Mao, H.; Alizadeh, M.; Menache, I.; Kandula, S. Resource management with deep reinforcement learning. In Proceedings of the

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Atlanta, GA, USA, 9–10 November 2016.

34.

Liu, F.; Li, S.; Zhang, L.; Zhou, C.; Ye, R.; Wang, Y.; Lu, J. 3DCNN-DQN-RNN: A deep reinforcement learning framework for

semantic parsing of large-scale 3d point clouds. In Proceedings of the IEEE International Conference on Computer Vision (ICCV),

Venice, Italy, 22–29 October 2017.

35.

Coronato, A.; Naeem, M.; De Pietro, G.; Paragliola, G. Reinforcement learning for intelligent healthcare applications: A survey.

Artif. Intell. Med. 2020,109, 101964. [CrossRef] [PubMed]

36.

Cinque, M.; Coronato, A.; Testa, A. A failure modes and effects analysis of mobile health monitoring systems. In Innovations and

Advances in Computer, Information, Systems Sciences, and Engineering; Springer: New York, NY, USA, 2013; pp. 569–582.

37.

Naeem, M.; Paragliola, G.; Coronato, A. A reinforcement learning and deep learning based intelligent system for the support of

impaired patients in home treatment. Expert Syst. Appl. 2021,168, 114285. [CrossRef]

38.

Alsheikh, M.A.; Niyato, D.; Lin, S.; Tan, H.P.; Han, Z. Mobile big data analytics using deep learning and apache spark. IEEE Netw.

2016,30, 22–29. [CrossRef]

39. Sutton, R.S.; Barto, A.G. Reinforcement Learning: An Introduction; A Bradford Book; MIT Press: Cambridge, MA, USA, 2018.

40.

Shah, S.I.H.; Coronato, A. Inverse Reinforcement Learning Through Max-Margin Algorithm. In Proceedings of the Intelligent

Environments 2021: Workshop Proceedings of the 17th International Conference on Intelligent Environments, Dubai, United

Arab Emirates, 21–24 June 2021; Volume 29, p. 190.

41.

Shah, S.I.H.; De Pietro, G. An Overview of Inverse Reinforcement Learning Techniques. In Proceedings of the Intelligent

Environments 2021: Workshop Proceedings of the 17th International Conference on Intelligent Environments, Dubai, United

Arab Emirates, 21–24 June 2021; Volume 29, p. 202.

42.

Shah, S.I.H.; Coronato, A. Learning Tasks in Intelligent Environments via Inverse Reinforcement Learning. In Proceedings of the

2021 17th International Conference on Intelligent Environments (IE), Dubai, United Arab Emirates, 21–24 June 2021; pp. 1–4.

43.

Naeem, M.; Rizvi, S.T.H.; Coronato, A. A Gentle Introduction to Reinforcement Learning and its Application in Different Fields.

IEEE Access 2020,8, 209320–209344. [CrossRef]

44.

Sarker, I.H. Deep Learning: A Comprehensive Overview on Techniques, Taxonomy, Applications and Research Directions. SN

Comput. Sci. 2021,2, 420. [CrossRef] [PubMed]

45.

Mathew, A.; Amudha, P.; Sivakumari, S. Deep Learning Techniques: An Overview. In International Conference on Advanced

Machine Learning Technologies and Applications; Springer: Singapore, 2020; pp. 599–608.

46.

Le, Q.V. A tutorial on deep learning part 2: Autoencoders, convolutional neural networks and recurrent neural networks. Google

Brain 2015,20, 1–20.

47.

Yamashita, R.; Nishio, M.; Do, R.K.G.; Togashi, K. Convolutional neural networks: an overview and application in radiology.

Insights Imaging 2018,9, 611–629. [CrossRef]

48.

Li, Z.; Liu, F.; Yang, W.; Peng, S.; Zhou, J. A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects.

IEEE Trans. Neural Netw. Learn. Syst. 2021, 1–21. [CrossRef]

49.

Bianchi, F.M.; Maiorino, E.; Kampffmeyer, M.C.; Rizzi, A.; Jenssen, R. An overview and comparative analysis of recurrent neural

networks for short term load forecasting. arXiv 2017, arXiv:1705.04378.

50.

Ian, G.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D. Generative Adversarial Nets. Avaliable online: https://arxiv.org/

pdf/1406.2661.pdf (accessed on 1 December 2021).

51. Salakhutdinov, R.; Hinton, G. Semantic hashing. Int. J. Approx. Reason. 2009,50, 969–978. [CrossRef]

52.

Slyusar, V.I.; Titov, I. Correction of characteristics of transmitting channels in an active digital antenna array. Radioelectron.

Commun. Syst. 2004,47, 9–13.

53.

Björnson, E.; Jorswieck, E. Optimal Resource Allocation in Coordinated Multi-Cell Systems; Now Publishers Inc.: Hanover, MA, USA,

2013.

54.

Gesbert, D.; Hanly, S.; Huang, H.; Shitz, S.S.; Simeone, O.; Yu, W. Multi-cell MIMO cooperative networks: A new look at

interference. IEEE J. Sel. Areas Commun. 2010,28, 1380–1408. [CrossRef]

55.

Karakayali, M.K.; Foschini, G.J.; Valenzuela, R.A. Network coordination for spectrally efﬁcient communications in cellular

systems. IEEE Wirel. Commun. 2006,13, 56–61. [CrossRef]

56.

Basnayaka, D.A.; Smith, P.J.; Martin, P.A. Performance analysis of macrodiversity MIMO systems with MMSE and ZF receivers in

ﬂat Rayleigh fading. IEEE Trans. Wirel. Commun. 2013,12, 2240–2251. [CrossRef]

57.

Marzetta, T.L. Noncooperative cellular wireless with unlimited numbers of base station antennas. IEEE Trans. Wirel. Commun.

2010,9, 3590–3600. [CrossRef]

58. Jakes, W.C. Mobile Microwave Communication; Wiley-IEEE Press: Piscataway, NJ, USA, 1974.

Sensors 2022,22, 309 32 of 41

59.

Lo, T.K. Maximum ratio transmission. In Proceedings of the 1999 IEEE international conference on communications (Cat. No.

99CH36311), Vancouver, BC, Canada, 6–10 June 1999; Volume 2, pp. 1310–1314.

60.

Yang, K.; Shen, C.; Liu, T. Deep Reinforcement Learning based Wireless Network Optimization: A Comparative Study. In

Proceedings of the IEEE INFOCOM 2020-IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS),

Toronto, ON, Canada, 6–9 July 2020; pp. 1248–1253.

61.

Ahmed, K.I.; Tabassum, H.; Hossain, E. Deep learning for radio resource allocation in multi-cell networks. IEEE Netw.

2019

33, 188–195. [CrossRef]

62.

Hassan, K.; Masarra, M.; Zwingelstein, M.; Dayoub, I. Channel estimation techniques for millimeter-wave communication

systems: Achievements and challenges. IEEE Open J. Commun. Soc. 2020,1, 1336–1363. [CrossRef]

63.

Heath, R.W.; Gonzalez-Prelcic, N.; Rangan, S.; Roh, W.; Sayeed, A.M. An overview of signal processing techniques for millimeter

wave MIMO systems. IEEE J. Sel. Top. Signal Process. 2016,10, 436–453. [CrossRef]

64.

Zhang, C.; Patras, P.; Haddadi, H. Deep learning in mobile and wireless networking: A survey. IEEE Commun. Surv. Tutor.

2019

21, 2224–2287. [CrossRef]

65.

Cerar, G.; Yetgin, H.; Mohorˇciˇc, M.; Fortuna, C. Machine learning for wireless link quality estimation: A survey. IEEE Commun.

Surv. Tutor. 2021,23, 696–728. [CrossRef]

66.

Chataut, R.; Akl, R. Massive MIMO systems for 5G and beyond networks—overview, recent trends, challenges, and future

research direction. Sensors 2020,20, 2753. [CrossRef]

67.

Luong, N.C.; Hoang, D.T.; Gong, S.; Niyato, D.; Wang, P.; Liang, Y.C.; Kim, D.I. Applications of deep reinforcement learning in

communications and networking: A survey. IEEE Commun. Surv. Tutor. 2019,21, 3133–3174. [CrossRef]

68.

Wang, M.; Gao, F.; Jin, S.; Lin, H. An overview of enhanced massive MIMO with array signal processing techniques. IEEE J. Sel.

Top. Signal Process. 2019,13, 886–901. [CrossRef]

69.

Rodríguez, E.; Otero, B.; Gutiérrez, N.; Canal, R. A Survey of Deep Learning Techniques for Cybersecurity in Mobile Networks.

IEEE Commun. Surv. Tutor. 2021,23, 1920–1955. [CrossRef]

70.

Arjoune, Y.; Faruque, S. Artiﬁcial intelligence for 5g wireless systems: Opportunities, challenges, and future research direction.

In Proceedings of the 2020 10th Annual Computing and Communication Workshop and Conference (CCWC), Las Vegas, NV,

USA, 6–8 January 2020; pp. 1023–1028.

71. Hu, Q.; Gao, F.; Zhang, H.; Li, G.Y.; Xu, Z. Understanding Deep MIMO Detection. arXiv 2021, arXiv:2105.05044.

72.

Wang, L.; Fortunati, S.; Greco, M.S.; Gini, F. Reinforcement learning-based waveform optimization for MIMO multi-target

detection. In Proceedings of the 2018 52nd Asilomar Conference on Signals, Systems, and Computers, Paciﬁc Grove, CA, USA,

28–31 October 2018; pp. 1329–1333.

73.

Ahmed, A.M.; Ahmad, A.A.; Fortunati, S.; Sezgin, A.; Greco, M.S.; Gini, F. A Reinforcement Learning based approach for

Multi-target Detection in Massive MIMO radar. arXiv 2020, arXiv:2005.04708.

74.

Samuel, N.; Diskin, T.; Wiesel, A. Deep MIMO detection. In Proceedings of the 2017 IEEE 18th International Workshop on Signal

Processing Advances in Wireless Communications (SPAWC), Lucca, Italy, 3–6 July 2017; pp. 1–5. [CrossRef]

75. Samuel, N.; Diskin, T.; Wiesel, A. Learning to Detect. arXiv 2018, arXiv:1805.07631.

76.

He, H.; Wen, C.K.; Jin, S.; Li, G.Y. A Model-Driven Deep Learning Network for MIMO Detection. In Proceedings of the 2018 IEEE

Global Conference on Signal and Information Processing (GlobalSIP), Chennai, India, 3–5 April 2018; pp. 584–588. [CrossRef]

77.

Un, M.W.; Shao, M.; Ma, W.K.; Ching, P.C. Deep Mimo Detection Using ADMM Unfolding. In Proceedings of the 2019 IEEE Data

Science Workshop (DSW), Toronto, ON, Canada, 2–5 June 2019; pp. 333–337. [CrossRef]

78.

Liu, X.; Li, Y. Deep MIMO Detection Based on Belief Propagation. In Proceedings of the 2018 IEEE Information Theory Workshop

(ITW), Riva del Garda, Italy, 25–29 November 2018; pp. 1–5. [CrossRef]

79.

Corlay, V.; Boutros, J.J.; Ciblat, P.; Brunel, L. Multilevel MIMO Detection with Deep Learning. In Proceedings of the 2018

52nd Asilomar Conference on Signals, Systems, and Computers, Paciﬁc Grove, CA, USA, 28–31 October 2018; pp. 1805–1809.

[CrossRef]

80.

Jia, Z.; Cheng, W.; Zhang, H. A partial learning-based detection scheme for massive MIMO. IEEE Wirel. Commun. Lett.

2019

8, 1137–1140. [CrossRef]

81.

Wen, C.K.; Shih, W.T.; Jin, S. Deep Learning for Massive MIMO CSI Feedback. IEEE Wirel. Commun. Lett.

2018

,7, 748–751.

[CrossRef]

82.

Wang, T.; Wen, C.K.; Jin, S.; Li, G.Y. Deep learning-based CSI feedback approach for time-varying massive MIMO channels. IEEE

Wirel. Commun. Lett. 2018,8, 416–419. [CrossRef]

83.

Ramjee, S.; Ju, S.; Yang, D.; Liu, X.; Gamal, A.E.; Eldar, Y.C. Fast deep learning for automatic modulation classiﬁcation. arXiv

2019, arXiv:1901.05850.

84.

Meng, F.; Chen, P.; Wu, L.; Wang, X. Automatic Modulation Classiﬁcation: A Deep Learning Enabled Approach. IEEE Trans. Veh.

Technol. 2018,67, 10760–10772. [CrossRef]

85.

Zhao, Z.; Vuran, M.C.; Guo, F.; Scott, S.D. Deep-waveform: A learned OFDM receiver based on deep complex-valued convolu-

tional networks. IEEE J. Sel. Areas Commun. 2021,39, 2407–2420. [CrossRef]

86.

Lu, Z.; Wang, J.; Song, J. Multi-resolution CSI Feedback with deep learning in Massive MIMO System. arXiv

2019

arXiv:1910.14322.

Sensors 2022,22, 309 33 of 41

87.

Jeon, Y.S.; Lee, N.; Poor, H.V. Robust data detection for MIMO systems with one-bit ADCs: A reinforcement learning approach.

IEEE Trans. Wirel. Commun. 2019,19, 1663–1676. [CrossRef]

88.

Jeon, Y.S.; So, M.; Lee, N. Reinforcement-learning-aided ML detector for uplink massive MIMO systems with low-precision

ADCs. In Proceedings of the 2018 IEEE Wireless Communications and Networking Conference (WCNC), Nanjing, China, 15–18

April 2018; pp. 1–6.

89.

Jeon, Y.S.; Lee, H.; Lee, N. Robust MLSD for wideband SIMO systems with one-bit ADCs: Reinforcement-learning approach. In

Proceedings of the 2018 IEEE International Conference on Communications Workshops (ICC Workshops), Kansas City, MO, USA,

20–24 May 2018; pp. 1–6.

90.

Zhang, Y.; Alrabeiah, M.; Alkhateeb, A. Deep learning for massive MIMO with 1-bit ADCs: When more antennas need fewer

pilots. IEEE Wirel. Commun. Lett. 2020,9, 1273–1277. [CrossRef]

91.

Klautau, A.; González-Prelcic, N.; Mezghani, A.; Heath, R.W. Detection and channel equalization with deep learning for low

resolution MIMO systems. In Proceedings of the 2018 52nd Asilomar Conference on Signals, Systems, and Computers, Paciﬁc

Grove, CA, USA, 28–31 October 2018; pp. 1836–1840.

92.

Peng, Y. LLL aided MIMO detection algorithm based on BP neural network optimization. In Proceedings of the 2021 2nd

International Conference on Artiﬁcial Intelligence and Information Systems, Chongqing, China, 28–30 May 2021; pp. 1–5.

93.

Gao, Y.; Niu, H.; Kaiser, T. Massive MIMO detection based on belief propagation in spatially correlated channels. In Proceedings

of the SCC 2017, 11th International ITG Conference on Systems, Communications and Coding, Hamburg, Germany, 6–9 February

2017; pp. 1–6.

94.

Zhang, Y.; Ge, L.; You, X.; Zhang, C. Belief propagation detection based on max-sum algorithm for massive MIMO systems. In

Proceedings of the 2017 9th International Conference on Wireless Communications and Signal Processing (WCSP), Changsha,

China, 11–13 October 2017; pp. 1–6.

95.

Tan, X.; Xu, W.; Be’ery, Y.; Zhang, Z.; You, X.; Zhang, C. Improving massive MIMO belief propagation detector with deep neural

network. arXiv 2018, arXiv:1804.01002.

96.

He, H.; Wen, C.K.; Jin, S.; Li, G.Y. Model-driven deep learning for joint MIMO channel estimation and signal detection. arXiv

2019, arXiv:1907.09439.

97.

Tan, X.; Zhong, Z.; Zhang, Z.; You, X.; Zhang, C. Low-complexity message passing MIMO detection algorithm with deep neural

network. In Proceedings of the 2018 IEEE Global Conference on Signal and Information Processing (GlobalSIP), Ottawa, ON,

Canada, 26–29 November 2018; pp. 559–563.

98.

Xu, D.; Du, Q. Space-Partition-Driven Learning Network for Efﬁcient MIMO Detection. Procedia Comput. Sci.

2021

,187, 507–511.

[CrossRef]

99.

Mohammad, A.; Masouros, C.; Andreopoulos, Y. Complexity-scalable neural-network-based MIMO detection with learnable

weight scaling. IEEE Trans. Commun. 2020,68, 6101–6113. [CrossRef]

100.

Pham, V.Q.; Dang, H.N.; Nguyen, T.V.; Nguyen, H.T. Performance of Deep Learning LDPC Coded Communications in Large

Scale MIMO Channels. In Proceedings of the 2019 6th NAFOSTED Conference on Information and Computer Science (NICS), Ho

Chi Minh City, Vietnam, 12–13 December 2019; pp. 214–218.

101.

Jeon, Y.S.; Lee, N.; Poor, H.V. Reinforcement-learning-aided detector for time-varying MIMO systems with one-bit ADCs. In

Proceedings of the 2019 IEEE Global Communications Conference (GLOBECOM), Taipei, Taiwan, 9–13 December 2019.

102.

Baek, M.S.; Kwak, S.; Jung, J.Y.; Kim, H.M.; Choi, D.J. Implementation methodologies of deep learning-based signal detection for

conventional MIMO transmitters. IEEE Trans. Broadcast. 2019,65, 636–642. [CrossRef]

103.

Chen, Z.; Li, D.; Xu, Y. Deep MIMO detection scheme for high-speed railways with wireless big data. In Proceedings of the 2019

IEEE 89th Vehicular Technology Conference (VTC2019-Spring), Kuala Lumpur, Malaysia, 28 April–1 May 2019; pp. 1–5.

104.

Takabe, S.; Imanishi, M.; Wadayama, T.; Hayashi, K. Deep learning-aided projected gradient detector for massive overloaded

MIMO channels. In Proceedings of the ICC 2019—2019 IEEE International Conference on Communications (ICC), Montreal, QC,

Canada, 20–24 May 2019; pp. 1–6.

105.

Hua, H.; Wang, X.; Xu, Y. Signal detection in uplink pilot-assisted multi-user MIMO systems with deep learning. In Proceedings

of the 2019 Computing, Communications and IoT Applications (ComComAp), Shenzhen, China, 26–28 October 2019; pp. 369–373.

106.

Zheng, P.; Zeng, Y.; Liu, Z.; Gong, Y. Deep Learning Based Trainable Approximate Message Passing for Massive MIMO Detection.

In Proceedings of the ICC 2020—2020 IEEE International Conference on Communications (ICC), Montreal, QC, Canada, 7–11

June 2020; pp. 1–6.

107.

Khani, M.; Alizadeh, M.; Hoydis, J.; Fleming, P. Adaptive neural signal detection for massive MIMO. IEEE Trans. Wirel. Commun.

2020,19, 5635–5648. [CrossRef]

108.

Scotti, A. Graph Neural Networks and Learned Approximate Message Passing Algorithms for Massive MIMO Detection. Avali-

able Online: https://www.researchgate.net/publication/342915312_Graph_Neural_Networks_for_Massive_MIMO_Detection

(accessed on 1 December 2021).

109.

Gao, G.; Dong, C.; Niu, K. Sparsely connected neural network for massive MIMO detection. In Proceedings of the 2018 IEEE 4th

International Conference on Computer and Communications (ICCC), Chengdu, China, 7–10 December 2018; pp. 397–402.

110.

Delgado, O.; Labeau, F. Deep Learning Decoder for MIMO Communications with Impulsive Noise. In Proceedings of the 2020

IEEE 17th Annual Consumer Communications & Networking Conference (CCNC), Las Vegas, NV, USA, 10–13 January 2020;

pp. 1–6.

Sensors 2022,22, 309 34 of 41

111.

Yu, Y.; Wang, J.; Guo, L. Multisegment Mapping Network for Massive MIMO Detection. Int. J. Antennas Propag.

2021

,2021,

9989634. [CrossRef]

112.

Nguyen, N.T.; Lee, K. Deep learning-aided tabu search detection for large MIMO systems. IEEE Trans. Wirel. Commun.

2020

19, 4262–4275. [CrossRef]

113.

Li, L.; Hou, H.; Meng, W. Convolutional-Neural-Network-Based Detection Algorithm for Uplink Multiuser Massive MIMO

Systems. IEEE Access 2020,8, 64250–64265. [CrossRef]

114.

Yan, X.; Long, F.; Wang, J.; Fu, N.; Ou, W.; Liu, B. Signal detection of MIMO-OFDM system based on auto encoder and extreme

learning machine. In Proceedings of the 2017 International Joint Conference on Neural Networks (IJCNN), Shenzhen, China,

14–19 May 2017; pp. 1602–1606.

115.

Li, F.; Zhou, M.; Li, H. A novel neural network optimized by quantum genetic algorithm for signal detection in MIMO-OFDM

systems. In Proceedings of the Computational Intelligence in Control and Automation (CICA), Orlando, FL, USA, 11–15 April

2011; pp. 170–177.

116.

Belgiovine, M.; Sankhe, K.; Bocanegra, C.; Roy, D.; Chowdhury, K.R. Deep Learning at the Edge for Channel Estimation in

Beyond-5G Massive MIMO. IEEE Wirel. Commun. 2021,28, 19–25. [CrossRef]

117.

Soltani, M.; Pourahmadi, V.; Mirzaei, A.; Sheikhzadeh, H. Deep learning-based channel estimation. IEEE Commun. Lett.

2019

23, 652–655. [CrossRef]

118.

Testa, A.; Coronato, A.; Cinque, M.; Augusto, J.C. Static veriﬁcation of wireless sensor networks with formal methods. In

Proceedings of the 2012 Eighth International Conference on Signal Image Technology and Internet Based Systems, Naples, Italy,

25–29 November 2012; pp. 587–594.

119.

Liu, Z.; Zhang, L.; Ding, Z. Overcoming the Channel Estimation Barrier in Massive MIMO Communication via Deep Learning.

IEEE Wirel. Commun. 2020,27, 104–111. [CrossRef]

120.

Huang, H.; Yang, J.; Huang, H.; Song, Y.; Gui, G. Deep learning for super-resolution channel estimation and DOA estimation

based massive MIMO system. IEEE Trans. Veh. Technol. 2018,67, 8549–8560. [CrossRef]

121.

Huang, H.; Gui, G.; Sari, H.; Adachi, F. Deep learning for super-resolution DOA estimation in massive MIMO systems. In

Proceedings of the 2018 IEEE 88th Vehicular Technology Conference (VTC-Fall), Chicago, IL, USA, 27–30 August 2018; pp. 1–5.

122.

Nie, S.; Akyildiz, I.F. Deep kernel learning-based channel estimation in ultra-massive MIMO communications at 0.06-10 THz. In

Proceedings of the 2019 IEEE Globecom Workshops (GC Wkshps), Taipei, Taiwan, 9–13 December 2019; pp. 1–6.

123.

Sun, H.; Zhao, Z.; Fu, X.; Hong, M. Limited feedback double directional massive MIMO channel estimation: From low-rank

modeling to deep learning. In Proceedings of the 2018 IEEE 19th International Workshop on Signal Processing Advances in

Wireless Communications (SPAWC), Kalamata, Greece, 25–28 June 2018; pp. 1–5.

124.

Gao, S.; Dong, P.; Pan, Z.; Li, G.Y. Deep learning based channel estimation for massive MIMO with mixed-resolution ADCs. IEEE

Commun. Lett. 2019,23, 1989–1993. [CrossRef]

125.

Zicheng, J.; Shen, G.; Nan, L.; Zhiwen, P.; Xiaohu, Y. Deep Learning-Based Channel Estimation for Massive-MIMO With

Mixed-Resolution ADCs and Low-Resolution Information Utilization. IEEE Access 2021,9, 54938–54950. [CrossRef]

126.

Nguyen, D.H. Neural network-optimized channel estimator and training signal design for MIMO systems with few-bit ADCs.

IEEE Signal Process. Lett. 2020,27, 1370–1374. [CrossRef]

127.

Takeda, M.Y.; Klautau, A.; Mezghani, A.; Heath, R.W. MIMO channel estimation with non-ideal ADCs: Deep learning versus

GAMP. In Proceedings of the 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing (MLSP), Gold

Coast, Australia, 13–16 October 2019; pp. 1–6.

128.

Yang, Y.; Gao, F.; Xing, C.; An, J.; Alkhateeb, A. Deep multimodal learning: Merging sensory data for massive MIMO channel

prediction. IEEE J. Sel. Areas Commun. 2020,39, 1885–1898. [CrossRef]

129.

Alrabeiah, M.; Alkhateeb, A. Deep learning for TDD and FDD massive MIMO: Mapping channels in space and frequency. In

Proceedings of the 2019 53rd Asilomar Conference on Signals, Systems, and Computers, Paciﬁc Grove, CA, USA, 3–6 November

2019; pp. 1465–1470.

130.

Tugfe Demir, Ö.; Björnson, E. Channel Estimation in Massive MIMO under Hardware Non-Linearities: Bayesian Methods versus

Deep Learning. arXiv 2019, arXiv:1911.07316.

131.

Le, H.A.; Van Chien, T.; Nguyen, T.H.; Choo, H.; Nguyen, V.D. Machine Learning-Based 5G-and-Beyond Channel Estimation for

MIMO-OFDM Communication Systems. Sensors 2021,21, 4861. [CrossRef]

132.

Zimaglia, E. Deep Learning Application to 5G Physical Layer for Channel Estimation and CSI Feedback Improvement. Ph.D.

Thesis, Politecnico di Torino, Torino, Italy, 2019.

133.

Helmy, H.; El Daysti, S.; Shatila, H.; Aboul-Dahab, M. Performance Enhancement of Massive MIMO Using Deep Learning-Based

Channel Estimation. In IOP Conference Series: Materials Science and Engineering; IOP Publishing: Bristol, UK, 2021; Volume 1051,

p. 012029.

134.

Hirose, H.; Ohtsuki, T.; Gui, G. Deep Learning-Based Channel Estimation for Massive MIMO Systems With Pilot Contamination.

IEEE Open J. Veh. Technol. 2020,2, 67–77. [CrossRef]

135.

Qiang, H.; Feifei, G.; Hao, Z.; Shi, J.; Ye, L.G. Deep Learning for MIMO Channel Estimation: Interpretation, Performance, and

Comparison. arXiv 2019, arXiv:1911.01918.

136.

Wen, C.K.; Jin, S.; Wong, K.K.; Chen, J.C.; Ting, P. Channel estimation for massive MIMO using Gaussian-mixture Bayesian

learning. IEEE Trans. Wirel. Commun. 2014,14, 1356–1368. [CrossRef]

Sensors 2022,22, 309 35 of 41

137.

Ding, Y.; Chiu, S.E.; Rao, B.D. Bayesian channel estimation algorithms for massive MIMO systems with hybrid analog-digital

processing and low-resolution ADCs. IEEE J. Sel. Top. Signal Process. 2018,12, 499–513. [CrossRef]

138.

Zhu, C.; Zheng, Z.; Jiang, B.; Zhong, W.; Gao, X. Bayesian channel estimation for massive MIMO communications. In Proceedings

of the 2016 IEEE 83rd Vehicular Technology Conference (VTC Spring), Nanjing, China, 15–18 May 2016; pp. 1–5.

139.

Chun, C.J.; Kang, J.M.; Kim, I.M. Deep learning-based joint pilot design and channel estimation for multiuser MIMO channels.

IEEE Commun. Lett. 2019,23, 1999–2003. [CrossRef]

140.

Xu, J.; Zhu, P.; Li, J.; You, X. Deep learning-based pilot design for multi-user distributed massive MIMO systems. IEEE Wirel.

Commun. Lett. 2019,8, 1016–1019. [CrossRef]

141.

Mehrabi, M.; Mohammadkarimi, M.; Ardakani, M.; Jing, Y. Decision Directed Channel Estimation Based on Deep Neural Network

k-Step Predictor for MIMO Communications in 5G. IEEE J. Sel. Areas Commun. 2019,37, 2443–2456. [CrossRef]

142.

Ma, W.; Qi, C.; Zhang, Z.; Cheng, J. Sparse channel estimation and hybrid precoding using deep learning for millimeter wave

massive MIMO. IEEE Trans. Commun. 2020,68, 2838–2849. [CrossRef]

143.

Kang, J.M.; Chun, C.J.; Kim, I.M. Deep learning based channel estimation for MIMO systems with received SNR feedback. IEEE

Access 2020,8, 121162–121181. [CrossRef]

144.

Jin, Y.; Zhang, J.; Jin, S.; Ai, B. Channel estimation for cell-free mmWave massive MIMO through deep learning. IEEE Trans. Veh.

Technol. 2019,68, 10325–10329. [CrossRef]

145.

Chen, Y.; Han, C. Deep cnn-based spherical-wave channel estimation for terahertz ultra-massive mimo systems. In Proceedings

of the GLOBECOM 2020—2020 IEEE Global Communications Conference, Taipei, Taiwan, 7–11 December 2020; pp. 1–6.

146.

Balevi, E.; Doshi, A.; Andrews, J.G. Massive MIMO channel estimation with an untrained deep neural network. IEEE Trans.

Wirel. Commun. 2020,19, 2079–2090. [CrossRef]

147.

Mthethwa, B.; Xu, H. Deep Learning-Based Wireless Channel Estimation for MIMO Uncoded Space-Time Labeling Diversity.

IEEE Access 2020,8, 224608–224620. [CrossRef]

148.

Liu, Y.; Chai, J.; Zhang, Y.; Liu, Z.; Jin, M.; Qiu, T. Low-complexity neural network based DOA estimation for wideband signals in

massive MIMO systems. AEU-Int. J. Electron. Commun. 2021,138, 153853. [CrossRef]

149.

Demir, Ö.T.; Björnson, E. Channel estimation under hardware impairments: Bayesian methods versus deep learning. In

Proceedings of the 2019 16th International Symposium on Wireless Communication Systems (ISWCS), Berlin, Germany, 27–30

August 2019; pp. 193–197.

150.

Qin, X.; Qu, F.; Zheng, Y.R. Bayesian iterative channel estimation and turbo equalization for multiple-input–multiple-output

underwater acoustic communications. IEEE J. Ocean. Eng. 2020,46, 326–337. [CrossRef]

151.

Ling, J.; Tan, X.; Yardibi, T.; Li, J.; Nordenvaad, M.L.; He, H.; Zhao, K. On Bayesian channel estimation and FFT-based symbol

detection in MIMO underwater acoustic communications. IEEE J. Ocean. Eng. 2013,39, 59–73. [CrossRef]

152.

Zhang, J.; He, H.; Yang, X.; Wen, C.K.; Jin, S.; Ma, X. Model-driven Deep Learning Based Turbo-MIMO Receiver. In Proceedings

of the 2020 IEEE 21st International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), Lucca, Italy,

26–29 May 2020; pp. 1–5.

153.

Sabeti, P.; Farhang, A.; Macaluso, I.; Marchetti, N.; Doyle, L. Blind channel estimation for massive mimo: A deep learning assisted

approach. In Proceedings of the ICC 2020—2020 IEEE International Conference on Communications (ICC), Montreal, QC,

Canada, 7–11 June 2020; pp. 1–6.

154.

Shohat, M.; Tsintsadze, G.; Shlezinger, N.; Eldar, Y.C. Deep quantization for MIMO channel estimation. In Proceedings of the

ICASSP 2019—2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Toronto, ON, Canada,

12–17 May 2019; pp. 3912–3916.

155.

Wang, X.; Hua, H.; Xu, Y. Pilot-assisted channel estimation and signal detection in uplink multi-user MIMO systems with deep

learning. IEEE Access 2020,8, 44936–44946. [CrossRef]

156.

Marinberg, B.; Cohen, A.; Ben-Dror, E.; Permuter, H.H. A Study on MIMO Channel Estimation by 2D and 3D Convolutional

Neural Networks. In Proceedings of the 2020 IEEE International Conference on Advanced Networks and Telecommunications

Systems (ANTS), New Delhi, India, 14–17 December 2020; pp. 1–6.

157.

Ma, X.; Gao, Z. Data-driven deep learning to design pilot and channel estimator for massive MIMO. IEEE Trans. Veh. Technol.

2020,69, 5677–5682. [CrossRef]

158.

Sohrabi, F.; Attiah, K.M.; Yu, W. Deep learning for distributed channel feedback and multiuser precoding in FDD massive MIMO.

IEEE Trans. Wirel. Commun. 2021,20, 4044–4057. [CrossRef]

159.

Chen, T.; Guo, J.; Wen, C.K.; Jin, S.; Li, G.Y.; Wang, X.; Hou, X. Deep learning for joint channel estimation and feedback in massive

MIMO systems. arXiv 2020, arXiv:2011.07242.

160.

Wang, C.X.; Di Renzo, M.; Stanczak, S.; Wang, S.; Larsson, E.G. Artiﬁcial intelligence enabled wireless networking for 5G and

beyond: Recent advances and future challenges. IEEE Wirel. Commun. 2020,27, 16–23. [CrossRef]

161.

Ayyalasomayajula, R.; Arun, A.; Wu, C.; Sharma, S.; Sethi, A.R.; Vasisht, D.; Bharadia, D. Deep learning based wireless localization

for indoor navigation. In Proceedings of the 26th Annual International Conference on Mobile Computing and Networking,

London, UK, 21–25 September 2020; pp. 1–14.

162.

Song, X.; Fan, X.; Xiang, C.; Ye, Q.; Liu, L.; Wang, Z.; He, X.; Yang, N.; Fang, G. A novel convolutional neural network based

indoor localization framework with WiFi ﬁngerprinting. IEEE Access 2019,7, 110698–110709. [CrossRef]

Sensors 2022,22, 309 36 of 41

163.

Zeng, X.; Zhang, F.; Wang, B.; Liu, K.R. Massive MIMO for High-Accuracy Target Localization and Tracking. IEEE Internet Things

J. 2021,8, 10131–10145. [CrossRef]

164.

Mukhtar, H. Machine Learning Enabled-Localization in 5G and LTE Using Image Classiﬁcation and Deep Learning. Ph.D. Thesis,

Université d’Ottawa/University of Ottawa, Ottawa, ON, Canada, 2021.

165.

Shao, W.; Luo, H.; Zhao, F.; Ma, Y.; Zhao, Z.; Crivello, A. Indoor positioning based on ﬁngerprint-image and deep learning. IEEE

Access 2018,6, 74699–74712. [CrossRef]

166.

Vieira, J.; Leitinger, E.; Sarajlic, M.; Li, X.; Tufvesson, F. Deep convolutional neural networks for massive MIMO ﬁngerprint-based

positioning. In Proceedings of the 2017 IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio

Communications (PIMRC), Helsinki, Finland, 8–13 October 2017; pp. 1–6.

167.

Sun, X.; Wu, C.; Gao, X.; Li, G.Y. Fingerprint-based localization for massive MIMO-OFDM system with deep convolutional neural

networks. IEEE Trans. Veh. Technol. 2019,68, 10846–10857. [CrossRef]

168.

Sun, X.; Wu, C.; Gao, X.; Li, G.Y. Deep Convolutional Neural Networks Enabled Fingerprint Localization for Massive MIMO-

OFDM System. In Proceedings of the 2019 IEEE Global Communications Conference (GLOBECOM), Taipei, Taiwan, 9–13

December 2019; pp. 1–6.

169.

Yang, Y.; Gao, F.; Li, G.Y.; Jian, M. Deep learning-based downlink channel prediction for FDD massive MIMO system. IEEE

Commun. Lett. 2019,23, 1994–1998. [CrossRef]

170.

Yang, Y.; Gao, F.; Zhong, Z.; Ai, B.; Alkhateeb, A. Deep transfer learning-based downlink channel prediction for FDD massive

MIMO systems. IEEE Trans. Commun. 2020,68, 7485–7497. [CrossRef]

171.

Osama, I.; Rihan, M.; Elhefnawy, M.; Eldolil, S. Deep Learning Based Hybrid Precoding Technique for Millimeter-Wave Massive

MIMO Systems. In Proceedings of the 2021 International Conference on Electronic Engineering (ICEEM), Menouf, Egypt, 3–4

July 2021; pp. 1–7.

172.

Li, X.; Alkhateeb, A. Deep learning for direct hybrid precoding in millimeter wave massive MIMO systems. In Proceedings of the

2019 53rd Asilomar Conference on Signals, Systems, and Computers, Paciﬁc Grove, CA, USA, 3-6 November 2019; pp. 800–805.

173.

Arnold, M.; Hoydis, J.; ten Brink, S. Novel massive MIMO channel sounding data applied to deep learning-based indoor

positioning. In Proceedings of the SCC 2019, 12th International ITG Conference on Systems, Communications and Coding,

Rostock, Germany, 11–14 February 2019; pp. 1–6.

174.

Widmaier, M.; Arnold, M.; Dorner, S.; Cammerer, S.; ten Brink, S. Towards practical indoor positioning based on massive mimo

systems. In Proceedings of the 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall), Honolulu, HI, USA, 22–25

September 2019; pp. 1–6.

175.

Arnold, M.; Dorner, S.; Cammerer, S.; Ten Brink, S. On deep learning-based massive MIMO indoor user localization. In Pro-

ceedings of the 2018 IEEE 19th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC),

Kalamata, Greece, 25–28 June 2018; pp. 1–5.

176. Sobehy, A. Machine Learning Based Localization in 5G. Ph.D. Thesis, Institut Polytechnique de Paris, Palaiseau, France, 2020.

177.

Sobehy, A.; Renault, É.; Mühlethaler, P. Generalization aspect of accurate machine learning models for CSI-based localization.

Ann. Telecommun. 2021, 1–13. [CrossRef] [PubMed]

178.

Moosavi, S.S.; Fortier, P. An Accurate, Robust and Low Dimensionality Deep Learning Localization Approach in DM-MIMO

Systems Based on RSS. Wirel. Pers. Commun. 2021. [CrossRef]

179.

De Bast, S.; Pollin, S. MaMIMO CSI-based positioning using CNNs: Peeking inside the black box. In Proceedings of the 2020

IEEE International Conference on Communications Workshops (ICC Workshops), Montreal, QC, Canada, 7–11 June 2020; pp. 1–6.

180.

De Bast, S.; Guevara, A.P.; Pollin, S. CSI-based positioning in massive MIMO systems using convolutional neural networks.

In Proceedings of the 2020 IEEE 91st Vehicular Technology Conference (VTC2020-Spring), Helsinki, Finland, 25–28 May 2020;

pp. 1–5.

181.

Sobehy, A.; Renault, E.; Muhlethaler, P. CSI based indoor localization using ensemble neural networks. In Proceedings of the

MLN 2019: 2nd IFIP International Conference on Machine Learning for Networking, Paris, France, 3–5 December 2019; Volume

12081, pp. 367–378.

182.

Cerar, G.; Švigelj, A.; Mohorˇciˇc, M.; Fortuna, C.; Javornik, T. Improving CSI-based Massive MIMO Indoor Positioning using

Convolutional Neural Network. arXiv 2021, arXiv:2102.03130.

183.

Wu, C.; Yi, X.; Wang, W.; You, L.; Huang, Q.; Gao, X.; Liu, Q. Learning to localize: A 3D CNN approach to user positioning in

massive MIMO-OFDM systems. IEEE Trans. Wirel. Commun. 2021,20, 4556–4570. [CrossRef]

184.

Wu, C.; Yi, X.; Wang, W.; Huang, Q.; Gao, X. 3D CNN-Enabled Positioning in 3D Massive MIMO-OFDM Systems. In Proceedings

of the ICC 2020—2020 IEEE International Conference on Communications (ICC), Montreal, QC, Canada, 7–11 June 2020; pp. 1–6.

[CrossRef]

185.

Hejazi, F.; Vuckovic, K.; Rahnavard, N. DyLoc: Dynamic Localization for Massive MIMO Using Predictive Recurrent Neural

Networks. arXiv 2021, arXiv:2101.07848.

186.

Gante, J.; Falcao, G.; Sousa, L. Deep learning architectures for accurate millimeter wave positioning in 5G. Neural Process. Lett.

2020,51, 487–514. [CrossRef]

187.

Patel, P.A. Millimeter Wave Positioning with Deep Learning. Master’s Thesis, California State University, Sacramento, CA, USA,

2020.

Sensors 2022,22, 309 37 of 41

188.

Liu, Y.J.; Tang, L.; Tong, S.; Chen, C.P.; Li, D.J. Reinforcement learning design-based adaptive tracking control with less learning

parameters for nonlinear discrete-time MIMO systems. IEEE Trans. Neural Netw. Learn. Syst. 2014,26, 165–176. [CrossRef]

189.

Mashhadi, M.B.; Gündüz, D. Deep learning for massive MIMO channel state acquisition and feedback. J. Indian Inst. Sci.

2020

100, 369–382. [CrossRef]

190.

Qi, C.; Dong, P.; Ma, W.; Zhang, H.; Zhang, Z.; Li, G.Y. Acquisition of channel state information for mmwave massive MIMO:

Traditional and machine learning-based approaches. arXiv 2020, arXiv:2006.08894.

191.

Muhan, C.; Jiajia, G.; Xiao, L.; Shi, J. An overview of the CSI feedback based on deep learning for massive MIMO systems. Chin. J.

Internet Things 2020,4, 33.

192.

Deng, D.; Li, X.; Zhao, M.; Rabie, K.M.; Kharel, R. Deep learning-based secure MIMO communications with imperfect CSI for

heterogeneous networks. Sensors 2020,20, 1730. [CrossRef]

193.

Li, X.; Wu, H. Spatio-temporal representation with deep neural recurrent network in MIMO CSI feedback. IEEE Wirel. Commun.

Lett. 2020,9, 653–657. [CrossRef]

194.

Lu, C.; Xu, W.; Shen, H.; Zhu, J.; Wang, K. MIMO channel information feedback using deep recurrent network. IEEE Commun.

Lett. 2018,23, 188–191. [CrossRef]

195.

Huang, C.; Alexandropoulos, G.C.; Zappone, A.; Yuen, C.; Debbah, M. Deep learning for UL/DL channel calibration in generic

massive MIMO systems. In Proceedings of the ICC 2019—2019 IEEE International Conference on Communications (ICC),

Shanghai, China, 20–24 May 2019; pp. 1–6.

196.

Liu, Z.; Zhang, L.; Ding, Z. Exploiting bi-directional channel reciprocity in deep learning for low rate massive MIMO CSI

feedback. IEEE Wirel. Commun. Lett. 2019,8, 889–892. [CrossRef]

197.

Ye, H.; Gao, F.; Qian, J.; Wang, H.; Li, G.Y. Deep learning-based denoise network for CSI feedback in FDD massive MIMO systems.

IEEE Commun. Lett. 2020,24, 1742–1746. [CrossRef]

198. Liang, P.; Fan, J.; Shen, W.; Qin, Z.; Li, G.Y. Deep learning and compressive sensing-based CSI feedback in FDD massive MIMO

systems. IEEE Trans. Veh. Technol. 2020,69, 9217–9222. [CrossRef]

199.

Jang, Y.; Kong, G.; Jung, M.; Choi, S.; Kim, I.M. Deep autoencoder based CSI feedback with feedback errors and feedback delay in

FDD massive MIMO systems. IEEE Wirel. Commun. Lett. 2019,8, 833–836. [CrossRef]

200.

Guo, J.; Wen, C.K.; Jin, S.; Li, G.Y. Convolutional neural network-based multiple-rate compressive sensing for massive MIMO CSI

feedback: Design, simulation, and analysis. IEEE Trans. Wirel. Commun. 2020,19, 2827–2840. [CrossRef]

201.

Chen, T.; Guo, J.; Jin, S.; Wen, C.K.; Li, G.Y. A novel quantization method for deep learning-based massive MIMO CSI feedback.

In Proceedings of the 2019 IEEE Global Conference on Signal and Information Processing (GlobalSIP), Ottawa, ON, Canada,

11–14 November 2019; pp. 1–5.

202.

Huang, X.L.; Wu, J.; Wen, Y.; Hu, F.; Wang, Y.; Jiang, T. Rate-adaptive feedback with Bayesian compressive sensing in multiuser

MIMO beamforming systems. IEEE Trans. Wirel. Commun. 2016,15, 4839–4851. [CrossRef]

203.

Mashhadi, M.B.; Yang, Q.; Gündüz, D. CNN-based analog CSI feedback in FDD MIMO-OFDM systems. In Proceedings of the

ICASSP 2020—2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain, 4–8

May 2020; pp. 8579–8583.

204.

Boloursaz Mashhadi, M.; Yang, Q.; Gunduz, D. CNN-based Analog CSI Feedback in FDD MIMO-OFDM Systems. arXiv

2019

arXiv:1910.10428.

205.

Yang, Q.; Mashhadi, M.B.; Gündüz, D. Deep convolutional compression for massive MIMO CSI feedback. In Proceedings of the

2019 IEEE 29th International Workshop on Machine Learning for Signal Processing (MLSP), Pittsburgh, PA, USA, 13–16 October

2019; pp. 1–6.

206.

Mashhadi, M.B.; Yang, Q.; Gündüz, D. Distributed deep convolutional compression for massive MIMO CSI feedback. IEEE Trans.

Wirel. Commun. 2020,20, 2621–2633. [CrossRef]

207.

Liu, Z.; Zhang, L.; Ding, Z. An efﬁcient deep learning framework for low rate massive MIMO CSI reporting. IEEE Trans. Commun.

2020,68, 4761–4772. [CrossRef]

208.

Guo, J.; Yang, X.; Wen, C.K.; Jin, S.; Li, G.Y. DL-based CSI feedback and cooperative recovery in massive MIMO. arXiv

2020

arXiv:2003.03303.

209.

Liao, Y.; Yao, H.; Hua, Y.; Li, C. CSI feedback based on deep learning for massive MIMO systems. IEEE Access

2019

,7, 86810–86820.

[CrossRef]

210.

Guo, J.; Wen, C.K.; Jin, S. Deep learning-based CSI feedback for beamforming in single-and multi-cell massive MIMO systems.

IEEE J. Sel. Areas Commun. 2020,39, 1872–1884. [CrossRef]

211.

Lu, Z.; Wang, J.; Song, J. Binary neural network aided CSI feedback in massive MIMO system. IEEE Wirel. Commun. Lett.

2021

10, 1305–1308. [CrossRef]

212.

Chen, M.; Guo, J.; Wen, C.K.; Jin, S.; Li, G.Y.; Yang, A. Deep Learning-based Implicit CSI Feedback in Massive MIMO. arXiv

2021

arXiv:2105.10100.

213.

Wang, J.; Gui, G.; Ohtsuki, T.; Adebisi, B.; Gacanin, H.; Sari, H. Compressive Sampled CSI Feedback Method Based on Deep

Learning for FDD Massive MIMO Systems. IEEE Trans. Commun. 2021,69, 5873–5885. [CrossRef]

214.

Cao, Z.; Shih, W.T.; Guo, J.; Wen, C.K.; Jin, S. Lightweight convolutional neural networks for CSI feedback in massive MIMO.

IEEE Commun. Lett. 2021,25, 2624–2628. [CrossRef]

Sensors 2022,22, 309 38 of 41

215.

Ji, S.; Li, M. CLNet: Complex Input Lightweight Neural Network Designed for Massive MIMO CSI Feedback. IEEE Wirel.

Commun. Lett. 2021,10, 2318–2322. [CrossRef]

216.

Arnold, M.; Dörner, S.; Cammerer, S.; Hoydis, J.; ten Brink, S. Towards practical FDD massive MIMO: CSI extrapolation driven

by deep learning and actual channel measurements. In Proceedings of the 2019 53rd Asilomar Conference on Signals, Systems,

and Computers, Paciﬁc Grove, CA, USA, 3–6 November 2019; pp. 1972–1976.

217.

Liu, Q.; Guo, J.; Wen, C.K.; Jin, S. Adversarial attack on DL-based massive MIMO CSI feedback. J. Commun. Netw.

2020

22, 230–235. [CrossRef]

218.

Lu, Z.; Zhang, X.; He, H.; Wang, J.; Song, J. Binarized Aggregated Network with Quantization: Flexible Deep Learning

Deployment for CSI Feedback in Massive MIMO System. arXiv 2021, arXiv:2105.00354.

219.

Yu, X.; Li, X.; Wu, H.; Bai, Y. DS-NLCsiNet: Exploiting non-local neural networks for massive MIMO CSI feedback. IEEE Commun.

Lett. 2020,24, 2790–2794. [CrossRef]

220.

Sun, Y.; Xu, W.; Fan, L.; Li, G.Y.; Karagiannidis, G.K. AnciNet: An efﬁcient deep learning approach for feedback compression of

estimated CSI in massive MIMO systems. IEEE Wirel. Commun. Lett. 2020,9, 2192–2196. [CrossRef]

221.

Guo, J.; Wen, C.K.; Jin, S. CAnet: Uplink-aided downlink channel acquisition in FDD massive MIMO using deep learning. arXiv

2021, arXiv:2101.04377.

222.

Liu, Z.; Del Rosario, M.; Ding, Z. A markovian model-driven deep learning framework for massive MIMO CSI feedback. IEEE

Trans. Wirel. Commun. 2021. [CrossRef]

223.

Fan, G.; Sun, J.; Gui, G.; Gacanin, H.; Adebisi, B.; Ohtsuki, T. Fully Convolutional Neural Network Based CSI Limited Feedback

for FDD Massive MIMO Systems. IEEE Trans. Cogn. Commun. Netw. 2021. [CrossRef]

224.

Qing, C.; Cai, B.; Yang, Q.; Wang, J.; Huang, C. Deep learning for CSI feedback based on superimposed coding. IEEE Access

2019

7, 93723–93733. [CrossRef]

225.

Zeng, J.; He, Z.; Sun, J.; Adebisi, B.; Gacanin, H.; Gui, G.; Adachi, F. Deep Transfer Learning for 5G Massive MIMO Downlink CSI

Feedback. In Proceedings of the 2021 IEEE Wireless Communications and Networking Conference (WCNC), Nanjing, China, 29

March 2021; pp. 1–5.

226.

Zhang, Y.; Wang, J.; Sun, J.; Adebisi, B.; Gacanin, H.; Gui, G.; Adachi, F. CV-3DCNN: Complex-valued deep learning for CSI

prediction in FDD massive MIMO systems. IEEE Wirel. Commun. Lett. 2020,10, 266–270. [CrossRef]

227.

Han, Y.; Li, M.; Jin, S.; Wen, C.K.; Ma, X. Deep learning-based FDD non-stationary massive MIMO downlink channel reconstruc-

tion. IEEE J. Sel. Areas Commun. 2020,38, 1980–1993. [CrossRef]

228.

He, H.; Wen, C.K.; Jin, S.; Li, G.Y. Deep learning-based channel estimation for beamspace mmWave massive MIMO systems.

IEEE Wirel. Commun. Lett. 2018,7, 852–855. [CrossRef]

229.

Wang, Q.; Feng, K. Precodernet: Hybrid beamforming for millimeter wave systems using deep reinforcement learning. arXiv

2019, arXiv:1907.13266.

230.

Wang, Q.; Feng, K.; Li, X.; Jin, S. PrecoderNet: Hybrid beamforming for millimeter wave systems with deep reinforcement

learning. IEEE Wirel. Commun. Lett. 2020,9, 1677–1681. [CrossRef]

231.

Alkhateeb, A.; Alex, S.; Varkey, P.; Li, Y.; Qu, Q.; Tujkovic, D. Deep learning coordinated beamforming for highly-mobile

millimeter wave systems. IEEE Access 2018,6, 37328–37348. [CrossRef]

232.

Huang, H.; Song, Y.; Yang, J.; Gui, G.; Adachi, F. Deep-learning-based millimeter-wave massive MIMO for hybrid precoding.

IEEE Trans. Veh. Technol. 2019,68, 3027–3032. [CrossRef]

233.

Alkhateeb, A. DeepMIMO: A generic deep learning dataset for millimeter wave and massive MIMO applications. arXiv

2019

arXiv:1902.06435.

234.

Arikawa, S.; Karasawa, Y. A simpliﬁed MIMO channel characteristics evaluation scheme based on ray tracing and its application

to indoor radio systems. IEEE Antennas Wirel. Propag. Lett. 2014,13, 1737–1740. [CrossRef]

235.

Klautau, A.; Batista, P.; González-Prelcic, N.; Wang, Y.; Heath, R.W. 5G MIMO data for machine learning: Application to

beam-selection using deep learning. In Proceedings of the 2018 Information Theory and Applications Workshop (ITA), Rockland,

ON, Canada, 30 May–2 June 2018; pp. 1–9.

236.

Dong, P.; Zhang, H.; Li, G.Y.; Gaspar, I.S.; NaderiAlizadeh, N. Deep CNN-based channel estimation for mmWave massive MIMO

systems. IEEE J. Sel. Top. Signal Process. 2019,13, 989–1000. [CrossRef]

237.

Dong, P.; Zhang, H.; Li, G.Y.; NaderiAlizadeh, N.; Gaspar, I.S. Deep CNN for wideband mmWave massive MIMO channel

estimation using frequency correlation. In Proceedings of the ICASSP 2019—2019 IEEE International Conference on Acoustics,

Speech and Signal Processing (ICASSP), Toronto, ON, Canada, 12–17 May 2019; pp. 4529–4533.

238.

Yu, X.; Shen, J.C.; Zhang, J.; Letaief, K.B. Alternating minimization algorithms for hybrid precoding in millimeter wave MIMO

systems. IEEE J. Sel. Top. Signal Process. 2016,10, 485–500. [CrossRef]

239.

Bao, X.; Feng, W.; Zheng, J.; Li, J. Deep CNN and equivalent channel based hybrid precoding for mmWave massive MIMO

systems. IEEE Access 2020,8, 19327–19335. [CrossRef]

240.

Sung, L.; Cho, D.H. Multi-User Hybrid Beamforming System Based on Deep Neural Network in Millimeter-Wave Communication.

IEEE Access 2020,8, 91616–91623. [CrossRef]

241.

Faragallah, O.S.; El-Sayed, H.S.; Mohamed, G. Performance Enhancement of MmWave MIMO Systems Using Deep Learning

Framework. IEEE Access 2021,9, 92460–92472. [CrossRef]

Sensors 2022,22, 309 39 of 41

242.

Elbir, A.M. CNN-based precoder and combiner design in mmWave MIMO systems. IEEE Commun. Lett.

2019

,23, 1240–1243.

[CrossRef]

243.

Elbir, A.M.; Papazafeiropoulos, A.K. Hybrid precoding for multiuser millimeter wave massive MIMO systems: A deep learning

approach. IEEE Trans. Veh. Technol. 2019,69, 552–563. [CrossRef]

244.

Zhang, Y.; Mu, Y.; Liu, Y.; Zhang, T.; Qian, Y. Deep learning-based beamspace channel estimation in mmWave massive MIMO

systems. IEEE Wirel. Commun. Lett. 2020,9, 2212–2215. [CrossRef]

245.

Wei, X.; Hu, C.; Dai, L. Deep learning for beamspace channel estimation in millimeter-wave massive MIMO systems. IEEE Trans.

Commun. 2020,69, 182–193. [CrossRef]

246.

Ma, W.; Qi, C.; Zhang, Z.; Cheng, J. Deep learning for compressed sensing based channel estimation in millimeter wave massive

mimo. In Proceedings of the Sweden 2019 11th International Conference on Wireless Communications and Signal Processing

(WCSP), Xi’an, China, 23–25 October 2019; pp. 1–6.

247.

He, H.; Wen, C.K.; Jin, S. Bayesian optimal data detector for hybrid mmWave MIMO-OFDM systems with low-resolution ADCs.

IEEE J. Sel. Top. Signal Process. 2018,12, 469–483. [CrossRef]

248.

Wang, X.; Gursoy, M.C. Multi-Agent Double Deep Q-Learning for Beamforming in mmWave MIMO Networks. In Proceedings of

the 2020 IEEE 31st Annual International Symposium on Personal, Indoor and Mobile Radio Communications, Helsinki, Finland,

31 August–3 September 2020; pp. 1–6.

249.

Abdallah, A.; Celik, A.; Mansour, M.M.; Eltawil, A.M. Deep Learning Based Frequency-Selective Channel Estimation for Hybrid

mmWave MIMO Systems. arXiv 2021, arXiv:2102.10847.

250.

Jin, Y.; Zhang, J.; Ai, B.; Zhang, X. Channel estimation for mmWave massive MIMO with convolutional blind denoising network.

IEEE Commun. Lett. 2019,24, 95–98. [CrossRef]

251.

Tang, H.; Wang, J.; He, L. Off-grid sparse bayesian learning-based channel estimation for mmwave massive mimo uplink. IEEE

Wirel. Commun. Lett. 2018,8, 45–48. [CrossRef]

252.

Mishra, A.; Rajoriya, A.; Jagannatham, A.K.; Ascheid, G. Sparse Bayesian learning-based channel estimation in millimeter wave

hybrid MIMO systems. In Proceedings of the 2017 IEEE 18th International Workshop on Signal Processing Advances in Wireless

Communications (SPAWC), Lucca, Italy, 3–6 July 2017; pp. 1–5.

253.

Elbir, A.M.; Papazafeiropoulos, A.; Kourtessis, P.; Chatzinotas, S. Deep channel learning for large intelligent surfaces aided

mm-wave massive MIMO systems. IEEE Wirel. Commun. Lett. 2020,9, 1447–1451. [CrossRef]

254.

Lu, Q.; Lin, T.; Zhu, Y. Channel Estimation and Hybrid Precoding for Millimeter Wave Communications: A Deep Learning-Based

Approach. IEEE Access 2021,9, 120924–120939. [CrossRef]

255.

Song, N.; Ye, C.; Hu, X.; Yang, T. Deep Learning based Low-Rank Channel Recovery for Hybrid Beamforming in Millimeter-Wave

Massive MIMO. In Proceedings of the 2020 IEEE Wireless Communications and Networking Conference (WCNC), Seoul, Korea,

25–28 May 2020; pp. 1–6.

256.

Naeem, M.; Bashir, S.; Ullah, Z.; Syed, A.A. A near optimal scheduling algorithm for efﬁcient radio resource management in

multi-user MIMO systems. Wirel. Pers. Commun. 2019,106, 1411–1427. [CrossRef]

257.

Naeem, M.; Bashir, S.; Khan, M.U.; Syed, A.A. Modiﬁed SINR based user selection for MU-MIMO systems. In Proceedings of the

2015 International Conference on Information and Communication Technologies (ICICT), Karachi, Pakistan, 12-13 December

2015; pp. 1–4.

258.

Naeem, M.; Khan, M.U.; Bashir, S.; Syed, A.A. Modiﬁed leakage based user selection for MU-MIMO systems. In Proceedings of

the 2015 13th International Conference on Frontiers of Information Technology (FIT), Islamabad, Pakistan, 14–16 December 2015;

pp. 320–323.

259.

Sanguinetti, L.; Zappone, A.; Debbah, M. Deep learning power allocation in massive MIMO. In Proceedings of the 2018 52nd

Asilomar conference on signals, systems, and computers, Paciﬁc Grove, CA, USA, 28–31 October 2018; pp. 1257–1261.

260.

Zhang, Y.; Kang, C.; Ma, T.; Teng, Y.; Guo, D. Power allocation in multi-cell networks using deep reinforcement learning. In

Proceedings of the 2018 IEEE 88th Vehicular Technology Conference (VTC-Fall), Chicago, IL, USA, 27–30 August 2018; pp. 1–6.

261.

Mismar, F.B.; Evans, B.L. Deep learning in downlink coordinated multipoint in new radio heterogeneous networks. IEEE Wirel.

Commun. Lett. 2019,8, 1040–1043. [CrossRef]

262.

Cao, G.; Lu, Z.; Wen, X.; Lei, T.; Hu, Z. AIF: An artiﬁcial intelligence framework for smart wireless network management. IEEE

Commun. Lett. 2017,22, 400–403. [CrossRef]

263.

Xiao, L.; Li, Y.; Dai, C.; Dai, H.; Poor, H.V. Reinforcement learning-based NOMA power allocation in the presence of smart

jamming. IEEE Trans. Veh. Technol. 2017,67, 3377–3389. [CrossRef]

264.

Cao, Y.; Zhang, G.; Li, G.; Zhang, J. A Deep Q-Network Based-Resource Allocation Scheme for Massive MIMO-NOMA. IEEE

Commun. Lett. 2021,25, 1544–1548. [CrossRef]

265.

Van Chien, T. Spatial Resource Allocation in Massive MIMO Communications: From Cellular to Cell-Free; Linköping University

Electronic Press: Linköping, Sweden, 2019; Volume 2036.

266.

Cui, W.; Shen, K.; Yu, W. Spatial deep learning for wireless scheduling. IEEE J. Sel. Areas Commun.

2019

,37, 1248–1261. [CrossRef]

267.

Van Chien, T.; Canh, T.N.; Bjornson, E.; Larsson, E.G. Power control in cellular massive MIMO with varying user activity: A deep

learning solution. IEEE Trans. Wirel. Commun. 2020,19, 5732–5748. [CrossRef]

Sensors 2022,22, 309 40 of 41

268.

Van Chien, T.; Bjornson, E.; Larsson, E.G. Sum spectral efﬁciency maximization in massive MIMO systems: Beneﬁts from deep

learning. In Proceedings of the ICC 2019—2019 IEEE International Conference on Communications (ICC), Shanghai, China,

20–24 May 2019; pp. 1–6.

269.

Xia, W.; Zheng, G.; Zhu, Y.; Zhang, J.; Wang, J.; Petropulu, A.P. A deep learning framework for optimization of MISO downlink

beamforming. IEEE Trans. Commun. 2019,68, 1866–1880. [CrossRef]

270.

Xia, W.; Zheng, G.; Zhu, Y.; Zhang, J.; Wang, J.; Petropulu, A.P. Deep learning based beamforming neural networks in downlink

MISO systems. In Proceedings of the 2019 IEEE International Conference on Communications Workshops (ICC Workshops),

Shanghai, China, 22–24 May 2019; pp. 1–5.

271.

Yang, Y.; Li, Y.; Li, K.; Zhao, S.; Chen, R.; Wang, J.; Ci, S. DECCO: Deep-learning enabled coverage and capacity optimization for

massive MIMO systems. IEEE Access 2018,6, 23361–23371. [CrossRef]

272.

Zappone, A.; Sanguinetti, L.; Debbah, M. User association and load balancing for massive MIMO through deep learning. In

Proceedings of the 2018 52nd Asilomar Conference on Signals, Systems, and Computers, Paciﬁc Grove, CA, USA, 28–31 October

2018; pp. 1262–1266.

273.

Kim, K.; Lee, J.; Choi, J. Deep learning based pilot allocation scheme (DL-PAS) for 5G massive MIMO system. IEEE Commun. Lett.

2018,22, 828–831. [CrossRef]

274.

Hoffmann, M.; Kliks, A.; Kryszkiewicz, P.; Koudouridis, G.P. A Reinforcement Learning Approach for Base Station On/Off

Switching in Heterogeneous M-MIMO Networks. In Proceedings of the 2020 IEEE 21st International Symposium on “A World of

Wireless, Mobile and Multimedia Networks” (WoWMoM), Cork, Ireland, 31 August–3 September 2020; pp. 170–172.

275.

Wijaya, M.A.; Fukawa, K.; Suzuki, H. Intercell-interference cancellation and neural network transmit power optimization for

MIMO channels. In Proceedings of the 2015 IEEE 82nd Vehicular Technology Conference (VTC2015-Fall), Boston, MA, USA, 6–9

September 2015; pp. 1–5.

276.

Ji, X.; Jin, M.; Liu, W.; Xu, Z. Inter-Cell Interference Suppression for MIMO-OFDM Systems Based on Complex-Valued Neural

Network. In Proceedings of the 2020 IEEE International Conference on Communications Workshops (ICC Workshops), Dublin,

Ireland, 7–11 June 2020; pp. 1–6.

277.

Maksymyuk, T.; Gazda, J.; Yaremko, O.; Nevinskiy, D. Deep learning based massive MIMO beamforming for 5G mobile network.

In Proceedings of the 2018 IEEE 4th International Symposium on Wireless Systems within the International Conferences on

Intelligent Data Acquisition and Advanced Computing Systems (IDAACS-SWS), Dortmund, Germany, 20–21 September 2018;

pp. 241–244.

278.

Carpi, F.; Häger, C.; Martalò, M.; Raheli, R.; Pﬁster, H.D. Reinforcement learning for channel coding: Learned bit-ﬂipping

decoding. In Proceedings of the 2019 57th Annual Allerton Conference on Communication, Control, and Computing (Allerton),

Monticello, IL, USA, 24–27 September 2019; pp. 922–929.

279.

Gruber, T.; Cammerer, S.; Hoydis, J.; ten Brink, S. On deep learning-based channel decoding. In Proceedings of the 2017 51st

Annual Conference on Information Sciences and Systems (CISS), Baltimore, MD, USA, 22–24 March 2017; pp. 1–6.

280.

Liang, F.; Shen, C.; Wu, F. An iterative BP-CNN architecture for channel decoding. IEEE J. Sel. Top. Signal Process.

2018

12, 144–159. [CrossRef]

281.

Nachmani, E.; Marciano, E.; Lugosch, L.; Gross, W.J.; Burshtein, D.; Be’ery, Y. Deep learning methods for improved decoding of

linear codes. IEEE J. Sel. Top. Signal Process. 2018,12, 119–131. [CrossRef]

282.

Kang, J.M.; Kim, I.M.; Chun, C.J. Deep learning-based MIMO-NOMA with imperfect SIC decoding. IEEE Syst. J.

2019

14, 3414–3417. [CrossRef]

283.

Wang, S.; Liu, H.; Gomes, P.H.; Krishnamachari, B. Deep reinforcement learning for dynamic multichannel access in wireless

networks. IEEE Trans. Cogn. Commun. Netw. 2018,4, 257–265. [CrossRef]

284.

O’Shea, T.J.; Erpek, T.; Clancy, T.C. Physical layer deep learning of encodings for the MIMO fading channel. In Proceedings of the

2017 55th Annual Allerton Conference on Communication, Control, and Computing (Allerton), Monticello, IL, USA, 3–6 October

2017; pp. 76–80.

285. O’Shea, T.J.; Erpek, T.; Clancy, T.C. Deep learning based MIMO communications. arXiv 2017, arXiv:1707.07980.

286.

O’shea, T.; Hoydis, J. An introduction to deep learning for the physical layer. IEEE Trans. Cogn. Commun. Netw.

2017

,3, 563–575.

[CrossRef]

287.

Arnold, M.; Dörner, S.; Cammerer, S.; Yan, S.; Hoydis, J.; Brink, S.t. Enabling FDD massive MIMO through deep learning-based

channel prediction. arXiv 2019, arXiv:1901.03664.

288.

Booth, J. Proof-of-Concept Prototype of Deep Learning Based Channel Mapping Using an Autonomous Channel Measurement

System. Ph.D. Thesis, Arizona State University, Tempe, AZ, USA, 2020.

289.

Bai, L.; Wang, C.X.; Huang, J.; Xu, Q.; Yang, Y.; Goussetis, G.; Sun, J.; Zhang, W. Predicting wireless mmWave massive MIMO

channel characteristics using machine learning algorithms. Wirel. Commun. Mob. Comput. 2018,2018, 9783863. [CrossRef]

290.

Fujihashi, T.; Koike-Akino, T.; Watanabe, T.; Orlik, P.V. Nonlinear equalization with deep learning for multi-purpose visual

MIMO communications. In Proceedings of the 2018 IEEE International Conference on Communications (ICC), Montreal, QC,

Canada, 20–24 May 2018; pp. 1–6.

291.

Cui, Y.; Li, S.; Zhang, W. Jointly sparse signal recovery and support recovery via deep learning with applications in MIMO-based

grant-free random access. IEEE J. Sel. Areas Commun. 2020,39, 788–803. [CrossRef]

Sensors 2022,22, 309 41 of 41

292.

Wang, Z.; Liu, L.; Zhang, H.; Xiao, G. Fault-tolerant controller design for a class of nonlinear MIMO discrete-time systems via

online reinforcement learning algorithm. IEEE Trans. Syst. Man Cybern. Syst. 2015,46, 611–622. [CrossRef]

293.

Yang, Q.; Yang, Z.; Sun, Y. Universal neural network control of MIMO uncertain nonlinear systems. IEEE Trans. Neural Netw.

Learn. Syst. 2012,23, 1163–1169. [CrossRef]

294.

Chen, M.; Ge, S.S.; How, B.V.E. Robust adaptive neural network control for a class of uncertain MIMO nonlinear systems with

input nonlinearities. IEEE Trans. Neural Netw. 2010,21, 796–812. [CrossRef]

295. Yuan, R.; Tan, X.; Fan, G.; Yi, J. Robust adaptive neural network control for a class of uncertain nonlinear systems with actuator

amplitude and rate saturations. Neurocomputing 2014,125, 72–80. [CrossRef]

296.

Chen, M.; Shao, S.Y.; Jiang, B. Adaptive neural control of uncertain nonlinear systems using disturbance observer. IEEE Trans.

Cybern. 2017,47, 3110–3123. [CrossRef] [PubMed]

297.

Kostarigka, A.K.; Rovithakis, G.A. Adaptive dynamic output feedback neural network control of uncertain MIMO nonlinear

systems with prescribed performance. IEEE Trans. Neural Netw. Learn. Syst. 2011,23, 138–149. [CrossRef] [PubMed]

298. Wang, C.; Liang, M.; Chai, Y. Adaptive neural network control of a class of fractional order uncertain nonlinear MIMO systems

with input constraints. Complexity 2019,2019, 1410278. [CrossRef]

299.

Chen, Z.; Li, Z.; Chen, C.P. Adaptive neural control of uncertain MIMO nonlinear systems with state and input constraints. IEEE

Trans. Neural Netw. Learn. Syst. 2016,28, 1318–1330. [CrossRef] [PubMed]

300.

Zhou, S.; Chen, M.; Ong, C.J.; Chen, P.C. Adaptive neural network control of uncertain MIMO nonlinear systems with input

saturation. Neural Comput. Appl. 2016,27, 1317–1325. [CrossRef]

301.

Wang, C.; Lin, Y. Robust adaptive neural control for a class of uncertain MIMO nonlinear systems. Int. J. Syst. Sci.

2015

46, 1934–1943. [CrossRef]

302.

Wang, Z.; Zhou, W.; Chen, L.; Zhou, F.; Zhu, F.; Fan, L. An adaptive deep learning-based UAV receiver design for coded MIMO

with correlated noise. Phys. Commun. 2021,47, 101365. [CrossRef]

303.

Ye, H.; Li, G.Y.; Juang, B.H. Power of deep learning for channel estimation and signal detection in OFDM systems. IEEE Wirel.

Commun. Lett. 2017,7, 114–117. [CrossRef]

304. Balevi, E.; Andrews, J.G. Deep learning-based channel estimation for high-dimensional signals. arXiv 2019, arXiv:1904.09346.

305.

Safari, M.S.; Pourahmadi, V.; Sodagari, S. Deep UL2DL: Channel knowledge transfer from uplink to downlink. arXiv

2018

arXiv:1812.07518.

306.

Zhang, Q.; Zhang, M.; Chen, T.; Sun, Z.; Ma, Y.; Yu, B. Recent advances in convolutional neural network acceleration.

Neurocomputing 2019,323, 37–51. [CrossRef]

307. Hinton, G.; Vinyals, O.; Dean, J. Distilling the knowledge in a neural network. arXiv 2015, arXiv:1503.02531.

308.

Ding, C.; Liao, S.; Wang, Y.; Li, Z.; Liu, N.; Zhuo, Y.; Wang, C.; Qian, X.; Bai, Y.; Yuan, G.; et al. Circnn: Accelerating and

compressing deep neural networks using block-circulant weight matrices. In Proceedings of the 50th Annual IEEE/ACM

International Symposium on Microarchitecture, Cambridge, MA, USA, 14–18 October 2017; pp. 395–408.

309.

He, Y.; Zhang, X.; Sun, J. Channel pruning for accelerating very deep neural networks. In Proceedings of the IEEE International

Conference on Computer Vision, Venice, Italy, 22–29 October 2017; pp. 1389–1397.

310.

Jaderberg, M.; Vedaldi, A.; Zisserman, A. Speeding up convolutional neural networks with low rank expansions. arXiv

2014

arXiv:1405.3866.

311.

Zhang, C.; Li, P.; Sun, G.; Guan, Y.; Xiao, B.; Cong, J. Optimizing fpga-based accelerator design for deep convolutional neural

networks. In Proceedings of the 2015 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays, Monterey,

CA, USA, 22–24 February 2015; pp. 161–170.

312.

Park, H.; Kim, D.; Ahn, J.; Yoo, S. Zero and data reuse-aware fast convolution for deep neural networks on GPU. In Proceedings

of the Eleventh IEEE/ACM/IFIP International Conference on Hardware/Software Codesign and System Synthesis, Pittsburgh,

PA, USA, 1–7 October 2016; pp. 1–10.

Towards Artificial Intelligence-Aided MIMO Detection for 6G Communication Systems: A Review of Current Trends, Challenges and Future Directions

Article

Full-text available

Nov 2023

Enhancing MIMO-OFDM channel estimation in 5G and beyond with conditional self-attention generative adversarial networks

Article

Full-text available

Jan 2024
WIREL NETW

Wireless networks need channel estimation (CE) to function well. Deep learning (DL) has improved 5G and future-generation network communication reliability and computational complexity. Due to its lack of statistical channel information, the least squares (LS) estimation approach has significant estimate mistakes despite its cost-effectiveness and widespread usage. The MIMO-OFDM-5G-CS-AGAN architecture improves CE for 5G and beyond networks. The suggested CE architecture uses DL to outperform LS. The suggested architecture has been specifically developed using a multipath channel profile and a MIMO system, to effectively address situations characterized by mobility-induced Doppler effects in 5G and future networks. Because of its adaptability and scalability, the suggested architecture can accommodate almost unlimited transceiver antennas. Conditional Self-Attention Generative Adversarial Networks (CS-AGAN) are at the heart of the proposed method because they reliably provide high-quality CE at low error rates. Python is used for the implementation of the suggested approach. In this case, we have the performance metrics symbol error rate (SER), bit error rate (BER), uplink sum rate, downlink sum rate, Normalized CE Error, and mean square error. When compared to current methods like MIMO-OFDM-5G-Bi-LSTM, MIMO-OFDM-5G-DNN, and MIMO-OFDM-5G-CGAN, the suggested approach achieves 13.4%, 13.02%, 12.21% lower BER and 11.23%, 6.3%, 9.08% lower SER.

Unlocking the power of recalling enhanced recurrent neural network: channel estimation and data analysis for 5G MIMO communication system

Article

Full-text available

Dec 2023
OPT QUANT ELECTRON

Multi-user detection techniques are crucial in minimizing the impact of Multi-Access Interference in Multiple-Input Multiple-Output Orthogonal Frequency-Division Multiplexing (MIMO-OFDM) systems. It is suggested to combine channel estimation with multi-user identification to get rid of different interferences and lower the Bit Error Rate. In this manuscript, Channel Estimation and Data Detection for 5G MIMO Communication in Fading Channels based on Recalling Enhanced Recurrent Neural Network are proposed. The suggested method prioritizes fast communication and strives to provide high reliability even when there is just a small amount of user equipment per base station. The study of the multipath Rayleigh fading channel in MIMO-OFDM systems has a difficulty with frequency selective channel estimate. The primary air interface for 4G and 5G broadband wireless communications is MIMO-OFDM. With its emphasis on the network's active users and its ability to reduce interference, MIMO-OFDM offers further advantages in multi-user identification. When the activity factor is unknown, the deep learning technique is used to identify the active users in the multi-access system. With a high data rate, the RERNN classifier is utilized to evaluate the unknown activity component. The BS receiver scans for active users and deciphers their transmissions. Both pilot and data symbols are used in activity detection. A unique pilot allocation strategy that reduces the common relation of the measurement matrix is proposed for optimum pilot placement. The sparse channel assessment is eventually verified, and the computational cost is decreased, by selecting the best pilot patterns with the aid of improved Mexican axolotl optimization algorithm. The CE and decoder techniques both regulate the impact of sounds and MAI. It accurately decodes the received signal without channel issues using Python implementation. Here the performance metrics like of Symbol Error Rate, Bit Error Rate, Uplink Sum Rate, Downlink Sum Rate, Normalized CE Error and Mean Square Error. The suggested approach demonstrates a reduction in BER by 23.34%, 17.02%, and 32.21%, as well as a reduction in SER by 21.09%, 12.32%, and 19.08%, when compared to conventional methodologies.

Deep learning-based receiver design for generalized frequency division multiplexing (GFDM)

Article

May 2024

Game Theory and Deep Learning Approach to Implement Scrabble

Conference Paper

Dec 2023

Illuminating Healthcare Management: A Comprehensive Review of IoT-Enabled Chronic Disease Monitoring

Article

Full-text available

Jan 2024

Present dynamic performance reputation and technical innovations in Internet of Things (IoT) technologies have endowed ultra-inexpensive, energy effcient, smart, and tiny IoT gadgets. IoT gadgets can be easily implanted inside, attached to, or placed around the chronic patient body, and they can be employed in several healthcare monitoring systems such as mobile, wearable, and implantable healthcare monitoring for chronic diseases. Healthcare monitoring for chronic diseases is one of the major applications of IoT and is also a typical challenging area. The rapidly rising proportion of patients with chronic diseases brought enormous pressure on governments and healthcare providers and required up-to-date long-term healthcare service and continuous monitoring. In this survey paper, we extensively review numerous industrial and non-commercial contemporary healthcare monitoring systems (HMS) and applications. We present a design layout of IoT-based HMS for chronic diseases. We presented societal and technological challenges associated with the design of IoT-based HMS and their solutions. To accomplish this, more than 80 different healthcare monitoring systems have been characterized and classied. Moreover, we describe contemporary healthcare monitoring networks and communication technologies. This review also presents the dynamic capabilities of key IoT technologies for chronic diseases healthcare monitoring to show dedicated research pathways to IoT researchers. Last, we deeply analyze different healthcare monitoring systems and describe open issues that will help researchers and healthcare system designers design future systems.

Optimization of Precoders for MIMO System for 6G Communication

Conference Paper

Nov 2023

Performance Evaluation of SISO-OFDM Channel Equalization Utilizing Deep Learning

Conference Paper

Jan 2024

Using deep learning (DL)--based channel equalization for orthogonal frequency division multiplexing (OFDM) communication systems involves leveraging neural networks to mitigate the effects of channel impairments and enhance the performance of the OFDM system. Time-varying and frequency-selective channels are difficult to equalize using conventional methods of zero-forcing (ZF) and minimum mean square error (MMSE) equalizers. The primary goal of this paper is the performance evaluation of DL-based channel equalizer to address the shortcomings of the ZF and MMSE equalizers in terms of modulation order, number of pilots, and cycle prefix (CP). Simulated results show that the DL-based channel equalizer can outperform the MMSE equalizer when increasing the number of pilots, with or without CP, for both low and high-frequency selective channel models, respectively. The conventional ZF and MMSE channel equalization techniques exhibit sensitivity to the absence of the cycle prefix. These schemes demonstrate satisfactory performance solely under the conditions of employing a complete set of pilot symbols (64 pilots) and operating in low-frequency selective channels. The performance of the DL-based channel equalizer surpasses that of the ZF and MMSE channel equalization methods, exhibiting significant robustness against CP removal, channel frequency selectivity, number of pilots, and various modulation orders such as QPSK, 16 QAM, and 64 QAM. This work contributed scientifically to evaluate the performance of a DL-based channel equalizer for the SISI-OFDM system using various parameters and compares

Deep Learning for Channel Estimation and Security Detection in Wireless Communication Systems

Conference Paper

Nov 2023

A quantitative analysis of MIMO channel performance in new generation networks: Prospects and limitations

Article

Sep 2023

The objective of this article is to analyze the behavior of channels in mobile radio transmission chains and propose a solution for reducing propagation attenuation and ensuring Quality of Service (QoS) in telecommunications. As network coverage and throughput demands continue to grow, a thorough characterization of channels is essential. The primary contribution of this work is to investigate the different phases of the digital transmission chain jointly to achieve optimal reliability and QoS. Spatial and temporal representations must be considered at the same level when characterizing and modeling MIMO system channels. Accurate and realistic modeling can be achieved through transmission channel characterization in the spatial domain, which is critical in a MIMO context to improve throughput and performance compared to conventional systems. The article explores the behavior of information carriers in the propagation environment and the impact of fading, diversity, multiplexing, and multipath phenomena.

Deep learning for joint channel estimation and feedback in massive MIMO systems

Article

Full-text available

Jan 2023

Performance Enhancement of Massive MIMO Using Deep Learning-Based Channel Estimation

Article

Full-text available

Feb 2021

Massive Multiple-Input Multiple-Output (massive MIMO) system relies on channel state information (CSI) feedback to perform precoding and achieve performance gain in frequency division duplex (FDD) networks. However, transmission of massive MIMO system is subject to excessive feedback overhead. In this paper, we propose a Deep Learning (DL) approach-based channel estimation technique to enhance the performance of massive MIMO system. This technique is used to enhance recovery quality and improve trade-off between compression ratio (CR) and complexity of massive MIMO system. The proposed technique is based upon using the Channel State Information Network combined with gated recurrent unit (CsiNet-GRU). Moreover, the dropout method is used in the proposed technique to reduce overfitting during the learning process. The simulation results demonstrate that the proposed CsiNet-GRU technique results in a significant improvement in performance when compared with existing techniques used in conjunction with massive MIMO systems.

Deep Learning-based Implicit CSI Feedback in Massive MIMO

Article

Full-text available

Dec 2021

Massive multiple-input multiple-output can obtain more performance gain by exploiting the downlink channel state information (CSI) at the base station (BS). Therefore, studying CSI feedback with limited communication resources in frequency-division duplexing systems is of great importance. Recently, deep learning (DL)-based CSI feedback has shown considerable potential. However, the existing DL-based explicit feedback schemes are difficult to deploy because current fifth-generation mobile communication protocols and systems are designed based on an implicit feedback mechanism. In this paper, we propose a DL-based implicit feedback architecture to inherit the low-overhead characteristic, which uses neural networks (NNs) to replace the precoding matrix indicator (PMI) encoding and decoding modules. By using environment information, the NNs can achieve a more refined mapping between the precoding matrix and the PMI compared with codebooks. The correlation between subbands is also used to further improve the feedback performance. Simulation results show that, for a single resource block (RB), the proposed architecture can save 25.0% - 40.0% of overhead compared with the Type I codebook under different antenna configurations. For a wideband system with 52 RBs, overhead can be saved by 30.7% and 48.0% compared with the Type II codebook when ignoring and considering extracting subband correlation, respectively.

Deep Learning Based Frequency-Selective Channel Estimation for Hybrid mmWave MIMO Systems

Article

Full-text available

Nov 2021

Millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) systems typically employ hybrid mixed signal processing to avoid expensive hardware and high training overheads. However, the lack of fully digital beamforming at mmWave bands imposes additional challenges in channel estimation. Prior art on hybrid architectures has mainly focused on greedy optimization algorithms to estimate frequency-flat narrowband mmWave channels, despite the fact that in practice, the large bandwidth associated with mmWave channels results in frequency-selective channels. In this paper, we consider a frequency-selective wideband mmWave system and propose two deep learning (DL) compressive sensing (CS) based algorithms for channel estimation. The proposed algorithms learn critical apriori information from training data to provide highly accurate channel estimates with low training overhead. In the first approach, a DL-CS based algorithm simultaneously estimates the channel supports in the frequency domain, which are then used for channel reconstruction. The second approach exploits the estimated supports to apply a low-complexity multi-resolution fine-tuning method to further enhance the estimation performance. Simulation results demonstrate that the proposed DL-based schemes significantly outperform conventional orthogonal matching pursuit (OMP) techniques in terms of the normalized mean-squared error (NMSE), computational complexity, and spectral efficiency, particularly in the low signal-to-noise ratio regime. When compared to OMP approaches that achieve an NMSE gap of {4-10} dB with respect to the Cramer Rao Lower Bound (CRLB), the proposed algorithms reduce the CRLB gap to only {1-1.5} dB, while reducing complexity by two orders of magnitude.

Learning Tasks in Intelligent Environments via Inverse Reinforcement Learning

Conference Paper

Full-text available

Jun 2021

Multisegment Mapping Network for Massive MIMO Detection

Article

Full-text available

Sep 2021
Int J Antenn Propag

The massive multiple-input multiple-output (MIMO) technology is one of the core technologies of 5G, which can significantly improve spectral efficiency. Because of the large number of massive MIMO antennas, the computational complexity of detection has increased significantly, which poses a significant challenge to traditional detection algorithms. However, the use of deep learning for massive MIMO detection can achieve a high degree of computational parallelism, and deep learning constitutes an important technical approach for solving the signal detection problem. This paper proposes a deep neural network for massive MIMO detection, named Multisegment Mapping Network (MsNet). MsNet is obtained by optimizing the prior detection networks that are termed as DetNet and ScNet. MsNet further simplifies the sparse connection structure and reduces network complexity, which also changes the coefficients of the residual structure in the network into trainable variables. In addition, this paper designs an activation function to improve the performance of massive MIMO detection in high-order modulation scenarios. The simulation results show that MsNet has better symbol error rate (SER) performance and both computational complexity and the number of training parameters are significantly reduced.

Channel Estimation and Hybrid Precoding for Millimeter Wave Communications: A Deep Learning-Based Approach

Article

Full-text available

Aug 2021

Hybrid analog and digital beamforming (HBF) has been regarded as a key technology for future millimeter wave (mmWave) communication systems due to its ability to obtain a good trade-off between achievable beamforming gain and hardware cost. In this paper, we investigate the channel estimation and hybrid precoding for mmWave MIMO systems with deep learning. We adopt the hierarchical codebook based algorithm for channel estimation as it requires limited number of pilot transmissions, and enhance its performance by proposing a new codebook design algorithm based on manifold optimization (MO). With the estimated channel state information (CSI) as the input, we develop a robust HBF network (HBF-Net) by applying convolutional layers and attention mechanism, which can be trained to generate a robust HBF matrix targeting at spectral efficiency maximization with imperfect CSI. To further improve the performance, we propose a joint channel estimation and HBF optimization network (CE-HBF-Net). Considering that the adaptively selected HBF vectors in the hierarchical codebook based channel estimation are different for different channel realizations, we skillfully propose an index assign-and-input method to efficiently feed such information to the CE-HBF-Net to reduce the network input dimensions and make the network trainable. Furthermore, we propose a signal self-attention mechanism to enable the CE-HBF-Net to intelligently assign larger weight coefficients to those signals that contribute more to channel estimation. Simulation results show that the well-designed HBF-Net and CE-HBF-Net outperform the conventional HBF algorithms with imperfect channel and exhibit robustness to mismatches between offline training and online deployment stages.

Binarized Aggregated Network With Quantization: Flexible Deep Learning Deployment for CSI Feedback in Massive MIMO Systems

Article

Jan 2022

Massive multiple-input multiple-output (MIMO) is one of the key techniques to achieve better spectrum and energy efficiency in 5G system. The channel state information (CSI) needs to be fed back from the user equipment to the base station in frequency division duplexing (FDD) mode. However, the overhead of the direct feedback is unacceptable due to the large antenna array in massive MIMO system. Recently, deep learning is widely adopted to the compressed CSI feedback task and proved to be effective. In this paper, a novel network named aggregated channel reconstruction network (ACRNet) is designed to boost the feedback performance with network aggregation and parametric rectified linear unit (PReLU) activation. The practical deployment of the feedback network in the communication system is also considered. Specifically, the elastic feedback scheme is proposed to flexibly adapt the network to meet different resource limitations. Besides, the network binarization technique is combined with the feature quantization for lightweight and practical deployment. Experiments show that the proposed ACRNet outperforms loads of previous state-of-the-art networks, providing a neat feedback solution with high performance, low cost and impressive flexibility.

CAnet: Uplink-aided Downlink Channel Acquisition in FDD Massive MIMO Using Deep Learning

Article

Oct 2021

In frequency-division duplexing systems, the downlink channel state information (CSI) acquisition scheme leads to high training and feedback overhead. In this work, we propose an uplink-aided downlink channel acquisition framework using deep learning to reduce such overhead. We consider the entire downlink CSI acquisition process, including the downlink pilot design, channel estimation, and feedback. First, we propose an adaptive pilot design module by exploiting the correlation in magnitude among bidirectional channels in the angular domain to improve channel estimation. Second, to avoid the bit allocation problem during the feedback module, we concatenate the complex channel and embed the uplink channel magnitude to the channel reconstruction at the base station. Finally, we combine the two modules and compare two popular uplink-aided downlink channel acquisition frameworks. One framework estimates and subsequently feeds back the channel at the user equipment. In the other framework, the user equipment directly feeds back the received pilot signals to the base station. Results reveal that with the help of the uplink channel, directly feeding back pilot signals can save approximately 20% of feedback bits. This work thus provides a guideline for future research.

Fully Convolutional Neural Network Based CSI Limited Feedback for FDD Massive MIMO Systems

Article

Oct 2021

Due to the lack of channel reciprocity in frequency division duplexity (FDD) massive multiple-input multiple-output (MIMO) systems, it is impossible to infer the downlink channel state information (CSI) directly from its reciprocal uplink CSI. Hence, the estimated downlink CSI needs to be continuously fed back to the base station (BS) from the user equipment (UE), consuming valuable bandwidth resources. This is exacerbated, in massive MIMO, with the increase of the antennas at the BS. This paper propose a fully convolutional neural network (FullyConv) to compress and decompress the downlink CSI. FullyConv will improve the reconstruction accuracy of downlink CSI and reduce the training parameters and computational resources. Besides, we add a quantization module in the encoder and a dequantization module in the decoder of the FullyConv to simulate a real feedback scenario. Experimental results demonstrate that the proposed FullyConv is better than the baseline on reconstruction performance and reduction of the storage and computational overhead. Furthermore, the FullyConv added quantization and dequantization modules is robust to quantization error in real feedback scenarios.