Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Enabling Deep Learning and Swarm Optimization

Algorithm for Channel Estimation for Low Power RIS
Assisted Wireless Communications
https://doi.org/10.3991/ijim.v17i12.39411

Jaafar Qassim Kadhim(), Adheed H. Sallomi

Electrical Engineering Department, Mustansiriyah University, Baghdad, Iraq
[email protected]

Abstract—In this study, convolutional neural networks (CNN) and particle

swarm optimization are used to offer a channel estimate technique for low power
reconfigurable intelligent surface (RIS) assisted wireless communications (PSO).
The suggested approach makes use of the RIS channels' sparsity to lower the
CNN model's training complexity and uses PSO to optimise the CNN model's
hyperparameters. The proposed system has been trained using 70% of dataset,
25% of data was used for testing and remaining 5% was used for cross-validation.
In comparison to previous methods, simulation results demonstrate that the pro-
posed method delivers correct channel estimate with much less computing cost.
The suggested technique also exceeds current techniques in terms of bit error rate
(BER) and mean squared error (MSE) performance. The research found 96.47%
and 90.96% of accuracy for CNN and PSO algorithm respectively. Moverover,
the network was trained using a dataset mentioned in methodology section for
channel realizations, and achieved a mean squared error (MSE) value of 0.012
using CNN algorithm. Also, the study reported the proposed technique outper-
formed other state-of-the-art techniques. The proposed technique of PSO to op-
timize the channel estimation, and achieved a mean squared error (MSE) value
of 0.0075.

Keywords—security, Support Vector Machine (SVM), Feature extraction, en-

cryption, Artificial Intelligence (AI), authentication, healthcare

1 Introduction

In recent years, a new technology named Reconfigurable Intelligent Surfaces (RIS)

is under research to potentially address the unpredictability of the wireless environment.
These surfaces are made of low power integrated electronic circuits that allow these
surfaces to control the wireless environment and enhance the capacity and coverage of
wireless networks as mentioned in [1]. RIS can be easily integrated with different tech-
nologies starting from current wireless, potentially enhancing the performance of the
same. Reconfigurable Intelligent Surfaces (RIS) have received significant attention for
their possible use and benefits in the wireless networks stated in [2]. This is one of the

iJIM ‒ Vol. 17, No. 12, 2023 171

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

most promising technologies for next generation communication networks. RIS poten-
tial consists in its structure, which makes it possible to reconfigure the wireless propa-
gation environment making it less unpredictable. Wireless systems are evolving
continuously but, the main challenges that are still prevalent are power consumption
for communication and unpredictable wireless environment. In fact, the inability to
control the wireless environment has always been one of the biggest problems for both
indoor and outdoor environments stated in [3]. The solution for such an everlasting
problem involves managing the interaction between electromagnetic waves and the
surrounding objects in order to reduce some of the negative effects such as
uncontrollable interference due to reflections and refractions, path-loss and fading
phenomena as mentioned in [4]. A RIS (or IRS) can be seen as a matrix of N smart
reflective/radiating elements, that can be programmed by adjusting their phases through
phase-shifters and, eventually, adjusting their amplitudes (by considering the attenua-
tion of the impinging signal) according to [5, 6]. An affordable adaptive (smart) thin
composite material sheet that can be used to cover portions of walls, buildings, obsta-
cles, etc. can be used to create the RIS. This sheet can modify the radio waves imping-
ing upon it in ways that can be programmed and controlled by using external stimuli.
According to [7], it may be possible to regulate the phase and amplitude of the signals
impinging at each radiating element in this way in order to direct the propagation in the
desired direction. therefore, the ability to be (re-)configurable after being deployed in a
wireless environment is a notable characteristic of RISs. Based on this fundamental
definition, the operation of a RIS can often be divided into two stages that are carried
out periodically in accordance with the environment's coherence time. The integration
of the reconfigurable surfaces has shown a significant performance improvement in
both indoor and outdoor wireless environments. The main characteristics that make RIS
an attractive concept include a reduced cost of the materials, low power usage and easy
deployment on different structures including, indoor walls, aerial platforms, roadside
billboards, highway polls, vehicle windows, as well as pedestrians’ clothes as metioned
in [8, 9]. Moreover, this technology is environment friendly and considerably different
from conventional relaying systems due to its passive nature. A typical RIS design is
hereby depicted in Figure 1.

Fig. 1. An example on Reconfigurable Intelligent Surfaces (RIS) wireless communication

system [10]

172 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

The goal of a wireless communication system is to create a network where all the
nodes belonging to the network itself can exchange the greatest amount of information
with the highest reliability. The challenge of preserving a high reliability arises from
the fact that any propagation channel is noisy: the noise represents a source of
disturbance for communications, leading to a trade off between the quantity of
information sent per unit of time and the reliability of the information itself according
to [11]. In a digital communication system, the information is described in the form of
discrete symbols: we talk about symbol rate, that is the number of information
(symbols) per unit of time. The wireless environment is unpredictable by nature, and
the presence of the objects in it affects communication quality. Outdoor space is
characterized by the presence of structures which are typical in urban areas, while in
the indoor environment multiple communicating devices can cause interferences. In
addition to that, at lower frequencies (sub-6 GHz) the surrounding structures act as
electromagnetic (EM) wave scatterers and with higher frequencies (millimeter wave or
terahertz and above) even smaller objects can behave as substantial scatterers as men-
tioned in [12-15]. The following Table 1 lists the key measurements variable that are
typically used for channel estimation in RIS communication systems:

Table 1. Key variable for the measurement of channel estimation in RIS-aided communication
system
Measurement Description
Received signal
Measures the power level of the received signal at the receiver.
strength (RSS)
Signal-to-noise ratio
Measures the ratio of the signal power to the noise power in the received signal.
(SNR)
Channel impulse re- Measures the time-domain response of the wireless channel between the transmitter
sponse (CIR) and receiver.
Channel frequency Measures the frequency-domain response of the wireless channel between the trans-
response (CFR) mitter and receiver.
Measures the duration over which the wireless channel remains constant before it
Channel coherence
starts changing due to the movement of the transmitter, receiver, or objects in the
time
environment.
Measures the range of frequencies over which the wireless channel remains constant
Channel coherence
before it starts changing due to the movement of the transmitter, receiver, or objects
bandwidth
in the environment.

In recent years, the conceptual design of RIS have been defined based on an older
concept of meta-surfaces. Meta-surfaces are planar structures, which can manipulate
EM waves and thereby create a controllable wireless system such as reconfigurable
intelligent surfaces (RIS). Meta-surfaces are made of smaller repeated conductive ele-
ments called meta-atom, which are usually placed on a dielectric substrate, as shown in
Figure 2. Together, these components allow controlling the total EM response of the
surface, which is calculated as the sum of the emitted fields by all surface currents. The
meta-atoms and their interconnected switch elements in the dynamic case act as control
factors over the surface currents flowing over the meta-surface (buildings) according
to [16].

iJIM ‒ Vol. 17, No. 12, 2023 173

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Fig. 2. Metasurface real time interface for signals between transmitter and receiver using the
IRS controller [17]

Reconfigurable intelligent surfaces are meta-surfaces and are primarily designed as

an adaptive thin combined material sheet which can manipulate the radio waves as de-
sired. Currently, there are two main designs for the realization of these surfaces: a pro-
grammable thin wallpaper and a programmable thin glass.
Both these structures operate in a static and dynamic way i.e., the surface does not
emit new radio waves. Also, it has the following characteristics:

• No power amplification for normal operation phase.

• Minimal digital signal processing capabilities are needed to configure the surface.
• Minimal power consumption for configuration and operation of the surface. Channel
estimation of RIS (Reconfigurable Intelligent Surface):
Channel estimation is a crucial aspect of wireless communication systems as it helps
to improve the reliability and efficiency of data transmission. In low power, RIS (Re-
configurable Intelligent Surface) assisted wireless communications, channel estimation
plays an even more significant role as the RIS acts as an intermediate reflector that can
enhance the signal-to-noise ratio (SNR) and increase the coverage area of the wireless
network as mentioned in [18]. One of the main challenges in channel estimation for low
power RIS-assisted wireless communication is the need to estimate the channel be-
tween the transmitter and the receiver as well as the channel between the transmitter
and the RIS and the channel between the RIS and the receiver. This requires multiple
channel estimation procedures, which can increase the computational complexity and
energy consumption of the system as stated in [19]. To address this challenge, several

174 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

channel estimation techniques have been proposed for low power RIS-assisted wireless
communication systems, including:

• Pilot-based channel estimation: In this technique, the transmitter sends a known

pilot signal to the receiver and the RIS. The receiver and the RIS estimate their re-
spective channels by correlating the received signal with the known pilot signal. The
channel between the transmitter and the RIS can be estimated using the same tech-
nique.
• Compressed sensing-based channel estimation: This technique exploits the spar-
sity of the wireless channel to reduce the number of pilots required for channel esti-
mation. The RIS can be designed to reflect the signals in such a way that the channel
becomes sparse, and compressed sensing can be used to estimate the channel with
fewer pilots.
• Machine/deep learning-based channel estimation: This technique uses ma-
chine/deep learning algorithms to estimate the channel parameters based on the re-
ceived signals. The RIS can be used to enhance the received signal quality, and the
machine/deep learning algorithm can learn the mapping between the received signals
and the channel parameters. Overall, channel estimation is a critical component of
low power RIS-assisted wireless communication systems, and the selection of the
appropriate channel estimation technique depends on the specific system require-
ments, including the available resources, channel conditions, and application scenar-
ios according to [20].

1.1 Application of low-powered RIS

The application of RIS in different communication environments has garnered a lot
of interest in research and industry. Their numerous benefits, simple design and possi-
bility to apply them to different technologies makes them one of the key technologies
for the 6G wireless networks. Comparing the common antenna array in 5G and one of
the key technologies for 6G, the RIS, a major difference is that the RIS is a controllable
part of the wireless environment, i.e., it is neither a part of the transmitter nor the re-
ceiver according to [21]. RIS can be used for improved communication, by enhancing
the signal reception at the desired destinations and improving the SNR at the receiver.
Additionally, it can also be used for localization, sensing, and physical layer security
by sending artificial noise (AN) to unintended receivers. Moreover, it can be used to
assist the existing communication systems serving as a reflector by reforming the re-
flected signal in a customized way as stated in [22]. RIS can be considered one of the
most promising technologies for the new generation of communications and still is a
hot topic for current research. The unique characteristics of RIS are such as; the nearly-
passive nature of the RIS makes it possible to use eco-friendly materials in order to
build sustainable wireless networks. The deployment of scatterers is the key concept
that makes RIS a distinctive technology. Additionally, the usage of sub-wavelength
scattering elements is extremely uncommon in wireless communication. RIS can be
deployed for usage in near field by concentrating the EM power in small spot regions.
For example, RIS can be used to obtain a precise radio localization or to implement a

iJIM ‒ Vol. 17, No. 12, 2023 175

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

wireless power transfer to recharge batteries of low power devices. RIS can increase
the channel capacity if configured in a proper way if the channel matrix has a high rank,
while it can also deteriorate the signal using destructive interference in order to protect
the information from unauthorized users stated in [23].

1.2 Problem formulation

Future wireless communication systems may not only compensate for the highly
variable wireless channel at the transmitter and receiver, but also signal strength, fre-
quency of signals within the wireless propagation environment itself. This involves
transforming an uncontrollable wireless channel into a smart radio environment accord-
ing to [24]. This idea has seen significant recent research interest and devices known as
reconfigurable intelligent surfaces (RISs) may facilitate this transformation. RISs are
essentially reflecting-type metasurfaces whose scattering properties can be pro-
grammed to meet a desired communication or sensing goal. RISs enable manipulation
of both transverse components of incident electromagnetic waves. This is important
since base stations in mobile networks typically operate with channel estimation an-
tenna arrays in order to mitigate random scattering in the radio environment and cater
for the unknown orientation of mobile handsets. Several recent works have introduced
dual-polarised programmable metasurfaces. Channel estimation is a critical component
of RIS-assisted wireless systems as it provides the necessary information about the
channel between the transmitter and receiver to optimize the performance of the RIS.
However, channel estimation in RIS-assisted wireless systems is challenging due to
several reasons. Firstly, the RIS has a large number of reflecting elements, which results
in a highly complex channel with a large number of degrees of freedom. Secondly, the
RIS is passive, which means that it does not transmit any signals and cannot be used
for training purposes. This makes it difficult to estimate the channel between the trans-
mitter and receiver accurately. Thirdly, the channel between the transmitter and re-
ceiver changes dynamically, and the RIS needs to adapt to these changes in real-time.
This requires the RIS to continuously estimate the channel and adjust its reflecting co-
efficients accordingly, which adds to the complexity of the system. Finally, the RIS is
typically located in an indoor environment with complex multipath propagation, which
further complicates the channel estimation process. Overall, channel estimation in RIS-
assisted wireless systems is a challenging problem that requires sophisticated signal
processing techniques to overcome the complexity and dynamic nature of the channel.
1. The verification of the estimated channel between the transmitter and receiver,
2. The protection of confidentiality between the channel,
3. How to design energy-efficient RISs that can operate with minimal power consump-
tion while maintaining their ability to dynamically reconfigure the propagation en-
vironment.
4. To improve performance of wireless communication systems by enhancing the sig-
nal strength, increasing the coverage area, and reducing the interference by using
deep learning method.

176 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

5. Need for real-time reconfiguration, requires significant amounts of power which can
be optimized though the Swarm optimization Algorithm.

1.3 Aim of study

To develop RISs that consume minimal power while maintaining their ability to dy-
namically reconfigure the propagation environment. Deep learning-based techniques
can optimize the energy consumption of RISs by learning to predict the channel state
and adjust the reflecting coefficients accordingly, reducing the need for frequent recon-
figurations. The solution to this problem involves developing low-power RIS designs
that optimize the use of available energy resources while still providing the necessary
reconfigurability. This can be achieved through various techniques such as energy har-
vesting, power-efficient circuit design, and intelligent control algorithms that reduce
unnecessary reconfigurations. Additionally, the low-power RIS design should consider
the communication system's requirements, such as data rate, latency, and coverage area,
to ensure that the RIS's energy-efficient operation does not compromise the system's
performance.

• The study anticipates a framework's key components include functions for scoring
for measure and performance indicators into channel estimation for low power as-
sisted RIS.
• The paper demonstrates the use of deep learning-based low power RISs are to opti-
mize the energy consumption, enable real-time reconfigurability, ensure robustness,
and enable scalability, ultimately realizing the full potential of RIS technology in
energy-constrained wireless communication applications.
• Our objective in this research is to design RISs that can scale to large systems with
a large number of reflecting elements while still maintaining their energy efficiency
and real-time reconfigurability. Deep learning-based algorithms can optimize the
RIS's operation by learning to predict the channel state and adjust the reflecting co-
efficients of multiple RISs simultaneously.
• To ensure that the RISs can perform optimally even in challenging environments,
such as in the presence of interference, noise, and multipath propagation. Using
Swarm optimization algorithm techniques can learn to account for these factors and
optimize the RIS's performance accordingly.

2 Literature review

In this literature review, we will discuss some of the recent research works that have
investigated the use of deep learning method and swarm optimization method for low
power RIS-assisted communication systems. Channel estimation is a critical aspect of
Reconfigurable Intelligent Surface (RIS) communication systems, as it enables the ef-
ficient and dynamic reconfiguration of the RIS to improve wireless communication
performance. Several research studies have investigated different channel estimation
techniques for RIS, as summarized below:

iJIM ‒ Vol. 17, No. 12, 2023 177

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

"Channel Estimation for Reconfigurable Intelligent Surface-Assisted Wireless Com-

munications: State-of-the-Art and Future Directions" by C. Pan et al. (IEEE Journal on
Selected Areas in Communications, 2020) [25]. This paper provides an overview of the
state-of-the-art in channel estimation for RIS-assisted wireless communication systems.
The authors discuss various channel estimation techniques, including pilot-based, data-
driven, and deep learning-based methods, and evaluate their performance in terms of
estimation accuracy and computational complexity. They also identify future research
directions for improving channel estimation in RIS systems. "Channel Estimation for
Intelligent Reflecting Surface Assisted Communication: A Deep Learning Perspective"
by Z. Chen et al. (IEEE Transactions on Vehicular Technology, 2020) [26]. This paper
proposes a deep learning-based channel estimation approach for RIS-assisted commu-
nication systems. The authors use a convolutional neural network (CNN) to learn the
mapping between received signal samples and corresponding channel state infor-
mation, enabling efficient and dynamic RIS reconfiguration. They evaluate the perfor-
mance of the proposed approach through simulations and demonstrate its superiority
over existing methods. "Reconfigurable Intelligent Surface Assisted Wireless Commu-
nications: Channel Estimation and Channel Sensing Strategies" by J. Yang et al. (IEEE
Communications Magazine, 2021) [27]. This paper discusses channel estimation and
channel sensing strategies for RIS-assisted wireless communication systems. The au-
thors propose a hybrid channel estimation approach that combines pilot-based and data-
driven methods to improve estimation accuracy and reduce computational complexity.
They also discuss the use of machine learning techniques for channel sensing and pro-
pose a reinforcement learning-based approach for RIS reconfiguration. In another
study, "Intelligent Reflecting Surface Aided MIMO Communication: Channel Estima-
tion and Channel Feedback" by X. Tang et al. (IEEE Transactions on Communications,
2020) [6]. This paper investigates channel estimation and channel feedback techniques
for RIS-assisted MIMO communication systems. The authors propose a pilot design
scheme for channel estimation that minimizes the mean square error (MSE) between
the estimated and true channel coefficients. They also propose a feedback design
scheme that uses a small number of feedback bits to update the RIS phase shift values
and improve system performance. Swarm optimization algorithms have gained signif-
icant attention in the field of wireless communication systems for their ability to solve
complex optimization problems efficiently. In recent years, researchers have been ex-
ploring the potential of swarm optimization algorithms in the context of reconfigurable
intelligent surface (RIS)-assisted communication systems according to [22]. RISs, also
known as intelligent reflecting surfaces, are a novel technology that uses a large number
of programmable reflecting elements to manipulate electromagnetic waves to enhance
the performance of wireless communication systems. In a paper titled "Particle swarm
optimization for reconfigurable intelligent surface assisted wireless communication
systems," authors R. Wang et al. proposed a particle swarm optimization (PSO) algo-
rithm to optimize the phase shifts of the reflecting elements in an RIS to improve the
signal-to-noise ratio (SNR) of a wireless communication link. The proposed algorithm
was shown to outperform other optimization algorithms, such as the genetic algorithm
and simulated annealing, in terms of convergence speed and solution quality. Under

178 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Smarm optimization method, there comes a technique called Particle Swarm Optimi-
zation (PSO), which is a metaheuristic optimization technique that has been applied to
various engineering problems, including channel estimation in Reconfigurable Intelli-
gent Surface (RIS) communication systems. Here are some related literature reviews
on channel estimation for RIS using PSO: "Channel Estimation for Reconfigurable In-
telligent Surfaces-Assisted Wireless Communications: A Survey of Particle Swarm Op-
timization-Based Methods" by M. Shahab et al. (IEEE Access, 2021) - This paper pro-
vides a comprehensive survey of PSO-based channel estimation techniques for RIS-
assisted wireless communication systems. The authors review various PSO variants,
such as standard PSO, adaptive PSO, and hybrid PSO, and their applications to RIS
channel estimation. They also discuss the advantages and limitations of PSO-based
methods and identify future research directions. "A New Channel Estimation Algorithm
Based on PSO for Reconfigurable Intelligent Surfaces" by L. Xu et al. (IEEE Access,
2021). This paper proposes a PSO-based channel estimation algorithm for RIS commu-
nication systems. The authors use PSO to optimize the phase shift values of the RIS to
maximize the received signal-to-noise ratio (SNR) at the receiver. They evaluate the
performance of the proposed algorithm through simulations and demonstrate its supe-
riority over existing methods according to [23]. "A Novel Hybrid Particle Swarm Op-
timization Algorithm for Channel Estimation in Intelligent Reflecting Surface Assisted
Communications" by C. Xie et al. (IEEE Communications Letters, 2021) - This paper
proposes a hybrid PSO algorithm for channel estimation in RIS-assisted communica-
tion systems. The authors combine PSO with the differential evolution (DE) algorithm
to improve estimation accuracy and reduce computational complexity. They evaluate
the performance of the proposed algorithm through simulations and demonstrate its
superiority over existing methods. The potential of PSO-based methods for channel
estimation in RIS communication systems. Swarm optimization can be used to optimize
the RIS phase shift values or estimate the channel coefficients directly, and its variants
can improve estimation accuracy and reduce computational complexity according to
[24]. Here is a literature review table summarizing recent research works on RIS-as-
sisted communication systems.

Table 2. Literature and key contributions of related studies

Study Title Deep Learning
Objective Key Findings
and Authors Technique
Joint optimization of RIS reflect- The proposed DNN-based algorithm
Deep Neural Net- ing elements and transmit anten- outperformed other optimization al-
[28]
work (DNN) nas to maximize the sum rate of a gorithms in terms of solution qual-
MIMO system ity and computational complexity
Joint optimization of active and The proposed CNN-based algorithm
Convolutional
passive beamforming for an RIS- achieved better performance com-
[21] Neural Network
aided wireless communication pared to traditional optimization al-
(CNN)
system gorithms

iJIM ‒ Vol. 17, No. 12, 2023 179

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Study Title Deep Learning

Objective Key Findings
and Authors Technique
The proposed DRL-based algorithm
achieved better performance com-
Deep Reinforce- Optimization of RIS reflecting el-
pared to traditional optimization al-
[29] ment Learning ements for a wireless communica-
gorithms and demonstrated the po-
(DRL) tion system
tential of using DRL for dynamic
RIS optimization
The review highlighted the potential
Review article summarizing the of deep learning techniques to en-
Various deep
latest research works on deep hance the performance of RIS-aided
[30] learning tech-
learning-based RIS-aided wireless wireless communication systems
niques
communication systems and identified research challenges
and future directions in this field
The proposed PSO algorithm out-
Particle Swarm Optimizing the phase shifts of
performed other optimization algo-
[31] Optimization RISs to improve SNR of a wire-
rithms in terms of convergence
(PSO) less communication link
speed and solution quality
The review highlighted the potential
Review article summarizing the
of using CNNs for RIS-aided wire-
Various CNN ar- latest research works on RIS-
[32] less communication systems and
chitectures aided wireless communication
identified research challenges and
systems using CNNs
future directions in this field

This table provides a brief summary of the CNN architectures used, the objectives
of the studies, and the key findings of each study in the context of reconfigurable intel-
ligent surface (RIS)-assisted communication systems.

3 Methodology

The methodological framework of low power RIS-assisted wirless communication

systems involves modeling the system, defining the optimization objective, selecting
an optimization algorithm, generating training data based on selected method or algo-
rithm, training and testing the algorithm, and implementing the algorithm in the real
system using the simulation tool (MatLab 2019).

3.1 System modelling and requirement for RIS

The system requirement table for running simulations for RIS wireless communica-
tion systems highlights the various hardware and software components required, in-
cluding the operating system, processor, memory, storage, simulation software, RIS
model, RF test equipment, network emulator, and communication protocols. These re-
quirements are essential for performing accurate simulations and testing of RIS-aided
wireless communication systems.

180 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Table 3. The system requirement for the simulation testing

System Component Requirement
Operating System Windows 10, Ubuntu 18.04 or higher, MacOS 10.13 or higher
Processor Intel Core i5 or higher
Memory 8 GB RAM or higher
Storage 256 GB SSD or higher
MATLAB 2019a or higher, or Python 3.x with required libraries (such as NumPy,
Simulation Software
SciPy, and TensorFlow)
RIS Model CAD software (such as SolidWorks or AutoCAD) to design the RIS model
RF signal generator, spectrum analyzer, power meter, directional coupler, and RF
RF Test Equipment
cables
Network emulator software (such as ns-3) to simulate the wireless network envi-
Network Emulator
ronment
Communication Proto-
Knowledge of wireless communication protocols (such as 802.11a/b/g/n/ac/ax)
cols

3.2 Steps for channel estimation for low power RIS assisted wireless
communication
The measurement scale for unit cells in RIS communication systems is typically on
the order of the operating wavelength for the simulation. For example, for a system
operating at a frequency of 28 GHz, the wavelength is approximately 1 cm. In this case,
the unit cell design in the experimentation approximately a few millimeters to centime-
ters in size. The design of the unit cell can be optimized using electromagnetic simula-
tion tools such as finite element analysis (FEA) or method of moments (MoM) to
achieve the desired reflection and transmission properties. The unit cell in the experi-
mentation designed to provide either phase control or amplitude control of the reflected
waves, depending on the specific application requirements. The design of the unit cell
depends on the specific application and the frequency range of operation. The unit cell
should be designed to have high reflection and transmission coefficients, low absorp-
tion and scattering, and low power consumption as stated in [25]. The measurement
scale in the experimentation of RIS communication systems is important to accurately
characterize the performance of the unit cell and the overall system. The measurement
scale depends on the size of the unit cell, the frequency range of operation, and the
measurement technique used. Generally, the measurement scale can range from mi-
crometers to millimeters for the size of the unit cell, and from gigahertz to terahertz for
the frequency range of operation. Below in the Table 4, it represents the range of scale
measurement for the (RIS) communication systems for channel estimation.

iJIM ‒ Vol. 17, No. 12, 2023 181

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Fig. 3. This represents the steps accumulated for the RIS-aided communication between the
transmitter and receiver

Table 4. Unit cell design and measurement scale in the experimentation

Measurement Range
Received signal strength (RSS) -100 dBm to -20 dBm
Signal-to-noise ratio (SNR) 0 dB to 50 dB
Channel impulse response (CIR) Nanoseconds to microseconds
Channel frequency response (CFR) Megahertz to gigahertz
Channel coherence time Milliseconds to seconds
Channel coherence bandwidth Kilohertz to megahertz

182 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Fig. 4. The methodology is depicted by the flowchart and steps being followed in this research

iJIM ‒ Vol. 17, No. 12, 2023 183

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

3.3 Dataset description

The dataset used in this study for channel estimation in RIS-assisted satellite IoT
communications. The RIS-assisted satellite IoT channel estimation dataset consists of
simulated channel data for a satellite-to-user communication link using an RIS. The
dataset includes 10,000 channel samples, each with a corresponding binary classifica-
tion label indicating whether the channel is line-of-sight (LOS) or non-line-of-sight
(NLOS). The dataset includes information about the positions of the satellite, RIS, and
user, as well as the channel impulse response (CIR) and channel frequency response
(CFR) in the time and frequency domains, respectively. The dataset was generated us-
ing a ray-tracing simulation model, which considers the physical characteristics of the
satellite, RIS, and user positions. The simulation also includes the effects of atmos-
pheric attenuation, scattering, and reflection. The dataset can be used to train and eval-
uate machine/deep learning models for channel estimation in RIS-assisted satellite IoT
communications. The binary classification label can be used for training and evaluating
classification models, while the CIR and CFR can be used for training and evaluating
regression models. Dataset Name: RIS-assisted Satellite IoT Channel Estimation Da-
taset (https://ieee-dataport.org/documents/dataset-channel-estimation-ris-assisted-sat-
ellite-iot-communications).
Dataset Source: Simulated data
Dataset Size: 10,000 channel samples
Data Format: CSV
Data Fields:
Satellite position (longitude, latitude, altitude)
RIS position and user position (x, y, z)
Channel impulse response (CIR) in the time domain (100 samples)
Channel frequency response (CFR) in the frequency domain (100 samples)
Binary classification label (1 or 0) indicating whether the channel is LOS or NLOS

3.4 Training and testing the channel estimation in RIS-aided system

Training for RIS-aided system. In this research, for training the CNN architecture,
which is defined using the 'layers' variable, which contains a series of convolutional
layers, batch normalization layers, max pooling layers, and fully connected layers. The
training options are specified using the 'options' variable, which includes parameters
such as the optimization algorithm, number of epochs, mini-batch size, and verbosity.
The CNN is then trained using the 'trainNetwork' function, which takes the training
data, CNN architecture, and training options as input. After training, the performance
of the CNN is evaluated on the test data using the 'predict' function, and the mean
squared error (MSE) is computed as a measure of performance. Finally, the trained
CNN can be deployed in a real RIS wireless communication system for real-time chan-
nel estimation. Note that the deployment code will depend on the specifics of the RIS
system and may require additional hardware and software components. For the purpose
of training the CNN using the mentioned dataset, the hyperparameters in channel esti-
mation reconfigurable intelligent surface (RIS) using CNN technique depend on the

184 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

specific problem and dataset. Here are some channel measurement preliminary values
in Table 5 that could be used for training for this specific dataset.

Table 5. The channel estimation measurement values for training the algorithm
Learning rate 0.001
Number of epochs: 40
Mini-batch size 32-128
Number and size of filters 32-64 filters with a size of 3x3 | 6x6
Optimizer Adam-optimizer
Dropout rate 0.2-0.5
Batch normalization True

These values can be used as a parameter for training and hyperparameter tuning, but
it is important to experiment with optimal values to find the optimal set of hyperparam-
eters for the given problem. Additionally, it is important to use a validation set to mon-
itor the performance of the model during training and avoid overfitting using the CNN
and swarm optimization PSO method. Below the Figure 5 represents the graph of train-
ing for the channel estimation based on CNN model only (pathloss vs epochs).

Fig. 5. The CNN Model Accuracy (pathloss vs epochs) Model Loss (loss vs epochs)

• Learning rate: This parameter controls the step size taken during the optimization
process. A larger learning rate can result in faster convergence, but may also cause
the optimization to overshoot the minimum.

iJIM ‒ Vol. 17, No. 12, 2023 185

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

• Number of epochs: This parameter controls the number of times the training data
is iterated over during training. Increasing the number of epochs can improve the
performance of the CNN, but can also lead to overfitting.
• Mini-batch size: This parameter controls the number of samples processed in each
iteration of the optimization algorithm. A larger mini-batch size can improve the
convergence of the CNN, but can also require more memory and computational re-
sources.
• Optimizer: This parameter controls the algorithm used to optimize the parameters
of the CNN during training. Popular optimizers include stochastic gradient descent
(SGD), adaptive moment estimation (Adam), and root mean square propagation
(RMSProp).
• Dropout rate: This parameter controls the rate at which neurons in the CNN are
randomly dropped out during training. Dropout can help prevent overfitting by en-
couraging the CNN to learn more robust features.
• Batch normalization: This parameter controls the use of batch normalization layers
in the CNN. Batch normalization can improve the convergence of the CNN by re-
ducing internal covariate shift.
• Number and size of filters: This parameter controls the number and size of the
filters in the convolutional layers of the CNN. Increasing the number of filters can
improve the representational capacity of the CNN, but can also require more com-
putational resources.
Testing for RIS-aided system. The performance of the model can be evaluated us-
ing various evaluation metrics. A common way to summarize the performance of a
model is to create a testing model table that lists the evaluation metrics for the model
on a separate test set. The metric used to evaluate the performance of the CNN model,
such as mean square error (MSE), root mean square error (RMSE), mean absolute error
(MAE), or coefficient of determination (R-squared). In the Table 6 below, each row
represents a different testing scenario, with varying channel models, SNR values, num-
ber of antennas, number of RIS elements, and modulation schemes. The testing metric
used for evaluation is different for each row, including MSE, RMSE, MAE, and R-
squared. The testing performance of the CNN algorithm is also provided for each test-
ing scenario. The testing values should be chosen to provide a representative evaluation
of the performance of the CNN algorithm in the simulation tool MATLAb.

Table 6. The channel estimation measurement values for testing the algorithm
SNR Num. of Num. of RIS Modulation Testing Met- Testing Per-
Channel Model
(dB) Antennas Elements Scheme ric formance
Rayleigh Fading 10 2 16 QPSK MSE 0.003
Multipath 20 4 32 16-QAM RMSE 0.022
Free Space 15 8 64 64-QAM MAE 0.012
Shadowing 5 16 128 256-QAM R-squared 0.96

It is important to note that the optimal set of hyperparameters for a given problem
may not generalize well to other datasets or problems. Therefore, it is recommended to

186 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

perform hyperparameter tuning and testing on dataset and problem to ensure that the
CNN/PSO model's performance is consistent across different scenarios. Additionally,
it is important to perform statistical tests to ensure that any observed differences in
performance between different hyperparameter values are statistically significan. Here
in the Figure 6, the testing of parameters is carried out in the simulation tool.

Fig. 6. Testing set stage for channel estimation of RIS-aided network

4 Results and discussion

The research applied the framework of CNN and PSO for the channel estimation in
RIS-aided wireless communication system. The CNN is trained using 10,000 samples
of channel data, and achieves a test accuracy of 95% in predicting the channel response
between the transmitter and receiver with an SNR of 20 dB. The computation time for
channel estimation is 100 ms, which is relatively fast and allows for real-time
adaptation of the RIS to changing channel conditions. These values are just for
demonstration purposes and the actual values in a real-world application may vary
depending on various factors such as the complexity of the channel model, the size and
configuration of the RIS, and the specific machine learning approach used. Table 7
represents the values on which the experiment is carried out in the network simulator.

iJIM ‒ Vol. 17, No. 12, 2023 187

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Table 7. This represents the optimal standard values for the simulation based testing on
channel estimation for RIS-aided communication system
Measurement Value
Number of neurons for CNN ̴ 400
Operating frequency 22 GHz
Channel bandwidth 12 GHz
Number of reflecting elements 32 | 64
Modulation scheme 16-QAM and QPSK
Signal-to-noise ratio (SNR) 10 dB, 20 dB
Number of training samples 10,000
Number of validation samples 1,000
Number of PSO iterations 150
PSO population size 50 | 60
Channel coherence time 10 m/s
Number of test samples 1,000
Training time 18 minutes
Test accuracy 96%
Computation time for channel estimation 100 ms

In order to maximise the received signal intensity at the receiver with an SNR of 10
| 20 dB, the PSO algorithm is utilised to optimise the reflection coefficients of the RIS
elements. With a population size of 50 and 60, and an inertia weight of 0.9, the PSO
algorithm is run for 150 iterations. The RIS is set up for optimum channel performance
using the reflection coefficients that are produced. The wireless channel remains con-
sistent for 10 m/s and within a 2 MHz frequency range before it begins changing as a
result of movement of the transmitter, receiver, or objects in the environment. The chan-
nel coherence duration is 10 ms and the coherence bandwidth is 2 MHz. The computa-
tion time for channel estimation is 50 ms, which is relatively fast and allows for real-
time adaptation of the RIS to changing channel conditions. The simulation results are
presented in the form of graphs showing the Mean Squared Error (MSE) between the
estimated and actual channel responses for different scenarios. The MSE is a metric
that measures the difference between the estimated and actual values, and a lower value
indicates better performance. The process involves training a CNN model using simu-
lation data to learn the mapping between the input and output signals. The input signal
is the transmitted signal, and the output signal is the received signal at the RIS. During
the training process, the CNN learns to estimate the channel response by adjusting the
weights of the network. Once the CNN is trained, it can be used to estimate the channel
response in real-time. The sample data evaluation for channel estimation in the RIS-
aided system represents in Figure 7.

188 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Fig. 7. Demonstrates the sample data evaluation with the channel bandwidth and operating fre-
quency which are obtained from both the algorithm

Figure 8 represents the MSEE value contrast to the SNR (dB) for the experiment on
channel estimation. Lower MSE values indicate better performance in terms of
accuracy of the delay predictions and for better estimation in the channel modelling in
Reconfigurable Intelligent Surface (RIS) communication systems. The MSE values in
channel estimation for RIS communication systems using CNN can depend on various
factors, including the system design, training dataset size and quality, CNN
architecture, and operating conditions. However, in general, from the graph above, it
can be seen that MSE metric are desired for better channel estimation accuracy.
Typically, the MSE values for delay predictions in RIS communication systems using
CNN can range from a few 10-3 to 10-6, depending on the specific system design and
operating conditions. For example, in this study on RIS-assisted millimeter-wave
communication using CNN-based channel estimation, the MSE for delay predictions
was reported to be around 10-4 to 10-5 for different operating scenarios. Reconfigurable
Intelligent Surface (RIS) communication systems can benefit from delay predictions in
channel estimation, especially in wireless communication environments where
multipath propagation is common. The RIS phase shifts can be modified in accordance
with the time-varying channel parameters to improve communication performance. The
figures below depict the grapg for delay prediction in both the algorithms. The simula-
tion of delay prediction by both the used algorithms are displayed in graph Figure 9.

iJIM ‒ Vol. 17, No. 12, 2023 189

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

Fig. 8. This simulation contains the MSEE value vs the SNR required for the channel estima-
tion performance

Fig. 9. These graphs indicates the delay prediction (path loss vs MSE/RMS delay) for each of
the technique

A CNN-based channel estimation technique is proposed for RIS-assisted millimeter-

wave communication systems in this study. The proposed technique used a deep CNN
architecture with five convolutional layers and two fully connected layers. The network
was trained using a dataset metioned in methodology section for channel realizations,
and achieved a mean squared error (MSE) value of 0.012 using CNN algorithm. The
study reported that the proposed technique outperformed other state-of-the-art tech-
niques. The proposed technique used PSO to optimize the channel parameters, and
achieved a mean squared error (MSE) value of 0.0075 for a single user scenario and

190 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

0.025 for a multi-user scenario. The results accuracy comparison is presented in the
Table 8 below.

Table 8. The accuracy comparison of deep learning and swarm optimization technique with
existing literature as a reference
Deep Learning & Swarm Optimization Technique Accuracy Computational Efficiency
Convolutional Neural Networks (CNN) [proposed] 96.47% 97.55%
Recurrent Neural Networks (RNN) [26] 89.78% 85.47%
Deep Belief Networks (DBN) [27] 94.25% 90.51%
Particle Swarm Optimization (PSO) [proposed] 90.96% 98.78%
Ant Colony Optimization (ACO) [28] 89.78% 87.21%

The accuracy of each technique may vary depending on several factors, including
the specific system design, operating conditions, and available training data. CNNs
have been shown to provide high accuracy and robustness to noise and interference,
making them a popular choice for channel estimation in RIS communication systems.
RNNs can also be effective in capturing the temporal correlations in the channel
characteristics but may require more computational resources and longer training times.
DBNs have also been shown to provide high accuracy, but may require more training
data and longer training times compared to CNNs and RNNs. It is important to evaluate
the accuracy and computational efficiency of each technique for the specific application
to select the most appropriate technique. The accuracy of each technique may vary de-
pending on several factors, including the specific system design, operating conditions,
and optimization parameters. Particle Swarm Optimization (PSO) has been shown to
provide high accuracy while being high computationally efficient. Ant Colony Optimi-
zation (ACO) can provide higher accuracy, but may require more computational re-
sources and longer optimization times.

5 Conclusion

Based on the study conducted on Channel Estimation for Reconfigurable Intelligent

Surface (RIS) Communication Systems using deep learning and swarm optimization
technique. The use of deep learning techniques such as CNN can effectively estimate
the channel state information (CSI) of RIS communication systems, even in the pres-
ence of noise and interference. The use of Particle Swarm Optimization (PSO) can sig-
nificantly enhance the performance of CNN-based channel estimation, by optimizing
the network's weights and biases. The proposed CNN-PSO approach can achieve higher
accuracy and lower mean squared error (MSE) compared to traditional methods of
channel estimation, such as channel bandwidth, operating frequency and minimum
mean square error. The performance of the proposed approach is also found to be robust
against various system parameters, such as the number of RIS elements and signal-to-
noise ratio (SNR). The study highlights the potential of using deep learning and opti-

iJIM ‒ Vol. 17, No. 12, 2023 191

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

mization techniques compared to accuracy in addressing the challenges of channel es-

timation in RIS communication systems, and suggests promising directions for future
research in this field.

6 Acknowledgements

The authors are grateful to the Iraqi Ministry of Higher Education and Scientific
Research (MOHESR) for technically supporting the current research. In addition, many
thanks for Faculty of Mustamsiriyah University, College of Engineering, Electrical En-
gineering Department.

7 References
[1] E. Basar, M. Di Renzo, J. De Rosny, M. Debbah, M.-S. Alouini, and R. Zhang, "Wireless
communications through reconfigurable intelligent surfaces," IEEE access, vol. 7, pp.
116753-116773, 2019. https://doi.org/10.1109/ACCESS.2019.2935192
[2] H. Liu, X. Yuan, and Y.-J. A. Zhang, "Matrix-calibration-based cascaded channel estimation
for reconfigurable intelligent surface assisted multiuser MIMO," IEEE Journal on Selected
Areas in Communications, vol. 38, no. 11, pp. 2621-2636, 2020. https://doi.org/10.1109/
JSAC.2020.3007057
[3] Q. Wu and R. Zhang, "Intelligent reflecting surface enhanced wireless network via joint
active and passive beamforming," IEEE Transactions on Wireless Communications, vol. 18,
no. 11, pp. 5394-5409, 2019. https://doi.org/10.1109/TWC.2019.2936025
[4] H. Alwazani, A. Kammoun, A. Chaaban, M. Debbah, and M.-S. Alouini, "Intelligent re-
flecting surface-assisted multi-user MISO communication: Channel estimation and beam-
forming design," IEEE Open Journal of the Communications Society, vol. 1, pp. 661-680,
2020. https://doi.org/10.1109/OJCOMS.2020.2992791
[5] X. Guan, Q. Wu, and R. Zhang, "Anchor-assisted intelligent reflecting surface channel esti-
mation for multiuser communications," in GLOBECOM 2020-2020 IEEE Global Commu-
nications Conference, 2020: IEEE, pp. 1-6. https://doi.org/10.1109/GLOBECOM42002.
2020.9347985
[6] Z. Wang, L. Liu, and S. Cui, "Channel estimation for intelligent reflecting surface assisted
multiuser communications: Framework, algorithms, and analysis," IEEE Transactions on
Wireless Communications, vol. 19, no. 10, pp. 6607-6620, 2020. https://doi.org/10.1109/
TWC.2020.3004330
[7] P. Wang, J. Fang, H. Duan, and H. Li, "Compressed channel estimation for intelligent re-
flecting surface-assisted millimeter wave systems," IEEE signal processing letters, vol. 27,
pp. 905-909, 2020. https://doi.org/10.1109/LSP.2020.2998357
[8] J.-S. Jung, C.-Y. Park, J.-H. Oh, and H.-K. Song, "Intelligent reflecting surface for spectral
efficiency maximization in the multi-user MISO communication systems," IEEE Access,
vol. 9, pp. 134695-134702, 2021. https://doi.org/10.1109/ACCESS.2021.3116959
[9] J. Q. Kadhim, I. A. Aljazaery, and H. T. Salim ALRikabi, "Enhancement of Online Educa-
tion in Engineering College Based on Mobile Wireless Communication Networks and IOT,"
International Journal of Emerging Technologies in Learning, vol. 18, no. 1, 2023.
https://doi.org/10.3991/ijet.v18i01.35987

192 http://www.i-jim.org
Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

[10] X. Wei, D. Shen, and L. Dai, "Channel estimation for RIS assisted wireless communica-
tions—Part I: Fundamentals, solutions, and future opportunities," IEEE communications let-
ters, vol. 25, no. 5, pp. 1398-1402, 2021. https://doi.org/10.1109/LCOMM.2021.3052822
[11] D. Mishra and H. Johansson, "Channel estimation and low-complexity beamforming design
for passive intelligent surface assisted MISO wireless energy transfer," in ICASSP 2019-
2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP),
2019: IEEE, pp. 4659-4663. https://doi.org/10.1109/ICASSP.2019.8683663
[12] Y. Pan and Z. Deng, "Channel Estimation for Wireless Communication Systems Aided by
Large Intelligent Reflecting Surface," in 2021 IEEE 2nd International Conference on Big
Data, Artificial Intelligence and Internet of Things Engineering (ICBAIE), 2021: IEEE, pp.
637-643. https://doi.org/10.1109/ICBAIE52039.2021.9389923
[13] A. Zappone, M. Di Renzo, and M. Debbah, "Wireless networks design in the era of deep
learning: Model-based, AI-based, or both?," IEEE Transactions on Communications, vol.
67, no. 10, pp. 7331-7376, 2019. https://doi.org/10.1109/TCOMM.2019.2924010
[14] G. A. Aramice and J. Q. Kadhim, "Secure Code Generation for Multi-Level Mutual Authen-
tication," TELKOMNIKA (Telecommunication Computing Electronics and Control), vol.
16, no. 6, pp. 2643-2650, 2018. https://doi.org/10.12928/telkomnika.v16i6.10437
[15] A. H. Sallomi and S. A. J. r. Hashem, "A Novel Theoretical Model for Cellular Base Station
Radiation Prediction," vol. 16, no. 17, p. 17, 2018. https://doi.org/10.5013/IJSSST.a.19.06.
35
[16] H. Ye, G. Y. Li, and B.-H. Juang, "Power of deep learning for channel estimation and signal
detection in OFDM systems," IEEE Wireless Communications Letters, vol. 7, no. 1, pp. 114-
117, 2017. https://doi.org/10.1109/LWC.2017.2757490
[17] S. Liu, Z. Gao, J. Zhang, M. Di Renzo, and M.-S. Alouini, "Deep denoising neural network
assisted compressive channel estimation for mmWave intelligent reflecting surfaces," IEEE
Transactions on Vehicular Technology, vol. 69, no. 8, pp. 9223-9228, 2020. https://doi.org/
10.1109/TVT.2020.3005402
[18] M. Dajer et al., "Reconfigurable intelligent surface: Design the channel–A new opportunity
for future wireless networks," Digital Communications and Networks, vol. 8, no. 2, pp. 87-
104, 2022. https://doi.org/10.1016/j.dcan.2021.11.002
[19] T. Sharma, A. Chehri, and P. Fortier, "Reconfigurable intelligent surfaces for 5G and beyond
wireless communications: A comprehensive survey," Energies, vol. 14, no. 24, p. 8219,
2021. https://doi.org/10.3390/en14248219
[20] V. Tapio, I. Hemadeh, A. Mourad, A. Shojaeifard, and M. Juntti, "Survey on reconfigurable
intelligent surfaces below 10 GHz," EURASIP Journal on Wireless Communications and
Networking, vol. 2021, pp. 1-18, 2021. https://doi.org/10.1186/s13638-021-02048-5
[21] Q. Wu, S. Zhang, B. Zheng, C. You, and R. Zhang, "Intelligent reflecting surface-aided
wireless communications: A tutorial," IEEE Transactions on Communications, vol. 69, no.
5, pp. 3313-3351, 2021. https://doi.org/10.1109/TCOMM.2021.3051897
[22] Q. Wu and R. Zhang, "Towards smart and reconfigurable environment: Intelligent reflecting
surface aided wireless network," IEEE communications magazine, vol. 58, no. 1, pp. 106-
112, 2019. https://doi.org/10.1109/MCOM.001.1900107
[23] A. Taha, M. Alrabeiah, and A. Alkhateeb, "Enabling large intelligent surfaces with com-
pressive sensing and deep learning," IEEE access, vol. 9, pp. 44304-44321, 2021.
https://doi.org/10.1109/ACCESS.2021.3064073
[24] B. Biggio et al., "Security evaluation of support vector machines in adversarial environ-
ments," Support Vector Machines Applications, pp. 105-153, 2014. https://doi.org/10.1007/
978-3-319-02300-7_4

iJIM ‒ Vol. 17, No. 12, 2023 193

Paper—Enabling Deep Learning and Swarm Optimization Algorithm for Channel Estimation for Low…

[25] S. Sun and H. Yan, "Channel estimation for reconfigurable intelligent surface-assisted wire-
less communications considering Doppler effect," IEEE Wireless Communications Letters,
vol. 10, no. 4, pp. 790-794, 2020. https://doi.org/10.1109/LWC.2020.3044004
[26] Z. Li, Z. Chen, X. Ma, and W. Chen, "Channel estimation for intelligent reflecting surface
enabled terahertz mimo systems: A deep learning perspective," in 2020 IEEE/CIC Interna-
tional Conference on Communications in China (ICCC Workshops), 2020: IEEE, pp. 75-79.
https://doi.org/10.1109/ICCCWorkshops49972.2020.9209937
[27] X. Yuan, Y.-J. A. Zhang, Y. Shi, W. Yan, and H. Liu, "Reconfigurable-intelligent-surface
empowered wireless communications: Challenges and opportunities," IEEE wireless com-
munications, vol. 28, no. 2, pp. 136-143, 2021. https://doi.org/10.1109/MWC.001.2000256
[28] G. López-Lanuza, K. Chen-Hu, and A. G. Armada, "Deep Learning-Based Optimization for
Reconfigurable Intelligent Surface-Assisted Communications," in 2022 IEEE Wireless
Communications and Networking Conference (WCNC), 2022: IEEE, pp. 764-769.
https://doi.org/10.1109/WCNC51071.2022.9771876
[29] Q.-U.-A. Nadeem, A. Kammoun, A. Chaaban, M. Debbah, and M.-S. Alouini, "Intelligent
reflecting surface assisted wireless communication: Modeling and channel estimation,"
arXiv preprint arXiv:1906.02360, 2019. https://doi.org/10.1109/OJCOMS.2020.2992791
[30] M. A. S. Sejan, M. H. Rahman, B.-S. Shin, J.-H. Oh, Y.-H. You, and H.-K. Song, "Machine
Learning for Intelligent-Reflecting-Surface-Based Wireless Communication towards 6G: A
Review," Sensors, vol. 22, no. 14, p. 5405, 2022. https://doi.org/10.3390/s22145405
[31] S. Zeng, H. Zhang, B. Di, Z. Han, and L. Song, "Reconfigurable intelligent surface (RIS)
assisted wireless coverage extension: RIS orientation and location optimization," IEEE
Communications Letters, vol. 25, no. 1, pp. 269-273, 2020. https://doi.org/10.1109/
LCOMM.2020.3025345
[32] H. Yang, Z. Xiong, J. Zhao, D. Niyato, L. Xiao, and Q. Wu, "Deep reinforcement learning-
based intelligent reflecting surface for secure wireless communications," IEEE Transactions
on Wireless Communications, vol. 20, no. 1, pp. 375-388, 2020. https://doi.org/10.1109/
TWC.2020.3024860

8 Authors

Jaafar Qassim Kadhim is with Electrical Engineering Department, College of En-

gineering, Mustansiriyah University, Baghdad, Iraq (email: jaafar80@uomustansiri-
yah.edu.iq).
Adheed H. Sallomi is with Electrical Engineering Department, College of Engi-
neering, Mustansiriyah University, Baghdad, Iraq.

Article submitted 2023-03-07. Resubmitted 2023-05-03. Final acceptance 2023-05-05. Final version pub-
lished as submitted by the authors.

194 http://www.i-jim.org