Energy-Efficient Near-Field Beamforming: A Review on Practical Channel Models

Abstract

The unprecedented expansion of wireless networks has resulted in spectrum sharing between numerous connected devices, demanding advanced interference management and higher energy consumption, which exacerbates the carbon footprint. Near-field communication emerges as a promising solution to these challenges as it enables precise signal focusing which reduces power consumption by providing higher spatial multiplexing gains. This review explores how near-field (NF) multiple-input multiple-output (MIMO) beamforming can enhance energy efficiency by optimizing beamfocusing and minimizing unnecessary energy expenditure. We discuss the latest advancements in near-field beamforming, emphasizing energy-efficient strategies and sustainable practices. Recognizing which practical channel models are better suited for near-field communication, we delve into the integration of Electromagnetic Information Theory (EIT) as a joint model for realistic applications. We also discuss the channel models for near-field beamforming, incorporating EIT to provide a comprehensive overview of current methodologies. We further analyze the strengths and limitations of existing channel models and discuss the state-of-the-art models which address existing gaps. We also explore opportunities for the practical deployment of energy-efficient near-field beamforming systems. By summarizing future research directions, this review aims to advance the understanding and application of sustainable energy practices in near-field communication technologies.

1. Introduction

1.1. Background

In recent years, interest in beamformers has significantly increased. With the advent of 6G, there is a trend of extreme capacity requirements to meet the growing demands for higher data rates and improved connectivity. This makes beamforming a promising technology for efficiently directing signals and enhancing overall network performance [1]. This trend is evidenced by the global beamformer market’s projected growth to USD 201,451.2 million by 2024, with sales expected to further escalate to USD 569,345.36 million by 2031. From 2024 to 2031, it will reflect a robust Compound Annual Growth Rate (CAGR) of 16.00% [2]. From the perspective of spatial diversity, traditional wireless communication systems have made good use of far-field spatial resources. As the need for data and connectivity increases and energy efficiency becomes more critical, near-field beamforming, an innovative field, holds a lot of promise for the future of the beamforming market [3]. Wireless communication systems are expected to gain a new physical spatial dimension with more study and application of near-field spatial resources. The near-field region will be a critical factor in 6G networks, necessitating research into new types of near-field technology [4]. In the field of near-field technology, alterations in electromagnetic wave propagation characteristics necessitate considering waves as spherical waves rather than plane waves. This new physical characteristic introduces a range of electromagnetic effects, including spatial non-stationarity, beamsplitting, tri-polarization, and evanescent waves [5]. As a result, many traditional communication algorithms may suffer significant performance degradation in 6G near-field scenarios or may not fully leverage these new physical properties [6]. A discrepancy between near-field propagation models and existing far-field communication technologies can significantly impair the performance of current far-field techniques in near-field regions [7]. Beamfocusing is a technique used in communication systems to direct the energy of a beam towards a specific target area. This ensures that the signal is concentrated and more effective [8]. Near-field beamforming enhances beamfocusing by reducing unnecessary energy consumption and improving the overall efficiency of communication systems [9].

By optimizing beamfocusing, near-field beamforming minimizes unnecessary energy consumption and improves the overall efficiency of communication systems [10] in the small near-field region illustrated in Figure 1. The latest advancements in near-field technology are characterized by the integration of dynamic metasurface antennas and ultra-large-scale antenna arrays (ULAAs), which allow for programmable control over beam modes and enhance signal processing abilities, thereby reducing reliance on specialized analog circuits [11]. By concentrating beams, especially in mid-frequency bands like the 10 GHz range, which provides a balance between path loss and coverage, this invention facilitates multiuser communication [12]. To enhance communication capacity and spatial resolution, intelligent surfaces have been further developed within communication networks and are currently under investigation [13]. The transition from 5G to 6G, where higher frequencies and bigger antenna apertures transform conventional far-field scenarios into near-field communication environments [14]. Near-field beamforming aims to enable accurate, high-quality communication in increasingly complex and dense network environments while being energy efficient [15]. This advanced technology enables more reliable connections and enhances network performance, which is anticipated to drive further advancements in wireless communication.

Beamtraining in the near field involves optimizing the orientation and shape of beams emitted by an antenna array to efficiently target and communicate with specific receivers [16]. Codebook design in the near field is the creation of a set of predefined beamforming vectors that guide the antenna array’s beamsteering [17]. The intricacies of beamtraining and codebook design must consider both angle and distance dimensions [18], which is among the difficulties that near-field technology must overcome. Near-field beams’ irregular shapes make it more challenging to decompose space orthogonally [19], which could result in blind spots and codebook redundancy [20]. Additionally, the high cost and power consumption of ULAAs pose significant challenges to their widespread implementation [21]. Significant energy resources are needed for these systems, which can strain power infrastructures and raise operating expenses [22]. Moreover, array gain loss due to near-field effects, like beamsplitting in wideband systems, makes system design more difficult [23]. Additionally, the combination of intelligent surfaces and dynamic metasurface antennas adds more technical challenges and complicates the system, which could result in increased energy consumption [24]. To effectively transition near-field communication technology from 5G to 6G and beyond, these challenges must be addressed in the future. This will guarantee that improvements in communication efficiency do not come at the expense of energy sustainability.

To effectively address these challenges, we undertake a comprehensive examination of techniques designed to resolve issues related to energy efficiency in the near field, and innovative solutions are being proposed. In particular, developing two-step beamforming [25] and ring-type codebooks [26] can enhance beam precision and gain, thus mitigating non-standardized beam patterns. Fractional Fourier transform vectors are used in codebook design to enhance near-field performance while preserving compatibility with current systems [27]. Furthermore, using deep learning for practical beamtraining [28] makes it possible to adapt more accurately and quickly to changing conditions, which in turn lowers complexity and operating expenses. ULAAs may become more practical for widespread deployment by investigating hybrid analog-digital architectures [29], which can further lower costs and power consumption. These methods address the technical challenges and aim to optimize energy usage, ensuring that near-field technology evolves in an energy-efficient and sustainable manner [30,31]. Improving energy efficiency in near-field MIMO systems is crucial for managing power resources effectively, ensuring strong signal quality and data rates without waste in close-proximity environments [19]. The challenges of near-field technology can be addressed by further developing these areas. This could create opportunities for future communication systems to utilize more effective and adaptable solutions.

1.2. Related Reviews for Energy-Efficient Techniques on Near-Field Communication

Over the years, researchers have significantly contributed to the study and review of energy-efficient techniques for enhancing near-field communication, the related reviews have been presented in Table 1. For instance, refs. [32,33] provided a detailed explanation of basic principles, channel modeling, and performance metrics. This comprehensive survey is valuable for its thorough coverage of near-field communication fundamentals. However, it falls short in addressing the integration of mixed-field channel modeling and near-field models, which are crucial for a more holistic understanding of the field. Similarly, ref. [34] reviewed new channel characteristics in near-field communication, but it did not delve into the practical applications and hardware complexities essential for real-world implementations.

In the area of hybrid beamforming and hardware requirements, ref. [35] introduced a hybrid beamforming architecture that enhances energy efficiency while minimizing hardware needs. This approach combines analog and digital beamforming techniques to optimize signal processing. By designing high-dimensional analog beamformers alongside low-dimensional digital beamformers in two stages, the method reduces computational complexity and simplifies channel estimation. However, the study lacks sufficient focus on channel state information techniques and practical channel models in near-field environments, which are necessary for a more comprehensive analysis. On the other hand, ref. [4] reviewed resource allocation issues and emphasizes the potential of large-scale MIMO systems in improving spectrum and energy efficiency. While this is an essential aspect of near-field communication, the study did not adequately address the complexities of channel models in diverse near-field environments.

Furthermore, analyzing point beamforming, energy efficiency, localization, and multiuser access technologies reveals additional strengths and limitations. Ref. [36] reviewed point beamforming in near-field applications, which is beneficial for understanding specific applications of this technology. However, it lacks attention to energy-efficient technologies and optimization algorithms, which are critical for a more complete review. Similarly, ref. [15] summarized the applications of energy efficiency, localization, channel estimation, and multiuser access technologies in near-field-driven 6G networks. While this survey provides valuable insights into the potential of these technologies, it did not focus on channel models and optimization algorithms in high-frequency bands. Additionally, ref. [22] emphasized the potential of enhancing spatial reuse gain and localization accuracy, but it overlooked channel models and estimation in diverse communication environments. Lastly, ref. [37] discussed the performance degradation of traditional far-field beamforming designs in near-field environments, but it did not focus on channel models and energy-efficient technologies. These limitations highlight the need for future research to develop and apply advanced channel models and optimization algorithms in complex near-field environments.

Table 1. List of related reviews.

Title	Year	Main Achievements	Limitations
[4]	2024	Reviewed resource allocation problems and highlighted the potential of massive MIMO systems in improving spectrum and energy efficiency.	Lack of focus on channel models and optimization techniques in complex near-field environments.
[36]	2024	Focused on reviewing “Spot Beamforming” in near-field applications. The review does not detail energy-efficient techniques or optimization algorithms.	The review does not detail the energy-efficient techniques or the optimization algorithms
[15]	2024	Summarized energy efficiency, localization, channel estimation, and multiuser access techniques in near-field driven 6G networks.	A review of insufficient channel modeling or optimization techniques in high frequency bands is not provided.
[22]	2024	Highlighted the potential in enhancing spatial multiplexing gain and positioning accuracy.	Lack of focus on channel models and channel estimation in diverse communication environment.
[34]	2024	Focused on reviewing the new channel characteristics of near-field communications.
[37]	2024	Introduced the performance degradation caused by traditional far-field beamforming designs in near-field environments and integrated sensing and communication (ISAC) systems.	The review does not detail the channel modeling techniques or the energy-efficient techniques.
[32]	2024	Detailed explanations of the fundamental principles, channel modeling, and performance metrics.	Insufficient hybrid-field channel modeling and lack of focus on integrating near-field models.
[33]	2023	Reviewed the fundamental near-field channel models and the focus on the near-field spherical wave propagation designs	No emphasis on the practical hardware costs and insufficient energy-efficient techniques provided.

Table 1. List of related reviews.

Title	Year	Main Achievements	Limitations
[4]	2024	Reviewed resource allocation problems and highlighted the potential of massive MIMO systems in improving spectrum and energy efficiency.	Lack of focus on channel models and optimization techniques in complex near-field environments.
[36]	2024	Focused on reviewing “Spot Beamforming” in near-field applications. The review does not detail energy-efficient techniques or optimization algorithms.	The review does not detail the energy-efficient techniques or the optimization algorithms
[15]	2024	Summarized energy efficiency, localization, channel estimation, and multiuser access techniques in near-field driven 6G networks.	A review of insufficient channel modeling or optimization techniques in high frequency bands is not provided.
[22]	2024	Highlighted the potential in enhancing spatial multiplexing gain and positioning accuracy.	Lack of focus on channel models and channel estimation in diverse communication environment.
[34]	2024	Focused on reviewing the new channel characteristics of near-field communications.
[37]	2024	Introduced the performance degradation caused by traditional far-field beamforming designs in near-field environments and integrated sensing and communication (ISAC) systems.	The review does not detail the channel modeling techniques or the energy-efficient techniques.
[32]	2024	Detailed explanations of the fundamental principles, channel modeling, and performance metrics.	Insufficient hybrid-field channel modeling and lack of focus on integrating near-field models.
[33]	2023	Reviewed the fundamental near-field channel models and the focus on the near-field spherical wave propagation designs	No emphasis on the practical hardware costs and insufficient energy-efficient techniques provided.

1.3. Motivation and Contribution

According to our review of the literature, most existing surveys have overlooked energy-efficient techniques in near-field communication. In addition, the current literature provides an incomplete overview of energy-efficient techniques. To bridge this gap, we aim to conduct a comprehensive review of these techniques for near-field communication, focusing on channel modeling, hybrid beamforming, resource allocation, and multiuser access technologies. Our objectives are twofold:

(a)

We examine current energy-efficient techniques for near-field communication, focusing on their characteristics, advantages, and limitations. In addition, we clarify the capabilities of advanced channel models and optimization algorithms within near-field environments.

(b)

We explore and highlight the characteristics, strengths, weaknesses, and potential of energy-efficient techniques documented in the existing literature. Additionally, we identify several unresolved fundamental research challenges in this rapidly evolving field to advance energy-efficient techniques and propose future research directions.

The contributions of this paper are fourfold and can be summarized as follows:

(1)

We offer comprehensive reviews of energy-efficient techniques in near-field communication, analyzing their characteristics, strengths, and potential.

(2)

We pinpoint the capabilities of advanced signal and channel models, including their roles in improving energy efficiency and reducing hardware requirements. We explain the reasons for leveraging these techniques in near-field communication and provide energy consumption models tailored to various scenarios within near-field communication.

(3)

We highlight existing energy-efficient techniques in near-field communication and discuss their applications in diverse communication environments.

(4)

We emphasize the challenges in near-field communication, such as the development of accurate channel models and the use of energy-efficient techniques.

1.4. Paper Organization

The remainder of this paper is organized as follows. Section 2 introduces near-field systems, addressing topics such as wave propagation regions, radiation in both near-field and far-field scenarios, as well as the challenges associated with near-field systems. In Section 3, we present techniques for energy-efficient beamforming and channel estimation specifically tailored for near-field communications. Section 4 focuses on efficient channel estimation techniques for near-field communication. Following this, Section 5 reviews and analyzes practical signal models designed to enhance the energy efficiency of near-field communications, specifically discussing Reconfigurable Intelligent Surfaces (RISs), holographic models, and CAP MIMO models. With a solid understanding of these models, Section 6 explores energy-efficient techniques derived from Electromagnetic Information Theory applicable to near-field communications. Finally, we summarize our findings and conclude our work in Section 7, as well as the challenges associated with near-field systems. The overall paper structure has been shown in Figure 2. The abbreviations are listed in

Table 2. List of abbreviations.

Abbr.	Full Form	Abbr.	Full Form
3GPP	3rd Generation Partnership Project	ADC	Analog-to-Digital Converter
AoAs	Angles of Arrival	BA	Beam Alignment
BER	Bit Error Rate	BS	Base Station
CAP	Continuous Aperture Phased	CAGR	Compound Annual Growth Rate
CAP-RISs	Continuous Aperture RISs	CLA	Cylindrical Antenna Array
CNN	Convolutional Neural Network	CRB	Cramér–Rao Bound
CS	Compressed Sensing	CSI	Channel State Information
D-NFCE	Deep-Learning-based Near-field Channel Estimation	DAC	Digital-to-Analog Converters
DFT	Discrete Fourier Transform	DMAs	Dynamic Metasurface Antennas
DoF	Degrees of Freedom	DPP	Delay Phase Precoding
EIT	Electromagnetic Information Theory	FSBL	Fast Sparse Bayesian Learning
FUs	Far-Field Users	GNN	Graph Neural Network
GS	Gerchberg–Saxton	HMA	Holographic Metasurface Antennas
HMIMO	Holographic MIMO	IOSs	Intelligent Omnisurfaces
ISAC	Integrated Sensing and Communication	LISs	Large Intelligent Surfaces
LISTA	Learning Iterative Shrinkage and a Thresholding Algorithm	LoS	Line of Sight
LS	Least Squares	MIMO	Multiple-input Multiple-output
MSE	Mean Square Error	MUSIC	Multiple Signal Classification
MVs	Mobile Vehicles	NB	Near-Field Beamsquint
NF	Near Field	NF-JCEL	Near-Field Channel Estimation and Localization
NOMA	Non-Orthogonal Multiple Access	NUs	Near-Field Users
OMP	Orthogonal Matching Pursuit	OFDM	Orthogonal Frequency Division Multiplexing
PD	Power Diffusion	PD-OMP	Power Diffusion aware Orthogonal Matching Pursuit
PDF	Phase Delay Focusing	PF-RCE	Polar-domain Frequency-dependent RIS-assisted Channel Estimation
P-SOMP	Polar-domain Simultaneous Orthogonal Matching Pursuit	PWE	Plane Wave Expansion
QoS	Quality of Service	RCRB	Root of Cramér–Rao Bound
RF	Radio Frequency	RIS	Reconfigurable Intelligent Surface
R-LS	Regularized Least Squares	SA-RIS	Subarray-based RIS
SIMs	Stacked Intelligent Metasurfaces	SINR	Signal-to-Interference-plus-Noise Ratio
SNR	Signal-to-Noise Ratio	STT	Sense-then-Train
THz	Terahertz	TPBE	Two-phase Angle and Distance Beam Estimator
ToI	Targets of Interest	TTD-RIS	True Time Delay RIS
UE	User Equipment	ULA	Uniform Linear Array
ULAAs	Ultra-large-scale Antenna Arrays	UPA	Uniform Planar Array
UPW	Uniform Plane Wave	V2V	Vehicle-to-Vehicle
VR	Visual Region	XL	Extremely Large-scale
XL-IRS	Extremely Large-scale Intelligent Reflecting Surface	XLARIS	Extremely Large-scale Aerial RIS

.

2. Near-Field Beamforming

Wireless systems utilize radio-frequency (RF) waves, which propagate using time-varying electric and magnetic fields. Their propagation mechanism is guided by Maxwell’s equations, proposed by James Clerk Maxwell [38,39], which posit that oscillating electric charges would generate RF waves traveling at speed c where [40]:

$$c={\displaystyle \frac{1}{\sqrt{{\mu }_0{ϵ}_0}}},$$

(1)

where ${\mu }_0$ is the permeability and ${ϵ}_0$ is the permittivity of free space. Permittivity is the measure of resistance which is provided by the space to the formation of an electric field. Hence, a higher permittivity results in weaker electric fields in a medium for the same amount of charge. Similarly, permeability is the measure of resistance, which is provided by the space to the formation of a magnetic field. A high-permeability material generates stronger magnetic fields for a given current. The oscillation and acceleration of electric charges, e.g., electrons in an antenna module, create an electric field that propagates at the speed of light. This charge oscillation simultaneously creates a magnetic field in the perpendicular direction. These transverse electromagnetic waves do not require a medium for propagation. The frequency of oscillation of the electric charges determines the characteristics of the generated RF wave.

2.1. Wave Propagation Regions

The characteristics of the generated wave differ with the distance from the source of wave generation. During outward propagation from the source, the wave experiences distinct phase and energy transitions, which give rise to distinct propagation regions with differing wavefront, phase, and energy characteristics, as shown in Figure 3 [41]. The regions are described as follows:

2.1.1. Reactive Near-Field Region

The reactive near-field region is the closest region to the source antenna. It is characterized by the electromagnetic fields, which are highly reactive and do not propagate as traveling waves [8]. Owing to this, this region is also called the non-radiating near-field region. The reactive region of the near-field spans only a few wavelengths of operation [42]. The fields in this region are 90 degrees out of phase with each other, and the energy does not radiate outwards; instead, it oscillates back to the antenna [42]. In the reactive near-field, the electromagnetic wavefronts are not fully developed. This creates a highly non-uniform power density and realizes spatial variations in the waves’ phase and amplitude. Owing to these characteristics, traditional plane-wave approximations are not valid in this region. The key characteristics of the reactive near-field region are described as follows:

• Energy: The energy is non-radiating and experiences rapid decay in field strength [42].
• Frequency: A higher frequency of operation extends the reactive near-field of the system [14].
• Wavefronts: Wavefronts are not entirely formed in this region and, therefore, cannot be approximated as planar waves.
• Power: The reactive near-field region has non-uniform power distribution with respect to spatial dimensions.
• Phase: The phase shifts in this region are non-linear and not approximated using simple linear equations.

The boundary of the reactive near-field region with the radiating near-field region is typically given by the Fresnel distance $d_f$, which is defined as follows [8]:

$$d_f=0.62\sqrt{{\displaystyle \frac{D^3}{\lambda }}},$$

(2)

where D is the aperture of the antenna array and $\lambda$ is the operating wavelength. This defines the maximum distance within which the electromagnetic fields are dominated by reactive energy storage instead of efficient radiation and propagation. This equation implies that large-aperture antennas (e.g., extremely large MIMO) have significantly extended non-radiating near-field range. It also implies that near-field modeling is particularly relevant in the design of millimeter-wave (mmWave) systems and terahertz (THz) communication systems, which have extremely high frequencies and, consequently, lower wavelengths.

2.1.2. Radiating Near-Field

The radiating near-field region is observed after the Fresnel distance. Here, the wavefronts are formed, and the energy does not oscillate back to the antenna source. Unlike the reactive near-field, the radiating near-field exhibits partial energy radiation. However, the wavefronts are spherical and exhibit a curvature. Hence, the phase variations across the aperture are nonlinear [14,43]. While the electromagnetic waves begin to radiate effectively, they do not behave as fully developed plane waves [41]. This enables energy focusing using beamforming, where energy can be concentrated at specific locations in space instead of angular directions. The key characteristics of this region are as follows:

• Energy: The energy is partially radiating.
• Frequency: A higher frequency of operation extends the radiating near field of the system [8,14].
• Wavefronts: Wavefronts are formed but cannot be approximated as traditional planar waves [14,44,45,46].
• Power: The radiating near-field region has uniform power distribution as the electromagnetic waves are not reactive.
• Phase: The phase shifts in this region are non-linear in relation to the antenna aperture [14].

The boundary of the radiating near-field and the far-field is defined by the Rayleigh distance $d_r$ [14]:

$$d_r={\displaystyle \frac{2D^2}{\lambda }}.$$

(3)

This distance describes the transition where wavefront curvature starts to diminish, and a planar wave approximation becomes increasingly valid. The equation implies that large-aperture antennas have far-reaching near-field regions. Furthermore, near-field modeling must be explored for mmWave communication, THz systems, and ELAAs to fully utilize the beamfocusing capabilities of near-field systems.

2.1.3. Far Field

The far field is the region where electromagnetic waves propagate as well-defined plane waves. In this region, the phase variations across the antenna aperture become linear, which enables simplified analytical modeling for signal processing. The plane wave approximations enable the modeling of antenna radiation patterns, as the radiated fields can maintain stable angular properties, which makes directivity and beamforming highly effective. In the far-field region, the signals are primarily defined by their angular properties. This is in contrast with the radiating near-field region, where both angle and distance influence the phase characteristics of the waves. The key characteristics of the far-field region are as follows:

• Energy: The energy is fully radiating, enabling the modeling of antenna radiation patterns.
• Frequency: A lower frequency of operation shortens the near-field region and, therefore, begins the far-field approximation closer to the antenna source.
• Wavefronts: Wavefronts are fully formed and can be approximated as planar waves [44].
• Power: The radiating near-field region has uniform power distribution as the electromagnetic waves are not reactive.
• Phase: The phase shifts in this region are linear with respect to the antenna aperture. This enables efficient beamforming.

Beyond the Rayleigh distance, electromagnetic waves exhibit predictable and stable propagation with planar wavefronts, making this region ideal for applications such as satellite communication, long-range radar, and far-field wireless sensing [14].

2.2. Challenges of Near-Field Systems

Researchers have recognized the limitations of using far-field plane wave approximations to accurately describe near-field electromagnetic phenomena. As a result, they are actively exploring alternative wavefront modeling approaches specifically designed for near-field communication systems. This ongoing research highlights the complexity of electromagnetic wave behavior near transceivers. Various modeling frameworks are continuously being developed and refined to better understand the dynamics of near-field propagation.

Electromagnetic wave propagation in near-field regimes is typically modeled using spherical wave representations, as demonstrated in recent studies [14,44]. These models take into account the intrinsic nonlinear phase characteristics that are evident in configurations near antennas, where spatial relationships significantly affect wave behavior. The multidimensional energy focusing capability of spherical wavefronts [44] enables optimization of beamforming across both angular and distance dimensions, offering potential advantages over traditional far-field techniques that are limited to angular beamsteering. These characteristics have inspired research into new communication paradigms that could theoretically improve spectral efficiency through multidimensional spatial resource allocation.

Recent research [46] has conducted an in-depth examination of spherical wave modeling in the context of near-field communication, revealing several practical challenges. Although the mathematical foundations of these models are robust, it is essential to acknowledge the heightened computational demands and the limited analytical simplicity associated with assessing array performance, particularly due to the distinctive characteristics of spherical waves emanating from diverse sources. Notably, complementary studies [45,46] have indicated that plane wave decomposition, despite its more complex parameterization compared to conventional far-field models, offers valuable insights into the capabilities of near-field communication. These findings suggest that utilizing plane wave representations may be advantageous in circumstances where a thorough understanding of field behavior and the preservation of computational efficiency are of paramount importance.

This paper primarily investigates electromagnetic phenomena in the radiating near-field region, where conventional uniform plane wave (UPW) approximations prove insufficient for characterizing wave propagation. The transition to spherical wavefront dynamics in this regime creates spatially dependent phase and amplitude variations governed by geometric relationships between transceivers. These distinctive propagation characteristics introduce complex spatial variations that necessitate advanced channel estimation, interference management, and hardware implementation, as shown in

Table 3. Challenges and opportunities in near-field MIMO systems.

Aspect	Challenges	Opportunities
Channel Modeling and Estimation	Spherical wave models complicate channel modeling and estimation. Distance-dependent channel variations increase and computational complexity increases.	Higher channel rank and distance-dependent channel variations enable higher multiplexing gains.
Beamforming and Beamtraining	Beamtraining is hard because the codebook size increases owing to the dual angular and distance sampling. The expanded codebook search space increases training overhead and computational complexity.	The dual angular and distance sampling creates beamfocusing opportunities that realize precise energy focusing in 3D spaces. This enables higher channel and system capacity.
Mutual Coupling Effects	Densely positioned antenna elements distort radiation patterns and reduce spatial multiplexing.	Advanced antenna isolation techniques can mitigate coupling and enhance system efficiency.
Interference Management	Complex interference patterns owing to the spherical wave model which require both angular and distance separation for interference mitigation.	Enhanced spatial DoF provide fine-grained control of interference and improve spectral efficiency.
Hardware Implementation	Higher computational complexity, which increases power consumption and requires high-precision RF chain components with extreme synchronization across large antenna arrays.	Energy-efficient hardware can enable feasible near-field signal processing and improve system performance.

. The details of the challenges are described as follows.

2.2.1. Near-Field Channel Modeling and Channel Estimation

Unlike conventional systems where all scatterers can be considered in either far-field or near-field regions exclusively, extremely large-scale (XL) MIMO channels exhibit a hybrid-field model where different scatterers simultaneously exist in different regions [47]. This hybrid near-field–far-field channel complicates channel modeling and necessitates advanced signal processing techniques. Composite channel models must be incorporated where planar wavefronts are considered for far-field components and spherical wavefronts are considered for near-field components of the channel [14,48]. In the near-field systems, the array steering vector becomes dependent on the distance between the base station and scatterers, in addition to the angle information. This increases the parameter space for channel estimation, increasing its computational overhead [44,48]. Furthermore, the non-uniform phase distribution across large aperture arrays immensely increases the computational complexity. In far-field models, channels exhibit sparsity in the angular dimension such that signals arriving at the receiver come from a limited number of dominant paths, enabling efficient sensing approaches [44]. However, near-field channels are not sparse, which introduces additional dimensions and increases complexity.

2.2.2. Beamforming and Beamtraining

In far-field systems, the codebooks for beamforming exclusively sample the angle with uniform angular grids. However, near-field systems depend on both the angular direction and the relative transmitter–receiver distance [8]. Therefore, codebooks must sample both angle and distance domains simultaneously, which dramatically increases the codebook cardinality. Furthermore, near-field codebooks often implement non-uniform grids to efficiently sample the expanded parameter space, which complicates the codebook design process. The expanded codebook directly impacts computational complexity and storage, creating implementation challenges for resource-constrained devices.

Owing to the expanded codebook, the beamtraining process becomes substantially more complex than its far-field equivalent. Traditional beamtraining methods become insufficient and suboptimal as they can only scan in the angular domain, whereas the near-field systems demand joint angle–distance searching. Sequential search approaches must be employed to efficiently navigate the expanded codebook parameter space. This massive increase in the search space consequences a higher training overhead and longer scanning times, which complicate system latency and throughput. Additionally, beamfocusing accuracy in the near-field region is sensitive to channel estimation errors. Minute errors in distance or angle estimation can lead to significant beam misalignment, which can critically impact system performance. Hence, robust training procedures need to be developed with enhanced error tolerance.

2.2.3. Mutual Coupling Effects

Mutual coupling refers to the electromagnetic interaction between the subsequent or neighboring elements of an antenna array. This effect is pronounced when the signals from one antenna element of an array induce undesired effects in nearby antennas. This alters the impedance characteristics and antenna radiation patterns, which leads to performance degradation. In near-field MIMO systems, the antennas are densely organized and placed close together, and therefore, mutual coupling effect is more pronounced [43,49,50]. This distorts beamforming, reduces spatial multiplexing, and introduces signal correlation, thus limiting overall system performance [51]. Mitigating this effect requires advanced antenna isolation techniques, which further complicate energy efficiency [49,50].

2.2.4. Interference

While far-field beamforming separates users and end devices through angular resolution, near-field systems require both angular and distance separation, which complicates interference management. The expanded spatial degrees-of-freedom (DoF) in near-field systems create opportunities for improved spatial multiplexing but introduce higher levels of interference owing to the spherical nature of wavefronts and the smaller separation between antennas, which create complicated interference patterns [49,50]. These patterns are more difficult to predict and mitigate compared to the more structured interference in far-field regions. This requires advanced interference management, which accounts for both angular and distance parameters of multiple end-devices simultaneously.

2.2.5. Hardware Implementation

Near-field MIMO systems have complex hardware requirements that go beyond those in conventional far-field systems. The increased computational complexity of near-field channel estimation, beamtraining, codebook design, etc. requires high processing power, which consequently increases hardware workload, increasing power consumption and thermal management requirements. Furthermore, near-field systems have higher precision requirements for phase and amplitude control across array elements, which necessitates improved RF chain designs with tighter tolerance. Additionally, signals received by different antennas can have very different power levels due to non-uniform radiation patterns in near-field region, and therefore, a higher dynamic range is required in RF components. For extremely large-scale arrays, the physical distribution of components introduces timing synchronization challenges that become more pronounced in near-field beamfocusing applications where precise phase relationships must be maintained across the antenna array aperture.

2.3. Opportunities in Near-Field Systems

Near-field MIMO provides significant opportunities that distinguish it from conventional far-field systems, as shown in Table 3. We describe the key opportunities in the following subsections.

2.3.1. High Spatial Multiplexing

The most significant advantage offered by near-field MIMO is the dramatic increase in spatial DoF, even in line-of-sight (LoS) scenarios. In wireless links, the rank of a channel matrix describes how many independent spatial data streams can be transmitted simultaneously without interference. Hence, a higher rank indicates that more independent transmission paths exist, enabling multiple data streams to be sent simultaneously. Due to distance-dependent variations in near-field systems, the channel matrix achieves a higher rank, thus enabling higher spatial multiplexing gains. The channels have increased DoF, which enable the simultaneous transmission of multiple data streams over a single path, dramatically improving spectral efficiency without relying on multipath propagation [44,48]. This represents a fundamental advantage over far-field MIMO systems, which depend on multipath propagation to achieve similar multiplexing gains.

2.3.2. Beamfocusing

Near-field MIMO systems provide a transformative approach to beamforming by enabling focal selectivity [8]. Unlike far-field systems, which can only direct energy toward specific angular directions by beamsteering, near-field beamfocusing with spherical wavefronts enables energy to be focused on specific spatial points with both angular and distance selectivity [8,47]. This three-dimensional beamfocusing ability creates unprecedented opportunities for communication and sensing. Even when multiple communication devices are placed at the same angles of incidence relative to the ELAA, near-field beamfocusing can separate them based on their different distances from the array. This indicates that near-field MIMO can support a higher density of users within a given coverage area in environments with limited angular separation between users. This ability can be a key-enabler of dense urban networks where users are frequently aligned along similar directions. The difference between beamsteering and beamfocusing is shown in Figure 4.

2.3.3. High-Precision Sensing

The larger antenna array aperture in near-field MIMO systems creates exceptional opportunities for wireless sensing applications. The spherical wavefront characteristics improve positioning performance by providing precise beamfocusing capabilities. Near-field MIMO also enables improved sensing of additional parameters, including velocity, orientation, and other geometric properties of targets-of-interest (ToIs). This is owing to the beamfocusing ability, which enables detailed parameter estimation compared to far-field equivalent systems. These abilities can enable advanced human–computer interaction, gesture recognition, and environmental monitoring applications.

3. Energy-Efficient Beamforming in Near-Field Systems

Energy-efficient beamforming is essential to activating the potential of near-field systems, as it enables precise signal focusing that enhances the quality and efficiency of wireless communication. In such applications, signals can be accurately directed, reducing interference and energy waste by concentrating transmission power exactly where it is needed. This precision is especially important in densely populated areas or environments with multiple devices, where efficient resource allocation significantly boosts system performance and lowers operational costs. Furthermore, energy-efficient beamforming supports sustainable practices by decreasing the overall energy consumption of wireless networks, contributing to global efforts to reduce the technology sector’s carbon footprint.

Beamforming is a multiantenna signal-processing technique that has become an integral component of 4G and 5G systems. It is expected to be included in the 6G systems, owing to their stringent quality-of-service demands. It enables precise directional control of RF waves such that the signal energies can be directed in the desired directions, thus providing enhanced receive signal quality [52], interference management [53], and spectral efficiency [54]. Beamforming utilizes multiple antenna elements and strategically manipulates the phase of individual signals at each individual antenna such that the system can form and steer beams at specific angles towards intended receivers while minimizing interference to other directions [52]. This directionality enables higher spatial resolution and fundamentally transforms how wireless signals propagate through space [54].

Different types of beamforming techniques can be realized in wireless systems, which provide different trade-offs between performance, computational complexity [55], flexibility [56], power consumption [57], cost [52], and hardware implementation [58]. Analog beamforming is the most power-efficient implementation of beamforming but provides the lowest flexibility in hardware implementation and has limited performance gains [59]. In analog beamformers, phase shifters directly manipulate the RF signals at each antenna element before they reach the power amplifiers. While analog beamforming provides spatial multiplexing gains, it has limited flexibility in beam pattern formation and can typically only support a single data stream. Hence, most purely analog beamformers can only create one directed beam of energy towards the desired angular orientations [29]. While this simplicity limits the performance gains, the hardware implementation is the least costly and can be of immense use to large antenna array systems. However, analog beamforming struggles to adapt to complex channel conditions and cannot support multiple users using the same time and frequency resources. Nevertheless, its power efficiency makes it a valuable component in hybrid beamforming architectures.

In contrast, digital beamforming provides maximum beamforming control at the cost of higher power consumption and complicated hardware architectures, which increase capital and maintenance costs [29]. In digital beamforming, each antenna element is connected to its own dedicated RF chain with an analog-to-digital converter (ADC) or digital-to-analog converters (DAC). This enables precise manipulation of the signals, which realize sophisticated beam patterns and provide simultaneous service for multiple data streams using the same time–frequency resources. The separation of each antenna’s RF chain provides remarkable beamforming flexibility and enables the formation of multiple beams. Hybrid beamforming has emerged as a compelling solution to the performance–cost trade-off. It combines analog and digital beamforming such that antenna elements are directly connected to phase shifters, but groups of these phase shifters are connected to dedicated RF chains. This substantially decreases hardware costs, computational requirements, and power consumption while still supporting multiple spatial streams. Despite the limited number of beams as compared to digital beamforming, hybrid beamforming has become the architecture of choice for 5G systems.

Unlike far-field MIMO systems, which assume the plane wave model, the spherical wave model of near-field systems complicates system design and energy efficiency. As near-field systems enable precise beamfocusing, they can provide higher spatial multiplexing gain, which can provide energy-efficient beamforming solutions. However, the precise beamfocusing action requires accurate channel state information, which is computationally complex in near-field systems. Hence, near-field systems require a precise balance of performance gains and energy consumption. This section provides a comprehensive account of energy-efficient designs in near-field systems.

3.1. Energy-Efficient Techniques in Near-Field MIMO

We offer a comprehensive discussion of energy-efficient techniques in near-field systems, as detailed in the next subsections.

3.1.1. Codebook Design and Beamtraining

Conventional codebook designs are not valid in near-field systems as the spherical wavefront is not negligible. Since near-field channel models do not have the plane wave assumption, the far-field codebooks that utilize angular sparsity are inadequate. Near-field systems require advanced codebook designs with joint angle–distance modeling, which immensely increases the computational complexity of beamtraining. To enable energy-efficient near-field MIMO systems, an efficient design of codebooks is imperative as it directly impacts beamforming gains. The increase in the number of antennas has resulted in the transition of EM waves from the FF to the NF. NF effects have higher computational costs owing to the large beamtraining overheads in angular and distance domains. Therefore, efficient beamtraining methods are imperative for sustainable NF MIMO system designs. Beamtraining is the process of finding the best beam alignment to maximize signal strength and communication efficiency. The transmitter utilizes different predefined beam patterns, or codewords, sequentially, and the receiver measures the signal strength for each codeword and reports the best-performing codeword back to the transmitter. Based on this feedback, the transmitter selects the optimal beam pattern that provides the strongest received signal. Each codeword represents a possible beam configuration in terms of both angle and distance. The latest works on near-field codebooks and beamtraining designs are summarized in

Table 4. Different types, associated works, methodologies, and motivations of near-field codebook design and beamtraining.

Type	Work	Methodology	Motivation
Codebook Design	[60,61,62,63,64,65,66,67]	NF Wavefront Modeling, DFT-Based Beam Optimization, and Beam Gain and Correlation Control	Conventional codebook designs are not valid in near-field systems as the spherical wavefront is not negligible, and the computational overhead is infeasible.
Beamtraining designs	[68,69,70,71]	Hybrid-field Training, Learning-Driven Codeword Optimization, and Vision-Aided Multimodal Sensing	Conventional beamtraining algorithms do not consider the range domain, and exhaustive search algorithms have high pilot overheads.
Hierarchical designs	[18,72,73,74,75,76,77]	Hierarchical Spatial Partitioning, DFT Off-Grid Hybridization, and Modular Leakage-Aware Optimization	Practical environments cannot have exhaustive angle and range designs. These domains should be resolved independently to enable feasible overheads.
Deep learning	[28,78,79]	Neural-Driven Training, Cross-Signal Codeword Estimation, and Hybrid Architecture Optimization	Conventional beamtraining and codebook designs can have high processing and pilot overheads. Data-driven and reinforcement-learning-based designs can mitigate these challenges.
Wideband, THz, and mmWave designs	[80,81,82,83,84,85]	Beamsplit Exploitation, Sparse Codebook Design, and Hybrid Domain Training	THz, mmWave, and wideband systems have non-negligible near-field effects. Conventional beamtraining and codebook designs are not feasible for these systems.
Sensing-based designs	[86,87]	Position-Aware Bayesian Adaptation and Active Sensing Optimization	The range information of near-field systems can assist in codebook design and beamtraining.

and described as follows:

Codebook Design.

Efficient codebook design was also explored in [60] for XL-MIMO technology. The authors utilized the eigenproblem based on the near-field electromagnetic wave transmission model and derived the general form of the eigenvectors associated with the near-field channel matrix. Based on the proposed codebook, they formulate a two-stage channel estimation scheme. Their scheme shows superior sparsification performance with lower leakage effects. As the 5G New Radio (NR) high-precision Type II codebook is not optimized for near-field characteristics, to satisfy the backward compatibility requirements of 3rd Generation Partnership Project (3GPP) releases, the authors in [61] formulated a plane wave expansion (PWE) method of focused beams. This method enables the generation of focused beams through the weighted superposition of multiple DFT beams. They formulate additional feedback contents with the NR Type II codebook, which accurately represents the near-field LoS CSI for feedback. Their results show that the proposed codebook achieves a performance within 2 dB of the optimal beamfocusing scheme. To address the problem of codebook mismatch between far-field and near-field systems, the authors in [62] designed a criterion of codebook design that maximizes the worst-case beam gain within the beam coverage. They also provided a closed-form expression of a near-field, full-dimensional codebook, which can enable spatial oversampling regardless of the number of antennas at the transceivers. Their results demonstrate that due to superior beam gain, the optimal beamfocusing scheme enhances accuracy in near-field beamtraining compared to existing codebooks, potentially achieving greater energy efficiency with the same energy usage.

To address the limitations of traditional codebooks, the authors in [63] proposed innovative codebook schemes based on the fitting formula of codewords’ quantization performance. They examined the quantization performance of codewords from uniform linear array (ULA) and uniform planar array (UPA) configurations. Subsequently, they proposed a ULA uniform codebook designed to maximize the minimum correlation, where correlation refers to the similarity between different codewords within the codebook. High correlation among codewords can lead to insufficiently distinct beams, resulting in quantization errors. Additionally, they introduced UPA uniform and dislocation codebook schemes. They showed that oversampling in the angular domain achieves higher accuracy and minimizes overhead in near-field channels. This approach contributes to energy efficiency by optimizing the use of resources, as the improved accuracy reduces the need for additional power to correct errors or compensate for signal degradation.

To fully utilize the higher spectral efficiency gains in near-field systems, the authors in [64] studied the system with an extremely large cylindrical antenna array (CLA). They exploited the geometric relationship between CLA and the user with the spherical-wavefront model and analyzed the beamforming gains in the elevation, azimuth, and distance domains. They also proposed a 3D CLA codebook to make beamfocusing more effective and efficient. Their results demonstrate that the proposed codebook can improve the system’s achievable rate. Similarly, in [65], the sampling method was used in each domain by controlling the correlation between different codewords to develop a 3D codebook for a near-field CLA system. DFT-based codebook design for near-field systems was also explored in [66]. Their results showed a decrease in near-field beamtraining overhead. A sparse DFT-based codebook was also proposed in [67]. The authors proposed a three-phase beamtraining scheme, which can enable 98.67% beamtraining overhead reduction as compared to the exhaustive-search scheme, which can improve energy efficiency.

Beamtraining Designs.

There are advancements in beamtraining techniques that not only enhance the performance and accuracy of communication systems but also contribute to energy efficiency. By reducing overhead and optimizing resource utilization, these methods lower the energy consumption required for effective beamforming, making the systems more sustainable and cost-effective.

In [68], the authors investigated the feasibility of utilizing far-field beamtraining for cross-field beam alignment. Initially, they demonstrated that far-field training can achieve an adequate signal-to-noise ratio (SNR) in both far-field and near-field scenarios. To enhance beam alignment precision, they proposed a two-phase angle and distance beam estimator (TPBE). Their method showed a 0.2% training overhead compared to a near-field exhaustive search. Beamtraining for XL-MIMO arrays was investigated in [69]. Due to an increase in the number of antennas resulting in excessive pilot overhead, efficient near-field beamtraining methods are imperative for the feasible functioning of next-generation systems. Beamtraining is further complicated by the expected high density of networks, which can have similar user channels that need to be considered for beamforming. Furthermore, the interference caused by the beams leaking to users in the same direction as the intended user needs to be addressed. To mitigate these issues, they proposed a graph neural network for the selection of the near-field codeword. This codeword was exploited for beam allocation, which consequently formulated the beamfocusing vectors. Their simulation results showed that the algorithm significantly reduces the pilot overhead and approaches the performance of an exhaustive search using 7% of the overhead, which could enhance energy efficiency.

As the problem of near-field beamtraining in XL-MIMO systems is challenging due to the need to consider both angular and distance information, the authors in [70] proposed a deep-learning-based approach to address this issue. They developed a deep neural network (DNN) that performs beamtraining using a near-field codebook containing both angular and distance information, allowing for the joint prediction of the optimal angle and distance. Additionally, they introduced a scheme with supplementary codewords to further enhance beamtraining performance by selecting additional codewords based on the probability vector output from the DNN. Their results demonstrated that the proposed schemes perform better than existing methods under the same training overhead. Specifically, the supplementary codewords scheme significantly improves the accuracy of codeword prediction by utilizing supplementary codewords, leading to higher normalized gain and achievable rate than previous approaches. The authors concluded that their approach effectively reduces the overhead associated with beamtraining in XL-MIMO systems while maintaining high performance with improved energy efficiency, and they plan to extend their work to multiuser scenarios in the future.

As the problem of high training overhead in XL-MIMO systems persists, the authors in [71] proposed a novel beamtraining scheme that leverages visual image information to aid in the process. They developed a vision image-aided beamtraining framework that integrates YOLOv5 and ResNet18 networks. The YOLOv5 network was used for object detection to extract the size and location information of mobile vehicles (MVs), while the ResNet18 network inferred the optimal beam index based on this extracted information without occupying in-band resources. Their results demonstrated that the proposed vision image-aided beamtraining scheme significantly outperforms traditional methods by improving training accuracy and reducing beamtraining overhead. The authors conducted simulations using a high-fidelity synthetic dataset that combines wireless and visual data, showing that the proposed scheme can achieve optimal codeword selection without consuming in-band resources. Additionally, the accuracy of the proposed scheme reaches 85% with the full training dataset, and the top-3 accuracy approaches nearly 100%. This indicates that the proposed method is highly effective in multivehicle scenarios, transforming the multiuser beamtraining problem into a single-user problem, which enhances performance along with the energy consumptions.

Hierarchical Designs.

The spherical wavefront properties of near-field MIMO systems can facilitate the identification of user locations within the Fresnel region. In these systems, beams can be directed and focused in both angle and distance dimensions, leading to large codebooks and extended beamtraining durations. To reduce the beam alignment (BA) overhead, the authors in [18] proposed a hierarchical codebook and BA scheme for near-field XL-MIMO systems using a novel spatial partition that accounts for the angle-offset effect. In this effect, the optimal beam direction changes with distance due to the curvature of near-field system wavefronts, thereby enhancing beam gain within the coverage area. The authors designed distance-coarse and focusing beams to construct high-level and low-level codebooks, respectively. The high-level codebook aids in angle dimension alignment, while the low-level codebook assists in distance dimension alignment. Utilizing the proposed codebook structure, they developed a two-stage tree search for BA, consisting of angle and distance aligning stages. The proposed codebook design effectively minimizes the BA error rate. Numerical results indicated that the scheme operates with only one percent of the overhead, achieving a lower BA error rate compared to exhaustive searching methods. Additionally, hierarchical codebook-based beamtraining was explored in [72] to further reduce beamtraining overhead. The authors addressed a hybrid near–far-field communication scenario and developed a two-stage hybrid field beamtraining scheme based on rough beam sweeping. They first obtained the 3 dB spatial angle width, which distinguishes between near-field users (NUs) and far-field users (FUs), and then utilized the hierarchical codebook to refine the beamtraining process. Their results show that hybrid field beamtraining can achieve near-field performance with significantly lower training overhead, which could improve energy efficiency.

In [73,74], the authors proposed a hierarchical beamtraining scheme to reduce the computational overhead without the need for extra hardware. They first devised the multiresolution codewords covering different angles and distances and then propose a Gerchberg–Saxton (GS)-based algorithm to acquire the theoretical codeword. This theoretical codeword provides the practical codeword using alternating optimization. In [77], the authors introduced a steering beam gain approximation method and designed a codebook to cover the Fresnel region. The Fresnel layer codebook operates as a lower-layer space, concentrating solely on the near-field region. In contrast, the upper-layer codebook employs beam rotation and relocation to position the beam pattern at a target location. Their hierarchical codebook design achieved higher average beamforming gains compared to conventional far-field codebooks. Their results show that the proposed scheme can achieve suboptimal performance without infeasible hardware costs, improving the near-field system energy efficiency.

As codebooks are coupled with both angles and distances of scatterers in near-field systems, they have unfavorable computational complexity and pilot overheads. To mitigate this complexity, the authors in [75] introduced a novel two-stage learning-based beamtraining protocol that independently manages angles and distances. They then presented a hierarchical codebook design with modular codeword units to prevent power leakage. Their results showed that the proposed algorithm is computationally simple and effectively alleviates power leakage. As the increased size of antenna arrays has made the planar wave assumption impractical, the design of near-field codebooks that consider the spherical wavefronts is imperative. Traditional near-field codebook designs typically rely on dedicated polar-domain codebooks and on-grid range estimation, leading to significant training overheads and high storage demands. To address this challenge, the authors in [76] proposed efficient beamtraining schemes utilizing off-grid range estimation with conventional discrete Fourier transform (DFT) codebooks. They introduce two schemes to jointly estimate user angle and range using the DFT codebook. The first scheme focuses on estimating user angles, while the second scheme estimates user range by minimizing the power ratio mean square error (MSE). Simulation results showed that the proposed method reduces overhead in near-field beamtraining and improves range estimation accuracy, and thereby, they can achieve high throughput for the same energy costs.

Deep-Learning-Based Designs.

The increase in the number of antennas of extremely large antenna arrays and the use of higher frequency signals greatly increases the reach of the radiative near field. The addition of the distance component in near-field makes the beamtraining operation extremely cumbersome. To mitigate the computational complexity, in [28], the authors used convolutional neural networks and historical data to optimize the beamforming vector to maximize the achievable rates. This technique substantially reduces the beamtraining overhead. The effectiveness of XL-MIMO systems is highly dependent on the efficient design of near-field codebook and beamtraining. To enable effective utilization of resources, the authors in [78] introduced a deep-learning-based beamtraining scheme that leverages the near-field channel model and corresponding codebook. This approach utilizes far-field wide beam signals to estimate the optimal near-field beam. They proposed two distinct schemes: the initial scheme directly estimates the near-field codeword using neural networks, while the enhanced scheme incorporates additional beam testing for refinement. Their findings demonstrate a substantial decrease in training overhead, which could lead to overall energy savings due to increased energy efficiency.

Beamtraining for mmWave ELAAs was explored in [79]. The authors utilized sub-6 GHz uplink pilot signals to estimate the optimal near-field codeword, thereby reducing pilot overhead. Additionally, they employed an innovative neural network architecture to improve the accuracy of beamtraining. The proposed network incorporates convolutional neural network (CNN) and graph neural network (GNN) modules to account for correlations between signals from different users and across subcarriers in orthogonal frequency division multiplexing (OFDM) systems. Numerical results demonstrated the superior performance of the proposed scheme compared to existing deep-learning-based schemes, which can improve energy efficiency.

Wideband, THz, and mmWave Designs.

Wideband XL-MIMO further complicates beamtraining and introduces a significant near-field beamsplit effect, which results in performance degradation. In [80], the authors demonstrated that this beamsplit effect can contribute to fast beamtraining in NF systems. They first utilize time delay beamforming to control the beamsplit effect such that the beams at different frequencies can flexibly be focused in the desired locality. This controlled beamsplit was exploited to develop a fast beamtraining scheme that generates beams focusing on a variety of locations at different frequencies. The simulation results demonstrated low training overheads using these near-field rainbow effects. The problem of high overhead wideband beamtraining was also investigated in [81]. The authors proposed a distance-dependent, beamsplit-based beamtraining method that reduces the training overhead. They first discussed the distance-dependent beamsplit effect, which enables simultaneous scans of the angular domain in multiple distance rings using beams at different frequencies. Leveraging this, they proposed a training method in which both angles and distances can be searched in one time slot. This significantly reduces the pilot overhead. Their results showed improvement in training overhead and beamforming gain, leading to better energy efficiency.

Beamtraining is an efficient beamforming strategy for THz communications; however, owing to the near-field effects, the codebook size increases infeasibly. To mitigate this, the authors in [82] studied a uniform circular array (UCA)-based THz near-field communication. They first proposed a model for characterizing the beam pattern of UCA and utilized it to develop a codebook design for near-field beamtraining. Numerical results showed that the proposed codebook outperforms conventional ULA-based beamtraining algorithms. To mitigate the beamtraining latency of THz systems, the authors in [83] proposed a subarray-based near-field beamtraining method that estimates the user location by performing frequency-dependent training. They simultaneously scanned multiple directions by generating multiple frequency-dependent beams and then estimating user angle and distance from the subarrays. Their scheme can provide 28% improvement in the data rate and 99% reduction in the latency, which could enhance the energy efficiency.

To overcome the high complexity and low identification accuracy of current beamtraining techniques, the study in [84] investigated mmWave near-field beamtraining. The authors developed a sparse polar-domain sparsity-based basis and utilized random hash functions to construct the near-field beamtraining codebook. This codebook was employed in a voting-based decision-making process to achieve correct beam alignment. The results indicated that the proposed scheme achieves 96.4% of the identification accuracy of exhaustive beamtraining while significantly reducing overhead to a logarithmic level. Downlink beamtraining of mmWave and THz systems was also investigated in [85]. The authors exploited hybrid far–near-field channels and obtained the best beam alignment on a two-dimensional angle-range domain. They utilized the approximate orthogonality of near-field steering vectors at the same effective distance and formulated multidirectional beamtraining, which can scan the space more efficiently. Their results showed promising advantages over state-of-the-art hybrid schemes in terms of performance and generality with improved energy efficiency.

Sensing-Aware Designs.

An active sense-then-train (STT) beamtraining method was proposed for near-field MIMO systems in [86]. The proposed method utilizes the additional degrees-of-freedom provided by the near-field spherical channel to improve the beamtraining process. Numerical results showed that STT can enhance the beamtraining performance in the near-field compared to the conventional far-field codebook-based methods and can perform fast and low-complexity beamtraining. A novel position-aware beamtraining algorithm was introduced for mmWave XL-MIMO systems in [87]. The authors used position information to reduce the beam search space and then employed an iterative measurement strategy that updates the polar-domain codebook’s posterior distribution using Bayes’ rule and dynamically adjusts the candidate beam until the termination condition is met. Their method achieved higher normalized beamforming gain, leading to better spectrum efficiency and energy efficiency.

3.1.2. Beamfocusing

Optimized beamfocusing is crucial for near-field MIMO systems as it enables precise energy concentration and spatial multiplexing gains, which optimizes power consumption. As near-field systems have more spatial degrees of freedom than far-field systems, precise precoding is challenging and experiences high computational overhead. State-of-the-art works are summarized in

Table 5. Different types, associated works, methodologies, and motivations of near-field beamfocusing.

Type	Work	Methodology	Motivation
Designs for sensing and communication	[37,88,89,90,91,92]	CRB-Driven Sensing-Comms Integration, Sparse Transceiver Architectures, Multidimensional Near-Field Differentiation, SINR-QoS Co-optimization Frameworks, Hybrid Antenna Optimization, and Predictive Beamforming	Integrated sensing and communication systems experience higher levels of interference and demand precise beamforming for successful dual functioning.
Near-field precoding	[93,94]	NOMA-Based Hybrid Field Coexistence, HMA Array Design and Precoder-Weighting Optimization, and Iterative Algorithms for HMA Optimization	Precise beamfocusing is challenging in near-field systems owing to the joint angle-range dependence of the channel, which results in high computational complexity.
Wideband, THz, and mmWave designs	[3,23,95,96,97]	Wideband Phase-Delay Correction, Frequency-Adaptive RIS Architectures, THz UM-MIMO System Design, and Decomposed Optimization Frameworks	Designing beams for mmWave and THz systems is challenging owing to the wide bandwidth of these systems. The wideband links may have optimized beams at the center frequency but have poor performance at frequencies that are further from the central frequency.
Physical layer security	[98,99,100]	Near-Field Security Parameter Analysis, Hybrid Beamforming for Secure Transmission, Distance-Domain Security Exploitation, and Eavesdropper Antenna Impact Mitigation	The high spatial multiplexing gains of near-field systems can enable advanced security features such as increased eavesdropping protection and jamming using precise beamforming.

.

Designs for Sensing and Communication Systems.

Integrated sensing and communication (ISAC) has been identified as a key use case for 6G systems. The dual function of ISAC systems increases the workload on the transceivers and necessitates power-efficient designs [88,101,102]. In [37], the authors explored NF beamforming in ISAC systems using XL arrays. They derived loss expressions caused by by far-field and near-field mismatches. They then analyzed the loss of performance caused by the traditional FF models in the NF for sensing and communication. They formulated an NF beamforming problem to maximize the sensing Signal-to-Interference-plus-Noise Ratio (SINR) while ensuring the required quality of service (QoS) and total power constraint. They utilized SDR and generalized Rayleigh entropy theory to solve the non-convex problem. Their results validate the effectiveness of the proposed scheme, demonstrating its capability to manage co-angle interference in sensing and communication. In [89], the authors introduced a novel near-field velocity sensing technique that enables the simultaneous estimation of both radial and transverse velocities of a moving target in a mono-static setup. Radial velocity measures how fast the target is approaching or receding from the station, while transverse velocity, perpendicular to radial velocity, indicates lateral movement. They used a maximum-likelihood-estimate (MLE) approach to estimate these velocities from echo signals, where MLE is a statistical method for estimating unknown parameters by maximizing the likelihood that the observed data fits a particular model. Additionally, they developed a predictive beamforming framework for moving communication users that eliminates the need for channel estimation. Their results showed that the proposed method is strong and adaptable to uncertainties, enabling the system to sustain throughput without using additional energy.

As the problem of integrating sensing and communication functions in wireless networks becomes increasingly important, the authors in [90] proposed a near-field ISAC framework, incorporating an additional distance dimension for both sensing and communication purposes—an aspect absent in traditional far-field systems. This framework involves deriving the Cramér–Rao bound for joint distance and angle sensing, which is minimized subject to a minimum communication rate requirement for each user. The authors explored both fully digital antennas and hybrid digital and analog antennas, obtaining a globally optimal solution for the former and a high-quality solution through two-stage optimization for the latter. Their results demonstrated that the additional distance dimension in the near-field ISAC framework provides a performance gain over far-field ISAC systems. The numerical results confirmed the effectiveness of the proposed framework, showing that even at high minimum communication rates, the root of the Cramér–Rao bound (RCRB) remains low, indicating a successful integration of near-field sensing and communications. The study also highlighted that hybrid digital and analog antennas, while reducing sensing performance compared to fully digital antennas, offer lower power consumption. In simple terms, the paper addressed the challenge of enhancing ISAC systems by leveraging near-field effects, which allow for joint distance and angle estimation, unlike far-field systems that only estimate angles. This approach minimizes interference between users by differentiating them based on distance, even when they are aligned in the same direction. The authors suggested that near-field ISAC systems can greatly benefit from incorporating this additional distance dimension, pointing to a promising direction for future research in wireless networks. By avoiding the need for extra energy to mitigate interference, this method could enhance overall energy efficiency.

In response to the growing issue of spectrum congestion caused by the rapid expansion of wideband communication, the authors in [92] proposed an innovative framework for wideband near-field ISAC that features a sparse transceiver design. This framework enables simultaneous multiuser downlink communications and multitarget localization. The authors derived the Cramér–Rao Bound (CRB) for direction-of-arrival (DOA) and distance estimations, with the goal of minimizing it while ensuring the QoS for all users is maintained. They also optimized the sparse transceiver array and the precoding matrix to reduce mutual coupling and system overhead. Their findings indicated that the proposed framework effectively supports both communication and sensing functions. The sparse transceiver design enhances sensing accuracy without compromising communication performance. The authors demonstrated that the wideband near-field ISAC design can balance communication and sensing by adjusting the communication SNR threshold. Notably, the sparse transceiver design significantly improves estimation accuracy, especially when using a smaller number of selected antennas. The study concluded that the proposed wideband near-field ISAC system provides an effective solution for future wireless networks. It addresses challenges related to large arrays and high frequencies in the near-field region. With accurate estimations, the system can focus beams on the desired positions precisely, thereby reducing energy waste and improving energy efficiency.

Near-Field Precoding.

Non-orthogonal multiple access (NOMA) is being explored in the context of its higher spectral efficiency. In [93], the authors explored near-field beamforming to not only serve the near-field region but also serve far-field users using the same preconfigured beams. The sum rate of the far-field users is maximized while fulfilling the quality-of-service requirements of the users in the near field. Their results showed that coexistence between near-field and far-field communications can be effectively supported via NOMA. The low cost and high energy efficiency of holographic metasurface antennas (HMA) make them a viable candidate for fulfilling the requirements of extremely large-scale MIMO systems. The authors in [94] investigated three different HMA-based arrays to minimize the total transmit power of each array while fulfilling a specific SINR requirement by optimizing the transmit precoders and HMA weighting matrices. The authors found that, in the special single-user scenario, the problem can be decomposed and solved using successive convex approximation to get the optimal solution. They showed that beamfocusing can be achieved using HMAs and a successive convex approximation alternating method of multipliers based alternating optimization can be used to optimize the multiuser case. They showed the superiority of their method compared to other benchmark schemes and also show that the MA-based array has lower power consumption and hardware cost compared to conventional schemes.

Wideband, THz, and mmWave Designs.

A 5G radio utilizes phased arrays to direct energy focusing towards specific directions and spatial regions. A phased array consists of multiple antennas, where the phase of the signal at each antenna element is precisely controlled. This allows for dynamic beamforming and electronic beamsteering, enabling real-time directional control of transmitted or received signals without mechanical movement. As a result, phased arrays are crucial for high-rate systems. While phased arrays effectively support far-field beamforming, designing beams for millimeter-wave and terahertz systems presents challenges due to their wide bandwidth. These wideband links may exhibit optimized beams at the center frequency but perform poorly at frequencies that deviate from it. This leads to a misfocus effect in near-field systems and a beamsquint effect in far-field systems. In far-field applications, different frequency components of the signal undergo varying phase shifts in wideband systems, causing the beam to point in slightly different directions for different frequencies. This is called the beamsquint effect. Conversely, in the near-field region, there is a misfocus of energy as the beam does not correctly converge at the intended receiver distance. To address these issues, the authors in [95] proposed a low-complexity beamforming technique for massive wideband phased arrays called InFocus. This method effectively mitigates both the beam misfocus and beamsquint effects in near-field and far-field systems. InFocus employs a spatial phase modulation function to counteract the misfocus effect, resulting in larger and more uniform beamforming gains in wideband systems, which ultimately leads to improved system throughput with the same amount of energy used.

The authors investigated a near-field wideband beamforming scheme for an RIS-aided MIMO system in [96]. They proposed a deep-learning-based end-to-end optimization framework to maximize the spectral efficiency of the system. A frequency-dependent hybrid precoding architecture is utilized to mitigate the near-field beamsplit effect using time delay units. Two RIS architectures are used to realize the frequency-dependent passive beamforming at the RIS. True time delay RIS (TTD-RIS) introduces true time delay units at each element of the RIS to precisely control the phase shifts for different frequency components. Virtual subarray-based RIS (SA-RIS) divides the RIS into multiple subarrays and applies a common phase shift to all elements within a subarray. Their results demonstrate the tradeoff between the beamforming gain and the hardware complexity of the different RIS architectures and have superior performance gains compared to conventional benchmarks. As large antenna arrays and high bandwidth systems suffer from wideband and near-field effects, the authors in [3] proposed a low-complexity frequency-aware beamforming solution for time-delay-based and phase-shift-based structures. The problems are decomposed into two subproblems to reduce the complexity. An online learning algorithm and a geometry-based algorithm are used to find optimal phase shift and time delay values. Simulation results show that this framework provides robust performance across a wide frequency range. THz communications are envisioned to be a key enabler of next-generation high-speed communication. While it provides immense data rates, it suffers from severe free-spreading and molecular-absorption losses, which limit the transmission distance. Ultra-massive MIMO can be utilized to generate high-gain directional beams to mitigate the effects of these losses. The authors in [3] discussed the system models and basic principles of ultra-massive MIMO beamforming with far-field and near-field assumptions in THz. They also discuss beamtraining and codebook design for THz UM-MIMO systems.

The beamsplit effect, arising from the extremely large array aperture and wide bandwidth of next-generation systems, results in beams at different frequencies being focused on different spatial locations, leading to a significant performance loss. To address these issues, the authors in [23] employed a piece-wise-far-field channel model to approximate the near-field wideband channel. They divided the ELAA into smaller subarrays, each modeled using a planar wave approximation. They then introduced a phase-delay focusing (PDF) method utilizing a delay phase precoding (DPP) architecture, which compensates for the phase discrepancies within the subarrays (intra-array) and between the subarrays (inter-array). PDF adjusts the phase shifters and time delays to ensure that signals at different frequencies remain focused at the same physical location. The DPP hardware architecture implements the PDF method by applying phase shifts to address intra-array phase differences and using time delays to keep signals synchronized in both phase and time across all subarrays. Their results show the effectiveness of the proposed PDF method in reducing the near-field beamsplit effect, which could enhance the system’s energy efficiency.

Physical Layer Security.

In [98], the authors explored the physical-layer security afforded by a particular effect, analyzing its impact on the power ratio between legitimate and malicious users while considering near-field effects, distance, and direction. They demonstrated that operating in the near-field can significantly enhance jamming rejection and the secrecy rate. To address the challenge of securing near-field communications against eavesdropping, the authors in [99] proposed an innovative secure transmission framework that utilizes beamfocusing to boost the secrecy capacity in near-field MIMO systems. They implemented a hybrid beamforming architecture at the base station to transmit confidential information to legitimate users while minimizing the risk of interception by eavesdroppers. This approach involves a two-stage algorithm: the first stage optimizes fully digital beamformers, and the second stage alternates between determining optimal analog beamformers and baseband digital beamformers using closed-form expressions. The results showed that the proposed hybrid beamforming scheme performs comparably to fully digital strategies, with secrecy performance mainly influenced by the distance between the eavesdropper and the legitimate user rather than the angle relative to the base station. Numerical results confirmed the scheme’s effectiveness, demonstrating that the near-field spherical-wave channel model offers new distance-domain security gains not present in far-field communications. Furthermore, the study reveals that increasing the number of antennas at the eavesdropper degrades network secrecy performance by enhancing its reception capability. Since this method provides enhanced security gains, it could improve system throughput, thus increasing overall system energy efficiency.

3.2. Challenges of Energy Efficiency in Near-Field MIMO

Near-field systems exhibit complex interactions with electromagnetic wavefronts. The key challenges for energy efficiency are described below:

3.2.1. System Scalability

The near-field region has become far-reaching, owing to the massive scale of antenna arrays, which have large physical dimensions. These arrays have thousands of elements and consume more power due to more active RF components and signal processors. Increasing the number of antennas increases the static power draw, which is the baseline power consumption of a wireless system without any active transmission or reception. A static power draw is crucial to keeping essential system components operational. This makes the design of sustainable NF systems challenging.

3.2.2. Dynamic Beamforming

The higher spatial resolution of near-field beamfocusing enables precise energy concentration at specific spatial coordinates, which enhances communication efficiency and interference management. However, this beamfocusing requires real-time control of precoders to account for the high variability of channel conditions caused by environment conditions and user mobility. Unlike far-field beamsteering, near-field beamfocusing must dynamically optimize both angular and distance parameters, which significantly increases computational complexity. This control creates a higher energy consumption footprint and makes sustainable NF designs challenging. Therefore, efficient beamforming algorithms are required to reduce computational energy overhead while maintaining system performance.

3.2.3. Processing Overhead

Since conventional far-field channel models rely on angular sparsity, the signals are assumed to arrive from a limited set of angular orientations [44]. Near-field propagation couples angles with distances and creates a parameter space expansion such that each channel path requires joint angle–distance estimation, which exponentially increases the search space. This increases the processing overhead, which consequently increases the energy consumption. Efficient parameter estimation methods must be integrated to develop feasible and scalable NF systems.

4. Efficient Channel Estimation Techniques for Near-Field Communication

As a necessary step in beamforming, highlighting the need for improved channel estimation in near-field communication is crucial because traditional far-field estimation methods rely on channel sparsity in the angle domain and use array steering vectors based solely on angular directions. These methods assume planar wave propagation, simplifying channel modeling, but this assumption becomes less accurate as array sizes increase and the near-field extends further. In near-field scenarios, sparsity occurs in both the angle and distance domains, necessitating more advanced estimation techniques to reconstruct the channel accurately. In addition, XL-MIMO systems may involve scatterers located in both far-field and near-field regions simultaneously, creating a hybrid channel model that adds complexity to the estimation process. Accurate channel estimation is essential for achieving high spectral efficiency in XL-MIMO systems, as beamforming performance is directly tied to the precision of channel state information (CSI) estimation. Without precise channel estimation, beams cannot effectively adapt to the spherical wavefront characteristics of near-field systems, leading to suboptimal energy focusing and diminished spatial multiplexing gains. We summarize the latest works that investigate near-field channel estimation in

Table 6. Different types, associated works, methodologies, and motivations of channel estimation.

Type	Work	Methodology	Motivation
Near-field	[44,103,104,105,106]	Geometry-Aided Pilot Reduction, Unsupervised Bayesian Learning, Sparsity-Driven Polar-Domain Estimation, Parametric MUSIC Enhancement, and Optimized Codebook Algorithms	Accurate channel estimation is necessary to enable high spectral efficiency of near-field systems.
Hybrid-Field	[48,107,108,109]	PD Mitigation, Dynamic Subarray Modeling, Hybrid-Field Component Separation, and Neural Network Joint Optimization	Conventional near-field or far-field only channel estimation methods are not suitable for hybrid near–far field communications realized by large-scale antenna arrays where the users and scatterers are randomly situated in both near-field and far-field zones.
NLoS/LoS	[110,111,112,113]	LoS/NLoS Path Decoupling, Theoretical Performance Bounds, Low-Rank Collaborative Estimation, and Condition-Adaptive Algorithms	Practical environments are not only limited to LoS links. Accurate channel estimation of mixed LoS/NLoS near-field systems is required for feasible deployments.
XL-MIMO	[47,114,115,116]	Parameter Decoupling with 2D-DFT, Polar-Domain Sparsity Exploitation, and Compressed Sensing with Deep Learning	Effective beamforming requires accurate CSI estimation, which is challenging in XL-MIMO owing to the extremely high pilot overhead.
THz or mmWave	[117,118,119,120,121]	Polar-Domain and Wideband Sparsity Exploitation, Deep Learning-Driven Parameter Extraction, Sparse Bayesian Learning for Hybrid Models, Covariance Matrix Decoupling, and NB-aware OMP and Federated Learning	THz and mmWave systems undergo severe attenuation. Large arrays are required to form feasible directed beams, which require precise channel estimation.

and describe them as follows:

4.1. Near-Field Channel Estimation

The increase in the number of antennas in next-generation wireless communication systems has further complicated channel estimation, such that CSI estimation from the user to the massive number of antennas at the base station incurs a large pilot overhead. This overhead becomes even more apparent with the increase in the number of users. To tackle this problem, the authors in [103] proposed a near-field channel reconstruction approach where a small number of downlink pilots and limited geometry information is used. The downlink pilot signals transmitted by the BS can be used by the UE (along with the geometry information) to estimate the angle-of-arrival and rotation angles. These can be communicated to the BS after quantization. This type of channel estimation does not scale with the increase in the number of users. The authors showed that 8-bit quantization and two downlink signals are enough for full near-field LoS channel reconstruction, given that the antenna placements are known. In [44], the authors studied the NF channel estimation problem by exploiting polar-domain sparsity. Owing to the spherical wavefronts of NF systems, the channel sparsity is not obtained in the angular domain. Hence, existing channel estimation schemes are not optimized for NF MIMO systems. The polar domain representation, in contrast, can simultaneously account for both angular and distance information. They also proposed on-grid and off-grid XL-MIMO channel estimation techniques. Their results showed that, in the NF region, their channel estimation techniques achieve better normalized mean squared error (NMSE), which quantifies the difference between the actual and estimated channel values. As the problem of near-field channel estimation in ultra-massive MIMO systems presents significant challenges due to the increased complexity of the near-field environment, the authors in [104] proposed a novel algorithm called the Newtonized Near-Field Channel Estimation (NNCE) algorithm. The authors addressed the limitations of existing approaches by offering a low-complexity, tuning-free, and convergence-guaranteed solution for near-field channel estimation. The authors leveraged the unique characteristics of near-field channels and designed a near-field codebook that incorporates both angle and distance parameters, which is crucial for accurately capturing the complex characteristics of near-field channels. Their results demonstrated that the proposed NNCE algorithm significantly outperforms existing state-of-the-art baselines in terms of estimation accuracy, particularly in high SNR regimes. The simulation results showed that the NNCE algorithm achieves lower NMSE compared to other methods, indicating its effectiveness in converging to the true solution. The authors concluded that their approach provides a robust solution for near-field channel estimation in UM-MIMO systems and suggested that future work could explore the integration of model-driven deep learning methods to further enhance performance. This is the general explanation for the sparse channel and squint effect: “Sparse channels” refer to communication channels where only a few paths or components carry significant energy, making them easier to estimate with fewer measurements, while the “squint effect” is a phenomenon where the direction of maximum radiation of an antenna array changes with frequency, which can complicate channel estimation.

As the problem of accurate channel estimation in extremely large aperture arrays is critical for exploiting beamforming gains, the authors in [105] proposed a parametric multiuser near-field channel estimation algorithm inspired by the Multiple Signal Classification (MUSIC) method. They devised a two-step MUSIC algorithm to estimate the DoA and leverage this data to determine the distances of user equipment (UE) from the base station (BS) array. The authors enhanced their parametric channel model using a least-squares-based estimation corrector, achieving accurate near-field channel estimation. Their results demonstrated that the proposed method significantly outperforms classical least-squares (LS) and regularized LS (R-LS) channel estimation methods in terms of normalized beamforming gain and NMSE. The simulation results show that the proposed scheme achieves better performance due to the effective phase correction method, which ensures precise phase alignment and correct scaling of the channel estimate with the actual channel. The authors concluded that their method is effective for multiuser parametric near-field channel estimation with minimal signaling, offering a substantial improvement over traditional techniques.

As the problem of efficiently estimating high-dimensional channels in near-field Holographic MIMO (HMIMO) systems presents significant challenges, attributable to the complex and dynamic characteristics of channel distributions, the authors in [106] proposed a Bayes-optimal unsupervised learning framework for channel estimation. They developed a method that leverages Stein’s score function of the received pilot signals and an estimated noise level to estimate HMIMO channels without relying on prior knowledge or supervision. The authors presented a neural network, trained using a denoising score-matching objective, to learn the parameterized score function. They also proposed a PCA-based algorithm to estimate noise levels by leveraging the low-rank spatial correlation characteristic of the near-field. Their results demonstrated that the proposed score-based estimator achieves near-Bayes-optimal performance, closely approaching the oracle MMSE performance bound, even without any oracle information. The method demonstrates robustness against various mismatches and exhibits rapid adaptability to dynamic electromagnetic environments. Additionally, the proposed framework significantly reduces computational complexity by eliminating the need for costly matrix inversions, making it highly efficient for practical applications.

4.2. Hybrid-Field Channel Estimation

Conventional near-field-only or far-field-only channel estimation methods are not suitable for hybrid near–far field communications realized by large-scale antenna arrays where the users and scatterers are randomly positioned within both near-field and far-field zones. To overcome this challenge, the authors in [107] described the power diffusion (PD) effect, which induces a discrepancy between the hybrid-field channel and existing channel estimation methods. This mismatch results from the diffusion of power of one path into another position, creating fake paths that hinder accurate channel estimation. The authors proposed a PD-aware orthogonal matching pursuit (PD-OMP) algorithm to address the power diffusion effect by identifying the PD range where path power disperses to other locations. Simulation results indicated that PD-OMP surpasses current state-of-the-art hybrid-field channel estimation methods. Conventional far-field assumption-based channel estimation is impractical in near-field systems, as next-generation arrays are expected to function in hybrid channels with numerous scatterers in the near-field and far-field regions of the system. To overcome these challenges, the authors in [108] presented a neural network-based joint optimization of beamforming and localization for near-field channel estimation. The initial network weights emulate the beamforming matrix during training, whereas the remainder of the network serves as the localization function. The location parameters are refined to provide accurate channel estimation. Their approach leverages range-dependent frequency selectivity characteristics of near-field systems to provide accurate channel estimation. Simulations results demonstrate the superior performance of the proposed algorithm compared to range-only channel estimation techniques. As existing FF estimation techniques are infeasible for the hybrid channel models of XL-MIMO, the authors in [48] proposed an efficient hybrid field channel estimation scheme through precise modeling of the XL-MIMO channel. They first created a hybrid XL-MIMO topology where the scatterers can be in the NF or FF region and then proposed a hybrid channel model to capture these scatterers. They then proposed a channel estimation technique where the FF and NF components are estimated with low pilot overhead, and estimated the far-field and near-field path components separately by employing distinct channel transform matrices. Their results showed that existing far-field and near-field channel estimation schemes are special cases of the proposed hybrid-field channel estimation method. As the challenge of accurately modeling RIS-assisted vehicle-to-vehicle (V2V) channels in both far-field and near-field conditions emerges, the authors in [109] proposed a hybrid far- and near-field stochastic channel model that incorporates a dynamic subarray partitioning scheme. They developed a 3D non-stationary MIMO end-to-end channel model that characterizes the small-scale fading characteristics of RIS-assisted V2V communication systems. This model dynamically divides the RIS array into smaller subarrays, allowing the planar wavefront assumption to be applied to each subarray, hence balancing modeling accuracy and complexity. Their results demonstrated that the proposed model achieves high accuracy and low complexity in channel modeling, outperforming traditional far-field models, especially when the RIS array dimension is large enough to place the mobile terminals in the near-field region. The study also highlighted the importance of considering the RIS location and the motion properties of the transceivers, as these factors significantly influence the spatial, temporal, and frequency correlation functions of the channel. The authors concluded that their subarray partitioning scheme provides a theoretical foundation for the design and evaluation of RIS-assisted V2V communication systems, offering a promising solution for future mobile networks. In the context of this paper, “small-scale fading” refers to rapid fluctuations in the amplitude and phase of a radio signal caused by interference from multiple paths, while “planar wavefront assumption” is a simplification used in far-field conditions where the wavefronts are considered flat due to the large distance from the source.

4.3. LoS/NLoS Channel Estimation

Since existing channel estimation methods do not fully model the simultaneous presence of mixed LoS and non-line-of-sight (NLoS) path components, in [110], the authors proposed a mixed LoS/NLoS NF XL-MIMO channel model where the FF components are modeled using free space propagation and NF components are modeled using the NF array response vectors. They also derived the Cramér–Rao lower bound of the proposed channel estimation solution to the LoS/NLoS model. The CRLB provides the best possible accuracy that any channel estimator can achieve under given noise and system conditions. The proposed two-stage channel estimation separately resolves the LoS and NLoS components and outperforms existing methods.

The problem of channel estimation in RIS-assisted mmWave multiuser MIMO systems is challenging due to the mixed LoS and NLoS near-field channels. The authors in [111] proposed a novel method called Piece-Wise Collaborative Low-Rank Approximation (PW-CLRA) to address this issue. Their proposed method divides the effective channel into piece-wise effective channels, each exhibiting a low-rank structure, and estimates these channels via collaborative low-rank approximation. The authors’ results demonstrated that the proposed PW-CLRA method significantly reduces the training overhead by approximately 70% compared to state-of-the-art methods while achieving the target accuracy. Additionally, the method provides stable performance with an increasing number of RIS reflection elements, making it suitable for large-scale RIS-assisted systems. The authors’ method effectively handles the high rank and non-sparsity of near-field LoS channels, which are typically challenging for traditional estimation methods. Through simulations, they validated the effectiveness of PW-CLRA across various channel environments, confirming its scalability and flexibility.

To address the same issue, the authors in [112] proposed a mixed LoS/NLoS near-field XL-MIMO channel model. Their formulation distinctly characterizes LoS and NLoS path components, employing geometric free space propagation to describe LoS and near-field array response vectors for NLoS. Furthermore, they proposed a two-stage channel estimation algorithm designed to independently estimate the LoS and NLoS components. Their results demonstrated that the proposed two-stage scheme outperforms existing methods in terms of NMSE performance, particularly in the near-field region. The authors also defined the MIMO Rayleigh distance (MIMO RD) to determine the boundary between near-field and far-field models, showing that their model degenerates into the far-field model when the distance exceeds the MIMO RD. This work highlights the importance of accurately modeling both LoS and NLoS components to improve channel estimation in XL-MIMO systems.

The problem of accurately estimating channels in 3D MIMO systems under varying LOS and NLOS conditions was addressed by the authors in [113]. They proposed a novel channel estimation scheme that identifies whether a user is in the LOS or NLOS condition and applies different estimation techniques accordingly. They utilized the distinct characteristics of LOS and NLOS channels, such as the concentration of energy in a single path for LOS and multiple paths for NLOS, to improve the accuracy of channel estimation. The proposed method involves an initial identification of the channel condition using a threshold-based approach and then applies enhanced DFT-based channel estimation schemes tailored for each condition. Their results demonstrated that the proposed scheme significantly outperforms traditional DFT-based channel estimation methods. Specifically, the new algorithm shows a 17.7% improvement in NLOS cases and a 35.7% improvement in LOS cases in terms of NMSE at a signal-to-noise ratio of 20 dB. The authors concluded that by incorporating LOS/NLOS identification and applying condition-specific estimation techniques, the accuracy of channel estimation in 3D MIMO systems can be greatly enhanced.

4.4. XL-MIMO Channel Estimation

To mitigate the computational issues of ELAA channel estimation, the authors in [114] proposed a low-complexity angle–distance channel estimation method for UPA. They decoupled the angle and distance parameters to obtain a two-dimensional discrete Fourier transform for angle measurements. They used a low complexity method with a closed-form solution for distance estimation. They also provided a numerical analysis to verify the superiority of their proposed method. To mitigate the computational overheads of XL-MIMO channel estimation, the authors in [47] proposed model-based deep learning algorithms for estimating the near-field channel of XL-MIMO communications. They utilized a spatial gridding-based sparse dictionary and formulated a compressed sensing task for XL-MIMO channel estimation. Compressed sensing utilizes channel sparsity to enable accurate channel estimation with fewer pilot signals. Traditional channel estimation techniques assume that numerous pilot signals must be transmitted to sample the channel response at all antennas. In practical environments, only a few dominant paths exist between the transmitter and receiver. Compressed sensing exploits this sparsity and drastically reduces the number of pilot signals required for channel estimation. The authors utilized learning iterative shrinkage and a thresholding algorithm (LISTA) to solve the formulated compressed sensing problem. To address the computational burden of the near-field region, they proposed a novel sparsifying dictionary learning-LISTA, which formulates the dictionary as a layer and embeds it into LISTA. Their algorithm’s results showed that our proposed algorithms outperform non-learning benchmark schemes. Since XL-MIMO arrays have an extremely large number of antennas, realizing fully digital precoding is infeasible owing to the high power consumption of RF chains. Therefore, high-frequency XL-MIMO arrays are designed to utilize the hybrid precoding architecture, which provides a balance between spatial multiplexing gains, energy consumption, and capital costs. Efficient and effective hybrid precoding requires accurate channel state information estimation, which is challenging in XL-MIMO owing to the extremely high pilot overhead. To mitigate this issue, the authors in [115] discussed near-field channel estimation and replaced the angle-domain representation of the far-field region with a polar-domain representation tailored to the near-field region. They first described the energy spread effect, which shows that the near-field channel is not sparse in the angle domain and, therefore, has high computational overhead. The proposed polar-domain representation of near-field channels accounts for both the angle and distance distortions caused by the spherical wavefronts. They exploited this sparsity to develop a polar-domain simultaneous orthogonal matching pursuit (P-SOMP) algorithm to estimate the near-field channel. Their results showed a better NMSE performance than existing schemes.

Given that acquiring CSI is a pivotal yet challenging task in XL-MIMO wireless systems due to the extensive number of BS antennas, the authors in [116] proposed a practical strategy for efficient channel estimation in XL-MIMO communication systems, considering a hybrid-field channel model. Their objective was to establish a criterion for determining the specific proportion of far-field and near-field channel paths, followed by acquiring the corresponding estimated channel components. The authors introduced an enhanced orthogonal matching pursuit (OMP) algorithm to tackle the challenge posed by the unknown proportions of far-field and near-field paths, which complicate the efficient acquisition of hybrid-field channels. They proposed a method that continuously adjusts the value of the proportion parameter to minimize the residual vector, which is the difference between the received signals and the estimated channel components. This approach allowed them to estimate the channels accurately using fewer training symbols. Their results showed that the proposed method can achieve better NMSE performance compared to existing benchmarks, even with fewer training symbols. The method was shown to be effective across various system scenarios, achieving NMSE values close to an ideal lower bound. This demonstrates the method’s superiority and effectiveness in improving estimation accuracy for different communication scenarios.

4.5. THz and mmWave Channel Estimation

While THz communication is emerging as a key enabler for next-generation high-rate applications, it undergoes severe attenuation and heavily depends on the existence of LoS links. RISs can modify the channel to mitigate these challenges. However, large-scale RISs are required to provide feasible performance gains. These physically large-scale RISs have extended near-field regions. Performance gains are highly sensitive to the availability of accurate CSI. Motivated by this, the authors in [117] proposed a near-field RIS-assisted wideband THz channel estimation scheme. They developed a polar-domain frequency-dependent RIS-assisted channel estimation (PF-RCE) method to exploit the polar-domain sparsity of near-field channels and the common support property of wideband systems. The common support property refers to the sparsity of channels across different frequency subbands. This can immensely improve the performance of compressed sensing-based channel estimation. Numerical evaluations show that the proposed PF-RCE scheme has better NMSE compared to conventional THz channel estimation methods.

In response to the issue of significant performance loss in channel estimation for THz ultra-massive MIMO systems within the near-field region, the authors in [118] introduced an advanced near-field channel estimation technique utilizing deep learning. They employed a CNN to extract essential channel parameters such as angles, distances, time delay, and complex gains, which are vital for precise channel estimation in the near-field region where signal radiation follows a spherical wavefront. This method, known as deep-learning-based near-field channel estimation (D-NFCE), exploits the spatially correlated features among adjacent antenna elements or subcarriers to achieve rapid and precise channel estimations with minimized pilot resource requirements. Their results demonstrated that the proposed D-NFCE technique significantly outperforms conventional channel estimation schemes, achieving about a 30% gain in normalized channel gain over existing compressed sensing (CS)-based approaches in realistic THz UM-MIMO environments. The D-NFCE also shows superior performance in terms of bit error rate (BER), achieving more than a 5 dB gain across all SNR regimes compared to other schemes. This improvement was attributed to the high-resolution channel estimates provided by D-NFCE, which reduce errors in subsequent stages like channel equalization and symbol detection. The squint effect refers to the frequency-dependent nature of the array steering vector in THz systems, which is significant due to the large bandwidth and affects the channel estimation process.

As the problem of channel estimation in XL-RIS-aided mmWave systems is challenging due to the high training overhead and computational complexity, the authors in [119] proposed a novel two-stage algorithm to address this issue. They conceptualized the XL-RIS-aided channel as a hybrid model that accounts for both near-field and far-field effects, and they formulated the channel estimation challenge as a sparse recovery problem. Sparse channels refer to channels where only a few paths significantly contribute to the signal, allowing for reduced complexity in estimation. The authors developed a joint channel estimation and visual region (VR) detection algorithm based on the Fast Sparse Bayesian Learning (FSBL) framework, which is robust to the near-field effect and spatial non-stationary. Their results demonstrated that the proposed algorithms outperform existing benchmark schemes in terms of NMSE due to the effective detection of VR information and the utilization of the shift common-support property among subchannels. The shift common-support property refers to the characteristic that subchannels associated with different paths share the same support after cyclically shifting, which can be exploited to enhance performance. The authors concluded that their approach significantly improves channel estimation accuracy and is beneficial for future 6G systems.

To tackle the significant challenges posed by near-field propagation channels in high-frequency THz communication systems for localization and channel estimation, the authors proposed a near-field channel estimation and localization (NF-JCEL) algorithm in [120]. They concentrated on the spherical wavefront propagation in the near field of the THz system, aided by a RIS. The authors derived a second-order Fresnel approximation for the near-field channel model and design a down-sampled Toeplitz covariance matrix to decouple and separately estimate UE distances and angles of arrival (AoAs). Employing a subspace-based method and one-dimensional search, they estimated AoAs and UE distances, while channel attenuation coefficients were determined through the LS method. Their results demonstrated that the proposed NF-JCEL algorithm outperforms the conventional far-field algorithm, achieving higher resolution accuracy in localization and channel estimation. The study highlighted the importance of considering the spherical wavefront feature for high-precision localization systems, especially in the context of extra-large RIS panels. Sparse channels refer to communication channels with a limited number of significant paths, which can complicate signal processing. The squint effect is a phenomenon where the beam direction changes with frequency, affecting the accuracy of beamforming in wideband systems.

As the problem of channel estimation in near-field THz communications is complicated by the presence of near-field beamsquint (NB), the authors in [121] proposed both model-based and model-free techniques to address this issue. They introduced a model-based approach using an OMP algorithm with an NB-aware dictionary, which accounts for angular and range deviations due to NB. Additionally, they proposed a model-free solution using federated learning to reduce complexity and training overhead in a multiuser scenario. The authors’ results demonstrated that their proposed techniques effectively estimate wideband THz channels, outperforming existing state-of-the-art methods. The NBA OMP approach achieves close-to-minimum MMSE performance with significantly reduced channel usage, while the FL-based method offers substantial communication overhead savings compared to centralized learning approaches. Sparse channels refer to channels with limited reflected path components, which are common in THz communications due to their high frequency and short transmission distances. The squint effect, or beamsquint, occurs when the directions of generated beams at different subcarriers differ, causing them to point in different directions.

5. Practical Modeling for Energy-Efficient Near-Field Communications

In the previous sections, we explained the benefits of various energy-efficient techniques for near-field communication. However, for these technologies to be successfully applied in the near-field domain, accurate models are essential. In this chapter, we will delve into the different effective system models developed to capture the complex propagation characteristics of near-field environments. We will also discuss effective system models and advanced techniques aimed at enhancing communication and energy performance. These models include RIS models, which optimize signal reflection to improve signal strength and coverage, and holographic MIMO models, which utilize holographic surfaces to control the propagation environment. Additionally, Continuous Aperture Phased (CAP) MIMO models focus on maximizing communication performance in near-field environments by using continuous aperture techniques to enhance signal strength and spatial resolution. It is essential to create a comprehensive framework for near-field communication that incorporates transmitters, relays, and channels. This framework should facilitate a practical approach to beamforming while reducing the number of RF chains. By doing so, it can significantly enhance energy efficiency in near-field communications.

5.1. Signal Models

This section will examine and evaluate existing signal models, emphasizing how they can increase near-field communications’ energy efficiency. We aim to identify the most effective methods for enhancing energy efficiency in near-field communication systems by analyzing the salient characteristics, guiding concepts, and performance indicators of RIS, Holographic MIMO, and CAP MIMO models. We will also review the most recent developments, difficulties, and prospects in this area, offering suggestions for future lines of inquiry and advancement.

5.1.1. RIS

RIS represents a significant advancement in metasurface technology, featuring integrated electronic circuits capable of programmable modulation of incoming electromagnetic fields. These surfaces are composed of one or several layers of planar structures, which can be readily fabricated using lithography and nano-printing techniques. Each unit within the RIS comprises reflect-arrays that leverage varactor diodes or other microelectromechanical systems, with resonant frequencies controlled electronically [122]. The development of RIS technology solves important issues related to complex and dynamic mobile communication settings and the intrinsic drawbacks of conventional technologies, especially in view of the evolving demands of 6G networks [123]. Through manipulation of electromagnetic wave propagation, RIS improves signal robustness [124] and compensates for channel fading [125], hence increasing communication reliability. RIS dramatically lowers the cost and power consumption often associated with massive MIMO technology by replacing a large number of base stations with inexpensive, passive reflective modules [126]. Furthermore, by passively reflecting electromagnetic waves, RIS successfully reduces the effects of channel fading and coverage gaps, changing the propagation environment and improving signal transmission in places with poor coverage [127]. As 6G networks demand greater spectrum efficiency and expansive coverage, RIS offers a sophisticated and flexible approach to optimizing channel quality [128], thereby meeting these elevated performance requirements. Together, RIS was introduced to advance the efficiency and reliability of communication systems through intelligent control of the electromagnetic environment.

In near-field communication, the application of RIS further enhances the efficiency and reliability of communication [129], the near-field RIS model has been shown in Figure 5, which shows how RIS could help to improve the communication quality by avoiding the obstructed channel. Higher stability and transmission rates are required for near-field communication, which usually entails transmission of short-range, high-frequency signals [130]. RIS exhibits greater adaptability in complex near-field environments by dynamically adjusting the signal propagation path through intelligent control of the electromagnetic environment [131,132,133]. In particular, RIS can improve signal directionality and transmission efficiency in near-field communication by achieving more accurate beamforming and beamsteering [134]. This capability will significantly benefit applications such as IoT devices [132,133,135,136] and smart homes [137] that require high-precision and low-latency communication. Furthermore, RIS will help create more effective and sustainable communication networks because of its low power consumption and affordability [138]. By decreasing reliance on conventional base stations, RIS improves the network’s scalability and flexibility while also lowering construction and operating costs [139], providing a strong basis for the advancement of communication technologies in the future. Additionally, by passively reflecting and intelligently directing signals, RIS minimizes the need for active signal amplification, which significantly reduces energy consumption [140]. This capability is particularly beneficial in densely populated areas or IoT ecosystems, where numerous devices require consistent connectivity without the burden of high power demands [141]. RIS’s ability to enhance signal propagation and reduce fading also means that devices can operate at lower power levels while maintaining high-quality communication. This results in prolonged battery life for mobile and IoT devices, contributing to a more sustainable and energy-efficient network environment [142]. Additionally, by reducing the dependence on numerous power-intensive base stations, RIS helps decrease the overall energy footprint of communication infrastructures [143]. This not only lowers operational costs but also aligns with global efforts to minimize environmental impact and promote green technology solutions. As the development of next-generation communication, RIS’s energy-saving properties will be instrumental in meeting the demands for sustainable and efficient connectivity.

As RISs are emerging as a promising technology for 6G systems, it is important to study their near-field regions to enable effective resource utilization. In [144], the authors introduced a low-complexity beamforming technique for massive wideband phased arrays, known as InFocus. This technique effectively mitigates beam misfocus and beamsquint effects when applied to both near-field and far-field systems. InFocus employs a spatial phase modulation function to address the misfocus effect, thereby delivering larger and approximately flat beamforming gains in wideband systems. In [145], a codebook-based beamtraining scheme was proposed for the RIS-aided Near-field communication systems. Their model can serve users located either in the near field or the far field regions without prior knowledge of their locations. The authors also presented a case study for large-scale RIS-enabled multiuser communication systems and extend the scheme to wideband systems.

The codebook design of extremely large-scale RIS systems was investigated in [17]. The authors proposed efficient near-field beamtraining schemes, which design the near-field codebook to match the near-field channel model. They first designed the codebook by utilizing the near-field cascaded array steering vector of XL-RIS. The optimal codeword for XL-RIS was obtained by exhaustive training. Then, to reduce the overhead of exhaustive training, the authors proposed a hierarchical near-field codebook and develop the corresponding hierarchical near-field beamtraining scheme. Simulation results showed that the proposed schemes outperform existing far-field training schemes. The conventional RISs are limited to a half-space coverage. To tackle this problem, intelligent omnisurfaces (IOSs) are being explored as novel metasurfaces that enable simultaneous signal reflection and refraction. To analyze their efficacy in next-generation wireless networks, in [146], the authors investigated codebook design and beamtraining within an intelligent omnisurface (IOS)-aided multiuser system. They developed a near-far field codebook to accommodate users in both near and far fields by leveraging the dual functionality of the IOS. This codebook facilitates multiuser beamtraining, with each codeword encompassing multiple areas to enable the simultaneous training of all users. The proposed scheme can improve the sum rate and throughput compared to state-of-the-art schemes.

UAV-mounted RISs can enable flexible deployment, coverage extension, energy saving, and interference mitigation in dense networks. Hence, next-generation systems are actively exploring the efficacy of extremely large-scale aerial RIS (XLARIS) in boosting network performance. However, end-to-end channel estimation is challenging in XLARIS due to the passive nature of RIS. To alleviate this challenge, the authors in [147] designed a novel positioning-based codebook (P-CB) for the near-field of the XLARIS with an affordable CB size. They exploited this codebook to develop a low-overhead beamtraining method. Simulation results validate the low overhead of the proposed scheme compared to other CB-based schemes. An indoor RIS-aided mmWave communication scenario was considered in [148], where the mmWave BS was located close to the RIS and the UE. The indoor system has a range of several meters. Therefore, the service is provided in the near-field region. The authors designed an RIS training codebook with UE localization information, which provides substantial performance gains. As an increase in the size of RISs realizes phase difference variations with distance in the near-field region, the authors in [149] proposed a 3D codebook for the near-field region of large RISs, which can flexibly balance the communication quality and computation overhead. Their scheme enables efficient IRS control with respect to the communication requirements.

For large-scale RISs, implementing discrete phase shifts with a small number of control bits is more practically feasible. However, these phase shifts result in power leakage. To mitigate this, the authors in [150] designed a ring-type codebook that can provide SNR enhancement and grating lobe suppression for 1-bit beamforming in the near-field region. The proposed scheme achieved a main grating lobe suppression ratio of 17.9 dB. Conventional RIS-aided NF systems consider RIS configurations that provide focused beams toward a grid of users to develop the codebooks. In [151], the authors designed variable-width beam codebooks for the NF region. They tuned the size and center of the RIS-mapped spherical surface, which consequently enables the formation of variable width beams. Their results showed that the proposed method can maintain a high average rate and expand the illumination region. In [8], the authors studied the impact of beamfocusing in the facilitation of a high-rate multiuser downlink MIMO system. They employ different antenna structures as the near-field region is highly dependent on the antenna array architectures. They first provided near-field channel models for the different antenna architectures and then formulated a beamfocusing problem for maximizing the achievable sum-rate in multiuser networks. Their results indicated that beamfocusing leads to improved rate performance in the near-field region. They also showed that NF beamfocusing can decrease co-channel interference in multiuser communication scenarios. The dynamic metasurface antennas (DMAs) enable the most focused beams among all considered antenna array architectures. DMAs are advanced antenna structures that utilize metasurfaces to shape and control electromagnetic waves dynamically at a fine-grained level. Unlike traditional phased arrays, which use discrete phase shifters and amplifiers, DMAs leverage wavelength tunable elements to manipulate the amplitude and phase of incoming and outgoing waves in real time.

Beamtraining for extremely large-scale intelligent reflecting surface (XL-IRS) was explored in [152]. The authors reduced the training overhead of two-dimensional exhaustive search by proposing a two-layer codebook-based near-field beamtraining method, which tackles the near-field distance and candidate location of users separately. Numerical results show that the proposed scheme can achieve higher estimation accuracy with smaller training overhead. The pilot overhead problem of XL-MIMO systems was also explored in [153]. Two deep-learning-based beamtraining schemes are proposed for near-field XL-RIS-assisted communication systems. Numerical results show that the proposed schemes can reduce the beamtraining overhead by approximately 95%.

RISs are emerging as a promising candidate for 6G systems, owing to their sustainable energy footprint. Conventional RISs are composed of a finite number of small, discrete reflective elements, which can independently adjust their phase shift, allowing the RIS to steer or focus reflected waves. While this enables high spectral efficiency, they can suffer from energy leakage and quantization errors, which leads to optimal near-field beamforming. Continuous Aperture RISs (CAP-RISs) are emerging as the key solution to these issues. In CAP-RISs, the entire surface acts as a continuously tunable reflective interface without discrete phase shifts. They can precisely control the reflected wavefront at every point on the surface, creating more accurate and efficient wave redirection, which helps enable near-field beamfocusing. In [154], the authors considered two types of receivers in the near-field of the RIS: a ULA receiver and a single antenna receiver. For the ULA receiver, the authors designed the RIS coefficient to convert planar waves into cylindrical waves to enable efficient near-field beamforming. Similarly, for the single antenna receiver, the authors designed an RIS coefficient to convert planar waves into spherical waves. They then proposed MLE and focal scanning methods to sense the location of the receiver based on the reflection coefficients and derive their position error bound. Their simulation results showed that the proposed scheme can reduce energy leakage and improve channel capacity. With the advent of physically large-scale Reconfigurable Intelligent Surfaces and active arrays, the near-field has become far-reaching, and it is plausible to have transmitters and receivers in the near-field of propagation. The RIS can be configured to form a beam towards a point in space with finite depth, instead of infinite focus, as observed in the far-field. Infinite focus indicates that a beam does not concentrate energy at a specific point. Instead, it maintains a constant angular direction. A finite depth beam converges at a particular point and then starts dispersing. Using finite depth beams, beamforming gain is maximized at a specific point and ensures the highest power delivery at the receiver location. To understand the different beams, the authors in [155] discussed the approximations behind the Fraunhofer distance and showed that it does not sufficiently characterize when near-field focusing is possible. They also obtained a distance range where finite-depth beamfocusing is possible.

RISs have been envisioned to incorporate a large number of elements, greatly increasing the range of their near field. In [156], the authors aimed to jointly optimize the reflection coefficients of several RISs and transmit beamforming vectors of a single base station in order to allow the positioning and velocity estimation of a single multiantenna user that can be present in the near-field or far-field. The RISs are assumed to be reconfigured less frequently than the BSs. The authors formed an optimization problem to reduce the user uncertainty area and compare it with a classical beamfocusing technique and a scheme that maximizes the communication rate. Through numerical results, the authors showed that the localization performance of the proposed method is better than that of the other methods that optimize communication. They suggested that the positioning performance of the user can be increased by using a larger number of bigger arrays if they are properly jointly optimized.

Large intelligent surfaces (LISs) are emerging as a promising technology for 6G systems, as they can transmit, receive, and reflect signals to enhance signal coverage and improve energy efficiency. In [157], the authors proposed a novel 3D NF beamforming design for large intelligent surfaces. They first derived the Fresnel near-field region for the LIS and then designed an analog 3D beamforming method that can recover array-gain losses in this near-field region. They also showed that the optimal 3D beamforming can be decomposed in 2D and 1D near-field beamforming using a decomposition theorem. The 2D beamforming method compensates phase variations from the azimuth and elevation angles, whereas the 1D near-field beamforming compensates remaining phase variations caused by distance differences on the LISs. This 2D + 1D beamforming design reduces the size of the codebook drastically. Their results showed that their proposed codebook can perform close to the optimal benchmarks. Stacked intelligent metasurfaces (SIMs) have emerged as a promising technology for 6G systems, as they have advanced degrees of freedom for precise inter-user interference suppression. As 6G systems are expected to have high-frequency propagation, the near-field region is increasingly relevant for SIM-enabled systems. In [158], the authors analyzed a SIM-aided multiuser MIMO system operating in the near-field region. They formulated a weighted sum rate maximization problem and designed the transmit power vector and the SIM coefficients while meeting the power budget constraints. The formulated problem is non-convex, and they utilized block coordinate descent to solve it. Their results showed that the near-field beamfocusing can improve the performance compared to far-field beamsteering.

5.1.2. Holographic MIMO

Holographic MIMO is an innovative approach to wireless communication that combines holographic and massive MIMO technologies to enhance spectrum efficiency and reliability [159]. Holographic MIMO employs active antenna elements to create highly directive beams for improved data transmission [160] and uses ultra-thin, planar metasurfaces with integrated circuits to actively manipulate electromagnetic waves, allowing for precise control of wavefronts [161]. This system features numerous tiny antenna elements capable of dynamic beamforming [162], which helps overcome signal fading and interference in complex environments. By optimizing the propagation environment, holographic MIMO significantly improves data rates and coverage [163], making it an efficient solution for the demands of 6G networks. It reduces infrastructure needs and power consumption, addressing the limitations of conventional MIMO systems while meeting the increasing demand for high data throughput [164].

Holographic MIMO technology offers significant advantages in terms of efficiency and energy savings in near-field communication. It enhances spatial resolution [165] and beamforming accuracy [166], allowing for more precise control of signal paths in complex near-field environments [167], which improves signal directionality and reduces interference, which is crucial for high-frequency, near-field communications [168]. Compared to passive RIS, holographic MIMO uses plenty of antennas for simultaneous transmission and reception, increasing reliability and data throughput [169]. Higher transmission rates and more reliable connections are the outcomes that can create reliable, low-latency communication. Additionally, holographic MIMO optimizes the use of passive elements, minimizing the need for active signal amplification, which significantly lowers power consumption and extends battery life for mobile and IoT devices [170]. As noted in the previous chapter, the wavefronts of electromagnetic waves are spherical, enabling holographic MIMO to achieve enhanced precision in beamforming and focusing capabilities [171]. Additionally, holographic MIMO systems with spatially constrained rectangular antenna apertures enhance spatial degrees of freedom and channel capacity in near-field scenarios by leveraging evanescent waves through Fourier plane-wave series expansion [172]. Furthermore, near-field HMIMO systems in unknown electromagnetic environments demonstrate robust adaptability and optimal performance in dynamic conditions, a low-complexity Bayes-optimal channel estimator for utilizing an unsupervised neural network trained with denoising score matching and principal component analysis-based noise estimation [106]. There has been exciting work on energy-efficient techniques for holographic MIMO, including the analysis of holographic MIMO channel capacity considering practical constraints like angle distribution and array aperture, using a wavenumber domain method for spectral density calculation, showing that capacity is heavily influenced by angle distribution at high SNR and limited by array aperture at high antenna densities [173]. As communication technology evolves, the advantages of holographic MIMO in energy savings and performance will be instrumental in meeting the demands for efficient and sustainable connectivity.

5.1.3. CAP MIMO

CAP MIMO has become a significant research topic representing continuous surfaces. Researchers investigating CAP MIMO have found that it surpasses traditional discrete MIMO systems by achieving higher mutual information [174] and demonstrating enhanced performance [175]. CAP MIMO offers fine-grained control over signal propagation through a continuous electromagnetic surface rather than discrete antenna arrays [176].

CAP has the potential to transform wireless communications by facilitating continuous control over the amplitude and phase of RF signals, thereby enabling precise radiation pattern alignment [177]. Furthermore, research has demonstrated that systems based on CAP achieve a superior weighted secrecy sum-rate compared to traditional discrete MIMO systems [178]. Optimized beamformer techniques have been proposed to enhance physical-layer security to maximize secrecy rates and minimize power requirements [179]. Additionally, deep learning for CAP framework offers superior spectral efficiency and reduced inference complexity relative to match-filtering and advanced Fourier-based discretization techniques. This approaches the theoretical performance limit of optimized beamforming in spatially discrete array systems as the antenna count within a fixed area approaches infinity [180]. An efficient solution based on the sum-rate maximization task for different beamforming structures has been proposed in near-field communication [8]. This seamless approach enhances beamforming resolution and dynamic adaptation across various signal environments, especially in near-field communication. However, as the aperture size and the spatial diversity requirements of transceiver antennas increase, it becomes necessary to incorporate CAP models for effectively characterizing near-field communication channels [32].

5.2. Limitations of Existing Signal Models

Despite its promising capabilities, CAP-MIMO, much like RIS and Holographic MIMO, faces challenges related to channel modeling and signal processing. Accurate channel estimation is critical for the effective deployment of these technologies, and traditional models may not fully capture the complexities of real-world environments. More precise modeling techniques are needed to address signal fading and interference, ensuring reliable communication performance. Furthermore, while these technologies offer significant energy savings and infrastructure reduction, more research is required to optimize their implementation and integration into existing networks. Addressing these limitations will be key to unlocking the full potential of CAP-MIMO in future communication systems.

6. Electromagnetic Information Theory Based Energy-Efficient Near-Field Communication

In the previous sections, we explored how effective system models can enhance the performance of near-field systems, with a foundation rooted in traditional information theory. However, to achieve a more accurate representation of near-field dynamics, it is essential to advance into the realm of Electromagnetic Information Theory. This approach allows for a deeper understanding of the interactions between electromagnetic waves and the environment, considering factors such as wave propagation, scattering, and absorption at a finer granularity. By incorporating electromagnetic principles, we can develop models that more precisely capture the complexities of near-field scenarios, leading to improved system design and performance optimization. This shift not only broadens the theoretical framework but also opens up new avenues for innovation in communication technologies that rely on near-field interactions.

In the field of wireless communication channel modeling, two primary methodologies are generally employed: the Green’s function-based approach for continuous models and the integral equation approach for discrete models [181]. To effectively integrate new technologies into the near field, it is essential to emphasize the characteristics of continuous channel modeling. Modeling channels on metasurfaces has consistently presented challenges in the near field. However, recent advancements in Electromagnetic Information Theory have successfully addressed many of these limitations.

In the radiating near field, where EIT provides a robust framework for characterizing field properties through Green’s function, energy propagation transitions to active radiation, where the Green’s function inherently encodes phase-coherent field information [182]. This characteristic enables precise spatial control of electromagnetic energy through beamforming techniques, thus bypassing the limitations of the reactive near field, such as energy localization, diffusive noise, and stochastic signal distortions. By leveraging the deterministic radiation properties of the radiating near field, EIT provides a theoretical foundation to enhance wireless communication capacity and reliability, offering new insights to overcome traditional near-field constraints.

The reactive near field is excluded from this analysis due to fundamental physical limitations that hinder the applicability of EIT. In the reactive near field, thermal transport modeling reveals a diffusion-dominated mechanism governed by the heat equation [183]. In this context, temperature fields evolve toward equilibrium via diffusive processes. The introduction of pulsed sources in the Green’s function triggers instantaneous thermal responses across various spatial points. However, rapid diffusion dynamics lead to energy localization and non-radiative behavior, distorting the structure of the Green’s function and disrupting the deterministic phase-amplitude relationships required for EIT. Additionally, strong localization effects [184] arise from wave confinement caused by multiple scattering. These effects are further complicated by the presence of absorption or gain, which causes intensity-dependent renormalization of the diffusion coefficient. Such nonlinearities introduce unpredictable fluctuations in signal correlation lengths and diffusion parameters, destabilizing channel modeling and rendering EIT inapplicable. Additionally, beamforming is inherently unfeasible in the reactive near field due to the dominance of evanescent waves, which decay exponentially and store energy non-radiatively, thereby eliminating the propagating wave coherence required for directional signal manipulation [23].

However, EIT provides a comprehensive framework that integrates electromagnetic wave theory with information theory [182] in the radiating near field, offering a more physically consistent model for evaluating and designing these communication systems. Captures the full-dimensional information of wireless channels, including time, frequency, space, polarization, and even orbital angular momentum, potentially offering higher multiplexing and diversity [185]. By considering the physical effects of EM wave propagation, EIT effectively models communication systems, taking into account both probabilistic and deterministic models [186]. This approach helps in understanding the fundamental limits and performing system designs more effectively and realistically, which is crucial for optimizing designs and improving spectral efficiency and capacity [185].

6.1. Basic Principle

Compared to discrete wireless communication models, EIT offers a deeper understanding of electromagnetic interactions by using Green’s function-based channel modeling approaches to improve the capacity of near-field channels [32]. As the central object of study in EIT, the Green’s function serves as a core mathematical construct that establishes the intrinsic linkage between electromagnetic field characterization and communication system design principles. The development of CAP MIMO architectures via EIT foundations has enabled the emergence of novel frameworks addressing channel estimation methodologies and energy-efficient optimization strategies [187], illustration has been shown in Figure 6. This progression demonstrates how continuous surface paradigms achieve performance enhancements through EIT-based approaches, particularly in scenarios requiring precise electromagnetic field manipulation.

6.2. Energy Sustainability

Over the past decade, advanced wireless technologies such as massive MIMO systems, millimeter-wave communications, and small cells have been developed to enhance network capacity and facilitate widespread wireless access. Despite their benefits, these technologies often face challenges such as high energy consumption and costly hardware, which can restrict their practical applications and implementation [188].

Consequently, energy sustainability has become a core design focus in the development of the 6G telecommunication standard, necessitating new physical layer technologies. EIT plays a crucial role in meeting these sustainability goals by providing a theoretical framework to optimize the efficiency and capacity of wireless communication systems. By leveraging principles from EIT, ref. [189] introduced an innovative concept for extracting antenna and propagation information parameters. This approach is particularly effective for cluster models with diverse distributions, sizes, and materials, offering a reliable method for directing the power and phase allocation of antenna units in complex near-field environments. Additionally, EIT incorporates orbital angular momentum technology, which focuses on near-field properties, to consider the physical aspects of electromagnetic waves, thereby enhancing signal processing and performance in practical scenarios [190].

Furthermore, ref. [4] summarized potential optimization tools for near-field resource allocation, including numerical optimization, heuristic optimization, and machine learning. In the future, these tools could be applied to solve power constraint challenges in conventional communication systems following the design of a codebook for the near-field by EIT.

6.3. EIT-Motivated Channel Modeling and Channel Estimation

Motivated by EIT, a new near-field channel model has been developed to better understand the performance limits of wireless communication systems [191], specifically for 6G. Conventional channel models for EIT in NLoS environments predominantly emphasize far-field modeling, which inadequately captures the channel characteristics under near-field conditions [5]. This limitation may result in notable inaccuracies in the analysis of DoF and capacity, alongside critical system design components such as channel estimation. Ref. [192] introduced a near-field channel model employing electromagnetic scattering theory and non-stationary Gaussian random fields to better characterize the channel. This innovative approach is imperative, as prior models, which are predicated on the assumption of spatial stationarity, are more suited to far-field contexts and insufficiently address the complexities arising from middle-band, millimeter-wave, and terahertz technologies or extremely large apertures. Hence, an accurate channel modeling framework for EIT in near-field scenarios is essential for the effective analysis of system capacity [193,194].

Channel estimation is indispensable in commonly used communication models to meet the sustainable energy requirements of near-field beamforming [37]. The channel estimation proposed based on EIT has become crucial to the whole communication process. Ref. [182] introduced a method for joint uplink–downlink channel estimation aimed at addressing the inherent power imbalance observed among various antenna ports associated with different polarizations. This approach utilizes realistic antenna patterns obtained via electromagnetic simulations, which are incorporated into the proposed channel model. The effectiveness of this channel estimation technique is demonstrated through system-level simulations conducted under realistic communication conditions. Moreover, the EIT-Cov covariance estimator and the EIT-MMSE channel estimator exhibit superior performance over their baseline algorithms, thereby demonstrating the benefits of EIT for wireless communication systems [195,196].

6.4. Limitations

The current limitations of EIT in wireless communication channel modeling are primarily related to the challenges of combining continuous channel modeling approaches, such as those based on Green’s functions, with discrete models like integral equations. Even while EIT offers a thorough framework that merges information theory and electromagnetic wave theory, it still has trouble precisely simulating channels on metasurfaces, particularly in near-field situations. These difficulties still exist despite recent developments because of the complex nature of electromagnetic interactions and the requirement for accurate modeling of continuous surfaces.

Further work is required to fully utilize these dimensions for higher multiplexing and diversity, even while EIT captures all of the dimensional information of wireless channels, including time, frequency, space, polarization, and even orbital angular momentum. Integrating probabilistic and deterministic models in EIT is crucial for understanding fundamental limitations and optimizing system design, yet this integration remains a complex task.

Energy sustainability is also a significant concern, as advanced wireless technologies like massive MIMO and millimeter-wave communications often encounter high energy consumption and expensive hardware challenges. EIT seeks to provide a theoretical framework to optimize the efficiency and capacity of wireless systems through practical implementation and validation of these models that are still in the early stages [182]. Developing new physical-layer technologies and optimization tools for near-field resource allocation is necessary to overcome these limitations and achieve the sustainability objectives of future communication standards.

7. Conclusions

This paper presents a comprehensive review of the emerging field of near-field communication, which the advent of ultra-high frequency bands and large-scale antenna arrays has propelled. We began by outlining the fundamental differences between near-field and far-field communication, emphasizing the unique properties of spherical wave propagation that near-field communication offers. These properties open up new possibilities for communication systems. Next, we explored advanced beamforming technologies, highlighting their role in achieving energy focusing at wavelength levels. This capability enhances transmission efficiency and reduces energy consumption. The complex nature of the near-field environment poses significant challenges, requiring innovation in three advanced technologies: Reconfigurable Intelligent Surfaces, Holographic MIMO, and Continuous Aperture MIMO. We examined how RIS utilizes passive reflective units for low-power wavefront reconstruction, while Holographic MIMO employs high-density active antenna arrays for three-dimensional spatial beamforming. In contrast, CAP MIMO addresses the limitations of traditional arrays through its continuous aperture structure, enabling seamless control of electromagnetic fields. A key factor in realizing these technological advancements is precise channel estimation, which depends on an accurate channel model. Electromagnetic Information Theory provides a robust mathematical framework by quantifying the effects of near-field electromagnetic coupling and spatial correlations. This theoretical development effectively tackles the challenges of channel modeling and sets the stage for unlocking the full potential of near-field communication systems. Additionally, we discussed current challenges and future directions in this research area. We aspire for this paper to inspire and guide future advancements in near-field communication, ultimately contributing to developing more efficient and effective communication systems.

This review paper highlights the need to develop accurate channel models and employ energy-efficient optimization algorithms, identifying these as open research challenges in near-field communication. It suggests possible directions for future research aimed at tackling these challenges. Additionally, the paper presents tailored energy consumption models specific to various near-field communication scenarios, offering insights not fully detailed in the abstract. Furthermore, it discusses the crucial roles of advanced signal and channel models in reducing hardware requirements and improving energy efficiency.

Below, we summarize and highlight the four main takeaways from this paper:

• Comprehensive Energy-Efficient Techniques: The paper reviews energy-efficient strategies in near-field communication, examining their characteristics, advantages, and potential future.
• Advanced Signal and Channel Models: This paper identifies the capabilities of sophisticated signal and channel models, explaining their role in enhancing energy efficiency and minimizing hardware requirements. The paper also introduces energy consumption models suited for various near-field communication scenarios.
• Diverse Applications of Existing Techniques: The research highlights existing energy-efficient techniques and discusses their applicability across different communication environments.
• Open Research Challenges: The paper outlines the open research challenges in near-field communication, such as developing accurate channel models and energy-efficient techniques.