Near-Field Communications: A Tutorial Review

Motivation

• 6G’s large arrays, high frequencies, and new antennas alter electromagnetic propagation, requiring near-field models that account for spherical rather than planar waves.The Rayleigh distance is 2D^2/λ, with D the aperture and λ the wavelength.Source: §I · pp. 1–2

Open Research Problems

• The paper identifies missing comprehensive NFC tutorials and calls for further research on channel-model validation and broader system scenarios.Existing overview articles did not cover fundamental channel models, antenna structures, and analytical foundations comprehensively.Source: B. Related Overview Articles · p. 3, C. Motivation and Contributions · p. 3

Channel Modelling

• The tutorial reviews spherical-wave channel models for spatially discrete antennas and a Green’s function-based model for continuous-aperture antennas.The reviewed spatially discrete models include MISO and MIMO settings, as well as USW and NUSW formulations.Source: C. Motivation and Contributions · p. 3

Beamfocusing and Antenna Architectures

• It studies near-field beamfocusing, phase-shift and true-time-delay architectures, metasurface-based approximations of continuous apertures, and beam training.Beam training can reduce the complexity of channel estimation and analog beamforming design.Source: C. Motivation and Contributions · pp. 3–4

Performance Analysis

• The performance framework covers deterministic LoS and statistical multipath channels through SNR, power-scaling, outage, capacity, and mutual-information analyses.The framework reports insights including diversity order, array gain, high-SNR slope, and high-SNR power offset.Source: C. Motivation and Contributions · p. 4

Near-Field Characteristics

• Near-field channels depend on both angle and distance, providing additional degrees of freedom compared with far-field channels.Far-field channels are special cases obtained by omitting distance-dependent phase terms.Source: 1) MISO Channel Model · p. 8

• Far-field LoS MIMO channels have rank one because their channel matrix decomposes into transmit- and receive-side array response vectors.Source: 2) MIMO Channel Model · p. 8
• Near-field LoS MIMO channels use accurate element-wise distances, yielding non-linear phase and typically high rank with high degrees of freedom.Source: 2) MIMO Channel Model · p. 8
• In near-field multipath MIMO, the NLoS channel combines transmitter- and receiver-side near-field array responses, while rich scattering can produce full rank.Source: 2) MIMO Channel Model · pp. 8–9

Parallel ULAs

• For parallel ULAs, Fresnel-based distance expansion introduces a term coupling transmit and receive antenna indices.Source: 2) MIMO Channel Model · p. 9

Parallel ULAs and UPAs

• The coupled component in parallel-ULA and parallel-UPA channels cannot be decomposed into separate array responses and results in higher near-field DoFs.Source: 2) MIMO Channel Model · pp. 9–10

USW Model

• The USW model assumes propagation beyond the uniform-power distance, so all channel gains are approximated as uniform.Source: 1) USW Model [30], [31] · p. 10

NUSW Model

• The NUSW model calculates each link’s channel gain separately when propagation is within the uniform-power distance or the aperture is extremely large.Source: 2) NUSW Model [33], [34] · p. 10, B. Non-Uniform Channel Model for SPD… · p. 10
• NUSW remains incomplete for considerable antenna arrays because it omits effective-aperture and polarization-mismatch losses.Source: 3) A General Model · p. 10

General Model

• The general near-field model combines free-space path loss, effective aperture loss, and polarization loss, all depending on transmit and receive positions.Source: 3) A General Model · pp. 10–11
• The general model subsumes USW and NUSW, while its polarization formulation applies to arbitrary receiving polarization and transmitting current directions.Source: 3) A General Model · p. 11

Uniform-Power Distance

• The uniform-power distance r_UPD is defined using a minimum weakest-to-strongest channel-gain ratio threshold Γ.Source: 4) Uniform-Power Distance · pp. 11–12
• When r ≥ r_UPD, channel gains are comparable and the USW model is sufficiently accurate.Source: 4) Uniform-Power Distance · p. 12
• With Γ = 0.95, the Fresnel distance is much smaller than the Rayleigh distance, so the Fresnel approximation is accurate across most of the near-field region.Source: 4) Uniform-Power Distance · p. 12

CAP Antennas

• CAP-antenna degrees of freedom increase with aperture size and depend on carrier frequency and communication distance.Source: C. Green’s Function-Based Channel… · pp. 12–13

Open Research Problems

• Open problems include compact statistical models for SPD and CAP antennas, reactive-near-field modelling, and empirical validation of NFC channel models.Source: D. Discussion and Open Research Problems · p. 13

Beamfocusing

• Near-field beamfocusing can mitigate interference between users sharing a direction, but its focusing region is limited and depth of focus is a key performance measure.Source: 1) Asymptotic Orthogonality · p. 14, 2) Depth of Focus · pp. 14–15

Narrowband Systems

• Hybrid beamforming balances near-field performance against RF-chain constraints, since fully digital processing is impractical and purely analog processing loses performance.The architecture combines analog beamfocusing with digital processing while keeping the number of power-hungry RF chains low.Source: B. Beamfocusing with SPD Antennas · p. 15, 1) Narrowband Systems · p. 15
• PS-based architectures connect RF chains to all antennas or antenna subsets, with required phase shifters numbering NRFN and N, respectively.Phase shifters impose a unit-modulus constraint because they adjust signal phase only.Source: 1) Narrowband Systems · p. 16
• Fully-digital approximation can be near-optimal, but its complexity grows substantially with antenna count and requires optimizing large-dimensional beamformers.The cited discussion notes practical difficulty for arrays such as N = 512.Source: 1) Narrowband Systems · p. 16
• Heuristic two-stage optimization lowers complexity by designing analog beams in closed form and optimizing a reduced-dimension digital beamformer.Analog beams focus on user locations through LoS paths, while digital processing handles the resulting equivalent channel.Source: 1) Narrowband Systems · p. 16

Wideband Systems

• Frequency-independent PS beamformers mismatch frequency-dependent wideband channels, producing near-field beam split and array-gain loss away from the central frequency.A beam designed for (θc, rc) at fc focuses at frequency-dependent locations for other subcarriers.Source: 2) Wideband Systems · p. 17
• TTDs mitigate beam split by creating frequency-dependent phase shifts, but replacing all phase shifters with TTDs is impractical because of higher cost and power consumption.Hybrid TTD architectures therefore combine a low-dimensional delay network with high-dimensional phase-shifter processing.Source: 2) Wideband Systems · p. 17

TTD-based Architectures

• Sub-connected TTD architectures reduce beam split within each smaller subarray and can require only one TTD per RF chain.The reduction follows from connecting each RF chain to a limited antenna subarray.Source: NRF ] · p. 18
• Fully-digital approximation for TTD architectures is more challenging because beamformers, phase shifts, and delays are deeply coupled across subcarriers.The approach can also be computationally expensive because it designs large-dimensional fully digital targets and optimizes large matrices.Source: NRF ] · p. 19
• TTD-based hybrid beamformers jointly design phase shifters and time delays so beams across subcarriers focus on the desired location.The heuristic two-stage design uses closed-form analog processing followed by low-dimensional digital optimization.Source: NRF ] · pp. 18–19

Metasurface Architectures

• Metasurface antennas approximate continuous apertures using closely spaced metamaterial elements that tune signal amplitude or phase while reducing power consumption.Their element spacing can be on the order of 1/10λ to 1/5λ, with signals fed through waveguides.Source: C. Beamfocusing with CAP Antennas · p. 20
• Metasurface-based hybrid beamforming uses RF-chain feeds, waveguides, and configurable metamaterial-element weights to form the radiated object signal.Element weights may follow continuous-amplitude, discrete-amplitude, or Lorentzian-constrained control models.Source: 1) Narrowband Systems · p. 20

MIMO Extensions

• Fully-connected hybrid beamforming needs only 2Ns RF chains to match fully digital performance, so additional RF chains increase power consumption without improving communication performance.The result motivates adapting RF-chain counts to near-field channel degrees of freedom.Source: D. MIMO Extensions · p. 21
• Dynamic RF-chain architectures formulate joint beamforming and power-consumption design as mixed-integer nonlinear programs.Global branch-and-bound solutions have exponential complexity, while ADMM and machine-learning methods can provide lower-complexity suboptimal solutions.Source: D. MIMO Extensions · p. 22

Beam Training

• Beam training is proposed to reduce the complexity of CSI acquisition and obtain high-quality analog beamformers in systems with extremely large arrays.It is presented as an alternative to conventional channel estimation.Source: E. Near-Field Beam Training · p. 22

Open Research Problems

• Open challenges include near-field channel estimation because angular sparsity no longer holds and distance adds another channel dimension.The appropriate sparsity domain for near-field estimation remains to be identified.Source: F. Discussion and Open Research Problems · p. 23
• Other open problems concern finite-resolution converters, multifunctional beamfocusing, hybrid near/far-field operation, and dynamic switching between field regions.These topics involve signal-constellation limits, differing functional objectives, mixed channel characteristics, and fixed field classification.Source: F. Discussion and Open Research Problems · p. 24

SNR expressions and scaling laws

• The paper derives closed-form received-SNR expressions and antenna-count power-scaling laws for USW, NUSW, and general LoS channel models.These results use effective channel coefficients for each model and support comparisons of asymptotic behavior.Source: 2) SNR Analysis for SPD Antennas · p. 25, NUSW model · p. 26

USW and NUSW models

• USW SNR scales linearly with N, while NUSW SNR scales logarithmically with N but still violates energy conservation as N approaches infinity.The differing laws arise from the uniform-versus-non-uniform amplitude assumptions in the two models.Source: 2) SNR Analysis for SPD Antennas · pp. 25–26

General channel model

• The general model’s asymptotic SNR approaches a constant rather than increasing without bound, and remains below the transmit SNR limit.Under the stated special polarization assumption, the bound is at most one-third of the total transmitted power.Source: NUSW model · p. 26

Numerical results

• For small and moderate N, all models predict approximately linear SNR growth; differences become noticeable at N = 10^6, corresponding to a 5.35 m × 5.35 m array.At sufficiently large N, the general model remains bounded while USW and NUSW approach infinity.Source: NUSW model · pp. 26–27

SNR expressions and scaling laws

• The paper derives CAP-antenna SNR expressions and surface-area scaling laws for USW, NUSW, and general near-field channel models.The surface area is defined as S_CAP = L_xL_z.Source: 3) SNR Analysis for CAP Antennas · p. 27

USW model

• USW SNR scales linearly with S_CAP, causing energy-conservation violation as S_CAP approaches infinity.The paper attributes this behavior to the USW model’s uniform-amplitude assumption.Source: 3) SNR Analysis for CAP Antennas · p. 27

NUSW model

• NUSW SNR increases logarithmically with S_CAP, yet can exceed the transmit SNR as S_CAP approaches infinity when projected-aperture and polarization variations are ignored.This asymptotic behavior also violates energy conservation.Source: 3) SNR Analysis for CAP Antennas · pp. 27–28

General channel model

• The general CAP model satisfies an energy-consistent finite asymptotic SNR below the transmit SNR as S_CAP approaches infinity.The result follows under the stated uniform polarization and induced-current direction assumption.Source: 3) SNR Analysis for CAP Antennas · pp. 27–28

Numerical results

• For small and moderate S_CAP, all considered models exhibit approximately linear SNR growth because the user is in the far field.The comparison highlights the need to model varying free-space path losses, projected apertures, and polarization losses across the CAP surface.Source: 3) SNR Analysis for CAP Antennas · p. 28

Channel-gain statistics

• The analysis derives the channel-gain PDF and CDF for correlated MISO Rayleigh channels before evaluating outage probability.The derivation uses the positive eigenvalues of the covariance matrix and special functions including Gamma and incomplete Gamma functions.Source: 2) Analysis of the OP for Rayleigh… · pp. 29–30

Closed-form outage probability

• The outage probability is obtained in closed form by substituting the channel-gain CDF into the outage definition.The resulting expression is stated as Theorem 7.Source: 2) Analysis of the OP for Rayleigh… · pp. 29–30

High-SNR behavior

• The high-SNR outage probability is expressed asymptotically in a standard power-law form, characterized by array gain and diversity gain.The diversity gain determines the high-SNR slope, while array gain specifies the power gain relative to a benchmark outage curve.Source: 2) Analysis of the OP for Rayleigh… · p. 30

Numerical validation

• Analytical and asymptotic outage results agree with simulations, and larger numbers of scatterers yield higher diversity orders.The asymptotic results approach the numerical results in the high-SNR regime.Source: 2) Analysis of the OP for Rayleigh… · p. 30, Figure 26 · p. 30

MIMO extension

• The framework extends to single-user MIMO, but evaluating its outage probability requires random matrix theory.Existing results indicate that high-SNR MIMO outage probability follows the same standard asymptotic form.Source: 2) Analysis of the OP for Rayleigh… · p. 30

Motivation and input models

• Finite-alphabet mutual information generally lacks a closed-form expression, motivating an approximation-based analysis of ergodic mutual information.Gaussian inputs reduce mutual information to log2(1 + γ), whereas practical QAM inputs require separate treatment.Source: 4) Analysis of the EMI for Rayleigh… · p. 32

Approximation accuracy

• The fitted mutual-information approximation has absolute error O(10^-4) and relative error O(10^-3) for the listed QAM constellations.These errors motivate using the fitted expression to approximate EMI.Source: 4) Analysis of the EMI for Rayleigh… · p. 32, Table VI · p. 33

High-SNR behavior

• At high SNR, finite-alphabet EMI converges to the input entropy, while Gaussian-input EMI grows without bound.The finite-alphabet convergence rate is characterized through array gain and diversity order.Source: 4) Analysis of the EMI for Rayleigh… · p. 33

Numerical and energy-efficiency results

• The approximated EMI agrees closely with simulations, and increasing EMI or spectral efficiency produces an energy-efficiency optimum because consumed power rises linearly while EMI rises sub-linearly.The energy-efficiency tradeoff is evaluated using a circuit-power consumption model.Source: 4) Analysis of the EMI for Rayleigh… · pp. 32–33

Channel-gain statistics

• The Rician channel-gain statistics are derived by decomposing the gain into a deterministic line-of-sight term and a weighted sum of noncentral chi-square variables.The derivation uses an eigenvalue decomposition of the covariance matrix and establishes PDF and CDF expressions.Source: 5) Analysis of the OP for Rician… · p. 35

Closed-form outage probability

• The outage probability is obtained in closed form from the derived CDF of the Rician channel gain.The result is stated in Theorem 10 after the channel-gain distribution is constructed.Source: 5) Analysis of the OP for Rician… · pp. 35–36

Finite-power zero outage

• For Rician fading, the outage probability is piecewise in transmit power and becomes zero at the finite threshold p0 = (2R−1)σ2/ã0.For Rayleigh fading, the corresponding deterministic term is zero, so zero outage requires infinitely high transmit power.Source: 5) Analysis of the OP for Rician… · p. 36

Comparison with Rayleigh fading

• Rician and Rayleigh fading have the same diversity order, but Rician fading has a larger array gain and therefore better outage performance.The comparison is obtained by contrasting their high-SNR outage forms.Source: 5) Analysis of the OP for Rician… · p. 36

Numerical validation

• Simulations agree with the analytical results, and in the LoS-dominated setting Rician fading achieves the same outage with much less transmit power than Rayleigh fading.The outage curve collapses to zero at the predicted finite-power threshold.Source: 5) Analysis of the OP for Rician… · p. 36, Figure 32 · p. 37, Figure 33 · p. 37

Closed-form capacity

• The Rician ECC analysis derives a closed-form expression by integrating the Shannon capacity over the channel-gain distribution.The procedure first analyzes channel-gain statistics and then uses the resulting CDF to obtain the ECC expression.Source: 6) Analysis of the ECC for Rician… · p. 37

High-SNR analysis

• A high-SNR asymptotic expression is derived for Rician ECC to characterize its behavior at sufficiently large transmit power.The asymptotic result is stated in Corollary 12.Source: 6) Analysis of the ECC for Rician… · p. 37

Comparison with Rayleigh fading

• Rician and Rayleigh fading have the same high-SNR slope, although their high-SNR power offsets require numerical comparison.The paper identifies the Rician high-SNR power offset through the expected logarithmic channel-gain term.Source: 6) Analysis of the ECC for Rician… · p. 37

Numerical validation

• Analytical and asymptotic results agree with simulations, and Rician fading achieves the same ECC with less transmit power than Rayleigh fading.The resulting Rician high-SNR power offset is smaller, with the gain attributed mainly to the strong line-of-sight component.Source: 6) Analysis of the ECC for Rician… · pp. 37–38, Figure 34 · p. 38

Analytical Derivation

• The EMI derivation analyzes the channel-gain statistics, obtains a closed-form approximation, and studies high-SNR asymptotics.These are presented as three successive analytical steps.Source: 7) Analysis of the EMI for Rician… · p. 38

High-SNR Behavior

• At high SNR, finite-alphabet EMI converges to the input entropy H_pX, with convergence rate proportional to the derived asymptotic rate.Source: 7) Analysis of the EMI for Rician… · p. 39
• Rician fading requires less power than Rayleigh fading to achieve the same EMI because its convergence rate is much faster.Source: 7) Analysis of the EMI for Rician… · p. 39

Numerical Results

• The asymptotic results approach numerical results in the high-SNR regime, while lower modulation orders yield faster convergence rates.Source: 7) Analysis of the EMI for Rician… · p. 39

Broader Applicability

• The analytical expressions are summarized for statistical multipath channels and extend to single-stream SIMO, MIMO, and multicast channels through the corresponding channel-gain statistics.Source: 8) Summary of the Analytical Results · p. 39

Discussion and Open Research Problems

• The tutorial analyzes NFC performance for deterministic and statistical near-field models, while identifying open problems in information-theoretic, system-level, and network-level analysis.Source: C. Discussion and Open Research Problems · pp. 39–40, §V · p. 40

CAP Electromagnetic Model

• The Green function models electromagnetic propagation from an electric current distribution to an observation point under a radiating near-field approximation.Source: Appendix A · p. 40

Field and Polarization

• The normalized electric field depends on the normalized electric current vector, while receive-mode polarization contributes a polarization loss factor.Source: Appendix A · pp. 40–41

Beamfocusing Depth

• The beamfocusing depth is derived from the range interval preserving the required normalized array-gain threshold.Source: Appendix B · p. 41

USW Received SNR

• For the USW model, received SNR incorporates effective aperture loss, polarization loss, and free-space path loss.Source: Appendix D · pp. 41–42
• Under USW, different transmit elements experience identical polarization mismatches, projected apertures, and free-space path losses.Source: Appendix D · p. 42

NUSW Model

• The NUSW model uses common polarization mismatches and projected apertures but different free-space path losses across transmit elements.Source: Appendix E · p. 42

Large-Array Approximation

• For large arrays, the NUSW received-SNR analysis approximates element sums using a double integral over the rectangular aperture.Source: Appendix F · p. 42

Asymptotic Scaling

• As the array dimensions grow, the USW asymptotic received SNR scales as O(log N).Source: Appendix F · p. 42

General Channel Model

• The general channel model allows different transmit elements to experience different free-space losses, projected apertures, and polarization mismatches.Source: Appendix G · p. 42

Large-Array Approximation

• The integration region is bounded by inscribed and circumscribed disks whose radii depend on the aperture dimensions and element spacing.Source: Appendix F · p. 42, Figure 37 · p. 43

CAP Extension

• The general received-SNR expression is obtained by integrating the element-level effective power gain over the continuous transmit aperture.Source: Appendix H · p. 43, Appendix I · p. 43

Channel-Gain Statistics

• Eigenvalue decomposition transforms the correlated channel into independent Gaussian components weighted by the positive eigenvalues of its covariance matrix.Source: Appendix J · p. 43
• The channel-gain distribution is obtained from weighted exponential random variables associated with the transformed Gaussian components.Source: Appendix J · p. 43

Asymptotic Analysis

• The outage-probability asymptotic expression follows by substituting an incomplete-Gamma-function expansion and omitting higher-order terms.Source: Appendix K · p. 44

Capacity Analysis

• The ergodic channel-capacity result is obtained by evaluating the expectation of log2(∥h∥2/σ2).Source: Appendix L · p. 44

EMI Analysis

• The EMI analysis uses MMSE properties and Mellin transforms to characterize finite-alphabet mutual information and its asymptotic behavior.Source: Appendix M · p. 44, M [MMSEX (t); rR + 1] · p. 44

• Unitary invariance preserves the complex Gaussian distribution, giving U˜h ∼CN(0, I) when UU^H = I and ˜h ∼CN(0, I).Source: Appendix N · p. 44
• The proof rewrites the relevant expressions and evaluates them using modified Bessel-function identities and cited integral formulas.Source: Appendix N · p. 45, Appendix O · p. 45, Appendix P · p. 45
• Mutual independence enables the Laplace-transform derivation of the PDF of ˜a, including its behavior as x approaches zero.Source: Appendix O · p. 45
• The derivations conclude after obtaining the referenced equations and completing the proof.Source: Appendix O · p. 45, Appendix P · p. 45, Appendix N · p. 45
• As p →∞, the proof uses the resulting asymptotic limits and Lemma 8 to derive the stated expressions, including equiprobable square M-QAM entropy.Source: Appendix P · p. 45