betaFeedback

WorksEngineering

Limitation of Bessel Beams in Near Field Communications: A Depth of Field Perspective

Abstract

Bessel beams have recently been explored as a candidate complement to focused beams in radiative near-field communications due to their non-diffracting and self-healing properties. However, recent work has shown that the practically useful range of finite-aperture Bessel beams can occupy only a fraction of the radiative near-field region, establishing the existence of a finite propagation limit. In this paper, we establish such a limitation analytically from a depth-of-field perspective for Bessel beams generated by finite circular and square apertures. To quantify the Bessel beam depth, we first derive analytical expressions for the normalized array gain and use them to establish absolute and peak-normalized 33-dB depth-of-field criteria. These criteria reveal critical cone-angle thresholds that partition the Bessel-beam configuration space into (i) undefined, (ii) finite, and (iii) infinite depth-of-field regimes. Our analysis further reveals that the maximum usable non-diffracting range under the proposed depth-of-field criterion is limited to ≈22%\approx22\% and ≈19%\approx19\% of the Fraunhofer distance for circular and square apertures, respectively. Numerical results confirm the accuracy of the derived limits for discrete planar arrays, providing practical engineering guidelines for the design of finite-aperture Bessel beams in future near-field communications.

Introduction

The ever-growing demand for higher data rates and network capacity has driven wireless systems toward higher frequencies, where larger bandwidths can be exploited [1]. However, increasing the signal frequency also introduces greater propagation losses, motivating the use of electrically large antenna arrays to maintain a sufficient link budget via highly directional transmissions. As wavelength decreases, while the physical size of the antenna arrays stays in the cm-scale, the so-called “radiative” near-field region can extend over tens or even hundreds of meters [2]. This effect is particularly profound for millimeter wave (mmWave) (30 – 300 GHz), and especially at Terahertz (THz) (0.3–3 THz) frequencies, where many future wireless links are expected to operate within the near-field regime [3].

Unlike in far-field scenarios, where electromagnetic propagation can be accurately described by planar waves, propagation in the near field becomes different [4]. Consequently, using conventional far-field beamforming based on plane-wave assumptions in the near field may result in significant performance degradation [5]. This has recently motivated growing interest in near-field specific beamforming [6]. A diverse family of techniques has been proposed with distinct properties, including beamfocusing [4], Bessel beams [7], Mathieu beams [8], Airy beams [9], Hermite–Gaussian beams [10], vortex beams [11], among others. However, the finite aperture of practical antenna arrays can limit the region over which these beams retain their intended characteristic properties. Therefore, understanding these limits is important for assessing the practical use of these beams in the radiative near field.

These limitations have been partially explored. Among others, in [12], a depth-of-field limitation was investigated for beamfocusing, revealing that the finite-depth focusing is achievable only for receiver locations within approximately 10%10\% of the Fraunhofer distance from a uniform planar array (UPA). Beyond this limit, the focused beam progressively loses its finite-depth behavior and effectively converges toward conventional far-field beamforming. These findings have informed the development of multi-user beamforming algorithms [13], range-angle beam training [14], codebook design [15], and near-field channel estimation and localization techniques [16].

This ≈10%\approx10\% limit for beamfocusing naturally raises the question of whether similar finite-depth limits exist for other beam types in the near field. Among the various near-field beams mentioned above, Bessel beams [7] have gained significant attention due to their defining propagation characteristics, namely the ability to maintain their beam profile over the propagation direction (non-diffracting) [17] and reconstruct after partial obstruction (self-healing) [18]. These properties make Bessel beams particularly attractive for providing robustness to user-location uncertainty [19, 20, 21] due to their non-diffracting property, and for blockage-resilient communications [20, 21, 22] due to their self-healing capability. This has spurred extensive research on Bessel beams, spanning beam synthesis, configuration design, experimental demonstrations, and localization applications, highlighting their potential to improve spectral efficiency in ultra-wideband THz systems [7], [23, 24, 25].

The growing maturity of Bessel beam-based communication systems makes understanding their propagation limits increasingly important. This motivates the following research question: Do finite-aperture Bessel beams also exhibit (i) the same, (ii) similar, or (iii) principally different propagation limit as beamfocusing?

Importantly, recent work in [26] provides evidence that such a limit may exist. The authors empirically measure the so-called “Effective Near-Field Range (ENFR)” of a Bessel beam, defined as a set of communication ranges where the Bessel beam shows at least a 3 dB average advantage in received power over conventional far-field beamforming. However, as the work in [26] primarily aims to empirically confirm the existence of such a limit, the authors do not provide any closed-form expressions for how to derive the limit for different Bessel beam configurations. Further, as the utilized ENFR metric averages the received power over the entire non-diffracting region, it does not directly reveal where exactly along the propagation direction the Bessel beam actually begins to lose its finite-depth behavior and deteriorate or how this transition depends on the beam configuration. This motivates the following questions explored in the present work: What determines the limiting distance of a finite-aperture Bessel beam, how does this limit constrain its feasible configuration space, and what is the maximum non-diffracting range over which a finite depth of field can be maintained?

To answer these questions, this paper builds upon the normalized array-gain criterion consistent with the ones originally developed for beamfocusing. Importantly, unlike ENFR, the normalized array gain retains the spatial evolution of the beam along the propagation direction, allowing us to characterize when its finite-depth behavior is lost. We use this evolution to identify the feasible range of Bessel-beam configurations, ultimately determining the maximum usable non-diffracting range of a Bessel beam for near-field communications.

The main contributions of this paper are thus as follows:

  • We derive analytical expressions for the normalized array gain of a phase-based Bessel beam generated by circular and square arrays. Building upon these expressions, we introduce absolute and peak-normalized 3-dB beamdepths and establish the conditions governing their existence and finiteness.

  • Using the derived existence and finiteness conditions, we obtain critical analytical thresholds that partition the Bessel beam configuration space into regimes with: (i) undefined, (ii) finite, and (iii) infinite depth-of-field for the circular array (CA) and square array (SA) cases. We hence provide a complete characterization of the transition between different Bessel beam-depth regimes.

  • Building upon the derived analytical boundary conditions, we establish the quantitative limit on the maximum usable non-diffracting range of a Bessel beam in near-field communications. We show that this limit: (i) is determined by the antenna aperture, geometry, and the signal frequency; and (ii) corresponds to only a fraction of the Fraunhofer distance (≈20%\approx20\%). Moreover, through numerical simulations we demonstrate the accuracy of the derived analytical limits for discrete antenna arrays.

The rest of the paper is organized as follows. Section II presents the scenario and the near-field channel model. Section III presents the normalized array gain for the CA and SA cases. Section IV defines the depth-of-field criterion and formulates the critical conditions. Section V utilizes the critical conditions to derive the corresponding Bessel beam configuration ranges and establish the maximum usable non-diffracting range. Section VI utilizes numerical simulations and shows the impact of the established analytical limit on practical Bessel beam based near-field links. Finally, Section VII concludes the paper. The key notations are summarized in Table 1.

NotationDescription
λ\lambdaCarrier wavelength
zzUE location measured from the array center
θ\thetaCone angle of the phase-based Bessel beam
ΠBB\Pi^{\mathrm{BB}}Bessel-beam phase profile applied at the array
Zmax⁡Z_{\max}Non-diffracting range, Zmax⁡=Reff/tan⁡(θ)Z_{\max}=R_{\mathrm{eff}}/\tan(\theta)
h(ρ,φ,z)h(\rho,\varphi,z)Channel coefficient between the UE and the point (ρ,φ)(\rho,\varphi) on the AP
Prx(θ,z)P_{\mathrm{rx}}(\theta,z)Total received power at the UE
AAArea; A=πR2A=\pi R^2 for the CA and A=DAP2A=D_{\mathrm{AP}}^2 for the SA
ρmax⁡\rho_{\max}Maximum radial distance from the origin to the array boundary
ReffR_{\mathrm{eff}}Effective aperture radius
Gnorm(θ,z)G_{\mathrm{norm}}(\theta,z)Normalized array gain
BD3dBabs(θ)\mathrm{BD}^{\mathrm{abs}}_{3\mathrm{dB}}(\theta)Absolute 3-dB beam depth
BD3dBrel(θ)\mathrm{BD}^{\mathrm{rel}}_{3\mathrm{dB}}(\theta)Peak-normalized 3-dB beam depth
RRRadius of the CA
dRAc,dFAcd_{\mathrm{RA}}^{c},d_{\mathrm{FA}}^{c}Reactive near-field boundary and Fraunhofer distance
Zmax⁡cZ_{\max}^{c}Non-diffracting range of a Bessel beam
μc\mu_cNormalized conic-phase parameter, μc=4Rsin⁡(θ)/λ\mu_c=4R\sin(\theta)/\lambda
τc\tau_cNormalized distance, τc=z/dFAc\tau_c=z/d_{\mathrm{FA}}^{c}
GnormcG_{\mathrm{norm}}^{c}CA normalized array gain
θcrit,A1c,θcrit,A2c\theta_{\mathrm{crit},A1}^{c},\theta_{\mathrm{crit},A2}^{c}Critical cone angle associated with BD3dBabs(θ)\mathrm{BD}^{\mathrm{abs}}_{3\mathrm{dB}}(\theta)
θcrit,Bc\theta_{\mathrm{crit},B}^{c}Critical cone angle associated with BD3dBrel(θ)\mathrm{BD}^{\mathrm{rel}}_{3\mathrm{dB}}(\theta)
DAPD_{\mathrm{AP}}Side length of the SA
dRAs,dFAsd_{\mathrm{RA}}^{s},d_{\mathrm{FA}}^{s}Reactive near-field boundary and Fraunhofer distance
Zmax⁡sZ_{\max}^{s}Non-diffracting range of a Bessel beam
μs\mu_sNormalized conic-phase parameter, μs=2DAPsin⁡(θ)/λ\mu_s=2D_{\mathrm{AP}}\sin(\theta)/\lambda
τs\tau_sNormalized distance, τs=z/dFAs\tau_s=z/d_{\mathrm{FA}}^{s}
GnormsG_{\mathrm{norm}}^{s}SA normalized array gain
θcrit,A1s,θcrit,A2s\theta_{\mathrm{crit},A1}^{s},\theta_{\mathrm{crit},A2}^{s}Critical cone angle associated with BD3dBabs(θ)\mathrm{BD}^{\mathrm{abs}}_{3\mathrm{dB}}(\theta)
θcrit,Bs\theta_{\mathrm{crit},B}^{s}Critical cone angle associated with BD3dBrel(θ)\mathrm{BD}^{\mathrm{rel}}_{3\mathrm{dB}}(\theta)

Table 1. List of notations

System Model

We consider a downlink communication scenario in which an AP equipped with a planar array with inter-element spacing dd serves a UE (point receiver) located in the radiative near-field of the transmitting array using a phase-based Bessel profile. We assume d≥λ/2d \ge\lambda/2, and mutual coupling is ignored. We consider two array geometries: (i) CA with radius RR and area A=πR2A = \pi R^2, containing Nc≈πR2/d2N_{\mathrm{c}} \approx\pi R^2/d^2 elements arranged in a square grid (see Figure 1a), and (ii) SA with side length DAPD_{\mathrm{AP}} and area A=DAP2A = D_{\mathrm{AP}}^2, containing Ns≈DAP2/d2N_{\mathrm{s}} \approx D_{\mathrm{AP}}^2/d^2 elements (see Figure 1b). For analytical tractability, the AP is represented using an ideal continuous spatial approximation of the array response. This assumption is justified due to the large number of antenna elements in electrically large arrays (D≫λD \gg\lambda), and is commonly used for analytical characterization [[12]]. The accuracy of this approximation is quantified in Section VI-C.

Figure 1. Downlink near-field AP–UE communication system utilizing a Bessel phase profile (Πρ,ϕBB\Pi_{\rho,\phi}^{\mathrm{BB}}) with cone angle θ\theta and non-diffracting range (Zmax⁡Z_{\max}).

The APs transmit with total power PtxP_{\mathrm{tx}} distributed equally across all antenna elements. Without loss of generality, both AP geometries are considered to be parallel to the xyxy-plane and centered at the origin. The UE is located on the AP boresight and displaced only along the zz-axis, enabling the analysis of the depth-of-field of a phase-based Bessel beam, as illustrated in Figure 1. Although the analysis is presented from the downlink perspective, the same formulation applies to the reciprocal uplink case by electromagnetic reciprocity [[27]].

Near-field channel model

We adopt a line-of-sight (LoS), non-uniform spherical-wave near-field channel model [28]. We operate in the radiative near-field region of the array such that dRA<z≤dFAd_{\mathrm{RA}}<z\le d_{\mathrm{FA}}. Here, dRAd_{\mathrm{RA}} and dFAd_{\mathrm{FA}} denote the standard reactive and radiative near-field boundaries of the array, respectively [29] (see Section III for array-specific formulas).

Let (ρ=x2+y2,ϕ=tan⁡−1(y/x),0)(\rho=\sqrt{x^2+y^2},\phi=\tan^{-1}(y/x),0), denote the point (x,y,0)(x,y,0) on the AP in cylindrical coordinates. The continuous spatial representation of the channel coefficient between the point (ρ,ϕ)(\rho,\phi) on the AP and UE located at (0,0,z)(0,0,z) is then

h(ρ,ϕ,z)=λ4πr(ρ,z)e−j2πλr(ρ,z),(1)h(\rho,\phi,z)=\frac{\lambda}{4\pi r(\rho,z)}e^{-j\frac{2\pi}{\lambda}r(\rho,z)}, \tag*{(1)}

where λ\lambda is the wavelength and r(ρ,z)=ρ2+z2r(\rho,z)=\sqrt{\rho^2+z^2} denotes the distance between the point (ρ,ϕ)(\rho,\phi) on the AP and the UE.

Bessel beam generation

To generate a Bessel beam using a phase-based technique, a spatially varying phase profile ΠBB(ρ,ϕ)\Pi^{\mathrm{BB}}(\rho,\phi) is applied across the AP aperture [19]:

ΠBB(ρ,ϕ)=2πλρsin⁡(θ)(2)\Pi^{\mathrm{BB}}(\rho,\phi)=\frac{2\pi}{\lambda}\rho\sin(\theta) \tag*{(2)}

Without loss of generality, the phase profile is referenced to the center of the array, i.e., the point ρ=0\rho=0 is assigned zero phase. This phase-based technique applies a conic phase profile with cone angle θ\theta at the transmitting array to generate a Bessel beam. Its geometrical non-diffracting range is given by Zmax⁡=Reff/tan⁡(θ)Z_{\max}=R_{\mathrm{eff}}/\tan(\theta). Here, ReffR_{\mathrm{eff}} is the effective radius of the AP array. For CA, Reff=RR_{\mathrm{eff}}=R, while for SA, Reff=DAP/2R_{\mathrm{eff}}=D_{\mathrm{AP}}/\sqrt{2}, equivalent to the radius of the circumscribed circle [23].

Received power

The total received power observed at the UE located at distance zz is obtained by coherently summing the contributions from all antenna elements on the AP, which corresponds to an integral under the continuous array approximation. Then, with uniform power allocation across the array aperture (A)(A), the received power PrxP_{\mathrm{rx}} at zz with Bessel phase profile (cone angle θ\theta) at the AP is given by

Prx(θ,z)=PtxA2∣∫Ah(ρ,ϕ,z)ejΠBB(ρ,ϕ)ρ dρ dϕ∣2,(3)P_{\mathrm{rx}}(\theta,z)=\frac{P_{\mathrm{tx}}}{A^2}\left|\int_{\mathcal{A}}h(\rho,\phi,z)e^{j\Pi^{\mathrm{BB}}(\rho,\phi)}\rho\,d\rho\,d\phi\right|^2, \tag*{(3)}

where A={(ρ,ϕ,0):0≤ρ≤ρmax⁡(ϕ),0≤ϕ≤2π}\mathcal{A}=\{(\rho,\phi,0):0\le\rho\le\rho_{\max}(\phi),0\le\phi\le2\pi\} denotes the array aperture, with ρmax⁡(ϕ)\rho_{\max}(\phi) representing the radial distance from the origin to the aperture boundary.

Normalized array gain of a phase-based Bessel beam

In [12], the normalized array gain is introduced to quantify the loss in coherent array gain in the radiative near field relative to the corresponding plane-wave array gain, given by

Gnorm(θ,z)=∣∫Ah(ρ,ϕ,z)ejΠBB(ρ,ϕ)ρ dρ dϕ∣2A∫A∣h(ρ,ϕ,z)∣2ρ dρ dϕ.(4)G_{\mathrm{norm}}(\theta,z)=\frac{\left|\int_{\mathcal{A}}h(\rho,\phi,z)e^{j\Pi^{\mathrm{BB}}(\rho,\phi)}\rho\,d\rho\,d\phi\right|^2}{A\int_{\mathcal{A}}|h(\rho,\phi,z)|^2\rho\,d\rho\,d\phi}. \tag*{(4)}

Accordingly, Gnorm=1G_{\mathrm{norm}}=1 corresponds to perfect phase alignment and maximum coherent array gain, while Gnorm<1G_{\mathrm{norm}}<1 captures the loss due to phase mismatch across the aperture. Although [12] introduces this for an uplink scenario, the same metric applies for downlink to characterize the coherent field generated by the transmit array at a given observation point.

The metric was originally developed for beamfocusing, for which Gnorm=1G_{\mathrm{norm}}=1 can be attained at the focal point. However, both beamfocusing and Bessel beams rely on controlling the distribution of energy along the propagation direction. This makes GnormG_{\mathrm{norm}} a natural metric for quantifying the degradation of the coherent gain of Bessel beams with distance. Its evolution along the propagation direction is therefore used here to characterize the practical depth-of-field of finite-aperture Bessel beams. In the following subsections, we derive GnormG_{\mathrm{norm}} for the CA and SA cases.

Normalized array gain of a circular array

Using the definition in (4), we first derive closed-form expression for the normalized array gain for a CA. The exact gain can be calculated using numerical integration of the exact expression of GnormG_{\mathrm{norm}}, where ρmax⁡=R\rho_{\max}=R. An analytical expression can be obtained using the Fresnel approximation [30]

Gnormc(μc,τc)=(8τc)2∣jπ(e−jπ2(12τc−μcτc)2−e−jπ2μc2τc)−μcτc(ΔCc(μc,τc)−jΔSc(μc,τc))∣2(5)G_{\mathrm{norm}}^{c}(\mu_c,\tau_c)=(8\tau_c)^2\left|\frac{j}{\pi}\left(e^{-j\frac{\pi}{2}\left(\frac{1}{2\sqrt{\tau_c}}-\mu_c\sqrt{\tau_c}\right)^2}-e^{-j\frac{\pi}{2}\mu_c^2\tau_c}\right)-\mu_c\sqrt{\tau_c}\left(\Delta C_c(\mu_c,\tau_c)-j\Delta S_c(\mu_c,\tau_c)\right)\right|^2 \tag*{(5)}
Gnorms(μs,τs)=(16τs)2∣∫0π/4jπ(e−jπ2(122τscos⁡(ϕ)−μs2τs)2−e−jπ2μs2(2τs))−μs2τs(ΔCs(ϕ,μs,τs)−jΔSs(ϕ,μs,τs)) dϕ∣2(6)G_{\mathrm{norm}}^{s}(\mu_s,\tau_s)=(16\tau_s)^2\left|\int_{0}^{\pi/4}\frac{j}{\pi}\left(e^{-j\frac{\pi}{2}\left(\frac{1}{2\sqrt{2\tau_s}\cos(\phi)}-\mu_s\sqrt{2\tau_s}\right)^2}-e^{-j\frac{\pi}{2}\mu_s^2(2\tau_s)}\right)-\mu_s\sqrt{2\tau_s}\left(\Delta C_s(\phi,\mu_s,\tau_s)-j\Delta S_s(\phi,\mu_s,\tau_s)\right)\,d\phi\right|^2 \tag*{(6)}
ρ2+z2≈z+ρ22z,\sqrt{\rho^2+z^2}\approx z+\frac{\rho^2}{2z},

which gives an error lower than 3.5×10−33.5\times10^{-3} if z≥2.5Rz\ge2.5R [12], reducing the axial normalized gain to

Gnormc(θ,z)=4R4∣∫0Rρe−j2πλ(ρ22z−ρsin⁡(θ)) dρ∣2.(7)G_{\mathrm{norm}}^{c}(\theta,z)=\frac{4}{R^4}\left|\int_{0}^{R}\rho e^{-j\frac{2\pi}{\lambda}\left(\frac{\rho^2}{2z}-\rho\sin(\theta)\right)}\,d\rho\right|^2. \tag*{(7)}

Now, to simplify (7), we define the radial integral II such that Gnormc(θ,z)=4∣I∣2/R4G_{\mathrm{norm}}^{c}(\theta,z)=4|I|^2/R^4, and

I=∫0Rρe−jαρ2+jβρ dρ,(8)I=\int_{0}^{R}\rho e^{-j\alpha\rho^2+j\beta\rho}\,d\rho, \tag*{(8)}

where α=π/(λz)\alpha=\pi/(\lambda z), and β=2πsin⁡(θ)/λ\beta=2\pi\sin(\theta)/\lambda. The phase term can be rewritten by completing the square. Then, we use the substitution u=ρ−zsin⁡(θ)u=\rho-z\sin(\theta). The corresponding limits are u1=−zsin⁡(θ)u_1=-z\sin(\theta), and u2=R−zsin⁡(θ)u_2=R-z\sin(\theta). Using ρ=u+zsin⁡(θ)=u+β/(2α)\rho=u+z\sin(\theta)=u+\beta/(2\alpha), the integral becomes

I=ejβ2/(4α)(∫u1u2ue−jαu2 du+β2α∫u1u2e−jαu2 du).(9)I=e^{j\beta^2/(4\alpha)}\left(\int_{u_1}^{u_2}ue^{-j\alpha u^2}\,du+\frac{\beta}{2\alpha}\int_{u_1}^{u_2}e^{-j\alpha u^2}\,du\right). \tag*{(9)}

The phase factor ejβ2/(4α)e^{j\beta^2/(4\alpha)} has unit magnitude and is independent of ϕ\phi, and therefore does not affect the final normalized gain. The first integral is elementary, therefore,

∫u1u2ue−jαu2 du=j2α(e−jαu22−e−jαu12).(10)\int_{u_1}^{u_2}ue^{-j\alpha u^2}\,du=\frac{j}{2\alpha}\left(e^{-j\alpha u_2^2}-e^{-j\alpha u_1^2}\right). \tag*{(10)}

For the second integral, define t=2α/πut=\sqrt{2\alpha/\pi}u. Then αu2=πt2/2\alpha u^2=\pi t^2/2 and du=π/(2α) dtdu=\sqrt{\pi/(2\alpha)}\,dt, and hence

∫u1u2e−jαu2 du=π2α(C(t2)−C(t1)−j(S(t2)−S(t1))),(11)\int_{u_1}^{u_2}e^{-j\alpha u^2}\,du=\sqrt{\frac{\pi}{2\alpha}}\left(C(t_2)-C(t_1)-j\left(S(t_2)-S(t_1)\right)\right), \tag*{(11)}

where C(⋅)C(\cdot) and S(⋅)S(\cdot) are the Fresnel integrals, defined as C(x)=∫0xcos⁡(πt2/2) dtC(x)=\int_0^x\cos(\pi t^2/2)\,dt, and S(x)=∫0xsin⁡(πt2/2) dtS(x)=\int_0^x\sin(\pi t^2/2)\,dt. We introduce the normalized variables μc=4Rsin⁡(θ)/λ\mu_c=4R\sin(\theta)/\lambda, and τc=z/dFAc\tau_c=z/d_{\mathrm{FA}}^{c}, where dFAc=8R2/λd_{\mathrm{FA}}^{c}=8R^2/\lambda is the Fraunhofer distance of the CA, and τc\tau_c measures the observation distance relative to the Fraunhofer distance and μc\mu_c measures the strength of the conic phase across the aperture.

The normalized limits can be written as t1=−μcτct_1=-\mu_c\sqrt{\tau_c}, and t2=1/(2τc)−μcτct_2=1/(2\sqrt{\tau_c})-\mu_c\sqrt{\tau_c}. Using the odd symmetry of the Fresnel integrals, the Fresnel differences can be written as C(t2)−C(t1)=−ΔCc(μc,τc)C(t_2)-C(t_1)=-\Delta C_c(\mu_c,\tau_c), and S(t2)−S(t1)=−ΔSc(μc,τc)S(t_2)-S(t_1)=-\Delta S_c(\mu_c,\tau_c). First, substituting (8) and (9) into (7) yields the radial integral II. Substituting (7) into (5) and expressing the results in terms of the normalized variables gives the closed-form expression at the top of the page in (10).

Normalized array gain of a square array

For a SA, the exact normalized array gain can similarly be evaluated through numerical integration of (4), where

ρmax⁡=DAP2max⁡{∣cos⁡(ϕ)∣,∣sin⁡(ϕ)∣}.(12)\rho_{\max}=\frac{D_{\mathrm{AP}}}{2\max\{|\cos(\phi)|,|\sin(\phi)|\}}. \tag*{(12)}

An analytical expression can be obtained by applying the same Fresnel approximation [30], which remains accurate across the array with errors below 3×10−33\times10^{-3} when z≥1.2DAPz\ge1.2D_{\mathrm{AP}} [12], and exploiting the eightfold symmetry of the SA. The normalized array gain reduces to

Gnorms(θ,z)=64DAP4∣∫0π/4∫0DAP/(2cos⁡(ϕ))ρe−j2πλ(ρ22z−ρsin⁡(θ)) dρ dϕ∣2.(13)G_{\mathrm{norm}}^{s}(\theta,z)=\frac{64}{D_{\mathrm{AP}}^4}\left|\int_{0}^{\pi/4}\int_{0}^{D_{\mathrm{AP}}/(2\cos(\phi))}\rho e^{-j\frac{2\pi}{\lambda}\left(\frac{\rho^2}{2z}-\rho\sin(\theta)\right)}\,d\rho\,d\phi\right|^2. \tag*{(13)}

Now, to simplify (12), we define the radial integral (I(ϕ))(I(\phi)) such that Gnorms(θ,z)=64∣∫0π/4I(ϕ)dϕ∣2/DAP4G_{\mathrm{norm}}^{s}(\theta,z)=64|\int_{0}^{\pi/4}I(\phi)d\phi|^2/D_{\mathrm{AP}}^4, given by

I(ϕ)=∫0q(ϕ)ρe−jαρ2+jβρ dρ,(14)I(\phi)=\int_{0}^{q(\phi)}\rho e^{-j\alpha\rho^2+j\beta\rho}\,d\rho, \tag*{(14)}

where q(ϕ)=DAP/(2cos⁡(ϕ))q(\phi)=D_{\mathrm{AP}}/(2\cos(\phi)) for 0≤ϕ≤π/40\le\phi\le\pi/4. Applying a similar technique as the CA, the phase term can be written by completing the square. Then, the integral admits the same form as the CA case (7), with the only difference being the upper integration limit u2=q(ϕ)−zsin⁡(θ)u_2=q(\phi)-z\sin(\theta). We introduce similar normalized variables μs=2DAPsin⁡(θ)/λ\mu_s=2D_{\mathrm{AP}}\sin(\theta)/\lambda, and τs=z/dFAs\tau_s=z/d_{\mathrm{FA}}^{s} where dFAs=4DAP2/λd_{\mathrm{FA}}^{s}=4D_{\mathrm{AP}}^2/\lambda is the Fraunhofer distance of the square array, and τs\tau_s and μs\mu_s have similar interpretations as the CA case. The normalized limits can then be written as t1=−μs2τst_1=-\mu_s\sqrt{2\tau_s}, and t2=1/(22τscos⁡(ϕ))−μs2τst_2=1/(2\sqrt{2\tau_s}\cos(\phi))-\mu_s\sqrt{2\tau_s}. Using the odd symmetry of the Fresnel integrals, the Fresnel differences can be written as C(t2)−C(t1)=−ΔCs(ϕ,μs,τs)C(t_2)-C(t_1)=-\Delta C_s(\phi,\mu_s,\tau_s), and S(t2)−S(t1)=−ΔSs(ϕ,μs,τs)S(t_2)-S(t_1)=-\Delta S_s(\phi,\mu_s,\tau_s).

Now substituting (13) into (12) and expressing the result in terms of the normalized variables gives the expression shown at the top of the page in (14). The key difference from the CA case is that, for the SA, the radial upper limit depends on the azimuth angle as ρmax⁡=DAP/(2cos⁡(ϕ))\rho_{\max}=D_{\mathrm{AP}}/(2\cos(\phi)) over 0≤ϕ≤π/40\le\phi\le\pi/4. Consequently, the radial integral has a closed-form primitive, but the SA gain retains an angular integral over ϕ\phi needing numerical evaluation.

Spatial evolution of the normalized array gain

The derived expressions (10) and (14) enable direct evaluation of the normalized array gain throughout the radiative near field. Figure 2 illustrates the resulting spatial evolution of GnormG_{\mathrm{norm}}. As the cone angle decreases, the high-gain region extends over a longer propagation distance while simultaneously broadening in the transverse direction. The figure also shows that contours of constant 0.5 normalized array gain naturally show the effective high-gain propagation region. In particular, the Gnorm=0.5G_{\mathrm{norm}}=0.5 (3 dB) contour (indicated in red) marks the points at which the coherent array gain has decreased by 3 dB relative to the plane-wave maximum.

Normalized array gain for a CA with $R=10$ cm at $f=125$ GHz for three cone angles

Figure 2. Normalized array gain for a CA with R=10R=10 cm at f=125f=125 GHz, illustrating the spreading behavior of finite-aperture Bessel beams for decreasing cone angle. Smaller cone angles produce a broader beam and a longer high-gain propagation region.

These observations show that the conventional non-diffracting range Zmax⁡Z_{\max}, while providing a geometric approximation of where the finite-aperture Bessel beam ceases to exist, does not capture how effectively the beam remains spatially confined and therefore does not directly characterize its communication performance. Instead, the normalized array gain reveals a gradual degradation in coherent gain as the cone angle decreases and the beam spreads, rather than an abrupt termination at Zmax⁡Z_{\max}. Motivated by this behavior, the following section adopts a 3-dB beam-depth metric based on the normalized array gain and derives the conditions under which this depth exists and remains finite.

Depth-of-field of a Bessel beam

Following the discussion from the previous section, we define the depth-of-field as the 3-dB beam depth along the propagation direction. Two definitions are considered: an absolute definition referenced to the plane-wave maximum, and a peak-normalized definition referenced to the maximum array gain achieved for the given cone angle. This section formalizes both definitions and derives the corresponding critical conditions.

Absolute 3-dB beam depth

The absolute definition compares the normalized gain directly with the fixed half-power level 0.5 (w.r.t. the plane-wave gain). For a fixed cone angle θ\theta, the half-power distance set is Z0.5abs(θ)={z>0:Gnorm(θ,z)≥0.5}\mathcal{Z}_{0.5}^{\mathrm{abs}}(\theta)=\{z>0:G_{\mathrm{norm}}(\theta,z)\geq0.5\}. The absolute 3-dB beam depth is then

BD3dBabs(θ)=[sup⁡Z0.5abs(θ)−inf⁡Z0.5abs(θ)].(15)\mathrm{BD}_{3\mathrm{dB}}^{\mathrm{abs}}(\theta)=\left[\sup\mathcal{Z}_{0.5}^{\mathrm{abs}}(\theta)-\inf\mathcal{Z}_{0.5}^{\mathrm{abs}}(\theta)\right]. \tag*{(15)}

This definition measures whether the Bessel beam maintains at least half of the ideal plane-wave gain. A finite absolute 3-dB beam depth requires two conditions.

Condition A1: For the 3-dB beam depth to be finite, under the absolute definition, the 3-dB tail must fall below 0.5, i.e.,

lim⁡z→∞Gnorm(θ,z)<0.5.(16)\lim_{z\to\infty}G_{\mathrm{norm}}(\theta,z)<0.5. \tag*{(16)}

Figure 3. Illustration of the absolute 3-dB beam depth behavior over UE location zz. The horizontal dotted line denotes the absolute half-power threshold.

Condition A2: For the 3-dB beam depth to exist, under the absolute definition, the axial peak must reach 0.5, i.e.,

max⁡z>0Gnorm(θ,z)≥0.5.(17)\max_{z>0}G_{\mathrm{norm}}(\theta,z)\geq0.5. \tag*{(17)}

Fig. 3 illustrates the two conditions for the absolute 3-dB beam depth. Condition A2 determines whether the normalized array gain reaches the half-power threshold, and therefore whether BD3dBabs(θ)\mathrm{BD}_{3\mathrm{dB}}^{\mathrm{abs}}(\theta) exists. If it does, Condition A1 determines whether the corresponding region is finite or infinite. The absolute definition is useful for comparing different Bessel-beam configurations or other beamforming techniques because all gains are measured relative to the plane-wave reference.

Peak-normalized 3-dB beam depth

The peak-normalized definition compares the gain with one half of its own axial maximum for a given cone angle θ\theta. Hence, the corresponding half-power set is Z0.5rel(θ)={z>0:Gnorm(θ,z)≥0.5max⁡z>0Gnorm(θ,z)}\mathcal{Z}_{0.5}^{\mathrm{rel}}(\theta)=\{z>0:G_{\mathrm{norm}}(\theta,z)\geq0.5\max_{z>0}G_{\mathrm{norm}}(\theta,z)\}. The peak-normalized 3-dB beam depth is then

BD3dBrel(θ)=[sup⁡Z0.5rel(θ)−inf⁡Z0.5rel(θ)].(18)\mathrm{BD}_{3\mathrm{dB}}^{\mathrm{rel}}(\theta)=\left[\sup\mathcal{Z}_{0.5}^{\mathrm{rel}}(\theta)-\inf\mathcal{Z}_{0.5}^{\mathrm{rel}}(\theta)\right]. \tag*{(18)}

Since the peak-normalized threshold is defined relative to the beam’s own maximum, the second requirement is always satisfied (analogous to Condition A2). However, the beam depth can still become infinite, which introduces a separate condition for the peak-normalized case.

Condition B: For the 3-dB beam depth to be finite, under the peak-normalized definition, we require the 3-dB tail to fall below half of the peak. Thus, the finite-depth condition is

lim⁡z→∞Gnorm(θ,z)<12(max⁡z>0Gnorm(θ,z)).(19)\lim_{z\to\infty}G_{\mathrm{norm}}(\theta,z)<\frac{1}{2}\left(\max_{z>0}G_{\mathrm{norm}}(\theta,z)\right). \tag*{(19)}

Figure 4 illustrates the condition for the peak-normalized 3-dB beam depth. Because the threshold is defined relative to the peak gain for each θ\theta, a half-power region always exists around the main peak. Condition B determines whether this region is finite: if the asymptotic tail falls below the half-power threshold, the beam depth is finite; otherwise, it is infinite. The peak-normalized definition therefore measures the propagation depth of a given Bessel beam relative to its own peak gain. The absolute definition reflects communication performance relative to the plane-wave maximum, whereas the peak-normalized definition measures the propagation depth of the Bessel beam relative to its own peak.

Figure 4. Illustration of peak-normalized 3-dB beam depth behavior over UE location zz. The horizontal dotted line denotes the relative half-power threshold.

Critical Angle Characterization

The 3-dB beam-depth definitions in Section IV establish critical conditions that separate the different depth-of-field regimes. In this section, we utilize these conditions to derive the corresponding critical cone angles for the CA and SA. These results also enable us to establish a limit on the maximum usable non-diffracting range.

Circular array

Evaluating Condition A1 for the CA case yields a lower bound on θ\theta over which a Bessel beam maintains a finite depth-of-field under the absolute 3-dB beam depth definition. Let Gnormc(μc,∞)G_{\mathrm{norm}}^{\mathrm{c}}(\mu_{\mathrm{c}},\infty) be defined as

Gnormc(μc,∞)=lim⁡τc→∞Gnormc(μc,τc).(20)G_{\mathrm{norm}}^{\mathrm{c}}(\mu_{\mathrm{c}},\infty)=\lim_{\tau_{\mathrm{c}}\to\infty}G_{\mathrm{norm}}^{\mathrm{c}}(\mu_{\mathrm{c}},\tau_{\mathrm{c}}). \tag*{(20)}

Using this definition we quantify the limits for a CA:

Proposition 1a. For a circular array under the absolute 3-dB beam depth definition, BD3dBabs\mathrm{BD}_{3\mathrm{dB}}^{\mathrm{abs}} is finite when the cone angle satisfies

θ>sin⁡−1(0.5533λR).(21)\theta>\sin^{-1}\left(0.5533\frac{\lambda}{R}\right). \tag*{(21)}

Proof: Evaluating the limit τ→∞\tau\to\infty for (10) (see Appendix A) gives

Gnormc(μc,∞)=ξ2+2−2cos⁡ξ−2ξsin⁡ξξ4,(22)G_{\mathrm{norm}}^{\mathrm{c}}(\mu_{\mathrm{c}},\infty)=\frac{\xi^{2}+2-2\cos\xi-2\xi\sin\xi}{\xi^{4}}, \tag*{(22)}

where ξ=πμc/2\xi=\pi\mu_{\mathrm{c}}/2. For the absolute definition, the beam depth becomes infinite whenever Condition A1 is not satisfied. Therefore, the corresponding asymptotic critical value μccrit,A1\mu_{\mathrm{c}}^{\mathrm{crit,A1}} is obtained by solving the equality Gnormc(μccrit,A1,∞)=0.5G_{\mathrm{norm}}^{\mathrm{c}}(\mu_{\mathrm{c}}^{\mathrm{crit,A1}},\infty)=0.5, where the first positive solution gives μccrit,A1=2π×3.4766‾≈2.2133\mu_{\mathrm{c}}^{\mathrm{crit,A1}}=\frac{2}{\pi}\times3.476\overline{6}\approx2.2133 (rounding error), with the corresponding cone angle

θcrit,A1=sin⁡−1(μccrit,A1λ4R)≈sin⁡−1(0.5533λR).(23)\theta_{\mathrm{crit,A1}}=\sin^{-1}\left(\frac{\mu_{\mathrm{c}}^{\mathrm{crit,A1}}\lambda}{4R}\right)\approx\sin^{-1}\left(0.5533\frac{\lambda}{R}\right). \tag*{(23)}

The critical cone angle associated with Condition A1 can be obtained directly from the limits of the normalized gain in (10), which yields Gnormc(μc,∞)G_{\mathrm{norm}}^{\mathrm{c}}(\mu_{\mathrm{c}},\infty). In contrast, conditions A2 and B depend on the maximum axial gain max⁡z>0Gnorm(θ,z)\max_{z>0}G_{\mathrm{norm}}(\theta,z), for which obtaining an analytical expression is difficult because the stationary-point condition is implicit (see Appendix C). Therefore, we utilize an algorithm to evaluate the peak gain numerically (see Appendix E) and utilize these values to determine the critical cone angles associated with these conditions. The accuracy of these results are presented in Section VI.

Under the absolute definition, Condition A2 gives an upper bound on θ\theta for a Bessel beam: if θ\theta is too large, the peak gain remains below 0.50.5, and BD3dBabs\mathrm{BD}_{3\mathrm{dB}}^{\mathrm{abs}} is empty.

Proposition 1b. For a circular array under the absolute 3-dB beam depth definition, BD3dBabs\mathrm{BD}_{3\mathrm{dB}}^{\mathrm{abs}} is defined only when the cone angle satisfies

θ≤sin⁡−1(4.056975λR).(24)\theta\le\sin^{-1}\left(4.056975\frac{\lambda}{R}\right). \tag*{(24)}

Proof: Evaluating Condition A2 using the peak-finding algorithm gives μccrit,A2≈16.2279\mu_{\mathrm{c}}^{\mathrm{crit,A2}}\approx16.2279, with the critical cone angle:

θcrit,A2=sin⁡−1(μccrit,A2λ4R)≈sin⁡−1(4.056975λR).(25)\theta_{\mathrm{crit,A2}}=\sin^{-1}\left(\frac{\mu_{\mathrm{c}}^{\mathrm{crit,A2}}\lambda}{4R}\right)\approx\sin^{-1}\left(4.056975\frac{\lambda}{R}\right). \tag*{(25)}

We next consider the peak-normalized definition. In this case, the beam depth is always nonempty. Condition B therefore imposes a lower bound on θ\theta, below which the peak-normalized beam depth becomes infinite.

Proposition 2. For a circular array under the peak-normalized 3-dB beam depth definition, BD3dBrel\mathrm{BD}_{3\mathrm{dB}}^{\mathrm{rel}} is finite when the cone angle satisfies

θ>sin⁡−1(0.5639λR).(26)\theta>\sin^{-1}\left(0.5639\frac{\lambda}{R}\right). \tag*{(26)}

Proof: Using the peak-finding algorithm, Condition B gives μccrit,B≈2.2558\mu_{\mathrm{c}}^{\mathrm{crit,B}}\approx2.2558. Therefore,

θcrit,B=sin⁡−1(μccrit,Bλ4R)≈sin⁡−1(0.5639λR).(27)\theta_{\mathrm{crit,B}}=\sin^{-1}\left(\frac{\mu_{\mathrm{c}}^{\mathrm{crit,B}}\lambda}{4R}\right)\approx\sin^{-1}\left(0.5639\frac{\lambda}{R}\right). \tag*{(27)}

Thus, Prop. 1a, 1b and 2 establish the critical cone angle thresholds that govern the existence and finiteness of both definitions of the 3-dB beam depth for the CA case. In the following subsection, we apply the same procedure to the SA case and quantify the corresponding critical cone angles.

Square array

Evaluating Condition A1 for the SA case, similarly yields a lower bound on θ\theta under the absolute 3-dB beam depth definition. Let Gnorms(μs,∞)G^{s}_{\mathrm{norm}}(\mu_s,\infty) be defined as

Gnorms(μs,∞)=lim⁡τs→∞Gnorms(μs,τs).(28)G^{s}_{\mathrm{norm}}(\mu_s,\infty)=\lim_{\tau_s\to\infty}G^{s}_{\mathrm{norm}}(\mu_s,\tau_s). \tag*{(28)}

Using this definition we can quantify the limits for a SA:

Proposition 3a. For a square array under the absolute 3-dB beam depth definition, BD3dBabs\mathrm{BD}^{\mathrm{abs}}_{3\mathrm{dB}} is finite when the cone angle satisfies

θ>sin⁡−1(0.91606λDAP).(29)\theta>\sin^{-1}\left(0.91606\frac{\lambda}{D_{\mathrm{AP}}}\right). \tag*{(29)}

Proof. Evaluating the limit τs→∞\tau_s\to\infty for (14) (see Appendix B) gives

Gnorms(μs,∞)=4∣∫0π/4sec⁡2ϕejξs(ϕ)[1−jξs(ϕ)]−1ξs2(ϕ) dϕ∣2,(30)G^{s}_{\mathrm{norm}}(\mu_s,\infty)=4\left|\int_{0}^{\pi/4}\sec^2\phi\frac{e^{j\xi_s(\phi)}[1-j\xi_s(\phi)]-1}{\xi_s^2(\phi)}\,\mathrm{d}\phi\right|^2, \tag*{(30)}

where ξs(ϕ)=πμs2cos⁡(ϕ)\xi_s(\phi)=\frac{\pi\mu_s}{2\cos(\phi)}. Unlike the CA case, (30) does not simplify to a closed-form real expression because the square-aperture boundary depends on the azimuthal direction through the cos⁡(ϕ)\cos(\phi) term. Therefore, the asymptotic threshold must be obtained numerically using (30). Analysis of the accuracy of the numerical evaluation is presented in Section VI.

Similar to the CA case, for the absolute definition, the beam depth becomes infinite whenever Gnorms(μs,∞)≥0.5G^{s}_{\mathrm{norm}}(\mu_s,\infty)\ge0.5. Therefore, the corresponding asymptotic critical value μscrit,A1\mu^{\mathrm{crit,A1}}_s is obtained from Gnorms(μscrit,A1,∞)=0.5G^{s}_{\mathrm{norm}}(\mu^{\mathrm{crit,A1}}_s,\infty)=0.5. Numerically solving this equation using (30) gives the first positive solution μscrit,A1≈1.83212\mu^{\mathrm{crit,A1}}_s\approx1.83212 (rounding error), with the corresponding cone angle

θcrit,A1s=sin⁡−1(μscrit,A1λ2DAP)≈sin⁡−1(0.91606λDAP).(31)\theta^{s}_{\mathrm{crit,A1}}=\sin^{-1}\left(\frac{\mu^{\mathrm{crit,A1}}_s\lambda}{2D_{\mathrm{AP}}}\right)\approx\sin^{-1}\left(0.91606\frac{\lambda}{D_{\mathrm{AP}}}\right). \tag*{(31)}

Similar to the CA case the max gain cannot be calculated analytically hence we utilize the peak-finding algorithm to quantify the critical cone angle associated with Condition A2, and B. Evaluating Condition A2 for a SA similarly, gives an upper bound on θ\theta:

Proposition 3b. For a square array under the absolute 3-dB beam depth definition, BD3dBabs\mathrm{BD}^{\mathrm{abs}}_{3\mathrm{dB}} is defined only when the cone angle satisfies

θ≤sin⁡−1(4.99875λDAP).(32)\theta\le\sin^{-1}\left(4.99875\frac{\lambda}{D_{\mathrm{AP}}}\right). \tag*{(32)}

Proof. Numerically solving Condition A2 using the peak-finding algorithm for the SA case gives μscrit,A2≈9.9975\mu^{\mathrm{crit,A2}}_s\approx9.9975, with the critical cone angle:

θcrit,A2s=sin⁡−1(μscrit,A2λ2DAP)≈sin⁡−1(4.99875λDAP).(33)\theta^{s}_{\mathrm{crit,A2}}=\sin^{-1}\left(\frac{\mu^{\mathrm{crit,A2}}_s\lambda}{2D_{\mathrm{AP}}}\right)\approx\sin^{-1}\left(4.99875\frac{\lambda}{D_{\mathrm{AP}}}\right). \tag*{(33)}

Finally, evaluating Condition B for a SA gives a lower bound on θ\theta for a Bessel beam under the peak-normalized definition, giving:

Proposition 4. For a square array under the peak-normalized 3-dB beam depth definition, BD3dBrel\mathrm{BD}^{\mathrm{rel}}_{3\mathrm{dB}} is finite when the cone angle satisfies

θ>sin⁡−1(0.9383λDAP).(34)\theta>\sin^{-1}\left(0.9383\frac{\lambda}{D_{\mathrm{AP}}}\right). \tag*{(34)}

Proof. Using the peak-finding algorithm, numerical calculation of Condition B gives μscrit,B≈1.8765\mu^{\mathrm{crit,B}}_s\approx1.8765. The corresponding cone angle is

θcrit,Bs=sin⁡−1(μscrit,Bλ2DAP)≈sin⁡−1(0.9383λDAP).(35)\theta^{s}_{\mathrm{crit,B}}=\sin^{-1}\left(\frac{\mu^{\mathrm{crit,B}}_s\lambda}{2D_{\mathrm{AP}}}\right)\approx\sin^{-1}\left(0.9383\frac{\lambda}{D_{\mathrm{AP}}}\right). \tag*{(35)}

Thus, Prop. 3a, 3b and 4 establish the critical cone angle thresholds for the SA case under the absolute and peak-normalized 3-dB beam-depth definitions. Together with the CA results in Subsection V-A, these thresholds show that a phase-based Bessel beam is usable only over a restricted cone angle interval. In the following subsection, we utilize the CA and SA results and define the operating ranges.

Usable range of Bessel beams

The results in Subsections V-A and V-B define the usable cone angle ranges for phase-based Bessel beams under the absolute and peak-normalized beam depth definitions. For the CA, the absolute 3-dB beam depth is nonempty and finite only when both Conditions A1 and A2 are satisfied. Hence, the usable cone angle range is a result of Prop. 1a and 1b,

sin⁡−1(0.5533λR)<θ≤sin⁡−1(4.056975λR).(36)\sin^{-1}\left(0.5533\frac{\lambda}{R}\right)<\theta\le\sin^{-1}\left(4.056975\frac{\lambda}{R}\right). \tag*{(36)}

Similarly, for the SA case, the corresponding usable cone angle range is a result of Prop. 3a and 3b,

sin⁡−1(0.91606λDAP)<θ≤sin⁡−1(4.99875λDAP).(37)\sin^{-1}\left(0.91606\frac{\lambda}{D_{\mathrm{AP}}}\right)<\theta\le\sin^{-1}\left(4.99875\frac{\lambda}{D_{\mathrm{AP}}}\right). \tag*{(37)}

In both cases, the lower bound on θ\theta corresponds to Condition A1, while the upper bound corresponds to Condition A2. For the peak-normalized definition, the half-power region is always nonempty, but it is finite only when Condition B is satisfied. Prop. 2 and Prop. 4 therefore define the corresponding usable ranges for the CA and SA, respectively.

From Prop. 1a, 2, 3a, and 4, we notice that the critical cone angles are proportional to λ/R\lambda/R or λ/DAP\lambda/D_{\mathrm{AP}}. Since practical communication systems typically satisfy λ≪R\lambda\ll R and λ≪DAP\lambda\ll D_{\mathrm{AP}}, i.e., being electrically large, the corresponding critical cone angles, θcrit,A1c\theta^{c}_{\mathrm{crit,A1}}, θcrit,A2c\theta^{c}_{\mathrm{crit,A2}}, θcrit,A1s\theta^{s}_{\mathrm{crit,A1}}, and θcrit,Bs\theta^{s}_{\mathrm{crit,B}}, lie in the small-angle regime. Consequently, when the approximation tan⁡(θ)≈sin⁡(θ)\tan(\theta)\approx\sin(\theta) holds, the maximum usable non-diffracting range can be expressed as Zmax⁡c≈dFAc/(2μc)Z^{c}_{\max}\approx d^{c}_{\mathrm{FA}}/(2\mu_c) for the CA and Zmax⁡s≈dFAs/(22μs)Z^{s}_{\max}\approx d^{s}_{\mathrm{FA}}/(2\sqrt{2}\mu_s) for the SA. The resulting critical values of Zmax⁡Z_{\max} are summarized in Table II.

PropositionCritical Zmax⁡Z_{\max}
Proposition 1aRtan⁡(sin⁡−1(0.5533λR))≈0.226dFAc\frac{R}{\tan\left(\sin^{-1}\left(0.5533\frac{\lambda}{R}\right)\right)} \approx0.226d_{\mathrm{FA}}^{\mathrm{c}}
Proposition 2Rtan⁡(sin⁡−1(0.5639λR))≈0.222dFAc\frac{R}{\tan\left(\sin^{-1}\left(0.5639\frac{\lambda}{R}\right)\right)} \approx0.222d_{\mathrm{FA}}^{\mathrm{c}}
Proposition 3aDAP2tan⁡(sin⁡−1(0.91606λDAP))≈0.193dFAs\frac{D_{\mathrm{AP}}}{\sqrt{2}\tan\left(\sin^{-1}\left(0.91606\frac{\lambda}{D_{\mathrm{AP}}}\right)\right)} \approx0.193d_{\mathrm{FA}}^{\mathrm{s}}
Proposition 4DAP2tan⁡(sin⁡−1(0.9383λDAP))≈0.188dFAs\frac{D_{\mathrm{AP}}}{\sqrt{2}\tan\left(\sin^{-1}\left(0.9383\frac{\lambda}{D_{\mathrm{AP}}}\right)\right)} \approx0.188d_{\mathrm{FA}}^{\mathrm{s}}

Table II. Critical geometric non-diffracting ranges.

These results show that both beam depth definitions impose a finite upper limit on the usable non-diffracting range. For the circular aperture, this limit is approximately 0.22dFAc0.22d_{\mathrm{FA}}^{\mathrm{c}} under both definitions. For the square aperture, the corresponding limit is approximately 0.19dFAs0.19d_{\mathrm{FA}}^{\mathrm{s}}. The close agreement between these limits arises because the peak normalized array gain remains close to the plane-wave reference near the critical cone angle. Consequently, the absolute threshold 0.5 and the peak-normalized threshold 0.5Gpeak0.5G_{\mathrm{peak}} are similar, causing the two beam-depth definitions to yield nearly identical limiting ranges. Thus, although the absolute limiting range scales with array size and wavelength through dFAd_{\mathrm{FA}}, its normalized value approaches a geometry-dependent constant.

Numerical analysis

In this section, we first verify the accuracy of the Fresnel approximations underlying the normalized array-gain expressions in (10) and (14) for the CA and SA, respectively. We then validate the critical cone angles obtained from Conditions A1, A2, and B which rely on the convergence of the numerical integration for the SA and the accuracy of the peak-finding algorithm (discussed in Appendices D and E, respectively). Building on these validations, we compare the analytical maximum usable non-diffracting range with the numerically obtained 3-dB beam depth, and then show the impact of discretization on the derived analytical limits. Finally, we investigate the engineering implications through comparisons between CA and SA, and practical Zmax⁡Z_{\max} limits for future near-field systems at varying array sizes and carrier frequencies.

Accuracy analysis

We first validate the Fresnel approximations used to derive the normalized array-gain expressions in (10) and (14). As shown in Sections III-A and III-B, the approximation error remains below 3.5×10−33.5 \times10^{-3} provided that the UE distance satisfies z≥2.5Rz \ge2.5R for the CA and z≥1.2DAPz \ge1.2D_{\mathrm{AP}} for the SA. Restricting the analysis to the radiative near-field reduces these conditions to R>2λR > 2\lambda and DAP>1.32λD_{\mathrm{AP}} > 1.32\lambda, respectively, which are readily satisfied for electrically large apertures.

Figs. 5a and 5b compare the exact normalized array gain with the proposed Fresnel-based expressions for representative array sizes (R,DAP={0.1,1.0} m)(R,D_{\mathrm{AP}}=\{0.1,1.0\}\ \mathrm{m}), cone angles (θ={0.5∘,0.25∘,0.1∘})(\theta=\{0.5^\circ,0.25^\circ,0.1^\circ\}), and a carrier frequency of 300 GHz. For both array geometries, the proposed expressions closely match the exact solution. This validates the Fresnel-based expressions

Figure 5a. Comparison of (4) with (10) for a circular array.

Figure 5b. Comparison of (4) with (14) for a square array.

Fig. 5. Fresnel approximation compared with the exact integral GnormG_{\mathrm{norm}} underlying the analytical evaluation of Condition A1 and the corresponding critical cone-angle thresholds. Conditions A2 and B additionally depend on the numerically evaluated peak array gain. The peak-finding algorithm used for their evaluation is validated in Appendix E, with relative errors in the critical angles below 0.01% for both the CA and SA.

Critical Zmax⁡Z_{\max} and the 3-dB beam depth

In this subsection, we compare the geometric non-diffracting range Zmax⁡Z_{\max} with the corresponding 3-dB beam depth. As discussed in Section IV, Zmax⁡Z_{\max} provides a geometric estimate of the finite-aperture non-diffracting range but does not capture the evolution of the beam’s spatial confinement, which is instead characterized by the 3-dB beam depth. Figs. 6 and 7 compare the analytical Zmax⁡Z_{\max} (dashed curves) with the numerically obtained 3-dB tail location (markers) for the CA and SA, respectively, at 300 and 500 GHz, with the insets highlighting the critical cone angles. We consider smaller angles to highlight the limiting cone angle regime.

For Conditions A1 and A2 (Figs. 6a and 7a), a finite absolute 3-dB beam depth exists only for θcrit,A1<θ≤θcrit,A2\theta_{\mathrm{crit,A1}} < \theta\le\theta_{\mathrm{crit,A2}}. As θ→θcrit,A1\theta\to\theta_{\mathrm{crit,A1}}, the beam depth diverges, marking the onset of the infinite depth regime predicted by Prop. 1a and 3a, whereas beyond θcrit,A2\theta_{\mathrm{crit,A2}} it becomes undefined because the normalized array gain no longer exceeds the absolute 3-dB threshold. The The limits of 0.22dFAc0.22d_{\mathrm{FA}}^{c} and 0.19dFAs0.19d_{\mathrm{FA}}^{s} intersect Zmax⁡Z_{\max} almost exactly at θcrit,A1\theta_{\mathrm{crit,A1}}, visually confirming the derived maximum usable non-diffracting ranges. Increasing the operating frequency shifts both A1 and A2 thresholds toward smaller angles and reduces the absolute angular width of the finite-depth interval.

For the peak-normalized criterion (Figure 6b and Figure 7b), the 3-dB beam depth remains defined for all cone angles, with θcrit,B\theta_{\mathrm{crit,B}} marking the transition from finite to infinite depth. Its intersection with Zmax⁡Z_{\max} again coincides closely with the corresponding asymptotic limit, confirming the predicted critical range. Increasing frequency lowers the single threshold θcrit,B\theta_{\mathrm{crit,B}}, thereby enlarging the set of cone angles satisfying the finite-depth condition. Finally, θcrit,B\theta_{\mathrm{crit,B}} lies close to θcrit,A1\theta_{\mathrm{crit,A1}} for both array geometries, showing that the two criteria identify nearly the same onset of infinite depth behavior.

Comparison between the geometric non-diffracting range and the 3-dB beam-depth-based usable range for the CA

Figure 6. Comparison between the geometric non-diffracting range Zmax⁡Z_{\max} and the 3-dB beam-depth-based usable range for the CA at R=0.5R = 0.5 m. (a) Conditions A1 and A2 based on the absolute 3-dB beam depth. (b) Condition B based on the peak-normalized 3-dB beam depth.

Comparison between the geometric non-diffracting range and the 3-dB beam-depth-based usable range for the SA

Figure 7. Comparison between the geometric non-diffracting range Zmax⁡Z_{\max} and the 3-dB beam-depth-based usable range for the SA at DAP=0.5D_{\mathrm{AP}} = 0.5 m. (a) Conditions A1 and A2 based on the absolute 3-dB beam depth. (b) Condition B based on the peak-normalized 3-dB beam depth.

Effect of discretization

For practical discrete array applications, we now evaluate the derived array limits numerically across different antenna element spacings, carrier frequencies, and array sizes. Figure 8a compares the critical non-diffracting range of the discrete circular array (antenna elements arranged in a square grid) with the analytical limit of 0.22dFAc0.22d_{\mathrm{FA}}^{c} from Prop. 1a for spacings from 0.5λ0.5\lambda to 5.5λ5.5\lambda. At half-wavelength spacing, the maximum relative error is only 0.16%0.16\%, remaining below 0.2%0.2\% up to ≈2.5λ\approx2.5\lambda and below 7%7\% even for the sparsest array.

Figure 8. Critical Zmax⁡Z_{\max} obtained numerically and analytically (under continuous array approximation) for discrete arrays at varying antenna element spacing.

Figure 8b shows the corresponding results for the discrete square array. At half-wavelength spacing, the maximum relative error remains approximately 0.3%0.3\%; however, the deviation increases with spacing, particularly for larger apertures and lower frequencies, reaching 87.8%87.8\% for the sparsest configurations. Increasing the operating frequency improves the accuracy, with errors below ≈5%\approx5\% even at the largest spacings.

Thus, the limits derived using continuous array approximation accurately predict both array geometries for half-wavelength discrete arrays. The critical non-diffracting range prediction for the circular array remains robust under sparse sampling, whereas the square array is more sensitive to discretization. For both geometries, agreement improves with increasing frequency and array size as the number of elements increases and the discrete aperture more closely resemble the continuous array approximation.

ditions A1 and B. The critical angles decrease with frequency and remain nearly identical between the two conditions. The inscribed and circumscribed circles provide the upper and lower bounds, respectively, while the SA lies between them and is most closely approximated by the equal-area circle.

The corresponding limiting Zmax⁡Z_{\max} values in Figure 9c follow the same geometric ordering, with the inscribed and circumscribed circles providing the lower and upper bounds, respectively. The SA limit is approximately 17.1% larger than that of the equal-area circle, while the circumscribed-circle limit is approximately 25.5% larger than the SA; the two circular bounds differ by more than 41.5%. Thus, the limiting non-diffracting range depends on array geometry, with the circular cases providing bounds for the SA and the equal-area circle giving the closest approximation.

Illustration of array geometries, critical cone angle, and critical non-diffracting range

Figure 9. Impact of array geometry on the critical cone angle and maximum usable non-diffracting range for CA and SA under different aperture criteria.

Effect of array geometry

Having established the critical cone angles and maximum usable non-diffracting ranges for the CA and SA, we next examine the impact of array geometry on these limits in Figure 9. Since a square aperture has no unique equivalent circular radius, we consider three common equivalence criteria: the inscribed (R=DAP/2R = D_{\mathrm{AP}}/2), equal-area (R=DAP/πR = D_{\mathrm{AP}}/\sqrt{\pi}), and circumscribed (R=DAP/2R = D_{\mathrm{AP}}/\sqrt{2}) circles, all illustrated in Figure 9a.

Fig. 9b compares the resulting critical cone angles for Conditions A1 and B. The critical angles decrease with frequency and remain nearly identical between the two conditions. The inscribed and circumscribed circles provide the upper and lower bounds, respectively, while the SA lies between them and is most closely approximated by the equal-area circle.

Implications for future near-field communication systems

We now illustrate the implications of the derived limits for different microwave, mmWave, and sub-THz transmitter configurations in Figure 10. To explore the fundamental limits numerically, we deliberately range the configurations from the microwave 3.5 GHz configuration all the way to futuristic large-scale sub-THz arrays operating at 300 GHz. Specifically, Figure 10a presents the results for the CA, while Figure 10b focuses on SA metrics. To be consistent with previous results, all cases employ half-wavelength-spaced discrete arrays (d=λ/2d = \lambda/2), with CA and SA having the same dFAd_{\mathrm{FA}} (i.e. R=DAP/2R = D_{\mathrm{AP}}/\sqrt{2}).

Figure 10. Comparing the conventional Fraunhofer distance, maximum Zmax⁡Z_{\max} of a Bessel beam with a finite depth-of-field, and the corresponding peak location.

Fig. 10a illustrates the maximum usable non-diffracting range, and corresponding axial peak location for different CAs. Although the maximum non-diffracting range with a finite depth-of-field occupies only a fraction of the Fraunhofer distance, it can still remain non-negligible for future near-field communication systems.

For instance, at 140 GHz, a circular indoor AP with R=0.3/2R = 0.3/\sqrt{2} m ≈21\approx21 cm achieves a critical Zmax⁡cZ_{\max}^{c} of approximately 38 m, with an axial peak near ≈23\approx23 m. Similar observations hold for the corresponding SAs in Figure 10b. Furthermore, for the same Fraunhofer distance, SAs exhibit slightly shorter maximum non-diffracting ranges than their circular counterparts, reaching approximately 32 m at 140 GHz and 193 m at 300 GHz with a 0.5×\times0.5 m transmitter array. Importantly, the established limit did not preclude the practical use of Bessel beams beyond those, but rather establish the maximum propagation range over which their finite-depth behavior can be exploited. Finally, the results in Figure 10b present a close match between the peak power locations at the finite depth-of-field boundary for the focused beam and the Bessel beam, indicating that there could be a similar general limit for any type of near-field beam to be explored in the future work.

Conclusion

In this work, we analytically characterized the finite-depth behavior of phase-based Bessel beams for near-field communications. First, we derived expressions for the normalized array gain of an ideal CA and a SA. We then derived absolute and peak-normalized 3-dB depth-of-field criteria for both circular and square geometries. Later, from the principal conditions governing the existence and finiteness of the Bessel beam depth in different setups, we obtained critical cone-angle thresholds that partitioned the Bessel beam configuration space into: (i) undefined, (ii) finite, and (iii) infinite depth-of-field regimes. Finally, these thresholds enabled us to determine the maximum usable non-diffracting range of a Bessel beam generated by a given array configuration.

The analysis shows that the conventional non-diffracting range Zmax⁡Z_{\max} alone does not determine whether a Bessel beam maintains a finite or infinite depth of field. The actual finite depth-of-field range depends on the critical Bessel cone angle, which is subject to: (i) the array aperture, (ii) the array shape, and (iii) the signal frequency. Importantly, for electrically large antennas, this leads to a maximum usable non-diffracting range of ≈0.22dFA\approx0.22d_{\mathrm{FA}} for the CA and ≈0.19dFA\approx0.19d_{\mathrm{FA}} for the SA. Hence, only a fraction of the radiative near-field (≈20%\approx20\%) can support finite-depth Bessel-beam operation, with the exact limit determined by the array geometry, aperture size, and carrier wavelength. The performed computer simulations cross-verify the analytical results and also confirm that the limits for the discrete half-wavelength spaced arrays closely follow the ones established analytically under the continuous array approximation. The results also show that, despite these limits, the usable non-diffracting range of a Bessel beam at

mmWave and THz frequencies still spans over up to several tens of meters. Our findings thus provide both theoretical limits and engineering insights on how to optimize the Bessel beam parameters for future near-field communication systems.

Derivation of Gnormc(μc,∞)G^{c}_{\mathrm{norm}}(\mu_c,\infty)

Starting from (10), the far-side behavior is obtained by expanding the two exponential terms and the Fresnel-integral differences for τ→∞\tau\to\infty. This gives

Gnormc(μc,τc)=(8τc)2∣e−jπ2μc2τc(1−jπμc2)ejπμc2−1τc1π2μc2+O(τc−2)∣2(38)G^{c}_{\mathrm{norm}}(\mu_c,\tau_c)=(8\tau_c)^2\left|\frac{e^{-j\frac{\pi}{2}\mu_c^2\tau_c}\left(1-j\frac{\pi\mu_c}{2}\right)e^{j\frac{\pi\mu_c}{2}}-1}{\tau_c}\frac{1}{\pi^2\mu_c^2}+\mathcal{O}\left(\tau_c^{-2}\right)\right|^2 \tag*{(38)}

Since ∣e−jπ2μc2τc∣=1\left|e^{-j\frac{\pi}{2}\mu_c^2\tau_c}\right|=1, the oscillatory phase does not affect the magnitude. Therefore, we get

Gnormc(μc,∞)=64∣(1−jπμc2)ejπμc2−1π2μc2∣2.(39)G^{c}_{\mathrm{norm}}(\mu_c,\infty)=64\left|\frac{\left(1-j\frac{\pi\mu_c}{2}\right)e^{j\frac{\pi\mu_c}{2}}-1}{\pi^2\mu_c^2}\right|^2. \tag*{(39)}

Defining ξ=πμc/2\xi=\pi\mu_c/2, (39) can be written as

Gnormc(μc,∞)=4∣ejξ(1−jξ)−1ξ2∣2.(40)G^{c}_{\mathrm{norm}}(\mu_c,\infty)=4\left|\frac{e^{j\xi}(1-j\xi)-1}{\xi^2}\right|^2. \tag*{(40)}

Thus, we can obtain the real-form in (22) using (40).

Derivation of Gnorms(μs,∞)G^{s}_{\mathrm{norm}}(\mu_s,\infty)

Starting from (14), the far-side behavior is obtained by expanding the two exponential terms and the Fresnel-integral differences for τs→∞\tau_s \to\infty. For a fixed ϕ\phi, let as(ϕ)=μs/2cos⁡(ϕ)a_s(\phi)=\mu_s/2\cos(\phi). Using the large-τs\tau_s expansion of the corresponding integrand, we obtain Gnorms(μs,τs)G^{s}_{\mathrm{norm}}(\mu_s,\tau_s)

(16τs)2∣e−jπμs2τs8τs∫0π/4sec⁡2ϕ[ejπas(ϕ)−1jπas(ϕ)+1−ejπas(ϕ)(jπas(ϕ))2]dϕ+O(τs−2)∣2(41)(16\tau_s)^2\left|\frac{e^{-j\pi\mu_s^2\tau_s}}{8\tau_s}\int_0^{\pi/4}\sec^2\phi\left[\frac{e^{j\pi a_s(\phi)}-1}{j\pi a_s(\phi)}+\frac{1-e^{j\pi a_s(\phi)}}{(j\pi a_s(\phi))^2}\right]d\phi+\mathcal{O}\left(\tau_s^{-2}\right)\right|^2 \tag*{(41)}

Since ∣e−jπμs2τs∣=1\left|e^{-j\pi\mu_s^2\tau_s}\right|=1, the oscillatory phase does not affect the magnitude. Therefore, we get

Gnorms(μs,∞)=4∣∫0π/4sec⁡2ϕ[ejπas(ϕ)jπas(ϕ)+1−ejπas(ϕ)(jπas(ϕ))2]dϕ∣2.(42)G^{s}_{\mathrm{norm}}(\mu_s,\infty)=4\left|\int_0^{\pi/4}\sec^2\phi\left[\frac{e^{j\pi a_s(\phi)}}{j\pi a_s(\phi)}+\frac{1-e^{j\pi a_s(\phi)}}{(j\pi a_s(\phi))^2}\right]d\phi\right|^2. \tag*{(42)}

Let ξs(ϕ)=πas(ϕ)=πμs2cos⁡(ϕ)\xi_s(\phi)=\pi a_s(\phi)=\frac{\pi\mu_s}{2\cos(\phi)}, then (42) can be written as

Gnorms(μs,∞)=4∣∫0π/4sec⁡2ϕ[ejξs(ϕ)jξs(ϕ)+1−ejξs(ϕ)(jξs(ϕ))2] dϕ∣2.(43)G_{\mathrm{norm}}^s(\mu_s,\infty)=4\left|\int_0^{\pi/4}\sec^2\phi\left[\frac{e^{j\xi_s(\phi)}}{j\xi_s(\phi)}+\frac{1-e^{j\xi_s(\phi)}}{(j\xi_s(\phi))^2}\right]\,d\phi\right|^2. \tag*{(43)}

Equivalently, by combining the two terms inside the integral, we obtain

Gnorms(μs,∞)=4∣∫0π/4sec⁡2ϕejξs(ϕ)[1−jξs(ϕ)]−1ξs2(ϕ) dϕ∣2.(44)G_{\mathrm{norm}}^s(\mu_s,\infty)=4\left|\int_0^{\pi/4}\sec^2\phi\frac{e^{j\xi_s(\phi)}[1-j\xi_s(\phi)]-1}{\xi_s^2(\phi)}\,d\phi\right|^2. \tag*{(44)}

Thus we can obtain (30). Unlike the circular array case, the square array retains the angular integral because the array boundary depends on the azimuthal direction through the factor cos⁡(ϕ)\cos(\phi).

Proof for Max Gain Being Implicit

Define the inner term of (10) as A(τc)A(\tau_c) for a fixed μc\mu_c, so that Gnormc(μc,τc)=64τc2∣A(τc)∣2G_{\mathrm{norm}}^c(\mu_c,\tau_c)=64\tau_c^2|A(\tau_c)|^2. Then

∂Gnormc∂τc=128τc∣A(τc)∣2+128τc2ℜ{A∗(τc)dA(τc)dτc}.(45)\frac{\partial G_{\mathrm{norm}}^c}{\partial\tau_c}=128\tau_c|A(\tau_c)|^2+128\tau_c^2\Re\left\{A^*(\tau_c)\frac{dA(\tau_c)}{d\tau_c}\right\}. \tag*{(45)}

Therefore, a stationary point τc∗\tau_c^* satisfies

∣A(τc∗)∣2+τc∗ℜ{A∗(τc∗)dA(τc)dτc∣τc=τc∗}=0.(46)|A(\tau_c^*)|^2+\tau_c^*\Re\left\{A^*(\tau_c^*)\left.\frac{dA(\tau_c)}{d\tau_c}\right|_{\tau_c=\tau_c^*}\right\}=0. \tag*{(46)}

Because A(τc)A(\tau_c) contains Fresnel-integral differences with τc\tau_c-dependent arguments, (46) does not yield a clean closed-form expression for τc∗\tau_c^* or Gmax⁡(μc)G_{\max}(\mu_c) for the CA case. A similar argument applies to (14); hence, a closed-form solution for Gmax⁡(μs)G_{\max}(\mu_s) is also unavailable for the SA.

Numerical Accuracy of the SA Angular Integral

The SA normalized array-gain expression (14) contains a one-dimensional angular integral, which is evaluated numerically using NϕN_\phi samples. This appendix validates the choice Nϕ=8001N_\phi=8001 adopted throughout the paper by comparing the resulting Condition A1 critical angle, θcrit,A1s\theta_{\mathrm{crit,A1}}^s, with both a high-resolution numerical reference computed using Nϕ=128001N_\phi=128001 and the analytical approximation reported in Prop. 3a.

Fig. 11 illustrates the convergence of θcrit,A1s\theta_{\mathrm{crit,A1}}^s for different array sizes and operating frequencies. The solid curves show the error relative to the high-resolution numerical reference, while the dotted curves show the error relative to Prop. 3a. At Nϕ=8001N_\phi=8001, the numerical integration error remains below 10−1410^{-14} degrees for all considered configurations, whereas the error relative to Prop. 3a is below 10−610^{-6} degrees and is dominated by the rounding of the analytical constant rather than the numerical integration. Therefore, Nϕ=8001N_\phi=8001 provides a practical tradeoff between accuracy and computation time when evaluating (14) and the critical cone angles.

Figure 11. Absolute error in θcrit,A1s\theta_{\mathrm{crit,A1}}^s versus the number of angular-integration samples NϕN_\phi, for DAP={0.01,0.1}D_{\mathrm{AP}}=\{0.01,0.1\} m with f=300f=300 GHz, and f={100,500}f=\{100,500\} GHz with DAP=0.3D_{\mathrm{AP}}=0.3 m (relative errors are <10−5%<10^{-5}\%).

Peak-Finding Algorithm
1:Input: Array type κ∈{c,s}\kappa\in\{c,s\}, array size, λ\lambda, θ\theta, NcoarseN_{\mathrm{coarse}}, ϵz\epsilon_z, and NϕN_\phi for the SA.
2:Output: zpeakz_{\mathrm{peak}} and GpeakG_{\mathrm{peak}}.
3:Compute dFAkd_{\mathrm{FA}}^k, Zmax⁡kZ_{\max}^k, and define G(z)=Gnorm(θ,z)G(z)=G_{\mathrm{norm}}(\theta,z) using the corresponding CA or SA expression.
4:Set zmin⁡z_{\min} and zmax⁡z_{\max}.
5:Generate a logarithmic grid {zn}n=1Ncoarse\{z_n\}_{n=1}^{N_{\mathrm{coarse}}} over [zmin⁡,zmax⁡][z_{\min},z_{\max}].
6:Evaluate Gn=G(zn)G_n=G(z_n) for all nn.
7:Find i∗=arg⁡max⁡nGni^*=\arg\max_n G_n.
8:if i∗=1i^*=1 then
9:B←[z1,z2]\mathcal{B}\leftarrow[z_1,z_2].
10:else if i∗=Ncoarsei^*=N_{\mathrm{coarse}} then
11:B←[zNcoarse−1,zNcoarse]\mathcal{B}\leftarrow[z_{N_{\mathrm{coarse}}-1},z_{N_{\mathrm{coarse}}}].
12:else
13:B←[zi∗−1,zi∗+1]\mathcal{B}\leftarrow[z_{i^*-1},z_{i^*+1}].
14:end if
15:Solve zpeak=arg⁡min⁡z∈B[−G(z)]z_{\mathrm{peak}}=\arg\min_{z\in\mathcal{B}}[-G(z)], using bounded scalar optimization with tolerance ϵz\epsilon_z.
16:Set Gpeak=G(zpeak)G_{\mathrm{peak}}=G(z_{\mathrm{peak}}).
17:return zpeakz_{\mathrm{peak}}, GpeakG_{\mathrm{peak}}, zpeak/dFAkz_{\mathrm{peak}}/d_{\mathrm{FA}}^k, and zpeak/Zmax⁡kz_{\mathrm{peak}}/Z_{\max}^k.

Table 1 (PDF p. 12). Algorithm 1 Peak-Finding Algorithm

Numerical Peak-Finding Algorithm

Conditions A2 and B require the maximum normalized array gain over the axial UE location for a given cone angle θ\theta, Gpeak(θ)=max⁡z∈[zmin⁡,zmax⁡]Gnorm(θ,z)G_{\mathrm{peak}}(\theta)=\max_{z\in[z_{\min},z_{\max}]}G_{\mathrm{norm}}(\theta,z). Since the corresponding stationary-point equation is implicit (see Appendix C), the maximum is evaluated numerically. For the CA, the closed-form expression (10) is used, while for the SA, (14) is evaluated using Nϕ=8001N_\phi=8001 (see Appendix D). The search interval is selected as zmin⁡=max⁡(dRAk,10−8Zmax⁡k)z_{\min}=\max(d_{\mathrm{RA}}^k,10^{-8}Z_{\max}^k) and zmax⁡=max⁡(3Zmax⁡k,dFAk)z_{\max}=\max(3Z_{\max}^k,d_{\mathrm{FA}}^k), ensuring that the dominant Bessel peak is contained within the optimization region.

Algorithm 1 employs a coarse-to-fine strategy. A coarse axial grid first identifies the dominant lobe, after which bounded optimization refines the peak location within the corresponding local bracket. Unlike exhaustive search, which refines the entire search interval, the proposed method performs high-resolution optimization only within the dominant lobe, substantially reducing the number of gain evaluations while eliminating the dominant grid-quantization error. Specifically, if Δzi\Delta z_i denotes the local grid spacing and εz\varepsilon_z the optimizer tolerance, the local gain error satisfies 0≤Gpeak−Gnorm(θ,z^)≤Ki2εz20 \leq G_{\mathrm{peak}}-G_{\mathrm{norm}}(\theta,\hat{z}) \leq\frac{K_i}{2}\varepsilon_z^2, where KiK_i denotes the maximum absolute curvature of GnormG_{\mathrm{norm}} over the local bracket containing the dominant peak, and which is smaller than the exhaustive-search bound whenever εz<Δzi/2\varepsilon_z < \Delta z_i/2. Throughout this work, the optimization tolerance is set to 10−1410^{-14} equivalent to relative error of 10−12%10^{-12}\%.

Peak-gain error versus distance-grid granularity for the SA

Figure 13. Peak-gain error versus distance-grid granularity for the SA.

The practical accuracy is evaluated in Figure 12 and 13. The reference solution is obtained by repeatedly refining candidate peaks until convergence. The solid curves correspond to direct exhaustive search, whereas the dashed curves show the proposed peak finder initialized with the same coarse grid. For both the CA and SA, the exhaustive-search error decreases with grid refinement, while the proposed method consistently remains below the prescribed numerical tolerance across all considered array sizes, operating frequencies, and cone angles. The slightly higher errors observed for the SA originate from the numerical angular integration in (6). Overall, these results demonstrate that the proposed peak finder achieves substantially higher accuracy than direct exhaustive search at the same initial grid resolution.

Figure 12. Peak-gain error versus distance-grid granularity for the CA.

References

  1. [1]International Telecommunication Union Radiocommunication Sector (ITU-R), “Recommendation ITU-R M.2160-0: Framework and Overall Objectives of the Future Development of IMT for 2030 and Beyond,” https://www.itu.int/rec/R-REC-M.2160-0-202311-I/en, Nov 2023.
  2. [2]Z. Wang et al., “A tutorial on extremely large-scale MIMO for 6G: Fundamentals, signal processing, and applications,” IEEE Communications Surveys & Tutorials, vol. 26, no. 3, pp. 1560–1605, Jan 2024.arxiv.org/abs/2307.07340
  3. [3]J. M. Jornet et al., “The evolution of applications, hardware design, and channel modeling for Terahertz (THz) band communications and sensing: Ready for 6G?” Proceedings of the IEEE, vol. 113, no. 9, pp. 920–951, Sep 2025.arxiv.org/abs/2406.06105
  4. [4]A. Singh, V. Petrov, P. Sen, and J. M. Jornet, “Near-field terahertz communications for 6G and beyond: From concepts to realizations,” IEEE Signal Proc. Mag., vol. 42, no. 1, pp. 106–125, 2025.DOI
  5. [5]K. Zhi, C. Pan, H. Ren, K. K. Chai, C.-X. Wang, R. Schober, and X. You, “Performance analysis and low-complexity design for XL-MIMO with near-field spatial non-stationarities,” IEEE Journal on Selected Areas in Communications, vol. 42, no. 6, pp. 1656–1672, 2024.arxiv.org/abs/2304.00172
  6. [6]M. Li, J. M. Jornet, D. M. Mittleman, and C. Han, “Beam manipulation for terahertz communications,” Comm. Engineering, vol. 5, no. 1, p. 83, 2026.
  7. [7]K. K. Mutai, C. Stults, Z. Fang, and D. M. Mittleman, “Generation of non-diffractive terahertz Bessel beams using 3D printed spiral zone plates,” Appl. Phys. Lett., Oct 2025.DOI
  8. [8]P. Zhang, Y. Hu, T. Li, D. Cannan, X. Yin, R. Morandotti, Z. Chen, and X. Zhang, “Nonparaxial Mathieu and Weber accelerating beams,” Phys. Rev. Lett., Nov 2012.DOI
  9. [9]H. Chen, A. Kludze, and Y. Ghasempour, “Curving around obstacles via nn-enabled wavefront shaping in sub-THz wireless networks,” in Proc. IEEE GLOBECOM, Dec 2024.DOI
  10. [10]C. Zhu, “Near field transmission using Hermite-Gaussian modes,” in Proc. IEEE APWC, Sep 2025.arxiv.org/abs/2505.10914
  11. [11]P. Mollahosseini and Y. Ghasempour, “Fast vortex beam alignment for OAM mode multiplexing in LOS MIMO networks,” IEEE Trans. on Wireless Commun., vol. 25, pp. 10546–10559, 2026.arxiv.org/abs/2508.19034
  12. [12]E. Björnson, Ö. T. Demir, and L. Sanguinetti, “A primer on near-field beamforming for arrays and reconfigurable intelligent surfaces,” in Proc. IEEE Asilomar Conf. on Signals, Systems, and Computers, Mar 2021.arxiv.org/abs/2110.06661
  13. [13]Z. Wang, X. Mu, and Y. Liu, “Beamfocusing optimization for near-field wideband multiuser communications,” IEEE Trans. on Wireless Commun., vol. 73, no. 1, pp. 555–572, 2024.DOI
  14. [14]J. Chen, F. Gao, M. Jian, and W. Yuan, “Hierarchical codebook design for near-field mmWave MIMO communication systems,” IEEE Wireless Commun. Lett., vol. 12, no. 11, pp. 1926–1930, 2023.arxiv.org/abs/2212.07847
  15. [15]X. Wei, L. Dai, Y. Zhao, G. Yu, and X. Duan, “Codebook design and beam training for extremely large-scale RIS: Far-field or near-field?” China Communications, vol. 19, no. 6, pp. 193–204, 2022.arxiv.org/abs/2109.10143
  16. [16]A. Kosasih, Ö. T. Demir, and E. Björnson, “Near-field beamfocusing, localization, and channel estimation with modular linear arrays,” IEEE Trans. on Wireless Commun., vol. 25, pp. 8784–8798, 2026.
  17. [17]J. Durmin, J. J. Miceli, and J. H. Eberly, “Diffraction-free beams,” Phys. Rev. Lett., Apr 1987.
  18. [18]A. Singh, V. Petrov, H. Guerboukha, I. V. Reddy, E. W. Knightly, D. M. Mittleman, and J. M. Jornet, “Wavefront engineering: Realizing efficient terahertz band communications in 6G and beyond,” IEEE Wireless Commun., Nov 2023.arxiv.org/abs/2305.12636
  19. [19]A. Singh, I. V. Reddy, D. Bodet, and J. M. Jornet, “Bessel beams for 6G - A performance analysis,” in Proc. IEEE Asilomar Conf. on Signals, Systems, and Computers, Mar 2022.DOI
  20. [20]Y. Ghasempour, A. Madanayake, and J. M. Jornet, “Spectrum management in near-field multi-user wireless communications,” IEEE Commun. Mag., vol. 63, no. 5, pp. 222–227, 2025.DOI
  21. [21]S. Uchimura, J. M. Jornet, and K. Ishibashi, “Optimization and characterization of near-field beams with uniform linear arrays,” IEEE Trans. on Wireless Commun., vol. 25, pp. 6904–6920, 2025.
  22. [22]—, “Optimal wavefronts for maximum ratio transmissions under path blockage effects,” IEEE Trans. on Wireless Commun., vol. 25, pp. 11051–11067, 2026.DOI
  23. [23]A. Jolly, V. Petrov, G. Fodor, and E. Björnson, “Bessel beam optimization for near-field THz communications under UE location uncertainty,” in Proc. IEEE VTC Fall, Jul 2026.arxiv.org/abs/2607.07069
  24. [24]I. V. A. K. Reddy et al., “Ultrabroadband terahertz-band communications with self-healing Bessel beams,” Commun. Engineering, vol. 2, no. 1, p. 70, 2023.DOI
  25. [25]V. Vazdnian and Y. Ghasempour, “NearLoc: Fast non-coherent near-field localization with Bessel beams,” in IEEE ICCSPA, June 2026.
  26. [26]A. Kludze, D. Bodet, J. M. Jornet, and Y. Ghasempour, “Distributed wavefront shaping in radiative near-field sub-terahertz wireless networks,” Nature Communications, vol. 17, no. 1, p. 5595, 2026.DOI
  27. [27]M. Neiman, “The principle of reciprocity in antenna theory,” Proc. of the IRE, Dec 1943.DOI
  28. [28]C. Ouyang, Z. Wang, Y. Chen, X. Mu, and P. Zhu, “A primer on near-field communications for next-generation multiple access,” Proc. of the IEEE, Aug 2024.arxiv.org/abs/2408.00952
  29. [29]K. T. Selvan and R. Janaswamy, “Fraunhofer and Fresnel distances: Unified derivation for aperture antennas,” IEEE Antenna and Prop. Mag., vol. 59, pp. 12–15, 2017.DOI
  30. [30]W. Southwell, “Validity of the Fresnel approximation in the near field,” Journal of the Optical Society of America, vol. 71, no. 1, pp. 7–14, 1981.DOI

Paper details

vlnc:W_fy4uqqwy
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 7

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…
Contents