PaperlayerView exact revision on arXiv

Source-linked AI summary

Towards 6G MIMO: Massive Spatial Multiplexing, Dense Arrays, and Interplay Between Electromagnetics and Processing

Emil Björnson, Chan-Byoung Chae, Robert W. Heath Jr., Thomas L. Marzetta, Amine Mezghani, Luca Sanguinetti, Fredrik Rusek, Miguel R. Castellanos, Dongsoo Jun, Özlem Tugfe Demir

arXiv:2401.02844v1cs.ITeess.SPmath.IT

TL;DR

Rising traffic and scarce spectrum motivate higher sum spectral efficiency, while larger 6G arrays raise questions about near-field propagation and spatial degrees of freedom. The paper surveys estimation and physically consistent modeling approaches; it concludes that increased antenna spacing can benefit single-user MIMO when it places propagation in the radiative near-field, where beamwidth can be smaller than the array.

Abstract

from arXiv · show

The increasing demand for wireless data transfer has been the driving force behind the widespread adoption of Massive MIMO (multiple-input multiple-output) technology in 5G. The next-generation MIMO technology is now being developed to cater to the new data traffic and performance expectations generated by new user devices and services in the next decade. The evolution towards "ultra-massive MIMO (UM-MIMO)" is not only about adding more antennas but will also uncover new propagation and hardware phenomena that can only be treated by jointly utilizing insights from the communication, electromagnetic (EM), and circuit theory areas. This article offers a comprehensive overview of the key benefits of the UM-MIMO technology and the associated challenges. It explores massive multiplexing facilitated by radiative near-field effects, characterizes the spatial degrees-of-freedom, and practical channel estimation schemes tailored for massive arrays. Moreover, we provide a tutorial on EM theory and circuit theory, and how it is used to obtain physically consistent antenna and channel models. Subsequently, the article describes different ways to implement massive and dense antenna arrays, and how to co-design antennas with signal processing. The main open research challenges are identified at the end.

I. INTRODUCTION

The introduction frames UM-MIMO as a response to rising traffic and the limits of simply enlarging 5G arrays: 6G arrays encounter near-field propagation, spatial degrees-of-freedom limits, and new modeling and implementation challenges. The near-field discussion explains how spherical wavefronts affect array gain and how many small antennas can preserve gain while exploiting those effects.

I. INTRODUCTION

  • Multiple antennas support adaptive directivity, robustness to fading, and spatial multiplexing, including simultaneous communication with multiple users or multiple layers to one user.The introduction describes beamforming as focusing signals at desired receivers while avoiding interference at specific locations.Source: §I · pp. 1–2
  • 6G traffic growth and scarce spectrum motivate higher sum spectral efficiency, while larger arrays bring near-field propagation, spatial degrees-of-freedom limits, and modeling and implementation challenges.The paper describes 100 times higher capacity than current networks as a target for next-generation MIMO.Source: §I · p. 2
  • The paper surveys radiative near-field multiplexing and channel estimation, then develops EM and circuit-theory models, array architectures, antenna–signal-processing co-design, and open challenges.The introduction identifies physically consistent channel models as part of this progression.Source: §I · pp. 2–3

II. MIMO COMMUNICATION IN THE RADIATIVE NEAR-FIELD

  • The radiative near-field has noticeable spherical wavefront curvature, unlike the nearly planar far-field; the reactive near-field involves non-radiated effects such as inductive coupling and evanescent fields.The paper focuses on the Fresnel region, not reactive-near-field technologies such as NFC.Source: §II · p. 3
  • At least 95% of maximum gain is achieved by 100 small antennas in the illustrated near-field case, versus 35% for one antenna covering the same surface.Combining the antenna signals requires receiver hardware, also needed for beamforming.Source: §II · p. 4
  • In the Fresnel region, phase variations across a large array can matter even when power variations are negligible; power variation must still be modeled for asymptotic large-array limits and non-line-of-sight settings.Geometric pathloss variation cannot be mitigated by signal processing or smaller antenna elements.Source: §II · pp. 4–5

A. Beamfocusing in the radiative near-field of arrays

The section characterizes beamfocusing by its angular and depth behavior. Angular beamwidth is limited in both the near- and far-field, while focusing closer than dF/10 produces finite beamdepth; farther focusing yields a beam extending to infinity.

A. Beamfocusing in the radiative near-field of arrays

  • With phase shifts aligned for constructive interference, M antennas provide M times the received power of one antenna at the same transmit power when pathloss is uniform across the array.The section states this array gain is the same in the far-field and radiative near-field.Source: A. Beamfocusing in the radiative… · p. 5
  • The angular array-gain pattern is independent of propagation distance, and its squared sinc taper gives half the array gain at a nonzero angular offset.The section approximates the antenna sum by an integral, tight when antenna spacing is small.Source: A. Beamfocusing in the radiative… · p. 5
  • Although angular beamwidth is the same at a given array size and spacing across distances, physical beamwidth is proportional to receiver distance and is smaller in the radiative near-field.Increasing the number of antennas per dimension narrows the beamwidth.Source: A. Beamfocusing in the radiative… · p. 6
  • For F < dF/10, the focused beam has finite half-power beamdepth; for F > dF/10, it continues to infinity.The finite-depth case applies to receivers in the radiative near-field.Source: A. Beamfocusing in the radiative… · p. 6, Figure 4 · p. 6

B. Near-field spatial multiplexing

Radiative near-field propagation can improve multiuser separation and support massive spatial multiplexing, while narrow beams can also enable multiple streams to one user. The examples show that performance depends on using a suitable propagation model and antenna spacing.

B. Near-field spatial multiplexing

  • Near-field focusing reduces beamwidth in width and depth, lowering interference risk and helping distinguish concurrent UE channels.The far-field approximation moves near-field UEs outward, widening and deepening beamfocus areas.Source: B. Near-field spatial multiplexing · pp. 6–7
  • LMMSE receive combining is used in the uplink example, with co-user interference treated as noise and the combining vector balancing signal amplification against interference rejection.The setup operates at 30 GHz with a 1 × 0.5 m BS array of 5000 antennas and UEs at 15–500 m.Source: B. Near-field spatial multiplexing · p. 7
  • With the exact near-field model, sum SE keeps growing as UEs increase, whereas the mismatched far-field model saturates around 200 UEs.The comparison uses a 5000-antenna setup; the exact model also produces higher average SE per UE.Source: B. Near-field spatial multiplexing · p. 7, Figure 5 · p. 7
  • For single-user MIMO with 16 antennas at each end, increasing BS spacing raises SE until a peak around 8 m, then sidelobes reduce it.At the peak, the channel has 16 equally large singular values, allowing a different narrow beam toward each receive antenna.Source: B. Near-field spatial multiplexing · p. 8

III. SPATIAL DEGREES-OF-FREEDOM

Spatial degrees-of-freedom (DoF) describe how many independent signal dimensions a channel can support. The section relates spectral DoF to bandwidth and spatial DoF to MIMO multiplexing, while explaining limits imposed by antenna counts and array area.

A. Spectral degrees-of-freedom

  • A signal in bandwidth B has B complex-valued samples per second as spectral DoF; with 16-QAM, each sample carries 4 bits, yielding 40 Mbps at 10 MHz.The passage describes the signal as a complex baseband representation of a real passband channel.Source: A. Spectral degrees-of-freedom · pp. 8–9

B. (Massive) MIMO: An information-theoretic perspective

  • MIMO creates parallel spatial channels for independent signals, adding spatial dimensions to spectral DoF; 6G data-rate growth motivates creating these dimensions.The passage frames 100× or 1000× growth as requiring new signal dimensions through MIMO.Source: B. (Massive) MIMO: An… · p. 9, A. Spectral degrees-of-freedom · p. 9
  • A single-user MIMO channel has at most min(Mr, Mt) spatial DoFs, giving min(Mr, Mt) × B total DoF per second; 64 antennas at each end offer potential for a 64-fold capacity increase over one antenna.The cited ideal-channel capacity result assumes equal singular values.Source: B. (Massive) MIMO: An… · p. 9
  • The spatial DoFs in multi-user MIMO cannot exceed min(M, K), while increasing M and K together at a preferably large fixed M/K can maintain near-constant per-UE SE as sum SE grows K-fold.Here M is the number of BS antennas and K the number of single-antenna UEs in the cited model.Source: B. (Massive) MIMO: An… · p. 10
  • Massive MIMO uses many more BS antennas than multiplexed devices, reducing interference and providing channel hardening; TDD uplink pilots support CSI acquisition for arbitrarily many BS antennas.The passage gives M/K ≥ 8 as an example antenna-to-device ratio.Source: B. (Massive) MIMO: An… · p. 10
  • With fixed bandwidth and transmission interval, spectral DoF are limited to TB, so increasing time samples cannot increase them indefinitely.The section uses this temporal-oversampling analogy to motivate a spatial sampling limit for arrays.Source: A. Spectral degrees-of-freedom · p. 8, B. (Massive) MIMO: An… · p. 10

C. Shannon-Nyquist sampling theorem for electric fields

The spatial sampling theorem uses the wavenumber support of electric fields to characterize the spatial degrees-of-freedom (DoF) available to an array. For planar arrays, λ/2 spacing captures the available DoF; closer spacing oversamples rather than increasing them.

C. Shannon-Nyquist sampling theorem for electric fields

  • Propagating electric fields have compact support in the wavenumber plane: a disk of radius κ, which is spatially band-limited.Scattering can further restrict that support.Source: C. Shannon-Nyquist sampling theorem… · p. 11
  • A planar aperture of area LxLy has spatial DoF proportional to LxLy/λ^2, with each λ^2 of aperture observing π DoF.The 2D count is lower than the product of the corresponding horizontal and vertical 1D counts by a factor of π/4 ≈ 0.79, because horizontal and vertical wavenumbers are correlated.Source: C. Shannon-Nyquist sampling theorem… · p. 11

D. Implications for MIMO Systems

  • Uniform planar arrays with λ/2 spacing capture all available spatial DoF, so conventional array spacing is sufficient.Reducing spacing below λ/2 does not increase DoF; it spatially oversamples the field.Source: D. Implications for MIMO Systems · pp. 11–12
  • For a ULA in isotropic scattering, reducing spacing to λ/4 or λ/6 reduces spatial DoF to M/2 or M/3, respectively.For a 64×64 UPA, even λ/2 spacing yields only 79% large eigenvalues.Source: D. Implications for MIMO Systems · p. 11
  • More antennas than the spatial DoF require can still help: smaller antennas have more isotropic radiation patterns, but their sizes should shrink while total antenna area stays fixed.Increasing spacing beyond λ/2 may also benefit near-field propagation, as described earlier in the paper.Source: D. Implications for MIMO Systems · pp. 11–12

IV. CHANNEL ESTIMATION FOR LARGE MIMO ARRAYS

This section describes channel estimation for large MIMO arrays, where pilot overhead grows with the number of antennas unless channel structure is exploited. It contrasts LS estimation with MMSE estimation, which uses spatial correlation to improve estimation and reduce the required pilot length.

IV. CHANNEL ESTIMATION FOR LARGE MIMO ARRAYS

  • Channel estimation uses pilots to acquire CSI for downlink beamforming and coherent uplink combining, but increasing UE antennas and multiplexed UEs increases pilot length.The section models the channel as correlated Rayleigh fading, with spatial correlation determined by propagation geometry and scattering.Source: §IV · pp. 12–13

A. Least-squares estimation

  • LS requires no channel statistics, but its pilot matrix must be invertible, requiring pilot length equal to the number of UE antennas.A unitary pilot matrix minimizes LS MSE by allocating equal power across the channel dimensions.Source: A. Least-squares estimation · p. 13

B. Minimum mean-squared error estimation

  • MMSE uses the known spatial correlation matrix to allocate pilot power across eigendirections, assigning more power to directions with larger eigenvalues.It can use any pilot length and transmits pilots along the strongest eigendirections; when pilot length equals correlation rank, longer pilots add energy but explore no new channel dimensions.Source: B. Minimum mean-squared error estimation · p. 14
  • The MMSE estimator outperforms LS by exploiting spatial correlation and enables good estimation quality at significantly smaller pilot lengths; Fig. 10 plots NMSE versus pilot length for isotropic and clustered scattering.The dots mark MMSE performance when pilot length equals the rank of the corresponding spatial correlation matrix.Source: B. Minimum mean-squared error estimation · p. 14, Figure 10 · p. 14

C. Reduced-subspace least-square estimation

Reduced-subspace least squares (RS-LS) estimates channels only within a known lower-dimensional subspace, reducing pilot requirements without UE-specific statistical information. Its performance depends on the subspace and pilot length, while smaller antenna spacing can increase spatial correlation and improve estimation.

C. Reduced-subspace least-square estimation

  • The RS-LS estimator projects the channel into a reduced subspace, estimates its coefficients there, and maps the estimate back to the original channel space.The subspace basis vectors are the columns of U, and the channel is represented as h = Uv.Source: C. Reduced-subspace least-square… · p. 15
  • RS-LS needs only τp ≥ r pilots for a subspace of dimension r, compared with τp ≥ M for conventional LS, and eliminates noise from unused channel dimensions.The subspace may be based on known spatial correlations or array-dependent low-rank behavior under isotropic scattering.Source: C. Reduced-subspace least-square… · p. 15
  • The optimized RS-LS MSE is (M/r)^2 times smaller than the LS MSE when τp = M.Source: C. Reduced-subspace least-square… · p. 15
  • As antenna spacing decreases in clustered scattering, RS-LS and MMSE NMSE decreases with increased spatial correlation; RS-LS with τp = M has lower NMSE than with τp = r.Using τp = r may still be preferable in practice because it requires fewer pilot resources; conventional LS is unaffected by spacing.Source: C. Reduced-subspace least-square… · p. 15

D. Compressed-sensing-based channel estimation

Compressed sensing can reduce pilot overhead when channels have sparse scattering but unknown statistics, while dense-array modeling requires EM and communication theory to be treated together. The section connects channel estimation to physically grounded antenna and field models.

D. Compressed-sensing-based channel estimation

  • Compressed sensing uses sparse-cluster information to jointly sense cluster locations and estimate their parameters, seeking good channel quality with relatively few pilots.The example builds a dictionary from array-response vectors over sampled azimuth and elevation angles; near-field use requires a polar-domain grid that also covers depth.Source: D. Compressed-sensing-based channel… · pp. 15–16
  • At τp ≥10, OMP-based estimation significantly outperforms LS and has the lowest NMSE, while its performance floor beyond 50 reflects limited dictionary size and on-grid estimation.At τp = r = 44, RS-LS has enough pilots to explore the reduced subspace; for less-sparse channels or longer pilots, RS-LS may be preferable because it is more computationally friendly.Source: Figure 12 · p. 16, D. Compressed-sensing-based channel… · p. 16

V. A LINEAR SYSTEM APPROACH TO ELECTROMAGNETIC THEORY

  • For large, densely spaced arrays, common assumptions such as i.i.d. Rayleigh fading, independent receiver interference, and negligible mutual coupling do not hold, motivating a union of EM and communication theory.Polarization must also be considered for rigorous spatial-DoF assessment.Source: §V · p. 16
  • The paper presents a linear-system approach to EM theory as suited to communication and signal-processing applications, grounded in antenna impedance, Maxwell’s equations, and outgoing plane-wave representations.The three principles characterize antenna systems by an impedance matrix, Maxwell’s equations as a linear space/time-invariant system, and external fields as superpositions of outgoing ordinary and evanescent plane waves.Source: §V · p. 17

A. Impedance matrix description of antennas

  • An antenna system in a linear time-invariant medium is represented by an impedance matrix relating antenna voltages and currents, with self-impedances on the diagonal and mutual impedances off-diagonal.The matrix is described as an exact antenna-system description; with significant mutual coupling, summing squared current magnitudes gives inaccurate transmitted power.Source: A. Impedance matrix description of… · p. 17

1) Linear space/time-invariant system:

  • Maxwell’s equations form a linear space/time-invariant system whose inputs are electric-current-density components and whose outputs are electric- and magnetic-field components.The paper obtains analytical field solutions by Fourier transforming the equations and solving algebraically.Source: 1) Linear space/time-invariant system · pp. 17–18

2) Plane-wave solution:

The section derives outgoing fields as plane waves and relates propagating and evanescent components to their roles in spatial degrees of freedom. It also presents the exact plane-wave expansion of a spherical wave and its practical advantages.

2) Plane-wave solution:

  • Outside a source distribution confined to |z| ≤ z0, the field is an outgoing plane wave on either side of the source.The fields are obtained by inverse wavenumber transformation, evaluating pole residues with contours closed in opposite half-planes for z > z0 and z < z0.Source: 2) Plane-wave solution · p. 18
  • Ordinary plane waves propagate, whereas evanescent waves decay exponentially with distance and carry only reactive power in the z-direction; only ordinary waves provide spatial DoF for radiative near- and far-field data.The outgoing external field depends on the current distribution through its plane-wave components.Source: 2) Plane-wave solution · p. 18

3) The plane-wave representation of the spherical wave:

  • A spherical wave can be represented exactly as an integral over propagating and evanescent plane waves.The paper establishes this plane-wave expansion after deriving the spherical-wave solution and applying the inverse wavenumber transform.Source: 3) The plane-wave representation of… · pp. 18–19
  • The plane-wave representation is favored for practical computation because fields can be computed with 2D DFTs, the waves are origin-free, and Cartesian-grid arrays have a simple description.It also supports modeling propagation through horizontally stratified media by expanding, propagating, and integrating plane-wave components.Source: 3) The plane-wave representation of… · p. 19

C. Mutual- and self-impedance

The EM treatment links antenna currents and fields to self- and mutual impedance, then applies plane-wave and polarization models to spatial degrees of freedom. For planar arrays, half-wavelength spacing suffices to control propagating-wave DoF, but one such array cannot distinguish all four waves sharing a horizontal wavenumber pair.

C. Mutual- and self-impedance

  • The mutual impedance between antennas is obtained from the power associated with one antenna’s current interacting with the field generated by the other; self-impedance has an infinite imaginary part but finite real part.The real part of self-impedance comes from half-residues associated with propagating plane waves.Source: C. Mutual- and self-impedance · pp. 19–20

D. Applications of EM theory to communications

  • The paper revisits spatial DoF by considering how arrays control polarimetric plane-wave spectra, with four linearly independent waves for each horizontal-wavenumber pair.These comprise h- and v-polarized waves in each of the plus- and minus-z directions.Source: 1) Polarization and degrees-of-freedom · p. 20, Figure 15 · p. 21

1) Polarization and degrees-of-freedom:

  • A planar array with spacing no greater than half a wavelength avoids aliasing propagating plane waves, and halving spacing while quadrupling antenna count does not add propagating waves.A λ/2-spaced UPA can control all spatial DoF, subject to its wavenumber resolution.Source: 1) Polarization and degrees-of-freedom · p. 21
  • A conventional UPA provides only one DoF per plane wave and cannot distinguish the four waves sharing (kx, ky): two propagation directions and two polarizations.The paper describes four separated planar arrays or a polarimetric planar array as possible designs for obtaining four DoF per plane wave.Source: 1) Polarization and degrees-of-freedom · p. 21, Figure 15 · p. 21

2) Are evanescent waves of any use?:

The section identifies extreme reactive-near-field operation and super-directivity as cases where evanescent waves may matter. It also notes practical constraints on realizing super-directive arrays.

2) Are evanescent waves of any use?

  • In extreme near-field operation, evanescent waves can be a significant component of the electromagnetic field.An increased array aperture also expands the reactive near-field, which may be used by some short-range systems.Source: 2) Are evanescent waves of any use? · pp. 21–22
  • In super-directive arrays, evanescent waves can carry real power along the array plane even though they decay exponentially in the z-direction.The cited example is a subwavelength-spaced planar array designed for strong mutual coupling.Source: 2) Are evanescent waves of any use? · p. 22
  • Super-directive arrays have not been realized on a large scale because their numerically large drive currents cause unacceptable I2R losses and an enormous local reactive field.For a linear array in end-fire, the stated limiting gain scales as M^2 as spacing approaches zero, compared with M at λ/2 spacing.Source: 2) Are evanescent waves of any use? · p. 22

3) Dense scattering propagation:

The section develops physically grounded models for scattering, thermal noise, wireless-link capacity, and antenna systems, connecting electromagnetic behavior to communication and circuit descriptions.

3) Dense scattering propagation:

  • A 4 × 4 scattering kernel relates transmitted and received plane-wave polarization amplitudes, yielding a physically and mathematically exact formulation.With evanescent waves neglected and independent complex-Gaussian kernel elements, the resulting stationary model is physically tenable and satisfies free-space Maxwell equations.Source: 3) Dense scattering propagation · p. 22
  • Isotropic scattering gives Green’s-function autocorrelation proportional to sinc(κ|pR|) · sinc(κ|pT|), equivalent to the Clarke model and close to i.i.d. Rayleigh fading.The stated correlation behavior also covers antenna spacings below λ/2 and planar arrays.Source: 3) Dense scattering propagation · p. 22

4) Rayleigh-Jeans-Clarke model for ambient thermal noise:

  • The Rayleigh-Jeans-Clarke spectrum models dark-sky noise: its temporal power-density spectrum is proportional to kBTω^2, and its spatial autocorrelation to sinc(κ|p|).It represents the minimum electromagnetic noise to which a receive antenna is subject.Source: 4) Rayleigh-Jeans-Clarke model for… · p. 22

5) Shannon capacity of a wireless link in a resonant chamber:

  • The Shannon-optimum transmit-power allocation avoids resonant frequencies, while capacity at fixed transmit power increases without bound as load resistance approaches infinity, despite infinite Johnson-noise density.The capacity example is a transmit–receive link inside a lossless resonant chamber.Source: 5) Shannon capacity of a wireless link… · pp. 22–23
  • Electromagnetic theory governs channel modeling, including spatial degrees of freedom, antenna–environment interactions, array design, and fundamental thermal noise.Array size, polarization, and adaptation to a scattering environment are identified as ways to expand spatial degrees of freedom.Source: 5) Shannon capacity of a wireless link… · p. 23

VI. CIRCUIT THEORY FOR PHYSICAL CHANNEL MODELING

  • A noisy multiport circuit model links array impedances, the propagation transimpedance, transmitter sources, receiver noise, and output voltages in an end-to-end MIMO description.The model uses superposition, power conservation, and reciprocity; practical systems may neglect transmit–receive back-scattering through a unilateral approximation.Source: §VI · p. 23

A. Antenna circuit models and their key properties

  • Antenna arrays are characterized jointly by radiation and scattering patterns and terminal multiport behavior; mutual coupling can be inferred from embedded, rather than isolated, far-field patterns.The port scattering, impedance, and pattern-coupling matrices share eigenvectors, though their eigenvalues differ.Source: A. Antenna circuit models and their… · pp. 23–24
  • Large-array circuit characterization is computationally and experimentally challenging, and the section introduces analytical techniques for modeling mutual coupling.For lossy antennas, the loss factor η ≤ 1 represents antenna efficiency.Source: A. Antenna circuit models and their… · p. 24

1) Infinite arrays:

Infinite periodic arrays offer a tractable approximation for large arrays, and their impedance representation better captures finite-array edge behavior than the admittance representation.

1) Infinite arrays:

  • The impedance description better approximates a finite array because open-circuiting its edge portions makes edge currents decay rapidly, unlike in the shorted-array case.Infinite periodic arrays are tractable because their elements have identical radiation properties and they have no edge effects.Source: 1) Infinite arrays · p. 25

2) Array of identical minimum scattering antennas:

The section surveys antenna and array models that account for scattering, physical power and noise, near-field propagation, and polarization, with discretized dipoles providing a flexible electromagnetic representation.

2) Array of identical minimum scattering antennas:

  • The extended isolated-pattern result may be inaccurate for closely spaced elements unless the antennas are minimum-scattering; for non-minimum-scattering elements it is valid when spacing exceeds λ.Embedded element patterns usually differ from isolated patterns because of nearby-element scattering.Source: 2) Array of identical minimum… · p. 25

B. Signal power

  • Radiated power is constrained by regulation and interference policies, whereas available power is constrained by device and technology limitations.The two are equivalent for a lossless array with perfect generator-to-antenna impedance matching; mismatch, structural loss, and technology limits complicate their characterization.Source: B. Signal power · pp. 25–26

2) The receive LNA model:

  • Mutual coupling generally makes receive noise spatially correlated, but uncorrelated-noise assumptions are more justified when LNA noise dominates or resistive losses dominate the receive impedance.The model treats LNAs as independently noisy, frequency-flat devices with unit gain.Source: 2) The receive LNA model · p. 26

VII. ANTENNA NEAR-FIELD AND POLARIZATION MODELING

  • Realistic array models must capture spherical-wave amplitude and phase variation and antenna polarization, including copolarization and cross-polarization gains.Cross-polarization discrimination describes how well a practical antenna isolates co-polarized signals from orthogonal cross-polarized signals.Source: §VII · p. 26

A. Electromagnetic-based array model

  • Antenna discretization models each segment as a Hertzian dipole, whose orientation-dependent polarization and distance-dependent field amplitudes capture properties absent from an isotropic point source.Summing segment fields yields the antenna response; the representation applies to arbitrary array structures that can be meshed and incorporates mutual coupling into the signal-dependent segment moments.Source: A. Electromagnetic-based array model · pp. 27–28

VIII. ANTENNA ARRAY DESIGN FOR UM-MIMO

UM-MIMO array design spans dense discrete arrays and continuous apertures, balancing spatial flexibility against hardware complexity while exploring architectures and signal-processing approaches.

VIII. ANTENNA ARRAY DESIGN FOR UM-MIMO

  • Very dense element spacing does not inherently increase spatial degrees of freedom, but supports candidate architectures for spatial efficiency and new functionality.UM-MIMO arrays may use spacings much smaller than λ/2; dense structures also raise coupling, estimation, hardware-cost, energy, and production-complexity challenges.Source: §VIII · p. 28

A. Uniform array-based radiation architecture

  • Fully digital arrays offer high spatial flexibility but require many RF chains, increasing complexity, energy consumption, and cost; hybrid arrays use fewer RF chains than antennas.Hybrid processing can have little spectral-efficiency impact when RF-chain count exceeds strong multipath-cluster count, particularly in line-of-sight scenarios.Source: A. Uniform array-based radiation… · pp. 28–29, Figure 20 · p. 29

B. Continuous aperture MIMO system

  • Continuous-aperture MIMO modulates information through a continuous antenna surface, offering transmission efficiency and application flexibility while retaining the spatial degrees of freedom of a same-size discrete array.Within those dimensions, its channel singular values can be improved; pattern division multiplexing converts continuous-pattern design into finite orthogonal-basis projection-length design.Source: B. Continuous aperture MIMO system · pp. 29–30, Figure 21 · p. 29

C. Fluid antenna system

The section surveys reconfigurable and metasurface-based arrays alongside practical hybrid-array and modular-processing designs. It highlights throughput and implementation constraints, deployment-dependent trade-offs, and unresolved challenges for UM-MIMO.

C. Fluid antenna system

  • Fluid antennas reconfigure element position, shape, polarization, or radiation pattern; multiple independently moving elements can be jointly optimized for multiplexing, but response speed and stability remain bottlenecks.The fluid can select among many spatially correlated ports to avoid deep fades, with diversity gains comparable to maximum-ratio processing; implementations may use liquid metals such as eGaIn.Source: C. Fluid antenna system · p. 30

D. Metasurface-inspired antenna system

  • DMAs use reconfigurable metadevices for low-power analog precoding without phase shifters and can pack arrays below half-wavelength spacing, but Lorentzian-limited weights constrain beamforming.Studies report reduced RF-chain counts while retaining most beamforming gain, and higher spectral efficiency and energy efficiency than conventional analog beamforming.Source: D. Metasurface-inspired antenna system · p. 31

IX. INTERPLAY BETWEEN ARRAY DESIGN AND SIGNAL PROCESSING

  • With UM-MIMO, RF-chain counts may not match antenna-element counts, making hybrid architectures practical; lossless hybrid operation requires at least η2D RF chains.The paper expects early 6G systems to use fully digital and hybrid architectures, while noting that fully digital designs may become benchmarks as array sizes grow.Source: §IX · p. 31
  • ≈5 Tbit/s at B = 100 MHz, fc = 3 GHz, b = 16, and A = 10 m2; at mmWave with 10 times more spectrum, RBBU ≈5000 Tbit/s, rates deemed infeasible in the foreseeable future.The authors identify high-performing but lossy hybrid architectures as a promising research direction.Source: §IX · p. 31

A. Modular UM-MIMO design

  • Modular arrays can activate only a fraction τ of their antenna area under a BBU interface-rate limit; the τ–µ choice depends on deployment and user distribution.In the four-panel example, τ = 0.5 activates two panels, each with two RF chains, for four BBU-connected chains; LPUs manage forwarding digitally without physical switches.Source: A. Modular UM-MIMO design · p. 32, Figure 23 · p. 32
  • Selecting τ, µ, and panel area requires deployment-specific system simulation; line-of-sight-dominant simulations found BBU inputs scale linearly with UE and active-panel counts, a result needing broader validation.The metric depends on propagation, user distribution, fixed system parameters, and LPU processing.Source: A. Modular UM-MIMO design · p. 32

X. FINAL REMARKS AND OPEN CHALLENGES

  • For arrays projected at 512–2048 antennas in the 6–15 GHz upper mid-band, open challenges include realistic-environment beamfocusing, channel estimation, energy use, array design, processing, and field trials.The paper also identifies fixed-complexity BBU processing, hardware distortions, and continuous spatial processing as open problems.Source: §X · pp. 32–33

Related papers

Loading 2401.02844v1…