Extremely Large Aperture Array Overview

• Extremely Large Aperture Arrays are systems with apertures so vast that conventional far-field models break down, requiring advanced near-field modeling.
• Novel beamforming methods enable finite-depth focusing and distance-domain multiplexing, improving signal separation and spectral efficiency.
• Architectural designs span modular, dense, and distributed arrays, balancing calibration, RF-chain constraints, and deployment challenges.

An extremely large aperture array (ELAA) is an array regime in which the physical aperture is so large that conventional modelling assumptions, such as uniform plane wave impingement, are no longer valid over much of the operational region. In wireless communications, ELAAs can significantly enhance beamforming gain and spectral efficiency, but the same aperture enlargement pushes many user equipments into the radiative near-field region, so spherical wavefronts, non-uniform amplitude, projected aperture, and spatial non-stationarity become first-order effects rather than perturbations (Si et al., 2024, Lu et al., 2021). In radio astronomy, aperture arrays provide electronically steered multi-beam operation, large instantaneous field of view, and very large collecting area without mechanical pointing, and modern systems such as LOFAR and EMBRACE demonstrate how digital beamforming and calibration scale to large apertures (Garrett, 2012, Torchinsky et al., 2016).

1. Historical emergence and terminology

Aperture arrays are as old as radio astronomy itself. Early aperture-array instruments were associated with Jansky’s broadside array, the detection of radio emission from the Sun and Jupiter, the first interferometric experiments, and discoveries including radio galaxies, quasars, and pulsars. The Cambridge “4-acre array” formed up to 16 independent beams and enabled early systematic surveys such as the 2C and 3C catalogues (Garrett, 2012). After the late 1960s, parabolic dishes largely displaced aperture arrays in astronomy for roughly four decades, principally because paraboloids enabled observations at centimetre, millimetre, and sub-millimetre wavelengths, supported sophisticated cooled low-noise receivers at the focus, and offered higher spatial resolution for source identification (Garrett, 2013).

The modern revival followed ICT advances in signal processing, digital electronics, high-speed networking, supercomputing, and mass storage. ASTRON’s developments from the mid-1990s onward—AAD, OSMA, THEA, and EMBRACE—reintroduced high-frequency aperture-array concepts into radio astronomy, while LOFAR, MWA, LWA, and PAPER established large electronically steered arrays as practical instruments (Garrett, 2012). In parallel, the wireless communications literature introduced closely related terminology—“extremely large-scale array” (XL-array) and “extremely large aperture massive MIMO” (xMaMIMO)—to designate base-station arrays whose dimensions are so large that conventional massive-MIMO stationarity assumptions fail and user-specific visibility regions appear across the aperture (Amiri et al., 2018).

This dual lineage matters because the term ELAA now spans at least two technically adjacent traditions. In radio astronomy it emphasizes collecting area, field of view, multibeaming, and calibration at scale. In wireless communications it emphasizes near-field propagation, finite-depth focusing, and array-aware transceiver design. The common denominator is not merely a large number of elements, but an aperture so large that electromagnetic propagation, beamforming, and system architecture must be reformulated.

2. Electromagnetic regime and modelling foundations

The defining modelling transition for ELAA systems is the breakdown of far-field approximations. A standard boundary is the Rayleigh distance,

$d_{\mathrm{Rayleigh}}=\frac{2D^2}{\lambda},$

with $D$ the array aperture and $\lambda$ the wavelength. Because $D$ is very large for ELAAs, the radiative near-field may extend to hundreds or thousands of meters, so many indoor and outdoor users lie in the Fresnel region rather than the Fraunhofer region (You et al., 2023, Ramezani et al., 2022). In this regime, the received phase at antenna $n$ is $\phi_n=-\frac{2\pi}{\lambda}d_n$, with $d_n$ a geometry-dependent distance that cannot in general be linearized into a direction-only model (You et al., 2023).

A unified ELAA communication model explicitly treats each element as having physical area rather than as a sizeless point. In the formulation of “Communicating with Extremely Large-Scale Array/Surface,” the occupation ratio is

$\xi=\frac{A}{d^2}\le 1,$

where $A$ is the area of each element and $d$ the inter-element spacing. The channel response is written as

$D$0

with the gain term incorporating distance variation and projected aperture across the array (Lu et al., 2021). This model unifies discrete arrays and continuous surfaces, and its closed-form SNR analysis shows that the relevant control variables are collective properties such as aperture size and occupation ratio rather than individual element properties.

For modular architectures, the non-uniform spherical wave (NUSW) model further includes phase variation, amplitude variation, and projected aperture across both modules and elements. This is necessary when modules are separated by inter-module distances much larger than the wavelength, a geometry motivated by deployability on common platforms such as building facades (Li et al., 2022). In this setting, the aperture is not only large but structurally heterogeneous, so inter-module spacing enters the performance equations directly.

Several papers argue that the classical Rayleigh distance is not sufficient by itself for communications design. “Near-Field Wideband Beamforming for Extremely Large Antenna Arrays” introduces an effective Rayleigh distance based on beamforming gain loss, while “Communicating with Extremely Large-Scale Array/Surface” introduces the uniform-power distance (UPD), defined through the ratio of the weakest to strongest received element power across the array (Cui et al., 2021, Lu et al., 2021). Together these proposals indicate that ELAA near-field classification is not a single-threshold issue: phase nonlinearity and power non-uniformity need not become critical at the same range.

3. Beam focusing, multiplexing, and spatial degrees of freedom

The principal electromagnetic capability unlocked by ELAA operation in the radiative near-field is finite-depth focusing. In conventional far-field beamforming, beams are effectively angle-resolved and of infinite depth. In the ELAA near-field, beams can be focused at specific ranges, so users located in the same angular direction but at different distances become separable (Ramezani et al., 2022). This is the foundation of distance-domain or depth-domain multiplexing.

A line-of-sight analysis gives a closed-form 3 dB depth-of-focus: $D$1 where $D$2 is the Fraunhofer array distance and $D$3 the focal distance (Ramezani et al., 2022). The same chapter states that the maximum number of users that can be spatially multiplexed is fundamentally determined by aperture area rather than by the number of antennas per se,

$D$4

This directly contradicts the common simplification that antenna count alone determines spatial multiplexing capability.

A more recent continuous-aperture analysis addresses the maximum spatial degree of freedom in the distance domain for collinear users. For a broadside, single-piece transmit aperture and a linear receive array spanning $D$5, the spatial DoF is

$D$6

where $D$7 and $D$8 are the minimal and maximal distances from the reference origin to the aperture boundary (Duong et al., 1 Jul 2025). A key conclusion is that the distance-domain DoF is governed predominantly by the extreme boundaries of the aperture rather than its detailed interior structure. The same framework shows that modular arrays can yield a DoF gain over a single-piece array under a physical-length constraint when sub-array main lobes do not overlap.

Wideband operation introduces an additional ELAA-specific phenomenon. “Near-Field Wideband Beamforming for Extremely Large Antenna Arrays” identifies near-field beam split, in which beams at different frequencies focus on distinct physical locations rather than merely distinct angular directions (Cui et al., 2021). To mitigate this, the paper partitions the array into subarrays and proposes phase-delay focusing (PDF) using a delay phase precoding (DPP) architecture: phase shifters compensate intra-subarray far-field phase discrepancy, while time delayers compensate inter-subarray near-field phase discrepancy. This result is important because it shows that ELAA near-field beamforming is not simply narrowband focusing repeated across subcarriers.

4. Beam training, estimation, and sensing

Beam management in ELAA systems is intrinsically two-dimensional in angle and range. A survey of near-field beam management identifies three main functions—beam training, beam tracking, and beam scheduling—and shows why far-field DFT codebooks are inadequate: in the near-field, a beam codeword must target a location, not only a direction (You et al., 2023). Proposed remedies include 2D polar-domain codebooks, two-phase training that first estimates angle and then range, hierarchical codebooks with logarithmic search scaling, partial-array training, and wideband “rainbow” beam training based on true-time-delay devices.

One concrete beam-training proposal is hash beam training. Instead of sweeping one DFT beam at a time, each training beam is formed as a composite superposition of several DFT beams selected by a hash function. The user receives reference signal received power measurements, accumulates them for each constituent DFT beam according to the known codebook, and chooses the beam with the largest accumulated value (Si et al., 2024). Two codebook designs are proposed. The “proposed random hash codebook” enforces fairness so that each DFT beam appears the same number of times, $D$9, and the “fixed hash codebook” is selected offline from many random fair candidates by simulated beam-alignment accuracy. In simulations with $\lambda$0 DFT beams and $\lambda$1 training beams, the method achieves beam-alignment accuracy close to exhaustive sweeping with half the training overhead, and the fixed codebook performs up to approximately $\lambda$2 better than the random codebook (Si et al., 2024).

Channel estimation also changes qualitatively in the ELAA near-field. “Parametric Near-Field Channel Estimation for Extremely Large Aperture Arrays” proposes a multi-user MUSIC-based estimator that first searches azimuth and elevation with a far-field approximation, then searches distance with the full near-field model, and finally applies a least-squares correction for residual per-user amplitude and phase mismatch (Kosasih et al., 2024). The reported outcome is improved normalized beamforming gain and normalized mean-square error relative to classical least-squares and MMSE baselines. A distinct geometry-driven approach, the reduced-subspace least-squares (RS-LS) estimator, exploits the fact that all user correlation matrices lie in the subspace spanned by the eigenvectors of the isotropic correlation matrix, so pilot design can be optimized without user-specific spatial correlation knowledge; the paper proves that an optimal pilot length exists and gives a low-SNR approximation for it (Alıcıoğlu et al., 2024).

Near-field dictionaries and sensing algorithms must also account for range dependence and spatial non-stationarity. For extremely large uniform planar arrays, a new polar-domain dictionary based on joint angle-distance sampling eliminates fully coherent columns that arise under separable angle and distance sampling, and its non-uniform distance sampling reduces maximum column coherence and improves user-equipment localization compared with uniform distance sampling (Demir et al., 2023). Under partial blockage, ELAA sensing becomes a joint estimation problem over channel parameters and visibility regions. “Joint Near-Field Sensing and Visibility Region Detection with Extremely Large Aperture Arrays” models blockage clustering with an Ising prior on the visibility indicator vector and alternates among channel-gain, steering-vector, amplitude, and binary visibility updates to recover both path geometry and blocked subapertures (Huang et al., 28 Feb 2025).

Limited-RF-chain architectures have motivated yet another class of ELAA estimators. “Near-Field Multiuser Localization Based on Extremely Large Antenna Array with Limited RF Chains” partitions the array into subarrays, imposes geometric constraints between user location and subarray responses, and uses message passing in the APLE-LM algorithm to approach the Bayesian Cramér-Rao Bound at high SNR under analog beamforming (Teng et al., 1 Jun 2025). This shows that ELAA sensing gains remain accessible even when fully digital RF scaling is impractical.

5. Architectures and representative systems

ELAA engineering spans dense, modular, sparse, and distributed realizations. In wireless communications, xMaMIMO receiver design often begins by partitioning the array into subarrays because only a user-specific visibility region contributes significant energy. A distributed linear data fusion receiver lets each subarray compute a local ZF estimate for all users and then fuses the outputs centrally, whereas a graph-based coded-random-access-inspired receiver exploits the sparse user–subarray visibility graph directly; in the best case its complexity scales as $\lambda$3, with $\lambda$4 the number of subarrays (Amiri et al., 2018). This architectural move is a direct consequence of ELAA spatial non-stationarity.

In astronomy, representative large-aperture systems demonstrate parallel lessons in hierarchical beamforming, multibeaming, and calibration:

System	Key characteristics	Relevance
LOFAR	Baselines up to thousands of km; operates at about 10–240 MHz; sub-arcsecond resolution at $\lambda$5 MHz	Large electronically steered aperture-array pathfinder for SKA (Garrett, 2012)
EMBRACE@Nançay	4,608 densely packed Vivaldi elements in an $\lambda$6 aperture; hierarchical analogue and digital beamforming	Dense aperture-array demonstrator for SKA-class technology (Torchinsky et al., 2016)
FAST Core Array	FAST plus 24 secondary 40-m antennas within 5 km; $\lambda$7 resolution at 1.4 GHz; phased array feed on FAST	Proposed high-sensitivity, high-resolution array extension (Jiang et al., 2024)

EMBRACE@Nançay is particularly informative as a dense-array demonstrator. It uses 4,608 densely packed Vivaldi elements, analogue beamforming in four hierarchical stages, and a LOFAR-type digital backend. The measured full width at half maximum in a 1171.4 MHz Cas A drift scan was $\lambda$8, close to the theoretical $\lambda$9, and sidelobes were more than 15 dB below the main lobe (Torchinsky et al., 2016). The system also demonstrated multibeaming, long calibration validity, and stable facility-style operation, while identifying power consumption and local-oscillator distribution as major scaling challenges.

The FAST Core Array represents a different large-aperture strategy: a highly sensitive central dish combined with nearby interferometric elements. The proposed design integrates FAST with 24 secondary 40-m antennas implanted within 5 km, provides about $D$0 angular resolution at 1.4 GHz, and is designed for broad frequency coverage and large field of view, assisted by a phased array feed mounted on FAST (Jiang et al., 2024). Although this architecture is not a dense ELAA in the wireless sense, it exemplifies the broader large-aperture principle of pairing enormous sensitivity with electronically intensive backends and multi-beam survey capability.

Modular ELAA communication arrays occupy the middle ground between dense contiguous apertures and widely distributed arrays. In the modular XL-array model, a large number of elements are arranged on a common platform in modules, with module spacing typically much larger than the inter-element spacing for ease of deployment. The resulting near-field SNR saturates as the number of modules grows, and the asymptotic ceiling decreases with larger inter-module spacing while increasing with module effective aperture (Li et al., 2022). Sparse ELAA radar configurations extend the same idea to sensing, where coherent or noncoherent fusion across widely separated subarrays supports single-snapshot far-field and near-field localization (Hu et al., 22 Sep 2025).

6. Limitations, trade-offs, and research directions

A recurring misconception is that ELAA performance is obtained by simply increasing antenna count. The modelling literature instead shows that aperture, occupation ratio, projected aperture, and geometry govern performance. Under the unified array/surface model, SNR grows with diminishing return and, in the infinite-array limit, saturates at

$D$1

rather than increasing linearly without bound (Lu et al., 2021). In the modular NUSW model, the near-field SNR likewise saturates, and wider inter-module spacing reduces the SNR ceiling even when the physical footprint grows (Li et al., 2022).

A second misconception is that far-field beam codebooks remain adequate after simple refinement. Near-field dictionary design shows otherwise: when angle and distance are sampled independently in a UPA dictionary, some column pairs can become highly or fully correlated, which degrades localization and sparse recovery; coupling distance sampling to angular coordinates lowers the maximum coherence but reduces dictionary size as the coherence threshold is tightened (Demir et al., 2023). Likewise, wideband ELAA beamforming cannot be treated as a narrowband problem replicated across subcarriers because near-field beam split causes different frequencies to focus on different physical locations (Cui et al., 2021).

Operational trade-offs appear throughout the ELAA stack. Hash beam training reduces overhead, but codebook design remains SNR-dependent through the optimal number of DFT beams per training beam $D$2, and fixed codebooks require offline scenario-specific optimization (Si et al., 2024). Decentralized xMaMIMO receivers reduce centralized complexity, but they rely on accurate CSI and careful handling of edge users and overlapping visibility regions (Amiri et al., 2018). Limited-RF-chain localization is feasible, but more RF chains and more generous partitioning improve performance (Teng et al., 1 Jun 2025). Dense astronomical arrays provide large field of view and multibeaming, but practical scaling must confront power consumption and local-oscillator distribution (Torchinsky et al., 2016).

The research direction suggested by these results is that ELAA design is fundamentally cross-layer. Electromagnetic modelling, codebook geometry, subarray partitioning, analog or hybrid hardware, and inference algorithms for localization and sensing all interact. This suggests that future ELAA systems will be judged less by raw element count than by how effectively they exploit aperture-induced near-field structure—finite-depth focusing, distance-domain separability, visibility-region sparsity, and electronically controlled multibeam operation—within the constraints of calibration, RF-chain count, synchronization, and deployment geometry.