Gigantic MIMO Systems Overview

• Gigantic MIMO systems are ultra-large antenna arrays that scale propagation physics and enable unprecedented spatial multiplexing and near-field beamforming.
• They improve spectral efficiency, localization, and sensing, making them a cornerstone for beyond-5G/6G networks at mid-band and mmWave frequencies.
• Their deployment demands innovative hardware, channel models, and beamforming algorithms to address near-field effects and scalability challenges.

Gigantic MIMO systems—sometimes referred to as ultra-massive, extra-large, or extremely large-scale MIMO—denote multiple-input multiple-output arrays whose number of elements increases by one or more orders of magnitude relative to classical "massive MIMO" systems, now comprising hundreds to thousands (and in some research directions, even more) of antennas per base station. These systems are emerging as a foundational technology for beyond-5G/6G wireless networks in the upper mid-band (7–24 GHz) and mmWave frequencies, promising order-of-magnitude improvements in spectral efficiency, spatial multiplexing, localization, and sensing. However, the sharp increase in scale introduces a new set of physical, architectural, and algorithmic phenomena whose management is critical for ultra-large array deployments.

1. Core Definitions, Scaling Regimes, and Propagation Principles

The regime of gigantic MIMO is defined by a ratio $M \gg K \gg 1$, where $M$ is the array size (hundreds to thousands of antennas), and $K$ the number of simultaneous users or streams (tens to low hundreds) (Björnson et al., 2024, Wang et al., 2022, Larsson et al., 2013, Rusek et al., 2012). Contrasted with "massive MIMO" (tens–hundreds, e.g., $M=64$–$128$), gigantic MIMO systems exploit the scaling of propagation physics, form factor, and spatial degrees of freedom (DoF):

• Spatial DoF and aperture scaling: To maintain constant free-space pathloss as frequency increases (shorter wavelengths), the number of elements per fixed physical aperture must scale as $1/\lambda^2$, yielding arrays with $N \sim 10^3$ at 15 GHz in the same area as $N \sim 100$ at 3.5 GHz (Björnson et al., 2024).
• Near-field vs. far-field: As aperture grows, the near-field (Fresnel) region,

$d_F = 2 D_{\text{array}}^2 / \lambda,$

expands, so typical users lie within it. This mandates spherical-wave channel models, with capacity scaling governed by joint angle–range focusing and "effective DoF" determined by the physical area, wavelength, and distance (Wang et al., 2022, Björnson et al., 2024).

• Channel hardening & orthogonality: In the gigantic MIMO scaling limit, user channels become near-orthogonal, with instantaneous gains that increasingly "harden" (variance vanishes) as $M \to \infty$ (Larsson et al., 2013, Rusek et al., 2012).

Gigantic MIMO encompasses both centralized arrays (e.g., full-dimension/planar or continuous holographic surfaces (Nadeem et al., 2016, Björnson et al., 2019)), distributed/extra-large-aperture layouts (arrays on buildings/facades, modular or cell-free (Wang et al., 2022, Amiri et al., 2018)), and doubly-massive (large arrays on both ends, e.g., in mmWave links) (Buzzi et al., 2016, Björnson et al., 2019).

2. Channel Modeling: Spherical Waves, Non-Stationarity, and Spatial DoF

When array dimensions become electrically large, conventional planar-wave (far-field) models become invalid. XL-MIMO and gigantic MIMO require:

• Spherical-wave channel models: Channel entries between array elements $m, n$ at positions $\mathbf{r}_m$, $\mathbf{t}_n$ are described by

$$H_{m, n} = \frac{\lambda}{4\pi r_{m, n}} \exp\left( -j \frac{2\pi}{\lambda} r_{m, n} \right), \qquad r_{m, n} = \|\mathbf{r}_m - \mathbf{t}_n\|$$

(Wang et al., 2022, Björnson et al., 2024).

• Spatial non-stationarity and visibility regions: Each user "sees" only a local subset of the aperture, with per-antenna path-loss terms and highly non-Toeplitz covariance matrices (Amiri et al., 2018, Wang et al., 2022, Björnson et al., 2019).
• Effective Degrees of Freedom (EDoF): The number of orthogonal spatial modes, quantifying the system's parallelism, can be estimated as

$$\mathrm{EDoF} = \frac{(\mathrm{trace}~\mathbf{R})^2}{\mathrm{trace}(\mathbf{R}^2)},$$

where $\mathbf{R}$ is the spatial covariance matrix (Wang et al., 2022). For line-of-sight (LoS) scenarios in the near field, DoF scales with physical area, wavelength, and inter-terminal distance (Wang et al., 2022).

Multipath richness and cluster/angular spread can further shape channel rank and attainable multiplexing (Buzzi et al., 2016), and mutual coupling effects become non-negligible for $\Delta \ll \lambda/2$ inter-element spacings, leading to new bandwidth/spatial trade-offs (Akrout et al., 2022).

3. Architectural Paradigms and Hardware Implications

Gigantic MIMO disrupts the viability of traditional one-RF-chain-per-element digital architectures. Multiple hardware approaches have been proposed and validated:

• Hybrid analog/digital beamforming: Few RF chains with analog beamforming networks (phase-shifter-based or hybrid Butler-matrix) reduce digital overhead but may limit beamforming flexibility (Buzzi et al., 2016, Wu et al., 15 Jan 2026, Nerini et al., 6 Jun 2025).
• Microwave Linear Analog Computers (MiLACs): Reconfigurable multi-port microwave networks realize analog-domain beamforming, with dramatic reductions in RF chain requirements, computational complexity, and power consumption (Nerini et al., 6 Jun 2025, Wu et al., 15 Jan 2026, Nerini et al., 18 Jun 2025).
  • MiLAC architectures (fully connected and stem-connected) achieve digital-like performance and capacity, offering circuit complexity scaling of $O(KN_T)$ (number of data streams, antennas), as against $O(N_T^2)$ for fully connected designs (Nerini et al., 18 Jun 2025, Wu et al., 15 Jan 2026).
  • Lossless, reciprocal MiLACs can be engineered so that their analog domain operations realize any beamforming matrix under a spectral norm constraint, or, in hybrid mode, exactly emulate any digital beamforming with only as many RF chains as data streams (Nerini et al., 6 Jun 2025, Wu et al., 15 Jan 2026).
• Switch-and-phase-shifter receiver architectures: Use arrays of switches and constant phase shifters rather than full RF chains, delivering robust performance with highly reduced hardware (Alkhateeb et al., 2016).
• Sparse/nonuniform and movable arrays: Array intelligence replaces sheer element count; irregular or site-optimized placement and mechanical reconfigurability can yield similar or improved sum rates with several times fewer RF chains (Björnson et al., 13 Jan 2026).
• Radio stripes and distributed subarray/XL-MIMO: Arrays partitioned into multiple subarrays, possibly distributed over a building or urban corridor, with local processing and decentralized data fusion (Amiri et al., 2018, Björnson et al., 2019).

Hardware Complexity and Robustness

The array size introduces linear or quadratic scaling in hardware cost and power under legacy architectures, but many hardware imperfections (e.g., increased distortion, lower ADC/DAC resolution, phase noise) can be tolerated with only sublinear loss in gigantic arrays. For additive distortion and noise, system performance remains robust if their variances increase at most as $O(N^{0.5})$ with antenna count $N$ (Björnson et al., 2014).

4. Signal Processing and Algorithmic Methods

Precoding and Beamforming

• Linear methods: Maximum Ratio Transmission (MRT), Zero-Forcing (ZF), Minimum Mean Square Error (MMSE) extend to gigantic arrays, with complexity exacerbated by matrix sizes but mitigated by the availability of low-dimensional subspace results and iterative/approximate inversion techniques (Wang et al., 2022, Wang et al., 2016, Nerini et al., 6 Jun 2025, Wu et al., 15 Jan 2026).
• Analog beamforming constraints: MiLACs impose spectral norm and orthogonality structure on feasible beamforming matrices; hybrid MiLAC-digital architectures regain full digital flexibility with minimal RF chains (Wu et al., 15 Jan 2026, Nerini et al., 6 Jun 2025).
• Near-field beamfocusing and range–angle focusing: Spherical-wave models enable beam patterns with both angle and range selectivity, improving performance for joint localization/sensing and dense user environments (Björnson et al., 2024, Wang et al., 2022).
• User scheduling and subarray selection: Modular array architectures with near-field user modeling necessitate geometry-informed scheduling algorithms; analytic IUI models and complexity-aware heuristics (such as rectangular/front-line schedulers) effectively optimize sum spectral efficiency (González-Coma et al., 10 Jan 2025).

Channel Estimation and Calibration

• Pilot contamination remains the key limiting factor: Interference from re-used pilot signals in multi-cell systems yields a hard limit on achievable SINR, not mitigated by further scaling $M$ (Larsson et al., 2013, Rusek et al., 2012, Wang et al., 2016).
• Reduced-complexity and subspace-aware CSI methods: Structure and sparsity in the gigantic MIMO channel covariance and "visibility regions" enable subarray-based processing, graph-based detection (inspired by coded random access), and low-rank or compressive estimation (Amiri et al., 2018, Wang et al., 2022).
• Reciprocity calibration and over-the-air alignment: TDD-based systems must calibrate hardware asymmetries across massive arrays; distributed calibration protocols and built-in hardware aids are employed (Yang et al., 2016, Björnson et al., 2019).

5. Capacity Limits, Performance Scaling, and Experimental Results

• Sum-rate and multiplicity scaling: In idealized i.i.d. Rayleigh fading, total spectral efficiency follows $K \log_2(1 + M \rho/K)$, suggesting that sum-rate grows logarithmically in $M$ with $K$ fixed, or linearly in $K$ with $M$ sufficiently large (Larsson et al., 2013, Rusek et al., 2012, Wang et al., 2016).
• MiLAC-aided systems: Achieve the same capacity as digital architectures at drastically reduced RF chain and computational budgets; closed-form designs of capacity-optimal analog networks realize the exact digital SVD/water-filling rate (Nerini et al., 6 Jun 2025, Wu et al., 15 Jan 2026, Nerini et al., 18 Jun 2025).
• Algorithmic complexity: Practical low-complexity algorithms (e.g., block-coordinate WMMSE) based on MiLAC architectures run in $O(NK^2 + K^3)$ per iteration, with convex relaxations and subspace dimension reductions offering further savings (Wu et al., 15 Jan 2026).
• Empirical demonstration: Field-tested 128-antenna TDD gigantic MIMO systems show 80 bps/Hz spectral efficiency and multi-hundred Mbps cell rates using QPSK/256-QAM, readily handled by distributed processing and reciprocity-calibrated hardware (Yang et al., 2016). Switch-based architectures, modular subarrays, and irregular placements all outperform their hardware-equivalent uniform array benchmarks (Alkhateeb et al., 2016, Björnson et al., 13 Jan 2026, González-Coma et al., 10 Jan 2025).

6. Emerging Applications and Research Directions

Gigantic MIMO unlocks new applications at the intersection of communications, localization, and sensing, including:

• Near-field localization and joint communications–sensing: Large apertures and spherical-wave models offer cm-level 3D positional and orientation accuracy, new forms of beamfocusing, and high-resolution MIMO radar (Björnson et al., 2024, Björnson et al., 2019, Amiri et al., 2018).
• Energy efficiency and green operation: Power can be scaled down as $1/M$ per antenna (with sufficiently large $M$) without loss in performance, supporting ultra-high energy efficiency (Larsson et al., 2013, Rusek et al., 2012, Wang et al., 2016).
• Antenna intelligence: Non-uniform, site-optimized, and mechanically reconfigurable ("movable") arrays achieve high DoF and sum rate with fewer elements, indicating a paradigm shift away from raw antenna abundance (Björnson et al., 13 Jan 2026).
• Quantum and AI-accelerated optimization: Future B6G systems are expected to incorporate variational quantum optimization frameworks for high-dimensional beamforming on the Stiefel manifold, and machine-learning augmented calibration, beam management, and CSI estimation (Rexhepi et al., 13 Apr 2025, Wang et al., 2022, Björnson et al., 2019).

Key Challenges

Current and open research questions include hardware calibration and mutual coupling in large arrays, high-rate and robust near-field channel models, computational scalability of precoding/detection for thousands of antennas, seamless integration of communications and sensing, and harnessing antenna intelligence for robust operation under regulatory and cost constraints (Björnson et al., 2024, Björnson et al., 13 Jan 2026, Wang et al., 2022, Björnson et al., 2019).

7. Physical Layer and Deployment Design: Trade-Offs and Guidelines

• Bandwidth gain via tightly coupled arrays: Optimum element spacing and tight mutual coupling yield a "bandwidth gain"—flat SNR and higher capacity over wider frequency bands than the conventional $\lambda/2$ spacing rule permits (Akrout et al., 2022).
• 2D planar and full-dimension arrays: Active antenna systems with 2D planar arrays (FD-MIMO) and closed-form spatial correlation functions enable compact, electronically steerable, and capacity-efficient deployments with hundreds to thousands of elements, controllable downtilt, and per-user beamforming (Nadeem et al., 2016).
• Power scaling and hardware cost: As $M \to \infty$, cost-effective design is achievable by relaxing hardware quality specs (EVM, noise figure, phase noise), following defined scaling laws for the acceptable growth in distortion and noise (Björnson et al., 2014). For N elements, EVM/noise can grow as $N^{0.5}$ while maintaining SINR saturation.

In sum, gigantic MIMO represents a paradigm shift in spatial wireless design, combining theoretical advances in near-field physics and capacity characterization, new hardware and architectural models for analog and distributed systems, and algorithmic innovations for scalable beamforming, CSI acquisition, and system optimization. Ongoing research is focused on closing the gap between ideal models and real-world deployments, particularly under hardware, cost, and energy constraints, and toward leveraging the unprecedented spatial resources for both classic communications and new physical-layer modalities.