XL-Dynamic Metasurface Antennas (XL-DMAs)

• XL-DMAs are extremely large-scale, tunable metasurface antennas that employ waveguide-fed architectures and sub-wavelength element density for advanced near-field 6G applications.
• They integrate free-space spherical-wave and guided-wave propagation, with Lorentzian-constrained element tuning and mutual coupling that enhances beam synthesis and controllability.
• They address wideband operation, dispersion, and analog dimensionality reduction challenges through innovative channel estimation, optimization frameworks, and calibration methods.

Extremely Large-Scale Dynamic Metasurface Antennas (XL-DMAs) are waveguide-fed, tunable metasurface apertures intended for extremely large-array operation with reduced power consumption and lower hardware cost than conventional fully digital or hybrid beamforming architectures. In the 6G near-field setting, they are characterized not only by electrically large apertures and sub-wavelength element density, but also by the fact that their effective channel is shaped jointly by free-space spherical-wave propagation and by guided-wave propagation inside the metasurface feeding structure (Zhang et al., 9 Aug 2025). This dual electromagnetic character distinguishes XL-DMAs from conventional XL-MIMO arrays and places hardware constraints—Lorentzian tunability, waveguide attenuation, mutual coupling, and analog dimensionality reduction—at the center of communication, localization, and imaging system design (Yang et al., 2024).

1. Definition, architecture, and distinguishing features

The basic DMA architecture described in the literature consists of multiple independent one-dimensional waveguides, often microstrips, each loaded with many sub-wavelength metamaterial elements and connected to a dedicated RF chain (Zhang et al., 9 Aug 2025). In communication-oriented formulations, if there are $N_d$ microstrips and $N_e$ radiating elements per microstrip, then the total number of radiating elements is $N=N_dN_e$ (Perović, 2024). XL-DMAs arise by scaling the number of waveguides and/or the number of elements per waveguide until near-field propagation becomes the relevant operating regime for practical distances (Zhang et al., 9 Aug 2025).

This hardware differs from phased arrays in several direct ways. XL-DMAs use sub-wavelength metamaterial elements, guided-wave-fed aperture excitation along waveguides, and resonant tunable responses rather than unconstrained phase shifters (Zhang et al., 9 Aug 2025). They also differ from reflectarrays and RISs because a DMA is an active antenna or transceiver aperture fed internally through waveguides and connected to RF chains, rather than a surface that mainly manipulates incident external waves by reflection (Zhang et al., 9 Aug 2025). The literature also places DMAs within the broader holographic MIMO family while emphasizing that a DMA is a specific realizable hardware instantiation based on waveguide-fed tunable metamaterial elements (Zhang et al., 9 Aug 2025).

A recurrent modeling feature is the Lorentzian constraint on element responses. In several DMA communication models, each tunable coefficient is restricted to

$$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$

or an equivalent sign convention, rather than being an arbitrary complex weight (Perović, 2024, Yang et al., 2022, Yang et al., 2024). This implies that amplitude and phase are coupled by resonant physics. A plausible implication is that XL-DMA beam synthesis should be regarded as constrained electromagnetic programming rather than conventional unconstrained array weighting.

The feed structure is equally fundamental. In DMA transmitter and receiver abstractions, signal propagation along a microstrip is modeled as

$h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$

capturing attenuation and phase progression before local element tuning is applied (Perović, 2024). This makes the analog transformation from RF chains to radiating elements architecture-dependent even before mutual coupling is introduced. For XL-DMAs, this suggests that aperture growth is inseparable from feed-network physics.

2. Near-field propagation and XL-DMA channel models

The near-field model emphasized for XL-DMAs has two components: the wireless channel between users and DMA elements, and the propagation channel inside the XL-DMA (Zhang et al., 9 Aug 2025). The first component departs from plane-wave models because the aperture is large enough that users or targets may lie in the radiative near field, where propagation is spherical-wave rather than planar-wave (Zhang et al., 9 Aug 2025). The second component captures microstrip or waveguide propagation and the resonant response of the metamaterial elements, and is the principal feature absent from standard XL-array near-field models (Zhang et al., 9 Aug 2025).

Near-field behavior is explicit in the channel-estimation literature for XL-DMAs. For an $M\times N$ oblong array with element $(m,n)$, the exact propagation distance used in the spherical-wave model is

$$\begin{aligned} r_{(n,m)} &= \Big( (r\sin\theta\cos\phi)^2 +(r\sin\theta\sin\phi-(n-1)d)^2 \ &\qquad\qquad +(r\cos\theta-(m-1)d)^2 \Big)^{1/2}, \end{aligned}$$

and the corresponding array manifold is

$$\mathbf{g}(\theta,\phi,r)= \frac{1}{\sqrt{MN}} \left[ e^{-j\frac{2\pi}{\lambda}(r_{(1,1)}-r)}, \dots, e^{-j\frac{2\pi}{\lambda}(r_{(M,N)}-r)} \right]^T$$

(Yang et al., 2024). This exact model retains the range-dependent quadratic phase terms omitted by planar-wave approximations.

A distinctive XL-DMA contribution is the “Oblong Approx.” model for oblong-shaped arrays. By neglecting the elevation quadratic term and the elevation–azimuth coupling term in the second-order Taylor expansion, the distance is approximated as

$$r_{(m,n)} \approx r-(n-1)d\varphi-(m-1)d\vartheta +\frac{((n-1)d)^2}{2r}(1-\varphi^2),$$

which yields the Kronecker-separable approximation

$N_e$0

(Yang et al., 2024). In this representation, the elevation direction is treated with a far-field steering vector and the azimuth direction with a near-field vector. The paper reports negligible model errors for oblong-shaped arrays and uses this separability to build a lower-complexity estimation framework (Yang et al., 2024).

The resulting uplink compressed measurement model for the $N_e$1-th microstrip is

$N_e$2

or equivalently

$N_e$3

with $N_e$4 capturing the microstrip attenuation and phase progression (Yang et al., 2024). This observation model makes explicit that XL-DMA channel acquisition is a compressed analog sensing problem, not a direct element-wise measurement problem.

3. Mutual coupling, controllability, and nonlinear aperture behavior

Mutual coupling in DMAs has traditionally been treated as harmful because it makes the mapping from configuration to field nonlinear, complicates characterization, and makes optimization harder (Prod'homme et al., 2024). In the coupled-dipole formalism specialized to DMAs, the self-consistent meta-atom moments satisfy

$N_e$5

and the radiated field in the region of interest is

$N_e$6

(Prod'homme et al., 2024). Here $N_e$7 encodes meta-atom interactions, and the inverse operator is the source of the nonlinear dependence on the programmable polarizabilities.

The principal revision to the conventional view is that stronger mutual coupling can improve controllability. The paper “Mutual Coupling in Dynamic Metasurface Antennas: Foe, but also Friend” shows that stronger coupling increases the sensitivity of the normalized radiation pattern to meta-atom tuning and can improve radiation-pattern synthesis fidelity, even while it decreases linear predictability (Prod'homme et al., 2024). In a chaotic-cavity-backed DMA with 64 continuously tunable meta-atoms around 10 GHz, sensitivity magnitude increased with coupling strength while the linearity metric decreased, and beam synthesis under strong coupling improved by about two orders of magnitude between no coupling and strong coupling in the reported color-scale comparison (Prod'homme et al., 2024).

Prototype work supports the practical importance of this coupling-aware view. A K-band end-to-end DMA wireless system with a $N_e$8 cavity-backed aperture and 96 binary-tunable meta-atoms achieved up to $N_e$9 discrimination between a desired transmitter and an undesired jammer when optimized with a mutual-coupling-aware model-agnostic procedure; an MC-unaware linear model yielded an undesired-channel null about $N=N_dN_e$0 worse in the shown cases (Yven et al., 11 Jun 2025). The same prototype reported zero observed bit errors out of 167,200 transmitted bits for all considered jamming strengths in the optimized configuration (Yven et al., 11 Jun 2025). Although that system is not itself extra-large, it establishes that coupling-aware control becomes decisive for delicate functions such as null steering. This suggests that for XL-DMAs, mutual coupling is not merely a parasitic artifact to be suppressed, but a structural mechanism that may be engineered if scalable calibration methods become available.

4. Wideband behavior, dispersion, and loss mechanisms

A second major distinction of XL-DMAs is that they are not naturally narrowband. In a millimeter-wave SIW-fed DMA with 16 complementary electric-inductive-capacitive meta-atoms, the literature shows that the aperture is inherently dispersive and can produce distinct radiation patterns across frequency, with this frequency diversity further controlled by binary holographic reconfiguration (Jabbar et al., 23 Oct 2025). The physical basis is the resonant Lorentzian response of the meta-atoms and their coupling to the guided mode. The relevant dispersion indicators are

$N=N_dN_e$1

with anomalous dispersion identified by

$N=N_dN_e$2

around resonance (Jabbar et al., 23 Oct 2025). The reported consequence is hybrid frequency-code diversity: for a fixed hologram, beam direction changes across 60, 61, and 62 GHz, and changing the hologram changes the pattern family again (Jabbar et al., 23 Oct 2025).

Wideband communication analysis with practical DMA characteristics reaches a related conclusion: waveguide attenuation, frequency-selective element response, and limited reconfigurability all degrade wideband beamforming, and DMAs can suffer more severe beam squint than phased arrays because both free-space and waveguide dispersion contribute to phase mismatch (Carlson et al., 9 Oct 2025). In that model, the effective channel is

$N=N_dN_e$3

the waveguide attenuation taper is

$N=N_dN_e$4

and each element response is generated by the magnetic polarizability

$N=N_dN_e$5

(Carlson et al., 9 Oct 2025). The paper proposes the heuristic factorization

$N=N_dN_e$6

separating frequency-selective mismatch, limited tuning bandwidth, and waveguide attenuation penalties (Carlson et al., 9 Oct 2025). A plausible implication is that per-strip wideband modeling is likely indispensable in XL-DMAs, where long guided paths magnify all three effects.

Losses are also central in near-field focusing. For lossy XL-DMAs, one paper models the microstrip propagation factor as

$N=N_dN_e$7

and shows that the maximum gain under perfect alignment is

$N=N_dN_e$8

with aperture-efficiency factor $N=N_dN_e$9 (Gavriilidis et al., 15 Mar 2025). The same work derives beam-depth limits and reports a non-monotonic effect of attenuation: moderate attenuation can shrink normalized beam depth, while sufficiently large attenuation broadens it and weakens near-field focusing capability (Gavriilidis et al., 15 Mar 2025). This revises the simpler view that losses are only a scalar gain penalty.

5. Signal processing, channel estimation, and optimization frameworks

XL-DMA signal processing is shaped by analog dimensionality reduction. In one communication abstraction, the radiated or received element-space signal is represented as

$$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$0

where $$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$1 captures guided-wave propagation along microstrips and $$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$2 contains the tunable element responses with the DMA block structure (Perović, 2024). This form reappears across transmission, localization, and low-resolution reception models, and it provides a useful first-order decomposition into feed-network propagation, local tunability, and low-dimensional RF excitation.

For near-field channel estimation, the EL-AZ-decoupled framework built on the Oblong Approx. treats azimuth–distance estimation as a distributed compressive sensing problem with shared support across microstrips: $$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$3 and then estimates elevation by a parallelizable one-dimensional search (Yang et al., 2024). The proposed off-grid distributed orthogonal least squares algorithm improves upon the on-grid decoupled version, and the paper reports linear runtime scaling with the number of metasurface elements $$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$4 (Yang et al., 2024). It also states that measurement-matrix optimization under the Lorentzian constraint improves NMSE relative to Gaussian random measurements, while the constraint itself degrades performance relative to ideal phased arrays (Yang et al., 2024).

Receiver-side hybrid analog/digital optimization has also been extended to XL-DMAs with low-resolution ADCs. In uplink multi-user symbol detection, the post-DMA observation is modeled as

$$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$5

followed by Bussgang-linearized quantization

$$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$6

and digital combining

$$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$7

(Pal et al., 8 Apr 2026). The design problem is cast as MSE minimization over the digital combiner and the Lorentzian-constrained block-diagonal DMA analog combiner, and is solved by alternating optimization with a semidefinite relaxation in the analog step (Pal et al., 8 Apr 2026). The paper reports that XL-DMA receivers can perform highly accurate multi-user symbol detection and exhibit attractive trade-offs between hardware complexity and MSE performance (Pal et al., 8 Apr 2026). This suggests that quantization-aware front-end design may be particularly important when XL-DMAs are used precisely to reduce RF-chain and ADC counts.

Beyond communication and estimation, near-field localization with DMAs has been framed as direct positioning under analog compression. For a user at $$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$8, the distance to element $$\mathcal{Q}=\left\{\frac{j+e^{j\phi}}{2}\,\middle|\, \phi\in[0,2\pi]\right\},$$9 is

$h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$0

and the DMA output obeys

$h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$1

(Yang et al., 2022). The paper derives a focusing-based tuning rule

$h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$2

then projects it to the Lorentzian set (Yang et al., 2022). In the reported 28 GHz simulation, a tuned dense DMA with $h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$3 spacing achieved performance similar to a fully digital array and at low SNR could be slightly better while using 5 RF chains instead of 240 (Yang et al., 2022). In the XL-DMA context, this positions localization-focused beam focusing as a core near-field use case rather than an ancillary one.

6. Applications, empirical evidence, and open problems

The most frequently cited applications of XL-DMAs are near-field communication, localization, and imaging (Zhang et al., 9 Aug 2025). In the communication case study of the XL-DMA perspective article, quarter-wavelength-spaced XL-DMAs at 28 GHz outperformed fully digital and hybrid arrays in achievable rate for a user located on the $h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$4-axis at distance 150 wavelengths, with the interpretation that denser element spacing permits more radiating elements in the same aperture and finer near-field focusing (Zhang et al., 9 Aug 2025). In the localization case study, quarter-wavelength-spaced XL-DMAs achieved better localization than hybrid arrays and performance close to fully digital arrays while using significantly fewer RF chains (Zhang et al., 9 Aug 2025). In the imaging case study, an XL-DMA with quarter-wavelength spacing was reported to be better than fully digital and closer to the reference image than the half-wavelength-spaced counterpart (Zhang et al., 9 Aug 2025). These are illustrative case studies rather than a universal scaling law, but they establish the application profile attached to XL-DMAs in current 6G discussions.

Related DMA work in energy beamforming shows why this application profile is plausible. In multiuser RF wireless power transfer, the transmit signal is modeled as

$h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$5

with received RF power at user $h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$6

$h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$7

(Azarbahram et al., 2023). The paper reports that required transmit power decreases as antenna length $h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$8 increases for both DMA and fully digital architectures, while for a fixed $h_{l,i}=e^{-\rho_{l,i}(\alpha_i+j\beta_i)},$9 cm the required transmit power for DMA generally increases with frequency whereas it remains almost constant with frequency for fully digital in the reported setup (Azarbahram et al., 2023). This suggests that XL-DMA scaling improves focusing capability but can also expose feed-network penalties that are absent from idealized array models.

The literature also identifies several persistent open problems. Accurate XL-DMA-specific channel modeling must jointly account for spherical-wave wireless propagation, intra-microstrip propagation, metamaterial frequency responses, and spatial non-stationarity (Zhang et al., 9 Aug 2025). Mutual coupling becomes more significant with high-density sub-wavelength integration, yet scalable calibration methods remain underdeveloped (Prod'homme et al., 2024, Yven et al., 11 Jun 2025). Wideband operation is particularly difficult because metamaterial responses are frequency selective and Lorentzian, while long guided paths induce additional dispersion (Zhang et al., 9 Aug 2025, Carlson et al., 9 Oct 2025). Current configuration methods based on iterative optimization or mapping optimization are described as complex and suboptimal, motivating AI-based XL-DMA configuration, distributed XL-DMAs, polarization diversity, and over-the-air hardware experiments as future directions (Zhang et al., 9 Aug 2025).

A final recurrent misconception is that XL-DMAs can be treated as ordinary XL arrays with a different combiner. The survey-style XL-DMA literature argues the opposite: the internal electromagnetic propagation is part of the communication channel itself (Zhang et al., 9 Aug 2025). The broader body of DMA work reinforces that conclusion by showing that mutual coupling, waveguide losses, Lorentzian tunability, and quantized or compressed front ends materially alter controllability, estimation, and communication performance (Prod'homme et al., 2024, Ramírez-Espinosa et al., 2022, Yang et al., 2024, Pal et al., 8 Apr 2026). Taken together, these results indicate that XL-DMAs are best understood not as a minor variant of phased arrays, but as integrated metasurface transceivers whose large-aperture behavior emerges from the interaction of guided waves, resonant meta-atoms, and near-field free-space propagation.