Electromagnetic Skins: Concepts & Applications

• Electromagnetic Skins (EMSs) are thin, engineered metasurfaces that locally modulate phase and amplitude to control wave propagation.
• EMSs leverage static, reconfigurable, and time-modulated designs and are optimized using analytical models and digital tools.
• Recent studies show EMS deployment enhances wireless coverage, sensing accuracy, and localization by tailoring electromagnetic responses.

Searching arXiv for papers on electromagnetic skins and closely related static/passive metasurface deployments, sensing, and channel-charting applications. I’m going to look up arXiv records for electromagnetic skins, smart electromagnetic environments, and static/non-reconfigurable metasurfaces relevant to communications, sensing, and localization. Electromagnetic skins (EMSs) are thin, engineered surfaces—typically metasurfaces—that impose prescribed local reflection or transmission properties in order to tailor electromagnetic propagation within smart electromagnetic environments. Across the recent literature, EMSs are modeled as apertures formed by sub-wavelength unit cells or meta-atoms, and they are used to redirect, focus, absorb, or diffuse waves in support of wireless coverage, opportunistic relaying, sensing, imaging, and localization. The term spans several subclasses, including static passive EMSs, reconfigurable passive EMSs, non-reconfigurable EMSs, one-time programmable EMSs, time-modulated EMSs, optically-transparent EMSs, and curved EMSs mounted on non-planar supports (Oliveri et al., 19 Jul 2025, Benoni et al., 2023).

1. Concept and taxonomy

EMSs are commonly defined as passive or reconfigurable metasurfaces that manipulate incident electromagnetic waves through spatially engineered local responses. In the static-passive case, the phase profile is fixed after fabrication or installation; in the reconfigurable case, tunable components such as varactors, PIN diodes, or MEMS enable run-time adaptation. A recurrent distinction in the literature is therefore between low-cost, passive, pre-configured panels and dynamically programmable reconfigurable intelligent surfaces (RISs), the latter requiring control networks, power, and signaling overhead (Maleki et al., 10 Aug 2025, Maleki et al., 2 Nov 2025).

Static passive EMSs differ not only from RISs but also from conventional metallic reflectors. A flat passive conductive screen enforces Snell’s law, whereas an EMS provides spatially varying reflection phases and can implement anomalous reflection, beam shaping, focal-spot control, or near-field focusing. This distinction is explicit in indoor smart electromagnetic environment experiments, where static-passive EMSs are described as monolithic, unbiased reflectors that require no biasing or calibration, while still steering energy into shadowed regions (Benoni et al., 2023). A related non-reconfigurable class appears in NLOS imaging, where the surface reflection coefficient is written as $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$, with the phase law fixed at manufacturing time (Bellini et al., 2024).

The taxonomy has broadened. One-time programmable EMSs (OTP-EMSs) reconcile passive-static operation with scenario-dependent configurability by embedding expendable components at the atomic level; after irreversible programming, the panel behaves as a static passive EMS with zero power consumption and zero maintenance (Oliveri et al., 19 Jul 2025). Time-modulated EMSs (TM-EMSs) introduce dynamic, periodic switching to generate harmonic beams for integrated sensing and communications (Poli et al., 11 May 2025). Optically-transparent opportunistic EMSs exploit existing insulating-glass panes as substrates for outdoor-to-indoor millimeter-wave transmission (Oliveri et al., 2023), while static passive curved EMSs conform to car doors and act as anomalous mirrors for vehicular relaying (Ayoubi et al., 2024).

2. Electromagnetic principles and analytical models

A common planar EMS model uses a diagonal reflection matrix,

$$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$

where $\phi_\ell$ is the phase imparted by element $\ell$. In phase-only formulations, amplitude variation and inter-element coupling are assumed negligible, so the EMS primarily controls phase. For a planar surface, generalized Snell’s law links the incident and outgoing wave vectors through the tangential phase gradient, reducing to a linear phase ramp $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$, sampled on the element positions $\mathbf{p}_\ell$ as $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$ (Maleki et al., 10 Aug 2025). In broader metasurface descriptions, the local reflection coefficient may be expressed as $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$, which makes explicit the joint amplitude-phase degrees of freedom when they are retained (Bellini et al., 2024).

Several analytical frameworks coexist. Surface-impedance and reflection-coefficient formulations write

$\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$

or their TE/TM oblique-incidence variants, and they are often used to interpret link budgets and anomalous-reflection synthesis (Benoni et al., 2023, Rocca et al., 2021). A more general description uses generalized sheet transition conditions, equivalent electric and magnetic surface currents, and local susceptibility tensors to connect unit-cell geometry to far-field or Fresnel-region radiation. The generalized analysis developed for anomalous reflection and focusing at 17.5 GHz derives a Fresnel-region reflected field in closed form and uses it to synthesize both far-field steering and near-field focusing within a unified formulation (Oliveri et al., 2022). In inverse-source approaches, the target field is first translated into an ideal current distribution, and the induced current is then decomposed into pre-image and null-space components to improve realizability with inexpensive passive layouts (Oliveri et al., 2024).

The physical consequence of phase control is especially clear in NLOS specular links. For flat passive conductive screens, the total path attenuation saturates to the image-theory free-space limit,

$$\mathcal{A}_{\infty}^{PCS}(\mathbf{r}_{RX})=\left[\frac{\lambda}{4\pi(r_{RX}+r_{TX})}\right]^2 G_{RX}G_{TX},$$

whereas the ideal EMS upper bound scales as

$$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$0

The threshold side length

$$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$1

therefore marks the minimum aperture needed for an EMS to outperform the asymptotic limit of a flat metallic reflector with the same geometry (Oliveri et al., 2022).

3. Design and synthesis methodologies

A large part of EMS research is organized as a multi-scale design problem. At macro-scale, the objective is to specify the desired field transformation: an anomalous-reflection beam, a near-field focal spot, a contour-constrained footprint, or a coverage gain in a region of interest. At micro-scale, the problem is to realize the corresponding current or phase profile with physically admissible unit cells. This logic underlies the System-by-Design paradigm, which appears across holographic smart skins, aperiodic passive smart skins, inexpensive inverse-source designs, and 1-bit reconfigurable passive EMSs (Oliveri et al., 2021, Oliveri et al., 2021, Oliveri et al., 2024, Oliveri et al., 2022).

In holographic and inverse-source formulations, the target field is encoded through reference currents satisfying a mask or footprint requirement. The Iterative Projection Technique is used in passive holographic skins, while the Quantized Iterative Projection Method is introduced for Single-Bit Reconfigurable Passive EMSs so that the reference current is already compatible with binary-state hardware. The subsequent micro-scale stage then minimizes the mismatch between the reference current and the current induced by a concrete layout of unit cells or binary states, often using particle swarm optimization, genetic search, or alternate minimization (Oliveri et al., 2021, Oliveri et al., 2022).

A second methodological axis concerns model acceleration. Digital twins based on Ordinary Kriging or Gaussian-process regression are used to predict local susceptibility tensors or coverage-related fitness functions from a reduced set of full-wave or ray-tracing simulations. In the AI-enhanced aperiodic design of passive smart skins, a $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$2 aperiodic small-scale model and $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$3 random layouts are used to train a local unit-cell digital twin that captures non-uniform coupling and edge effects (Oliveri et al., 2021). In low-cost inkjet-printed EMS synthesis, a full-wave database of passive unit cells replaces repeated geometry solves within the optimization loop (Oliveri et al., 2024).

A third design family is codebook-based. In channel-charting enhancement, each EMS is assigned a discrete set of DFT-like phase gradients, and the optimal joint configuration is selected by exhaustive or greedy search over $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$4 combinations, using upper quantiles of localization error, trustworthiness, and continuity as the objective (Maleki et al., 10 Aug 2025, Maleki et al., 2 Nov 2025). In curved vehicular relays, the continuous surface is partitioned into modules, each assigned a phase profile from a codebook of azimuth angle pairs, and a genetic algorithm searches the Cartesian product of module assignments (Ayoubi et al., 2024). One-time programmable EMSs apply the same logic at fabrication time: a closed-form rule selects whether each expendable fuse remains intact or is irreversibly burnt so that the realized local phase best approximates the desired anomalous-reflection field (Oliveri et al., 19 Jul 2025).

4. Communications, coverage enhancement, and smart-environment planning

The most established application class is coverage enhancement in built environments. A large-scale indoor demonstration of a Smart ElectroMagnetic Environment based on static-passive EMSs at 5.64 GHz reported average received-power gains of $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$5 dB and $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$6 dB, with peak gains of $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$7 dB and $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$8 dB, respectively. In hallway “A”, the measured region of interest shrank from $$\Phi=\operatorname{diag}(e^{j\phi_1},\ldots,e^{j\phi_L}),$$9 to $\phi_\ell$0, corresponding to $\phi_\ell$1; the measured below-threshold probability improved from $\phi_\ell$2 to $\phi_\ell$3, average download throughput increased by $\phi_\ell$4, average upload by $\phi_\ell$5, and download latency decreased by $\phi_\ell$6 (Benoni et al., 2023).

In urban millimeter-wave coverage planning, EMSs are treated as tile-based reflective metasurfaces installed on admissible facade regions and optimized to maximize coverage while minimizing the number of deployed tiles. In a $\phi_\ell$7 GHz setting, a $\phi_\ell$8 facade partitioned into $\phi_\ell$9 tiles with side $\ell$0 m yielded a best-coverage solution with $\ell$1 tiles and total installed area $\ell$2, sufficient to raise the entire $\ell$3 area of interest above the threshold $\ell$4 dB (Rocca et al., 2021). A related city-scale planning framework formulates EMS deployment as a binary optimization over candidate facades, uses a digital twin of the propagation scenario to evaluate thresholded coverage penalties, and reported that at $\ell$5 dBm only $\ell$6 EMSs were needed to eliminate outages in one region of interest and reduce the outage probability in another from $\ell$7 to $\ell$8 (Benoni et al., 2021).

EMSs also appear as one component of heterogeneous smart electromagnetic environments. In a planning framework that jointly considers static passive EMSs, reconfigurable passive EMSs, smart repeaters, and integrated access-and-backhaul nodes, EMSs are the passive smart electromagnetic entities mounted on facades to recover or enhance coverage while balancing installation cost and energy consumption. In one real-world scenario, the best-compromise solution used $\ell$9 static passive EMSs, $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$0 reconfigurable passive EMS, and $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$1 smart repeaters, achieving blind-spot area reductions of $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$2 and $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$3 at two time instants with $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$4 and $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$5 W (Benoni et al., 10 Sep 2025).

Beyond opaque reflective panels, EMSs have been extended to transmission through existing windows. Optically-transparent opportunistic EMSs designed on standard “4-10-4” insulated glass at $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$6 GHz achieved optical transmittance $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$7, TE phase coverage $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$8 deg, and worst-case transmittance magnitude $\Phi(r)=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top r$9 dB. For normal transmission, the OTO-EMS improved the transmitted power by $\mathbf{p}_\ell$0 dB relative to a non-patterned insulated-glass panel of the same size, and it remained only $\mathbf{p}_\ell$1 dB below the ideal hollow-window reference (Oliveri et al., 2023). In mobile environments, static passive curved EMSs mounted on vehicle doors were shown to raise opportunistic-relay connectivity from $\mathbf{p}_\ell$2 for specular CEMS to $\mathbf{p}_\ell$3 for optimized anomalous CEMS, while in low-traffic regimes high penetration improves network connectivity by more than $\mathbf{p}_\ell$4 (Ayoubi et al., 2024).

5. Sensing, localization, imaging, and channel-aware learning

EMSs have become increasingly important in sensing. In near-field radar imaging under non-line-of-sight conditions, a modular non-reconfigurable EMS illuminated by a moving source synthesizes a wide effective aperture $\mathbf{p}_\ell$5, enabling multi-view focusing over an occluded region. With $\mathbf{p}_\ell$6 GHz, $\mathbf{p}_\ell$7 GHz, and $\mathbf{p}_\ell$8, the reported system achieved SNR $\mathbf{p}_\ell$9 dB in $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$0 of the NLOS area and reduced cross-range resolution from about $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$1 cm for the standalone radar to about $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$2 cm with the EMS-enabled aperture mapping (Bellini et al., 2024).

For vehicular localization, EMSs mounted on the roof are used as high-reflectivity markers that make vehicles behave like point targets instead of extended targets. In the RIS-based formulation, a roof marker with area $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$3 matches a corner reflector of side $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$4 within a $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$5 dB margin; for $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$6 cm, this yields a practical size of about $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$7 cm$$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$8. The same work shows that wideband and spatially wideband effects become central at $$\phi_\ell=\Phi_0+(\mathbf{k}_o-\mathbf{k}_i)^\top \mathbf{p}_\ell$$9–$\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$0 GHz, and that the localization performance depends sensitively on the trade-off between EMS aperture, beam squint, and angle uncertainty (Tagliaferri et al., 2023).

A distinct research line uses static EMSs to improve channel charting. In a $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$1 GHz 3D ray-traced urban canyon with two $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$2 EMS panels placed on facing building walls, the optimized codebook-based EMS configuration reduced the $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$3th-percentile localization error from over $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$4 m without EMS to less than $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$5 m, while trustworthiness and continuity empirical CDFs shifted favorably across the $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$6th–$\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$7th percentiles (Maleki et al., 10 Aug 2025). A subsequent study confirmed the same mechanism with both semi-supervised t-SNE and semi-supervised autoencoders, reporting $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$8th-percentile positioning error reductions from $\Gamma(x,y)=A(x,y)e^{j\phi(x,y)}$9 m to $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$0 m for t-SNE and from $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$1 m to $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$2 m for the autoencoder at $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$3 supervision, together with a reduction of catastrophic trajectory outliers from about $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$4 to $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$5 (Maleki et al., 2 Nov 2025). The key finding is that channel-charting quality depends on a balance between SNR and spatial dissimilarity: gain-focused codebooks can erase location fingerprints, whereas optimized static EMSs enrich multipath while preserving discriminative covariance structure.

Time-modulated EMSs extend EMS-enabled sensing to integrated sensing and communications. By periodically switching local reflection coefficients, a TM-EMS generates a sum beam at $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$6 and a difference beam at $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$7, so that a receiving base station can separate them by frequency and apply monopulse-like processing. Numerical and experimental studies at $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$8 GHz showed simultaneous beam formation with sidelobe control, and in one $\Gamma=\frac{Z_s-\eta}{Z_s+\eta},$9 array example the ratio $$\mathcal{A}_{\infty}^{PCS}(\mathbf{r}_{RX})=\left[\frac{\lambda}{4\pi(r_{RX}+r_{TX})}\right]^2 G_{RX}G_{TX},$$0 ranged from $$\mathcal{A}_{\infty}^{PCS}(\mathbf{r}_{RX})=\left[\frac{\lambda}{4\pi(r_{RX}+r_{TX})}\right]^2 G_{RX}G_{TX},$$1 to $$\mathcal{A}_{\infty}^{PCS}(\mathbf{r}_{RX})=\left[\frac{\lambda}{4\pi(r_{RX}+r_{TX})}\right]^2 G_{RX}G_{TX},$$2 as the base-station direction varied, enabling angle discrimination without modifying the RF hardware beyond frequency separation (Poli et al., 11 May 2025).

6. Practical constraints, common misconceptions, and outlook

Several misconceptions recur in discussions of EMSs. First, EMSs are not synonymous with RISs. Static EMSs are designed to operate without run-time control, which is why they naturally fit unsupervised or semi-supervised tasks such as channel charting, where requiring user positions or CSI to configure the surface would create a circular dependency (Maleki et al., 10 Aug 2025). Second, an EMS is not merely a metallic reflector. Even in static passive form, the surface is patterned to impose a spatially varying phase law that supports anomalous reflection, focusing, or contour-shaped footprints (Benoni et al., 2023). Third, static does not necessarily imply unconfigurable: OTP-EMSs are configurable once at installation and then operate passively thereafter (Oliveri et al., 19 Jul 2025).

The main implementation constraints are consistent across the literature. Performance is conditioned by phase quantization, limited amplitude control, unit-cell spacing, bandwidth, illumination angle, and the validity of local periodicity or deterministic ray-tracing assumptions. Many studies adopt phase-only models with negligible amplitude variation and negligible inter-element coupling; others use diagonal susceptibility tensors or modular linear phase gradients for tractability (Oliveri et al., 2022, Maleki et al., 10 Aug 2025). Hardware nonidealities, manufacturing tolerances, weather, temporal dynamics, and geometry mismatch are frequently acknowledged but not explicitly modeled in the baseline optimization loop (Bellini et al., 2024, Maleki et al., 2 Nov 2025). In wideband sensing, large apertures improve reflectivity but intensify beam squint; in vehicular relays, increasing azimuthal resolution is typically more beneficial than increasing elevation resolution (Ayoubi et al., 2024, Tagliaferri et al., 2023).

A plausible implication is that EMS research is moving from isolated surface design toward infrastructure co-design. Planning frameworks already optimize EMS number, placement, orientation, and interaction with other smart electromagnetic entities (Benoni et al., 10 Sep 2025). Design frameworks already integrate inverse problems, digital twins, and offline codebooks (Oliveri et al., 2021, Oliveri et al., 2024). Emerging variants such as optically-transparent, curved, one-time programmable, and time-modulated EMSs indicate that the central question is no longer whether passive wave-control surfaces can be deployed, but how static, programmable-once, and dynamically modulated skins should be matched to geometry, cost, and task structure in communications and sensing.