Smart Electromagnetic Environment (SEME)

• SEME is a paradigm that treats physical objects as active electromagnetic controllers to dynamically shape radio-frequency propagation.
• It leverages devices like metasurfaces, smart skins, and RIS to enhance wireless coverage, capacity, and sensing with integrated control algorithms.
• SEME design combines physical layer models with digital twins and optimization methods to deliver cost-effective, adaptive, and secure electromagnetic environments.

Smart Electromagnetic Environment (SEME) denotes a wireless-infrastructure paradigm in which the propagation medium is treated as an additional design degree of freedom rather than as a passive source of fading, blockage, clutter, or multipath distortion. In SEME, walls, facades, ceilings, vehicles, street furniture, and other objects can be equipped with field-manipulating devices, while control algorithms jointly shape electromagnetic propagation to meet coverage, capacity, efficiency, sensing, or safety objectives. The literature spans passive and reconfigurable metasurfaces, electromagnetic skins (EMSs), reconfigurable intelligent surfaces (RISs), smart repeaters, integrated access-and-backhaul nodes, and even source-synthesis strategies that exploit already-present scatterers without adding new surfaces (Kisseleff et al., 2020, Barbuto et al., 2021, Benoni et al., 10 Sep 2025, Rù et al., 2024).

1. Conceptual scope and evolution

SEME is defined in multiple, closely related ways across the literature. One formulation describes it as turning every object and scatterer in a wireless scenario into an active “degree of freedom” for shaping radio-frequency propagation, rather than treating such objects as uncontrollable impairments. Another defines it as the ensemble of reconfigurable electromagnetic objects whose response is jointly optimized to shape multi-user, multi-cell wireless channels so as to meet city-wide performance, efficiency, and safety targets. A third formulation emphasizes that the physical propagation medium itself becomes an additional design degree of freedom for restoring or enhancing coverage in Regions of Interest while respecting installation-cost and energy-consumption constraints (Rocca et al., 2021, Kisseleff et al., 2020, Benoni et al., 10 Sep 2025).

The historical framing of SEME is tied to a broader shift in wireless-system design. In the “Metasurfaces 3.0” perspective, the field moved from passive, homogeneous, periodic arrangements, to spatially modulated metasurfaces, and then to programmable, space-time-modulated, and even cognitive surfaces. Within this progression, the environment is no longer merely adapted to; it becomes part of the network and, in some formulations, an active, reconfigurable entity that can sense, process, and manipulate electromagnetic waves (Barbuto et al., 2021).

A longer-range conceptual precursor appears in the taxonomy of symbiotic EM sensors. There, the emphasis is on cooperative coexistence between sensing and communications systems through commensal and mutualistic relationships, such as passive radar illuminated by FM broadcast towers or database-assisted sharing in TV White Space ecosystems. That framework does not use the later metasurface-centric language, but it already treats the electromagnetic environment as something to be intentionally organized rather than merely endured. This suggests that SEME is not solely a metasurface deployment model; it is also a system-level view of coordinated electromagnetic coexistence and control (Inggs et al., 2017).

A common misconception is that SEME is synonymous with RIS deployment alone. The cited literature is broader: it includes static passive smart skins, modular reflecting EMS tiles, active relays, heterogeneous deployments, environmental electromagnetic illusion, and opportunistic source synthesis that uses existing buildings as virtual reflectors without adding reconfigurable metasurfaces (Flamini et al., 2022, Taghvaee et al., 2024, Rù et al., 2024).

2. Electromagnetic and system-theoretic foundations

At the unit-cell level, a central abstraction is the programmable reflection coefficient

$$\Gamma(\theta,\phi,\omega)=A(\theta,\phi,\omega)e^{j\Phi(\theta,\phi,\omega)},$$

where $A(\theta,\phi,\omega)\in[0,1]$ is the amplitude response and $\Phi(\theta,\phi,\omega)\in[0,2\pi)$ is the phase shift imparted by the element. In many RIS designs, $A\approx 1$ is fixed and control is exerted primarily through phase. This local description underpins anomalous reflection, beam steering, focusing, and broader wavefront engineering (Kisseleff et al., 2020).

At the communication-link level, a standard end-to-end channel model writes the effective downlink channel as

$$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$

where $H_{BU}$ models direct BS-to-user links, $H_{BR}^{(n)}$ the BS-to-RIS channels, $\Theta^{(n)}$ the diagonal reflection matrix of RIS $n$, and $H_{RU}^{(n)}$ the RIS-to-user channels. This formulation makes explicit that SEME augments, rather than replaces, the baseline channel with programmable boundary interactions (Kisseleff et al., 2020).

For coverage-oriented analysis, ray-tracing-based received-power models are frequently used. In the indoor Wi-Fi experimental study, the quasi-optical power at a location $A(\theta,\phi,\omega)\in[0,1]$0 is written as

$A(\theta,\phi,\omega)\in[0,1]$1

where $A(\theta,\phi,\omega)\in[0,1]$2 is the total field, $A(\theta,\phi,\omega)\in[0,1]$3 the handset gain, and $A(\theta,\phi,\omega)\in[0,1]$4 the free-space impedance. The same study defines a received-power CDF, gain maps, and the region-of-interest reduction metric

$A(\theta,\phi,\omega)\in[0,1]$5

These quantities translate field manipulation into service-level performance indicators (Benoni et al., 2023).

A more general theoretical treatment models any linear electromagnetic environment with boundary conditions as a space-variant linear feedback filter. After expanding the induced currents and tangential fields on each electromagnetic object in a finite basis, the global response is described by coupled propagation and boundary matrices, leading to

$A(\theta,\phi,\omega)\in[0,1]$6

Here, $A(\theta,\phi,\omega)\in[0,1]$7 collects the Green-operator coupling blocks and $A(\theta,\phi,\omega)\in[0,1]$8 the local constitutive or boundary operators of the objects. This formulation is important because it unifies multi-object, multi-surface programmable environments within a single operator-theoretic framework and shows that programmability enters through the boundary operators themselves (Dardari, 2023).

The same electromagnetic formalism also supports non-communication objectives. In environmental electromagnetic illusion, generalized sheet transition conditions and transfer-matrix models are used to synthesize metasurface-loaded environments whose total reflection mimics a virtual object or suppresses the signature of a real one without coating that object directly. A plausible implication is that SEME extends naturally from coverage engineering into security, deception, and sensing-control applications (Taghvaee et al., 2022, Taghvaee et al., 2024).

3. Enabling devices and architectural realizations

One of the simplest SEME realizations uses static-passive electromagnetic skins (SP-EMSs). In the indoor Wi-Fi demonstration, an SP-EMS is a two-layer printed-circuit metasurface whose unit cell consists of two square metallic patches on FR-4 boards backed by a ground plane. By tailoring the patch dimensions across an $A(\theta,\phi,\omega)\in[0,1]$9 array with interelement spacing $\Phi(\theta,\phi,\omega)\in[0,2\pi)$0, the skin implements a prescribed anomalous reflection law from an access-point barycenter toward a target dead-zone. In this realization, the building wall itself becomes a passive beam-shaping component of the network (Benoni et al., 2023).

A related urban design approach partitions an admissible facade region $\Phi(\theta,\phi,\omega)\in[0,2\pi)$1 into minimum-size square tiles $\Phi(\theta,\phi,\omega)\in[0,2\pi)$2, each of side $\Phi(\theta,\phi,\omega)\in[0,2\pi)$3 and area $\Phi(\theta,\phi,\omega)\in[0,2\pi)$4. Each tile behaves like a small passive reflecting panel that re-radiates a pencil beam toward a designated point of interest in the area of interest. The modularity of this scheme is significant because it makes the SEME synthesis problem combinatorial: one optimizes not only the phase law but also which tiles are installed (Rocca et al., 2021).

RIS-based realizations provide dynamic control through electronically tunable unit cells. In the Smart Cities viewpoint, each RIS is a planar array of subwavelength unit cells loaded with tunable impedance networks such as varactors or PIN diodes. In heterogeneous mmWave SEME deployments, RISs are described as nearly passive programmable reflectors with surface sizes 25×25 cm² or 50×50 cm², fully duplex operation, zero-noise amplification, and power consumption below 2 W. These devices occupy an intermediate point between fully static skins and active relays (Kisseleff et al., 2020, Flamini et al., 2022).

Later hardware extends this model in two directions. The adaptive metasurface (AMS) integrates sensing and manipulation within each unit cell: a 6×6 array uses an H-shaped slot antenna to couple a fraction of the incident wave into a sensing branch, while a varactor-loaded phase shifter tunes the reflected phase through $\Phi(\theta,\phi,\omega)\in[0,2\pi)$5–$\Phi(\theta,\phi,\omega)\in[0,2\pi)$6 with $\Phi(\theta,\phi,\omega)\in[0,2\pi)$7. Each unit reports sensed magnitude and phase through an RF switch matrix to a USRP receiver, eliminating the need for external sensors. This architecture embodies a closed-loop SEME in hardware rather than only in external control software (Yang et al., 2024).

At the opposite end of the complexity spectrum are one-time programmable passive electromagnetic skins (OTP-EMSs). These use copper patches on Rogers 4350B whose opposite edges are shunted to ground through surface-mount fuses that can be irreversibly blown. The resulting meta-atom is passive-static in operation but configurable once during installation. The explicit design goal is to combine modular fabrication, one-time configurable reflection properties, passive-static operation, and zero maintenance (Oliveri et al., 19 Jul 2025).

SEME architectures also include active Smart EM Entities. In the industrial viewpoint and in later heterogeneous planning work, smart repeaters are non-regenerative amplify-and-forward relays with beamforming antenna faces, while IAB nodes are regenerative relays or micro-BTSs that support in-band wireless backhaul. Static passive smart skins, reconfigurable passive skins, smart repeaters, and IAB nodes are treated as complementary rather than competing technologies (Flamini et al., 2022, Benoni et al., 10 Sep 2025).

4. Design, optimization, and digital-twin methodologies

SEME design problems are typically formulated as nonconvex, mixed discrete-continuous optimizations. In the modular reflecting-skin formulation, the decision variable is a binary vector $\Phi(\theta,\phi,\omega)\in[0,2\pi)$8 indicating whether each tile is installed. The objectives are conflicting: coverage-error decreases as more tiles are added, while deployment complexity increases. The paper therefore uses Binary NSGA-II GA to search for Pareto-optimal layouts, returning multiple non-dominated solutions ordered by increasing complexity (Rocca et al., 2021).

System-by-Design (SbD) is a recurrent methodology for large-scale SEME planning. In the urban QoS-planning work, SbD decomposes the process into EM-Skin Design, Problem Formulation, Fitness-Function Evaluation, and Solution-Space Exploration. A Digital Twin based on Gaussian-Process regression is trained on a limited number of full ray-tracing evaluations and then used inside a Binary Genetic Algorithm. For the reported setup, each full-wave ray-tracing evaluation takes approximately $\Phi(\theta,\phi,\omega)\in[0,2\pi)$9 s, brute-force enumeration would take about $A\approx 1$0 days, and replacing around $A\approx 1$1 evaluations with the Gaussian-process surrogate saves about $A\approx 1$2 of the total time (Benoni et al., 2021).

An even more detailed passive-skin synthesis route separates macro-scale current design from micro-scale unit-cell realization. The AI-enhanced aperiodic micro-scale design of passive EM skins first solves an inverse-source problem for an optimal surface-current distribution and then uses a Local UC Digital Twin, built through Ordinary Kriging on an explicitly aperiodic small-scale model, to predict the effective susceptibilities of irregularly assembled unit cells. This removes the local periodicity assumption and incorporates edge effects and non-uniform mutual coupling into the synthesis loop (Oliveri et al., 2021).

Not all SEME optimization acts on surfaces. Opportunistic Source Synthesis (OSS) alters the excitation phases of an existing BTS antenna array so that the surrounding urban scatterers act as opportunistic sources for a target Region of Interest. Its Embedded-plus-Environment Patterns method precomputes the field radiated by each individual antenna element within the full scattering environment and reconstructs trial array patterns analytically during optimization. The decision variables are the element phases, the cost function measures mismatch to a target power map over the Region of Interest, and Particle Swarm Optimization is used with a stagnation-based termination rule (Rù et al., 2024).

Closed-loop adaptive control can dispense with prior geometric knowledge when element-wise channel information is available. In the AMS formulation, the objective is to maximize the cascaded transmitter-to-AMS-to-receiver channel gain, and the closed-form solution is

$A\approx 1$3

Because the coefficients $A\approx 1$4 and $A\approx 1$5 are measured rather than inferred from a prior scene model, the optimization is data-driven at the surface itself (Yang et al., 2024).

The most recent direction casts SEME control as a Markov decision process supported by a semantic electromagnetic world model. In metaEI-WM, the system state includes reconstructed geometry, semantic objects, room topology, material assignments, metasurface poses, and source positions; the action is the metasurface coding matrix; and in-silico rollout is combined with a time-reversal plus modified Gerchberg–Saxton inversion and optional local greedy bit flips. The framework further uses Model-Agnostic Meta-Learning for zero- or one-step adaptation in unseen environments (Liu et al., 2 Jul 2026).

5. Empirical demonstrations and measured gains

The most detailed large-scale indoor validation focuses on a 5.64 GHz Wi-Fi deployment on the second floor of the Mesiano building at UniTrento, approximately $A\approx 1$6 m² in total, with five Aruba AP-304 access points. In Hallway A, the reference condition had $A\approx 1$7 of the hallway below the $A\approx 1$8 dBm threshold, corresponding to $A\approx 1$9 m². With a wall-mounted $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$0-unit-cell SP-EMS of side approximately $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$1 m, the predicted dead-zone area dropped to $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$2 m², corresponding to $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$3, and the threshold CDF metric dropped from $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$4 to $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$5. The measured results were closely aligned: average received-power gain $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$6 dB versus predicted $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$7 dB, maximum gain $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$8 dB versus predicted $$H_{\mathrm{eff}}(\Theta)=H_{BU}+\sum_{n=1}^{N} H_{RU}^{(n)}\Theta^{(n)}H_{BR}^{(n)},$$9 dB, measured ROI $H_{BU}$0 m² with $H_{BU}$1, and measured $H_{BU}$2 (Benoni et al., 2023).

The same study reports user-layer improvements in the dead-zone through OOKLA Speedtest. Peak download increased from $H_{BU}$3 Mbps to $H_{BU}$4 Mbps, average download improved by $H_{BU}$5, average download latency dropped from $H_{BU}$6 ms to $H_{BU}$7 ms, upload increased from $H_{BU}$8 Mbps to $H_{BU}$9 Mbps, and upload latency dropped from $H_{BR}^{(n)}$0 ms to $H_{BR}^{(n)}$1 ms. A tolerance analysis under installation misplacements $H_{BR}^{(n)}$2 m still yielded $H_{BR}^{(n)}$3 in the worst case, with nominal performance at $H_{BR}^{(n)}$4 (Benoni et al., 2023).

The same indoor work also includes a Total Cost of Ownership comparison. For the Hallway A SP-EMS solution, the reported values are $H_{BR}^{(n)}$5100$H_{BR}^{(n)}$6\mathcal{C}_c^{SEME}=\$H_{BR}^{(n)}$7, $H_{BR}^{(n)}$8, and $H_{BR}^{(n)}$9, giving $\Theta^{(n)}$0105$\Theta^{(n)}$1\$\Theta^{(n)}$2 plus %%%%73$H_{BU}$074$\Theta^{(n)}$5 versus $\Theta^{(n)}$61\,350$\Theta^{(n)}$7\$\Theta^{(n)}$8 saved, i.e. an approximately $\Theta^{(n)}$9 reduction (Benoni et al., 2023).

Adaptive hardware has also been experimentally validated. The AMS prototype reports a $n$0 phase-shift range with reflection magnitude at least $n$1, coupling factor approximately $n$2 dB, and DOA-estimation error below $n$3 for $n$4. In integrated sensing and communication scenarios, the adaptive AMS yields approximately $n$5 dB received-power gain over a metal plate in an NLOS obstacle case, approximately $n$6 dB in a corridor multipath case, and approximately $n$7 dB in a simple LOS case. QPSK constellations that collapse under a metal plate are well recovered under AMS control, and signal-to-noise ratios improve by more than $n$8 dB in the worst-case complex environment (Yang et al., 2024).

Time-modulated skins extend SEME from coverage enhancement to joint localization and communication. A 16×16 TM-EMS printed on FR4 and controlled by a Raspberry Pi produced measured harmonic patterns at $n$9 and $H_{RU}^{(n)}$0 that match simulations within $H_{RU}^{(n)}$1 dB, with measured monopulse ratio $H_{RU}^{(n)}$2 and communication-link gain above $H_{RU}^{(n)}$3 dBi in the sum beam. The same line of work reports that a 10×10 idealized design can achieve sidelobe level no higher than $H_{RU}^{(n)}$4 dB at both harmonic beams (Poli et al., 11 May 2025).

One-time programmable passive skins have likewise been validated experimentally. At $H_{RU}^{(n)}$5 GHz, fabricated panels of 10×10, 15×15, 20×20, and 30×30 cells showed reflection magnitude above $H_{RU}^{(n)}$6 dB in both logic states and phase shift of approximately $H_{RU}^{(n)}$7 at $H_{RU}^{(n)}$8. For a 30×30 array, the main lobe was steered to $H_{RU}^{(n)}$9 with side-lobe level approximately $A(\theta,\phi,\omega)\in[0,1]$00 dB, and the measured 10×10 panel showed peak within $A(\theta,\phi,\omega)\in[0,1]$01 dB of prediction and pointing error below $A(\theta,\phi,\omega)\in[0,1]$02 (Oliveri et al., 19 Jul 2025).

The world-model direction has also been evaluated beyond simulation. In three indoor testbeds, metaEI-WM reports average measured power gains of $A(\theta,\phi,\omega)\in[0,1]$03 dB, $A(\theta,\phi,\omega)\in[0,1]$04 dB, and $A(\theta,\phi,\omega)\in[0,1]$05 dB for non-line-of-sight enhancement, bit-error-rate reductions of $A(\theta,\phi,\omega)\in[0,1]$06, $A(\theta,\phi,\omega)\in[0,1]$07, and $A(\theta,\phi,\omega)\in[0,1]$08 orders of magnitude, Willmott IOA above $A(\theta,\phi,\omega)\in[0,1]$09 before tuning and above $A(\theta,\phi,\omega)\in[0,1]$10 after tuning, and end-to-end latency below $A(\theta,\phi,\omega)\in[0,1]$11 ms from natural-language command to metasurface actuation (Liu et al., 2 Jul 2026).

6. Applications, limits, and open directions

Coverage extension remains the most mature application. Urban mmWave studies motivate SEME as an alternative to simply increasing the number of base stations, which would raise cost, power consumption, and ElectroMagnetic Field levels. Indoor and outdoor planning studies show that passive skins can recover blind spots, while heterogeneous deployments combining passive and active Smart EM Entities offer stronger performance-cost-energy trade-offs than homogeneous ones (Flamini et al., 2022, Benoni et al., 10 Sep 2025).

Electromagnetic-exposure management is another explicit use case. The Reduced EMF Exposure Area framework defines the subset of locations where, for a required QoS, the uplink transmit power needed by the user equipment with RIS assistance is lower than without RIS while still meeting the SNR target. In the reported realistic indoor-to-outdoor ray-tracing scenario at $A(\theta,\phi,\omega)\in[0,1]$12 GHz, the maximum user-transmit-power reduction is approximately $A(\theta,\phi,\omega)\in[0,1]$13 dB, and the key placement rule is to prioritize a strong RIS-to-BS link rather than merely proximity to the target user cluster (Phan-Huy et al., 2022).

SEME is also being used for sensing, symbiotic communications, and physiological monitoring. OSS uses the existing urban clutter as opportunistic sources. Time-modulated skins generate sum and difference beams for monopulse-style localization while maintaining a communication link. MetaEI-WM reports approximately $A(\theta,\phi,\omega)\in[0,1]$14 kbps secondary backscatter rate in symbiotic communications and respiratory monitoring with RMSE below $A(\theta,\phi,\omega)\in[0,1]$15 breaths/min and correlation above $A(\theta,\phi,\omega)\in[0,1]$16 against a wearable reference (Rù et al., 2024, Poli et al., 11 May 2025, Liu et al., 2 Jul 2026).

Security-oriented formulations appear under electromagnetic illusion and smart EMI environments. Here the environment, rather than the protected object, is engineered so that a remote observer measures the reflection signature of a virtual object or of a different material configuration. The literature explicitly lists camouflaging, deceptive sensing, radar cognition control, and defence security as application domains. This suggests that SEME should be understood as a general electromagnetic scene-synthesis paradigm, not only as a communications-coverage tool (Taghvaee et al., 2022, Taghvaee et al., 2024).

The main constraints are also consistent across studies. Realistic RIS elements exhibit amplitude loss, finite phase resolution, mutual coupling, and angular-spectrum limitations; cascaded channel estimation scales poorly; control signaling overhead can become substantial in mobile scenarios; passive panels still require powered control electronics if they are reconfigurable; large arrays create calibration and update-latency bottlenecks; and static passive skins cannot adapt in operation, while Digital Twin surrogates may require retraining when the environment changes (Kisseleff et al., 2020, Barbuto et al., 2021, Benoni et al., 2021).

Current research directions therefore move along several axes simultaneously: heterogeneous SEME planning with EMSs, smart repeaters, and IAB nodes; autonomous sensing-and-actuation surfaces; low-overhead control and estimation; physics-informed world models; one-time programmable passive hardware; time modulation for integrated sensing and communications; and software-defined electromagnetic illusion. Taken together, these strands indicate that SEME is evolving from a coverage-enhancement concept into a general framework for programmable, measurable, and optimizable electromagnetic space (Yang et al., 2024, Oliveri et al., 19 Jul 2025, Benoni et al., 10 Sep 2025, Liu et al., 2 Jul 2026).