Optical Reconfigurable Intelligent Surfaces (O-RIS)

• O-RIS are reconfigurable optical metasurface platforms that modulate amplitude and phase to create a programmable virtual line-of-sight in diverse communication systems.
• They enhance performance in free-space optical links, resonant-beam SWIPT, and quantum networks through coherent beam steering, diversity gain, and optimized channel modeling.
• Transparent O-RIS designs integrate switch-controlled metasurfaces on windows, balancing high electromagnetic reconfigurability at mmWave frequencies with optical clarity.

Searching arXiv for recent and foundational papers on optical reconfigurable intelligent surfaces, including FSO, quantum, and transparent/window-integrated RIS. Optical reconfigurable intelligent surface (O-RIS) denotes a passive, reconfigurable optical metasurface or mirror-array platform that imposes programmable amplitude and phase transformations on incident waves to redirect, reshape, or coherently combine propagation paths. In the current literature, the term is used in two closely related but non-identical senses: first, for RIS operating in optical or infrared free-space optical (FSO) links, resonant-beam systems, and quantum optical networks; second, for optically transparent RIS deployed at radio or millimeter-wave frequencies while preserving visible transparency for installation on windows and other transparent surfaces (Chapala et al., 2021, Fang et al., 2023, Chehimi et al., 2024, Yang et al., 7 Sep 2025). Across these settings, the common function is the creation of a programmable virtual line-of-sight (LoS), but the underlying propagation models range from scalar diffraction and cavity optics to statistical fading theory and far-field reflectarray synthesis.

1. Scope, terminology, and canonical architectures

In RIS-empowered FSO communication, the surface is modeled as a passive metasurface with $N$ tunable elements, each controlling phase, and possibly amplitude, to shape the incident optical wave toward the receiver. Under perfect phase compensation, coherent combining is assumed, yielding a direct-link model

$y=h_{\mathrm{DL}}s+\nu$

and a RIS-assisted model

$y=\sum_{i=1}^{N} h_i g_i s+\nu,$

with independent but non-identically distributed source–RIS channels $h_i$ and RIS–destination channels $g_i$, and additive white Gaussian noise $\nu$ at the receiver (Chapala et al., 2021).

In resonant-beam simultaneous wireless information and power transfer (SWIPT), the O-RIS is inserted into a spatially separated laser cavity operating at $\lambda=1064\ \mathrm{nm}$. Each RIS element is represented by a complex reflection coefficient

$$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$

and, in the reported realization, $\beta$ is set to $1$ to maximize energy transfer. The RIS is reconfigured from channel state information inferred by a CMOS sensor observing the intra-cavity field, so that a blocked LoS path is replaced by a non-line-of-sight (NLOS) resonant path (Fang et al., 2023).

In FSO quantum networking, the O-RIS is a passive, reconfigurable optical metasurface or mirror array that establishes a virtual LoS between a quantum base station and blocked users. A single RIS is placed in a feasible three-dimensional region $y=h_{\mathrm{DL}}s+\nu$0, partitioned into $y=h_{\mathrm{DL}}s+\nu$1 sub-surfaces, and each sub-surface is phase-programmed to direct entangled photons toward its intended user under far-field operation (Chehimi et al., 2024).

A distinct but related usage appears in optically transparent RIS for 5G mmWave systems. There, the surface is a via-less stack of two screen-printed silver-nanowire conductive layers on PEN films laminated to transparent glass: the top layer forms the resonant, switch-loaded metasurface, and the bottom layer is a reflective ground plane. The device is optically transparent yet electromagnetically reconfigurable at $y=h_{\mathrm{DL}}s+\nu$2–$y=h_{\mathrm{DL}}s+\nu$3 (Yang et al., 7 Sep 2025).

Domain	O-RIS role	Representative formulation
FSO link analysis	Coherent redirection and diversity gain	$y=h_{\mathrm{DL}}s+\nu$4
Resonant-beam SWIPT	NLOS cavity closure and beam steering	$y=h_{\mathrm{DL}}s+\nu$5
FSO quantum network	Virtual LoS for entanglement distribution	RIS partitioned into $y=h_{\mathrm{DL}}s+\nu$6 sub-surfaces
Transparent mmWave RIS	Window-integrated beam redirection with visible transparency	Screen-printed AgNW on PEN/glass

This terminological breadth is a recurrent source of ambiguity. In the optical-FSO literature, O-RIS usually means operation on optical or infrared beams; in the transparent mmWave literature, it denotes optical transparency of a surface whose electromagnetic operation is not itself optical. This suggests that “O-RIS” is best interpreted contextually rather than as a single hardware class.

2. Propagation physics and impairment models

The FSO performance-analysis literature factorizes each hop into fog, turbulence, and pointing-error terms: $y=h_{\mathrm{DL}}s+\nu$7 Two detection methods are unified through

$y=h_{\mathrm{DL}}s+\nu$8

with $y=h_{\mathrm{DL}}s+\nu$9 for heterodyne detection (HD) and $y=\sum_{i=1}^{N} h_i g_i s+\nu,$0 for intensity modulation/direct detection (IM/DD) (Chapala et al., 2021).

For zero-boresight pointing errors, based on the RIS-specific model built upon Farid and Hranilovic, the pointing gain has PDF

$y=\sum_{i=1}^{N} h_i g_i s+\nu,$1

where $y=\sum_{i=1}^{N} h_i g_i s+\nu,$2, $y=\sum_{i=1}^{N} h_i g_i s+\nu,$3, and

$y=\sum_{i=1}^{N} h_i g_i s+\nu,$4

For direct FSO, $y=\sum_{i=1}^{N} h_i g_i s+\nu,$5; for RIS-assisted FSO,

$y=\sum_{i=1}^{N} h_i g_i s+\nu,$6

Atmospheric turbulence is unified through Fisher–Snedecor $y=\sum_{i=1}^{N} h_i g_i s+\nu,$7, Gamma–Gamma (GG), and Málaga $y=\sum_{i=1}^{N} h_i g_i s+\nu,$8 families. In the reported interpretation, $y=\sum_{i=1}^{N} h_i g_i s+\nu,$9 is a tractable model capturing a range of turbulence conditions, GG is widely accepted for moderate-to-strong scintillation, and $h_i$0 covers weak to saturated turbulence regimes for plane and spherical waves (Chapala et al., 2021).

Fog path-loss is treated in both deterministic and stochastic forms. The deterministic Beer–Lambert component is

$h_i$1

where $h_i$2 is visibility and $h_i$3 is the particle-size exponent. The random-fog gain has PDF

$h_i$4

with shape $h_i$5 and scale-related parameter $h_i$6 (Chapala et al., 2021).

Resonant-beam SWIPT employs a deterministic scalar-diffraction description rather than a fading model. The basic transfer kernel is the Rayleigh–Sommerfeld propagator

$h_i$7

and the RIS acts multiplicatively on the incident field: $h_i$8 The phase $h_i$9 is tied to the angle-of-arrival and angle-of-departure geometry so that the reconfigured path satisfies the round-trip resonance condition (Fang et al., 2023).

In FSO quantum networks, the end-to-end RIS-assisted gain to user $g_i$0 is modeled as

$g_i$1

where $g_i$2 is RIS reflection efficiency, $g_i$3 is receiver responsivity, $g_i$4 is atmospheric attenuation, $g_i$5 is Gamma–Gamma turbulence, and $g_i$6 is pointing error. The atmospheric term is written in dB/km form as

$g_i$7

while the turbulence term uses a Gamma–Gamma PDF and the pointing term uses a zero-boresight-type truncated power-law PDF parameterized by $g_i$8 and $g_i$9 (Chehimi et al., 2024).

The common structural feature is multiplicative degradation across hops and impairments. A plausible implication is that O-RIS design problems are inseparable from geometric placement, beam divergence, aperture capture, and jitter control, even when the surface itself is treated as passive.

3. Unified statistical analysis of RIS-empowered FSO links

A major analytical development is the unified performance framework for RIS-empowered FSO under Fisher–Snedecor $\nu$0, GG, and Málaga $\nu$1 turbulence, zero-boresight pointing errors, deterministic or random fog path-loss, and both HD and IM/DD (Chapala et al., 2021). The PDF of the joint turbulence-pointing term $\nu$2 is expressed in a unified Meijer-$\nu$3 form, and, after multiplication by random fog, the direct-link channel PDF and CDF are obtained in exact closed form. For RIS-assisted links, with $\nu$4 the product of independent links at element $\nu$5 and $\nu$6, the resultant channel PDF and CDF are given by multivariate Fox’s $\nu$7 functions.

The SNR-domain transformation is

$\nu$8

From this transform, the framework derives exact outage probability, average BER, ergodic capacity, and SNR moments for both direct-link FSO and RIS-assisted FSO. The average BER uses the Ansari et al. identity

$\nu$9

with $\lambda=1064\ \mathrm{nm}$0 specified by the binary modulation format (Chapala et al., 2021).

The ergodic capacity is defined as

$\lambda=1064\ \mathrm{nm}$1

with $\lambda=1064\ \mathrm{nm}$2 for HD and $\lambda=1064\ \mathrm{nm}$3 for IM/DD. High-SNR asymptotics then yield diversity-order expressions. For RIS-assisted outage,

$\lambda=1064\ \mathrm{nm}$4

and the reported interpretation is that diversity scales linearly with $\lambda=1064\ \mathrm{nm}$5 under independent per-element channels and perfect RIS phase alignment (Chapala et al., 2021).

Monte Carlo validation with $\lambda=1064\ \mathrm{nm}$6 channel realizations closely matches the analytical PDFs, CDFs, and performance metrics. The comparative results are explicit. Under moderate fog at $\lambda=1064\ \mathrm{nm}$7, average SNR degradation is approximately $\lambda=1064\ \mathrm{nm}$8 for HD and approximately $\lambda=1064\ \mathrm{nm}$9 for IM/DD. At $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$0, IM/DD can suffer an approximately $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$1 ergodic-capacity reduction under fog, versus approximately $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$2 for HD. For light fog and IM/DD, increasing $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$3 from $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$4 to $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$5 yields an approximately $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$6 transmit-power gain to reach $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$7. With $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$8, RIS-assisted FSO achieves approximately $$H_s=\beta e^{j\theta},\qquad \beta\in[0,1],\ \theta\in[0,2\pi),$$9 higher spectral efficiency than direct-link FSO under light fog, even when the direct link may have higher average SNR, which the paper attributes to coherent multi-element combining and the sum-of-products channel structure (Chapala et al., 2021).

These results place O-RIS within a classical reliability-and-capacity framework rather than only a geometric steering framework. They also clarify that the performance benefit is not only path restoration under blockage; it is also diversity enhancement under turbulence, pointing loss, and fog.

4. O-RIS in resonant-beam SWIPT and NLOS cavity closure

In resonant-beam SWIPT, the O-RIS is embedded inside a spatially separated laser cavity composed of a transmitter, a receiver, and free-space propagation segments. The transmitter includes a pump power source exciting an Nd:YVO$\beta$0 gain medium, an input cat’s-eye retro-reflector with $\beta$1, and a CMOS sensor plus controller. The receiver includes an output cat’s-eye retro-reflector with $\beta$2, and the transmitted optical power is split with ratio $\beta$3 into a photovoltaic charging branch and an avalanche photodiode communication branch (Fang et al., 2023).

The free-space round-trip transfer with the RIS in the loop is written as

$\beta$4

where $\beta$5 and $\beta$6 are the transmitter–RIS and RIS–receiver propagation operators, and $\beta$7 is the RIS reflection. The geometry uses the slant distances $\beta$8 and $\beta$9, and the phase is set from the angle-of-arrival $1$0 and angle-of-departure $1$1. No explicit numerical phase optimization is formulated; the controller sets $1$2 deterministically from geometry and optical field sensing to “close” the cavity under NLOS (Fang et al., 2023).

Laser oscillation obeys the threshold condition

$1$3

and the steady mode is obtained through Fox–Li iteration. SWIPT is then defined by

$1$4

with APD current

$1$5

and spectral efficiency, following Lapidoth et al. and the paper’s formulation,

$1$6

The overall architecture couples optical-field propagation, power transfer, and communication within a single resonant structure (Fang et al., 2023).

The reported numerical behavior is strongly geometry-dependent. The maximum NLOS energy efficiency can achieve $1$7 within a transfer distance of $1$8, a translation distance of $1$9, and a rotation angle of $y=h_{\mathrm{DL}}s+\nu$00. Horizontal RIS placement near the midpoint $y=h_{\mathrm{DL}}s+\nu$01 marginally improves the overall round-trip efficiency $y=h_{\mathrm{DL}}s+\nu$02 and output power. At $y=h_{\mathrm{DL}}s+\nu$03, the reported values are $y=h_{\mathrm{DL}}s+\nu$04, $y=h_{\mathrm{DL}}s+\nu$05, and $y=h_{\mathrm{DL}}s+\nu$06 for $y=h_{\mathrm{DL}}s+\nu$07, $y=h_{\mathrm{DL}}s+\nu$08, and $y=h_{\mathrm{DL}}s+\nu$09, with corresponding $y=h_{\mathrm{DL}}s+\nu$10, $y=h_{\mathrm{DL}}s+\nu$11, and $y=h_{\mathrm{DL}}s+\nu$12. Vertical offset increases diffraction loss: at $y=h_{\mathrm{DL}}s+\nu$13, $y=h_{\mathrm{DL}}s+\nu$14 falls from approximately $y=h_{\mathrm{DL}}s+\nu$15 at $y=h_{\mathrm{DL}}s+\nu$16 to approximately $y=h_{\mathrm{DL}}s+\nu$17 at $y=h_{\mathrm{DL}}s+\nu$18 (Fang et al., 2023).

The system is also used for blockage detection. An obstruction is modeled as an invasion plane that shrinks the effective aperture. As $y=h_{\mathrm{DL}}s+\nu$19 increases from $y=h_{\mathrm{DL}}s+\nu$20 to $y=h_{\mathrm{DL}}s+\nu$21 for reflector radius $y=h_{\mathrm{DL}}s+\nu$22, $y=h_{\mathrm{DL}}s+\nu$23 decreases from approximately $y=h_{\mathrm{DL}}s+\nu$24 to approximately $y=h_{\mathrm{DL}}s+\nu$25, and $y=h_{\mathrm{DL}}s+\nu$26 falls to zero at $y=h_{\mathrm{DL}}s+\nu$27. This supports the claim that the resonant-beam architecture is intrinsically safe in the sense used by the paper: intrusion suppresses oscillation rather than sustaining hazardous output power (Fang et al., 2023).

5. O-RIS for entanglement distribution in FSO quantum networks

In terrestrial quantum networking, the O-RIS provides a virtual LoS between a quantum base station and blocked users for discrete-variable entanglement distribution. The network is star-shaped, with one QBS, $y=h_{\mathrm{DL}}s+\nu$28 users, and a single RIS located at $y=h_{\mathrm{DL}}s+\nu$29. The end-to-end path length for user $y=h_{\mathrm{DL}}s+\nu$30 is

$y=h_{\mathrm{DL}}s+\nu$31

and the RIS is partitioned into $y=h_{\mathrm{DL}}s+\nu$32 sub-surfaces, one per user (Chehimi et al., 2024).

Each source state is Bell-diagonal: $y=h_{\mathrm{DL}}s+\nu$33 with fidelity $y=h_{\mathrm{DL}}s+\nu$34 relative to $y=h_{\mathrm{DL}}s+\nu$35. The end-to-end success event is defined by the threshold condition $y=h_{\mathrm{DL}}s+\nu$36, where

$y=h_{\mathrm{DL}}s+\nu$37

Atmospheric attenuation is modeled as $y=h_{\mathrm{DL}}s+\nu$38, turbulence uses Gamma–Gamma fading with shape parameters $y=h_{\mathrm{DL}}s+\nu$39 derived from the Rytov variance, and pointing error is parameterized through $y=h_{\mathrm{DL}}s+\nu$40, $y=h_{\mathrm{DL}}s+\nu$41, and $y=h_{\mathrm{DL}}s+\nu$42 (Chehimi et al., 2024).

The end-to-end entanglement rate is

$y=h_{\mathrm{DL}}s+\nu$43

where $y=h_{\mathrm{DL}}s+\nu$44 is given in Theorem 1 as a Meijer-$y=h_{\mathrm{DL}}s+\nu$45-function expression. Fidelity degradation is decomposed into two quantum channels: a depolarizing channel for the stored matter qubit,

$y=h_{\mathrm{DL}}s+\nu$46

and a phase-damping channel for the flying qubit,

$y=h_{\mathrm{DL}}s+\nu$47

The final Bell-diagonal coefficients $y=h_{\mathrm{DL}}s+\nu$48 then yield the end-to-end fidelity $y=h_{\mathrm{DL}}s+\nu$49 (Chehimi et al., 2024).

The joint design problem maximizes a weighted fairness index

$y=h_{\mathrm{DL}}s+\nu$50

over RIS placement and initial entanglement rates, subject to rate, fidelity, capacity, and placement constraints. The paper solves this nonconvex problem with simulated annealing. In the reported scenarios, the algorithm reaches solutions within $y=h_{\mathrm{DL}}s+\nu$51 of exhaustive search while reducing computation time by more than $y=h_{\mathrm{DL}}s+\nu$52 (Chehimi et al., 2024).

The system-level conclusions are quantitative. Baseline algorithms that ignore fidelity lead to a drop of at least $y=h_{\mathrm{DL}}s+\nu$53 in users’ end-to-end fidelities relative to minimum requirements, whereas the proposed framework satisfies all minimum fidelity constraints. The framework also achieves a $y=h_{\mathrm{DL}}s+\nu$54 enhancement in fairness compared to baseline rate-maximizing approaches. Weather is reported to have a more significant effect than pointing errors and turbulence: changing from sunny conditions with $y=h_{\mathrm{DL}}s+\nu$55 to rain with $y=h_{\mathrm{DL}}s+\nu$56 reduces the end-to-end sum rate by approximately $y=h_{\mathrm{DL}}s+\nu$57, while the cited increases in pointing jitter and strong turbulence reduce it by approximately $y=h_{\mathrm{DL}}s+\nu$58 and $y=h_{\mathrm{DL}}s+\nu$59, respectively. Under strong turbulence, far users become infeasible beyond approximately $y=h_{\mathrm{DL}}s+\nu$60 when the minimum fidelity requirement is $y=h_{\mathrm{DL}}s+\nu$61 (Chehimi et al., 2024).

This body of work broadens O-RIS from a classical link-budget enhancer to a quantum-network control variable. The surface is no longer evaluated only through SNR or BER, but through success probability, Bell-state fidelity, and fairness under heterogeneous application constraints.

6. Transparent wideband O-RIS at mmWave and cross-cutting limitations

The transparent-RIS branch of the literature uses O-RIS to denote optically transparent electromagnetic surfaces intended for window integration. The reported implementation uses screen-printed silver nanowires with conductivity $y=h_{\mathrm{DL}}s+\nu$62 and effective layer thickness $y=h_{\mathrm{DL}}s+\nu$63, which implies $y=h_{\mathrm{DL}}s+\nu$64 under the conventional relation $y=h_{\mathrm{DL}}s+\nu$65. The substrates are AF32 glass with $y=h_{\mathrm{DL}}s+\nu$66, $y=h_{\mathrm{DL}}s+\nu$67, and PEN film with $y=h_{\mathrm{DL}}s+\nu$68, $y=h_{\mathrm{DL}}s+\nu$69. The array is $y=h_{\mathrm{DL}}s+\nu$70, each column serially connects $y=h_{\mathrm{DL}}s+\nu$71 unit cells, and the $y=h_{\mathrm{DL}}s+\nu$72 columns are individually addressable through an ESP32 controller and an NMOS sink of $y=h_{\mathrm{DL}}s+\nu$73 per column (Yang et al., 7 Sep 2025).

The unit cell uses two H-shaped resonators serially connected along the $y=h_{\mathrm{DL}}s+\nu$74-axis, with a MA4AGBLP912 PIN diode at the center of each H-resonator. The switch model is $y=h_{\mathrm{DL}}s+\nu$75, $y=h_{\mathrm{DL}}s+\nu$76, and $y=h_{\mathrm{DL}}s+\nu$77. Simulated reflection performance is reported as $y=h_{\mathrm{DL}}s+\nu$78 in the OFF state across $y=h_{\mathrm{DL}}s+\nu$79–$y=h_{\mathrm{DL}}s+\nu$80, while the ON-state $y=h_{\mathrm{DL}}s+\nu$81 is at least $y=h_{\mathrm{DL}}s+\nu$82 after $y=h_{\mathrm{DL}}s+\nu$83 and exceeds $y=h_{\mathrm{DL}}s+\nu$84 beyond $y=h_{\mathrm{DL}}s+\nu$85. The ON/OFF phase difference remains near $y=h_{\mathrm{DL}}s+\nu$86, varying between approximately $y=h_{\mathrm{DL}}s+\nu$87 and $y=h_{\mathrm{DL}}s+\nu$88, which realizes $y=h_{\mathrm{DL}}s+\nu$89-bit phase quantization. The RIS works from $y=h_{\mathrm{DL}}s+\nu$90 to $y=h_{\mathrm{DL}}s+\nu$91 with $y=h_{\mathrm{DL}}s+\nu$92 angular stability, and the measurement results show up to $y=h_{\mathrm{DL}}s+\nu$93 gain enhancement within a $y=h_{\mathrm{DL}}s+\nu$94 angle range while maintaining high optical transparency and large bandwidth (Yang et al., 7 Sep 2025).

The device is fabricated by screen printing with an AUREL 900 PA printer, a $y=h_{\mathrm{DL}}s+\nu$95-mesh stencil, $y=h_{\mathrm{DL}}s+\nu$96 wire diameter, $y=h_{\mathrm{DL}}s+\nu$97 emulsion thickness, and $y=h_{\mathrm{DL}}s+\nu$98 print speed, followed by drying at $y=h_{\mathrm{DL}}s+\nu$99 for $y=\sum_{i=1}^{N} h_i g_i s+\nu,$00 minutes and photonic sintering with a Novacentrix Pulse Forge at $y=\sum_{i=1}^{N} h_i g_i s+\nu,$01 lamp distance, $y=\sum_{i=1}^{N} h_i g_i s+\nu,$02 pulse length, and $y=\sum_{i=1}^{N} h_i g_i s+\nu,$03 pulse rate (Yang et al., 7 Sep 2025).

Several limitations recur across the O-RIS literature. RIS-assisted FSO performance analysis often assumes perfect phase compensation, zero-boresight pointing errors, independent per-element channels, and no explicit amplitude control or insertion loss (Chapala et al., 2021). Resonant-beam SWIPT idealizes the RIS with $y=\sum_{i=1}^{N} h_i g_i s+\nu,$04 and continuous phase control, and neglects atmospheric turbulence in its scalar-diffraction model (Fang et al., 2023). Quantum-network analysis assumes far-field RIS steering, reflection efficiency $y=\sum_{i=1}^{N} h_i g_i s+\nu,$05 depending on bias voltage and wavelength, and does not explicitly model phase quantization or finite element size (Chehimi et al., 2024). Transparent mmWave O-RIS uses $y=\sum_{i=1}^{N} h_i g_i s+\nu,$06-bit phase control, which the paper identifies as limiting sidelobe control and maximum beamforming gain relative to multi-bit designs (Yang et al., 7 Sep 2025).

A common misconception is that all O-RIS results concern the same physical regime. The literature instead spans optical-frequency FSO beam control, intra-cavity resonant optics, quantum-state distribution, and visually transparent but mmWave-operating reflectarrays. Another misconception is that O-RIS primarily solves blockage alone. The cited works show a broader role: in classical FSO it changes diversity order and outage behavior; in resonant-beam SWIPT it couples NLOS restoration with safety and charging; in quantum networks it enters directly into fidelity-constrained optimization; and in transparent mmWave systems it enables architecturally unobtrusive deployment on glass (Chapala et al., 2021, Fang et al., 2023, Chehimi et al., 2024, Yang et al., 7 Sep 2025).