Transmissive Reconfigurable Intelligent Surface (TRIS)

• TRIS is a programmable metasurface that shapes electromagnetic waves in transmission via element-wise phase shifts, enabling coherent beamforming and improved coverage.
• Electromagnetic and channel models reveal that TRIS optimizes signal quality through quantized phase control and addresses cascaded path loss in multi-hop links.
• Hardware realizations of TRIS include passive and active designs that demonstrate enhanced data rates, energy efficiency, and effective through-wall beamforming.

A Transmissive Reconfigurable Intelligent Surface (TRIS) is a programmable metasurface that shapes electromagnetic waves in transmission rather than reflection. In the RIS taxonomy, it is the transmissive counterpart to reflective RIS and the pure-transmission limiting case of transmissive–reflective architectures: the incident wave penetrates the surface, acquires element-wise controllable phase shifts and possibly amplitude shaping, and is re-radiated on the opposite side of the panel. In idealized form, a transmissive RIS satisfies $\Gamma^r=0,\Gamma^t=1$, whereas reflective RIS satisfies $\Gamma^t=0,\Gamma^r=1$; hybrid RIS allows both with $(\Gamma^t)^2+(\Gamma^r)^2=1$ (Zeng et al., 2021). The broader RIS survey literature does not always use the acronym “TRIS” explicitly, but it identifies transmissive operation as part of transmissive–reflective RIS and as an operational mode of an “ultimate RIS” supporting active-transmit and passive-transmit states (Basar et al., 2021).

1. Conceptual placement within the RIS taxonomy

TRIS belongs to the general class of reconfigurable intelligent surfaces modeled element-wise by

$y_n = p_n e^{j\phi_n} x_n,$

where $x_n$ is the impinging complex baseband signal, $y_n$ is the outgoing signal, $p_n$ is an amplitude factor, and $\phi_n$ is a programmable phase. In passive operation, $0\le p_n\le 1$; in active operation, $p_n>1$ (Basar et al., 2021). For transmissive operation, the same abstraction is interpreted as a transmission matrix rather than a reflection matrix, so the surface acts as a programmable transmissive aperture rather than a reflector (Basar et al., 2021).

The central architectural distinction is topological. In reflective RIS, the transmitter and receiver are on the same side of the surface; in TRIS, they are on opposite sides, and the surface primarily transmits energy through itself. The 2021 comparative letter formalizes this distinction by defining reflection and transmission zones separated by the RIS plane and showing that TRIS is naturally matched to users in the transmission zone $\Gamma^t=0,\Gamma^r=1$0, especially when direct links are blocked and the surface can be integrated into walls, facades, or smart windows (Zeng et al., 2021). The 2021 RIS survey further emphasizes that transmissive–reflective RIS can provide full $\Gamma^t=0,\Gamma^r=1$1 coverage by supporting both electric polarization currents and magnetic currents; a pure TRIS is the case in which the surface is engineered to primarily transmit or refract rather than reflect (Basar et al., 2021).

This placement clarifies a recurring misconception. TRIS is not simply a reflective RIS viewed from the opposite side. Reflective RIS typically uses a conductive, opaque substrate and prevents penetration, whereas TRIS removes the metallic backplane and uses transmissive meta-atoms so that the wave passes through the structure and is shaped in the forward half-space (Zeng et al., 2021).

2. Electromagnetic and channel models

At the electromagnetic level, TRIS is typically modeled by a per-element transmissive coefficient. In the compact base-station formulation,

$\Gamma^t=0,\Gamma^r=1$2

denotes the transmissive phase shifter of the $\Gamma^t=0,\Gamma^r=1$3-th unit, where $\Gamma^t=0,\Gamma^r=1$4 captures transmission loss and $\Gamma^t=0,\Gamma^r=1$5 is the programmable phase (Boloori et al., 3 Nov 2025). If the incident field at element location $\Gamma^t=0,\Gamma^r=1$6 is $\Gamma^t=0,\Gamma^r=1$7, the transmitted field contribution is approximated by

$\Gamma^t=0,\Gamma^r=1$8

so wavefront shaping is achieved by compensating the propagation phase from the source to the surface and from the surface to the target (Boloori et al., 3 Nov 2025).

For a near-field source and far-field user, the received signal can be written as

$\Gamma^t=0,\Gamma^r=1$9

with $(\Gamma^t)^2+(\Gamma^r)^2=1$0 the source-to-TRIS channel, $(\Gamma^t)^2+(\Gamma^r)^2=1$1 the TRIS-to-user channel, and $(\Gamma^t)^2+(\Gamma^r)^2=1$2. The corresponding SNR is

$(\Gamma^t)^2+(\Gamma^r)^2=1$3

and the ideal continuous-phase choice is

$(\Gamma^t)^2+(\Gamma^r)^2=1$4

which yields coherent combining at the user (Boloori et al., 3 Nov 2025). In low-resolution hardware, the phase is quantized to a finite set $(\Gamma^t)^2+(\Gamma^r)^2=1$5, and both the 2025 and 2026 MA–TRIS studies use nearest-point quantization of the ideal phase to model discrete TRIS control (Boloori et al., 3 Nov 2025, Boloori et al., 13 Mar 2026).

At the link-budget level, passive RIS-assisted propagation exhibits the familiar multiplicative or “double” path loss. For LOS-dominated source–surface and surface–destination links with unavailable direct path, the passive RIS SNR scales as

$(\Gamma^t)^2+(\Gamma^r)^2=1$6

showing coherent $(\Gamma^t)^2+(\Gamma^r)^2=1$7 gain but also multiplicative decay in the two hop distances (Basar et al., 2021). This observation applies to passive TRIS as well: transmissive operation does not eliminate the cascaded path-loss structure by itself. That point is essential because TRIS is sometimes conflated with an active relay; passive TRIS shapes the field but does not regenerate the signal (Basar et al., 2021).

Active transmissive RIS modifies this picture by adding amplification inside the meta-atom. The active TRIS prototype at 2.6 GHz models each unit by

$(\Gamma^t)^2+(\Gamma^r)^2=1$8

where $(\Gamma^t)^2+(\Gamma^r)^2=1$9 represents residual passive loss, $y_n = p_n e^{j\phi_n} x_n,$0 is amplifier gain controlled by bias, and $y_n = p_n e^{j\phi_n} x_n,$1 is programmable phase (Song et al., 2024). To predict link power, that work introduces a dual radar cross section-based path-loss model,

$y_n = p_n e^{j\phi_n} x_n,$2

which is validated experimentally against measured distance and angle sweeps (Song et al., 2024).

3. Physical architectures and hardware realizations

The physical realization of TRIS spans passive transmitarrays, active amplifying metasurfaces, and integrated transceiver panels. A foundational architectural distinction is the substrate: reflective RIS is built on a conductive, opaque substrate, whereas transmissive or transmissive–reflective RIS requires a transparent substrate and meta-atoms capable of controlling the transmitted field (Basar et al., 2021). In the survey account, transmissive–reflective operation relies on magnetic RIS elements that support both electric polarization currents and magnetic currents, enabling simultaneous reflection and refraction (Basar et al., 2021).

A widely cited passive implementation is the 27 GHz, 2-bit transmissive RIS prototype comprising $y_n = p_n e^{j\phi_n} x_n,$3 elements, each using a penetration structure with a 1-bit current-reversible dipole and a $y_n = p_n e^{j\phi_n} x_n,$4 digital phase shifter based on a quadrature hybrid coupler (Tang et al., 2022). The prototype reports an average insertion loss of approximately $y_n = p_n e^{j\phi_n} x_n,$5 dB per state, a maximum broadside gain of $y_n = p_n e^{j\phi_n} x_n,$6 dBi at 27 GHz, and two-dimensional beam scanning up to $y_n = p_n e^{j\phi_n} x_n,$7 (Tang et al., 2022). In system experiments, the same platform increases the achievable data rate from 1024 Mbps to 1683 Mbps, reduces required transmit power by 8.2 dB for comparable performance, and restores a blocked 27 GHz link to 1683 Mbps through obstacle-penetrating transmissive beamforming (Tang et al., 2022).

Active implementations introduce gain at the element level. The 2024 active TRIS work develops a 2-bit programmable amplifying transmissive RIS operating at 2.6 GHz, in which each element uses top and bottom radiation patches, a microstrip line, a power amplifier, a quadrifilar directional coupler, and an SP4T switch (Song et al., 2024). The fabricated $y_n = p_n e^{j\phi_n} x_n,$8 array validates four phase states $y_n = p_n e^{j\phi_n} x_n,$9, $x_n$0, $x_n$1, and $x_n$2, beam steering over $x_n$3, and a measured signal power gain of 11.9 dB compared with the traditional passive RIS in the reported design (Song et al., 2024).

A different hardware direction appears in through-the-wall sensing. The TRIS-HAR system uses a 1-bit, $x_n$4, 5.8 GHz transmissive surface of physical size $x_n$5, controlled through 196 bias lines and four logic boards. After greedy row/column configuration, it reports an array gain of 11.7 dBi and raises received power from $x_n$6 dBm to $x_n$7 dBm in the reported laboratory link (Liu et al., 2024). This demonstrates that even 1-bit transmissive control can materially change the propagation geometry in a wall-penetrating setting (Liu et al., 2024).

4. TRIS as a transmitter and transceiver architecture

A major shift in the TRIS literature is the move from “environmental” deployment to transmitter-side integration. The TRIS-enabled transceiver architecture, denoted TRTC, uses a single horn feed, a passive transmissive RIS aperture, and a controller that implements time-modulation arrays. In downlink, the horn emits an information-free carrier, while the controller loads multi-user information onto the surface by time-modulating the transmissive elements; in uplink, OFDMA user signals pass through the TRIS toward the horn and controller for reception (Li et al., 2024). Within this architecture, the effective control vector is

$x_n$8

with $x_n$9 the TRIS beamforming matrix and $y_n$0 the user-stream vector, and the time-modulation array maps element-wise complex control symbols $y_n$1 into switching parameters $y_n$2 (Li et al., 2024).

A closely related transmitter-centric formulation appears in cognitive RSMA networks, where a single feed antenna illuminates a transmissive RIS and the controller uses time-modulated arrays to realize complex element coefficients. The effective TRIS-side precoding matrix $y_n$3 supports a common stream and private streams under interference constraints toward primary users, with DC programming and SCA used to balance spectral efficiency and energy efficiency (Liu et al., 2023). In secure ISAC, the same general idea is extended so that the common RSMA stream acts simultaneously as useful signal and artificial noise, while transmissive beamforming and timeslot allocation are jointly optimized under secrecy outage, detection, and beampattern constraints (Liu et al., 2024).

Near-field TRIS architectures further exploit the geometry between feed and surface. In the movable-antenna TRIS base-station design, a single movable antenna is placed in the near field of the TRIS, and the joint choice of antenna position and quantized TRIS phase is optimized for SNR (Boloori et al., 3 Nov 2025). Under a 20 GHz setup with user distance 12.5 m, the reported results give, for $y_n$4 elements, approximately 46.6 dB SNR for continuous phase, 45.7 dB for 2-bit TRIS with antenna optimization, and 43.3 dB for 1-bit TRIS with antenna optimization; for a $y_n$5 TRIS, the average relative SNR is approximately 0.70, 0.82, 0.95, and 0.99 of the continuous-phase upper bound for 1-, 2-, 3-, and 4-bit phase shifters, respectively (Boloori et al., 3 Nov 2025). The 2026 extension replaces search by an alternating optimization framework with gradient ascent for antenna position and quantized phase alignment for the TRIS, again emphasizing that near-field geometry can mitigate coarse phase quantization (Boloori et al., 13 Mar 2026).

5. Performance regimes, applications, and comparative behavior

TRIS performance is highly geometry-dependent. In the multi-user comparative analysis of reflective, transmissive, and hybrid RIS, the pure transmissive sum-rate upper bound is

$y_n$6

which increases with the number $y_n$7 of users in the transmission zone, whereas the reflective counterpart decreases with $y_n$8 (Zeng et al., 2021). That paper establishes a threshold $y_n$9 such that TRIS outperforms reflective RIS when transmission-side users are sufficiently numerous, while hybrid RIS can dominate when the number of elements is very large and users must be served on both sides (Zeng et al., 2021). This makes clear that TRIS is not universally preferable; its strength lies in opposite-side service geometry.

The technology has been explored across communications, sensing, and joint information–power transfer. In human activity recognition, the TRIS-HAR system improves recognition accuracy from 85.00% to 98.06% in laboratory conditions and maintains 98.33% and 97.78% accuracy in office and meeting-room environments, respectively, by creating stronger and more informative through-the-wall channels for the HiMamba state-space model (Liu et al., 2024). In cognitive radar, an RL-driven TRIS transmitter using a SARSA agent adapts the phase matrix $p_n$0 and is reported to match, or for large numbers of elements even surpass, conventional cognitive MIMO radar in multi-target detection while using far fewer transmit RF chains (Umra et al., 17 Sep 2025).

In SWIPT, TRIS is treated as a transmissive transceiver that jointly serves information-decoding and energy-harvesting users through active beamforming under per-element power constraints. One line of work reformulates the beam design into SDP plus SCA and penalty-based updates (Guo et al., 15 Nov 2025), while another uses WMMSE, MM, SOCP, and ADMM to obtain a parallelizable low-complexity algorithm without performance degradation (Guo et al., 15 Nov 2025). A related spatial-modulation architecture activates TRIS columns rather than conventional antennas, derives closed-form ABEP upper bounds, and reports that the improved TRIS-SM scheme outperforms conventional SM in reliability (Zhu et al., 2024).

Finally, practical beam synthesis itself has become a TRIS application area. A geometrical-optics model together with constrained Max–min optimization enables simultaneous multi-beam synthesis and directional suppression, outperforming SDR–SDP and QuantRand in the reported simulations and prototype experiments, with explicit relevance to physical-layer security and interference mitigation (Xiong et al., 2024).

6. Open problems and research directions

Several research fronts remain unsettled. Hardware design remains central: low-loss transparent substrates, magnetic meta-atoms, fast switching networks, scalable control buses, and aesthetically acceptable integration into walls and windows are repeatedly identified as limiting factors for practical deployment (Basar et al., 2021). Active TRIS adds another layer of concerns, including power consumption, thermal management, stability, linearity, and electromagnetic exposure, particularly when scaling from tens to hundreds or thousands of amplified units (Song et al., 2024).

Channel modeling and calibration are likewise incomplete. The RIS survey identifies the need for accurate dual-output scattering models, standalone operation with sparse sensors and embedded intelligence, and near-field/far-field modeling across different service regions (Basar et al., 2021). The near-field MA–TRIS studies highlight additional gaps in wideband behavior, dispersion, imperfect CSI, and hardware-aware optimization under quantized phase control (Boloori et al., 3 Nov 2025, Boloori et al., 13 Mar 2026). In the transmitter-centric TRTC architecture, uplink channel estimation must often infer cascaded near-field horn–TRIS and far-field TRIS–user structure without conventional reciprocity, which motivates codebook-based, message-passing, and separable-cascade estimators (Li et al., 2024).

System-level control is another open theme. TRIS has already been coupled with RSMA, TMA, SARSA-based radar adaptation, WMMSE beamforming, and state-space sensing models, but those integrations are largely scenario-specific. This suggests that future work will likely emphasize cross-layer TRIS control, robust online optimization, and task-oriented reconfiguration rather than static beam steering alone. The broader RIS vision of PHY slicing—where transmissive, reflective, active, passive, and absorptive subsurfaces are dynamically resized—positions TRIS as one operational mode inside a more general programmable aperture architecture rather than an isolated hardware category (Basar et al., 2021).