Reconfigurable Intelligent Base Station (RIBS)

• RIBS is a base station architecture that integrates an intelligent surface with active antennas to enable both active beamforming and passive wave control.
• Architectural variants include reflective backside, transmissive frontside, and surrounding panel setups, each balancing trade-offs in propagation loss, alignment sensitivity, and hardware cost.
• Joint active–passive optimization with accurate CSI acquisition is critical in RIBS to shape composite channels, manage interference, and improve multiplexing gains.

Reconfigurable Intelligent Base Station (RIBS) denotes a class of base-station architectures in which the propagation interface around the BS is reconfigured in real time, typically by integrating an intelligent surface with the BS antenna system or, in an earlier full-duplex formulation, by switching among preset antenna modes at the BS. In the RIS-centric interpretation, the BS does not rely only on active beamforming; it also performs passive or semi-passive wave control through reflective, transmissive, surrounding, continuous, or near-field integrated surfaces, with the aim of shaping coverage, suppressing interference, improving spectral efficiency, and reducing the number of required RF chains relative to conventional fully digital mMIMO (Huang et al., 2024, Interdonato et al., 2022, Beyraghi et al., 11 Jul 2025, Yang et al., 2016).

1. Conceptual scope and evolution

The contemporary RIBS concept is rooted in intelligent-surface-integrated base stations, where a software-controlled metasurface is co-located with the BS antenna array, either inside the radome or on the tower or facade, and enables “reconfigurable propagation” through passive phase or amplitude tuning of the impinging field. In this formulation, the surroundings of the BS become part of the transceiver, so that direct and reflected or transmitted paths are jointly shaped for coverage extension, interference management, and rate enhancement. Passive IS provides low-cost, energy-efficient passive beamforming; semi-passive variants add low-power sensing for CSI acquisition; active and beyond-diagonal intelligent surfaces are identified as future extensions that enlarge the control space at the expense of cost, energy consumption, and noise amplification (Huang et al., 2024).

A second, closely related lineage places a non-large active array at short distance from a RIS, so that the active array supplies a limited number of RF chains while the RIS supplies a large number of controllable electromagnetic degrees of freedom. In this configuration, the effective user signatures become $\bar{h}_k = HPh_k$, and the design objective is not merely beam steering but the sculpting of composite channels so that a small active array can approach the multiplexing behavior of much larger conventional arrays. This is the sense in which RIBS was explicitly introduced as an antenna structure able to approach massive MIMO performance with reduced hardware resources (Interdonato et al., 2022).

An older usage of the broader idea appears in full-duplex cellular networks with reconfigurable antennas at the BS transmitter and receiver. There, reconfigurability is realized not by a metasurface but by preset transmit and receive modes, selected over time as $\alpha(t)$ and $\beta(t)$, to induce favorable channel fluctuations and interference structures. The central mechanism is blind or partially informed interference management rather than passive wavefront synthesis, but the architectural principle is still a base station whose effective channel is altered by controlled hardware reconfiguration (Yang et al., 2016).

2. Architectural variants

Several hardware realizations fall under the RIBS umbrella. They differ primarily in where the intelligent surface is placed relative to the active array, whether the surface is reflective or transmissive, whether the BS-side electromagnetic interaction is near-field or far-field, and whether reconfigurability is purely electronic or also mechanical.

Variant	Configuration	Reported properties
Backside-IS	Reflective panel behind the BS array	Good surface scalability; moderate loss; moderate performance
Frontside-IS	Transmissive metasurface in front of the BS array	High within single-hop designs; sensitive to transmissive loss and alignment
Surrounding-IS	Four reflective panels on the sides of the radome	Single and double reflections; very high under equal total aperture
Array-plus-RIS RIBS	Non-large active array at short distance from a RIS	Near-field multi-rank coupling; fewer RF chains
MA-TRIS RIBS	Single movable antenna behind a transmissive RIS	Near-field joint optimization; 1–2-bit phase shifters
Continuous-surface RIBS	BS tightly integrated with a large, possibly continuous surface	Optimal continuous phase profile; upper bound for discrete RIS

The three IS-integrated BS architectures formalized in the 6G survey are backside-IS, frontside-IS, and surrounding-IS. Backside-IS is reflect-array-like, mounted behind the BS array, and offers good surface scalability but suffers moderate extra propagation loss because of BS–IS distance. Frontside-IS is EM-lens-like and transmissive; it can achieve high performance within single-hop designs, but transmissive loss and alignment are more critical. Surrounding-IS deploys multiple reflective panels inside the radome, perpendicular to the BS array, thereby creating both single- and double-reflection paths; under equal total aperture, it attains the highest sum-rate in the reported numerical comparisons because short inter-panel distances strengthen effective passive combining (Huang et al., 2024).

A compact near-field transmissive realization is the movable-antenna TRIS design. It combines a single active movable antenna with a transmissive RIS of $N=N_x\times N_y$ nearly passive elements, places the antenna and the TRIS in each other’s near field, and jointly optimizes antenna position and quantized transmissive phases. At $20$ GHz with $d=\lambda/2$, a $16\times 16$ TRIS has edge length $120$ mm and Rayleigh distance about $1.92$ m, which is used as the near-/far-field boundary in the reported results. The physical arrangement is explicitly compact: a thin front-facing TRIS panel together with a single movable RF chain behind it (Boloori et al., 3 Nov 2025).

A limit case of surface integration is the continuous reconfigurable intelligent surface. Here the RIS does not consist of finitely many discrete elements but can implement a phase rotation continuously over the whole aperture. The reported interpretation is twofold: it is a feasible design for future metamaterials, and it is the limiting case of a discrete RIS when the number of elements in a fixed area tends to infinity. In the RIBS setting, this continuous-surface analysis provides performance-optimal phase laws, geometric scaling relations, outage approximations, and an upper bound for dense discrete implementations (Inwood et al., 17 Jun 2025).

3. Electromagnetic and channel models

The standard RIS-integrated BS signal model is cascaded. For a single-reflection architecture with BS beamformer $w$ and diagonal surface response $\alpha(t)$0, the received signal is

$\alpha(t)$1

where $\alpha(t)$2 is the direct BS–UE channel, $\alpha(t)$3 is the BS–IS channel, and $\alpha(t)$4 is the IS–UE channel. In surrounding-IS architectures, additional double-reflection terms appear:

$\alpha(t)$5

These models also distinguish propagation regimes: user–BS and user–IS links are treated as far-field uniform plane waves, whereas BS–IS and inter-IS links are near-field and require uniform spherical-wave modeling because the arrays are separated by only several to tens of wavelengths (Huang et al., 2024).

In the array-plus-RIS RIBS model, the BS–RIS channel matrix $\alpha(t)$6 is deterministic and explicitly spherical-wave, with per-element amplitude decay proportional to $\alpha(t)$7 and phase proportional to $\alpha(t)$8. A key modeling assumption is “element-wise far-field, surface-wise near-field”: each active antenna element and RIS element are in mutual far-field, but the whole active array and the whole RIS are not required to be in mutual far-field. This condition is crucial because it allows $\alpha(t)$9 to have rank greater than one, which in turn gives a small active array nontrivial spatial multiplexing capability through the RIS (Interdonato et al., 2022).

For continuous-surface RIBS, the single-user uplink model is

$\beta(t)$0

where the RIS–BS link is LoS, the UE–RIS and UE–BS links are correlated Rayleigh fading, and the RIS reflection coefficient is a continuous function $\beta(t)$1. The SNR-maximizing phase law is

$\beta(t)$2

so the surface performs local cancellation of the UE–RIS phase together with a single global offset aligning the reflected field to the BS steering vector (Inwood et al., 17 Jun 2025).

For MA–TRIS RIBS, the model is explicitly near-field and geometry-driven. If the movable antenna is at position $\beta(t)$3 and the TRIS element phases are $\beta(t)$4, the instantaneous SNR is

$\beta(t)$5

When each phase equals the ideal propagation-compensating value $\beta(t)$6, the upper bound becomes

$\beta(t)$7

This model emphasizes that the controllable object in a near-field RIBS is not only a phase vector but also the active source position relative to the aperture (Boloori et al., 3 Nov 2025).

The reconfigurable-antenna full-duplex model is structurally different. The downlink user $\beta(t)$8 receives

$\beta(t)$9

while the BS receives

$N=N_x\times N_y$0

Here reconfigurability enters through the selected antenna modes rather than a metasurface state matrix, but the role of hardware control is analogous: the BS engineers the effective interference geometry by changing its radiation and reception patterns over time (Yang et al., 2016).

4. Control, estimation, and optimization

Joint active–passive optimization is the central algorithmic problem in RIBS. In the IS-integrated BS formulation, the canonical objective is

$N=N_x\times N_y$1

with extensions to sum-rate and fairness in multiuser systems. Alternating optimization is the dominant approach: optimize the BS beamformer for fixed surface state, then update the surface for fixed beamformer. For surrounding-IS, element-wise successive refinement is particularly relevant because double-reflection terms create strong inter-panel coupling. When explicit CSI is unavailable or too costly, codebook-based control is used; the reported survey contrasts DFT codebooks, random or IRPA designs, and a robust sector codebook that maximizes average effective channel power over angular sectors and reduces training overhead (Huang et al., 2024).

In array-plus-RIS RIBS, channel estimation is nontrivial because typically $N=N_x\times N_y$2. The reported procedure uses $N=N_x\times N_y$3 RIS configurations during pilot transmission, stacks the resulting observations, and performs a truncated SVD of the effective sensing matrix $N=N_x\times N_y$4 because its singular values decay rapidly. The dominant $N=N_x\times N_y$5-dimensional subspace is then estimated by reduced-dimension LMMSE. Downlink transmission uses MRT based on the estimated composite channel, while RIS configuration is optimized by minimizing the pairwise cross-correlation objective

$N=N_x\times N_y$6

Passive RIS optimization is handled by alternating minimization over diagonal entries under $N=N_x\times N_y$7; active RIS optimization is handled by projected gradient descent on the unit sphere together with a power split $N=N_x\times N_y$8 between array transmission and RIS amplification, followed by max–min SINR power control solved by bisection over linear feasibility constraints (Interdonato et al., 2022).

The continuous-surface RIBS model produces an unusually simple control law. The optimal phase profile has a decomposed structure consisting of a spatially varying cancellation term $N=N_x\times N_y$9 plus a single global offset $20$0. This yields a low-dimensional control interpretation: one common offset aligns the reflected field with the BS steering vector, while the rest of the surface locally removes the incoming phase. The associated optimal SNR is

$20$1

with

$20$2

Mean SNR, a Jensen upper bound on mean spectral efficiency, a Gamma approximation of outage, and the coefficient of variation for channel hardening are all derived in closed or approximate form from the first two moments of this quantity (Inwood et al., 17 Jun 2025).

MA–TRIS RIBS replaces functional optimization over a continuous aperture by search over a discrete set of candidate antenna positions. The reported alternating optimization controller samples the movement region into $20$3 positions, computes the ideal continuous phase at each position, quantizes it to the $20$4-bit set

$20$5

evaluates the SNR, and returns the maximizing position and quantized phase profile. Its complexity is $20$6. A central insight is that motion compensates low-resolution phase quantization: by changing $20$7 across the aperture, motion changes the ideal phase pattern and can reduce the residual quantization errors $20$8 for many elements simultaneously (Boloori et al., 3 Nov 2025).

The full-duplex reconfigurable-antenna line uses a different control mechanism, based on deterministic mode schedules and IDFT-based precoding. Under no CSIT, receive-mode cycling at the BS and universal uplink precoding align user-to-user interference into a one-dimensional subspace at each downlink user, while the BS collects linearly independent uplink observations across time. Under partial CSIT, transmit- and receive-mode cycling are combined with zero-forcing in orthogonal IDFT subspaces. The essential control problem is the scheduling of $20$9 and $d=\lambda/2$0 rather than the tuning of a surface matrix (Yang et al., 2016).

5. Performance characteristics and scaling laws

The reported gains of RIBS depend strongly on architecture and propagation regime, but several scaling laws recur. In the full-duplex reconfigurable-antenna setting, the sum degrees of freedom under no CSIT is

$d=\lambda/2$1

with $d=\lambda/2$2, and under partial CSIT the sum DoF equals $d=\lambda/2$3 when $d=\lambda/2$4, $d=\lambda/2$5, $d=\lambda/2$6, and $d=\lambda/2$7. The symmetric example $d=\lambda/2$8 yields $d=\lambda/2$9 DoF without CSIT and $16\times 16$0 DoF with partial CSIT, compared with $16\times 16$1 for the half-duplex baseline. Finite-SNR simulations with residual self-interference modeled as additional noise still show sum-rate gains over half-duplex operation (Yang et al., 2016).

In the array-plus-RIS formulation, the headline quantitative result is that a RIBS with $16\times 16$2 active antennas and $16\times 16$3 RIS elements can outperform conventional non-RIS MIMO systems with substantially more active antennas. For $16\times 16$4, passive RIBS outperforms a legacy system unless the latter has at least $16\times 16$5 active antennas, while active RIBS with directional active antennas and $16\times 16$6 outperforms legacy MIMO with $16\times 16$7 antennas. For $16\times 16$8, the legacy system requires about $16\times 16$9 active antennas to surpass passive RIBS and about $120$0 to surpass active RIBS. The same study reports that $120$1-bit phase quantization causes non-negligible degradation, whereas $120$2 bits causes only small degradation (Interdonato et al., 2022).

For continuous-surface RIBS, the dominant RIS term in the SNR is $120$3, and the second moment of $120$4 scales with surface area $120$5. The reported interpretation is that linear SNR grows approximately quadratically with area because $120$6 dominates for large apertures. Shape also matters: elongated apertures reduce average point-to-point correlation across the surface and therefore improve hardening and narrow the outage CDF. In the numerical results with $120$7 and a $120$8 GHz carrier, the Jensen bound $120$9 is tight for moderate or low correlation, with less than $1.92$0 error at $1.92$1, but its error rises to $1.92$2–$1.92$3 at $1.92$4. The channel-hardening metric $1.92$5 decreases as correlation falls, RIS area increases, and RIS–BS distance decreases (Inwood et al., 17 Jun 2025).

For MA–TRIS RIBS, the reported SNR scaling is consistent with coherent aperture addition. With continuous phases, SNR increases from $1.92$6 dB at $1.92$7 to $1.92$8 dB at $1.92$9; with $w$0-bit phases and MA optimization it increases from $w$1 dB to $w$2 dB; with $w$3-bit phases and MA optimization it increases from $w$4 dB to $w$5 dB. At $w$6, mobility provides gains of $w$7 dB for $w$8-bit control and $w$9 dB for $\alpha(t)$00-bit control relative to a fixed antenna. The gap to the continuous upper bound at $\alpha(t)$01 is about $\alpha(t)$02 dB for $\alpha(t)$03-bit optimization and about $\alpha(t)$04 dB for $\alpha(t)$05-bit optimization. The average relative performance with MA optimization is about $\alpha(t)$06–$\alpha(t)$07 of the continuous SNR for $\alpha(t)$08-bit TRIS, depending on aperture size. SNR also decreases monotonically as MA–TRIS separation grows from the near field toward the far field; for $\alpha(t)$09, it drops from about $\alpha(t)$10 dB at $\alpha(t)$11 m to about $\alpha(t)$12 dB at $\alpha(t)$13 m in the continuous case (Boloori et al., 3 Nov 2025).

At the system level, the IS-integrated BS survey reports that all three architectures outperform a conventional BS without IS, and that surrounding-IS achieves the highest sum-rate under equal total aperture because it exploits double-reflection channels, reduced propagation and reflection loss, and stronger passive beamforming gain. In the same line of work, a robust sector codebook outperforms DFT and random or IRPA codebooks across architectures while using far fewer training pilots. Separately, site-specific evaluation by ray tracing shows that optimized RIS configurations outperform random RIS settings with median sum spectral-efficiency gains above $\alpha(t)$14, and that the SIONNA-based ray-tracing model predicts substantially higher performance than the statistical 3GPP-like model in the evaluated urban scenario, roughly doubling the median sum SE when the RIS is optimally configured. The reported interpretation is that ray tracing captures site-specific constructive multipath that the statistical model misses, though the same work explicitly notes that empirical validation is still required (Huang et al., 2024, Beyraghi et al., 11 Jul 2025).

6. Practical constraints, misconceptions, and research directions

A frequent misconception is that RIBS refers to a single hardware topology. The literature instead uses the term for several related architectures: reflective backside panels, transmissive frontside panels, surrounding reflective panels, a small active array placed near a RIS, a movable antenna behind a TRIS, continuous or near-continuous surfaces, and even, in the earlier full-duplex line, reconfigurable transmit and receive antenna modes without a metasurface. What unifies these forms is not the physical embodiment but the fact that the BS-side electromagnetic interface is itself a controllable design variable (Huang et al., 2024, Boloori et al., 3 Nov 2025, Yang et al., 2016).

A second misconception is that enlarging the intelligent surface automatically yields massive-MIMO-like hardening in every RIBS. In the array-plus-RIS model, the favorable-propagation and channel-hardening metrics satisfy

$\alpha(t)$15

and are stated not to improve simply by increasing $\alpha(t)$16, the number of RIS elements. By contrast, the continuous-surface single-user uplink model reports decreasing $\alpha(t)$17 with larger area and lower correlation. This suggests that hardening behavior is architecture- and metric-dependent rather than universal (Interdonato et al., 2022, Inwood et al., 17 Jun 2025).

A third misconception is that RIBS eliminates the need for estimation, calibration, or active control. The survey and system papers repeatedly identify CSI acquisition, mutual coupling, finite phase resolution, dynamic noise in active RIS, polarization mismatch, switching speed, controller signaling, and calibration overhead as central design issues. The continuous-surface results assume LoS from RIS to BS and correlated Rayleigh fading elsewhere; the MA–TRIS design assumes accurate geometry or localization; the active-RIS sum-SE optimization and the ray-tracing study both assume perfect CSI; and the array-plus-RIS formulation relies on repeated training over $\alpha(t)$18 configurations plus dominant-subspace estimation (Huang et al., 2024, Interdonato et al., 2022, Beyraghi et al., 11 Jul 2025).

The current literature also leaves several boundaries explicit. Wideband beam squint, multiuser precoding for MA–TRIS, near-field calibration pipelines, active intelligent surfaces with noise amplification, beyond-diagonal intelligent surfaces, IS-integrated terminals, hybrid BS/terminal deployments, ISAC, distributed MA/TRIS, and hybrid transmissive–reflective panels are all identified as open directions. The ray-tracing study further emphasizes that the observed advantage of site-specific modeling over a statistical channel model is still a prediction rather than a validated deployment rule. A plausible implication is that future RIBS research will be driven as much by measurement methodology, calibration, and hardware-aware control as by idealized optimization (Huang et al., 2024, Boloori et al., 3 Nov 2025, Beyraghi et al., 11 Jul 2025).

In aggregate, RIBS represents a shift in base-station design from active-array-only architectures toward jointly controllable electromagnetic front ends. Depending on the realization, the reconfigurable object may be a reflective or transmissive metasurface, a continuous aperture, a movable source behind a transmissive panel, or a bank of preset antenna modes. Across these variants, the recurring technical themes are near-field BS–surface coupling, joint active–passive optimization, geometry-aware control, and a trade-off between aperture gain, RF-chain count, estimation overhead, and hardware realism.