Distributed RIMSA: Intelligent Metasurface Antennas

• Distributed RIMSA are advanced metasurface antenna systems featuring programmable electromagnetic apertures that function directly as radiating, receiving, or hybrid elements in integrated transceiver chains.
• They exploit distributed multi-panel cooperation and joint analog-digital optimization to achieve high spectral efficiency, improved coverage, and flexible integrated sensing and communication capabilities.
• The architectures, including DMA, RDARS, and transmissive RMS, emphasize hardware calibration, synchronization, and scalable control, addressing practical challenges for 6G network deployments.

Searching arXiv for papers on RIMSA, RDARS, dynamic metasurface antennas, and related distributed metasurface antenna systems. Distributed Reconfigurable Intelligent Metasurface Antennas (RIMSA) denote a class of metasurface-based antenna systems in which programmable electromagnetic apertures function not only as environment-shaping surfaces but also as radiating, receiving, or hybrid antenna front ends, and in distributed deployments multiple such apertures cooperate across space for communication, sensing, or integrated sensing and communication (ISAC). In the literature synthesized here, the concept appears through several closely related realizations: active waveguide-fed dynamic metasurface antennas (DMAs) for extreme massive MIMO, radiating RIMSA arrays with parallel feeding and unit-modulus phase control, transmissive reconfigurable metasurfaces used as low-RF-chain multi-antenna systems, and RDARS architectures that switch elements between passive reflection mode and BS-connected antenna mode (Shlezinger et al., 2020, Wei et al., 23 Jun 2025, Li et al., 2021, Ma et al., 2023, Wang et al., 22 Apr 2026). Taken together, these works establish distributed RIMSA as a networked metasurface-antenna paradigm that combines aperture-scale electromagnetic programmability, low-power element control, and multi-panel coordination to pursue high spectral efficiency, coverage enhancement, and flexible sensing in 6G-class systems (Hodge et al., 2023, Wang et al., 2023).

1. Conceptual scope and taxonomy

The central taxonomic distinction in this area is between a passive reconfigurable intelligent surface (RIS) and a metasurface antenna. A passive RIS is treated as a nearly passive reflector whose local impedance profile reshapes impinging waves without implementing amplification or baseband processing, whereas a DMA is an active, waveguide-fed metasurface functioning directly as a transmit/receive antenna array connected to RF chains and performing analog signal processing through its guided-wave architecture and tunable metamaterial elements (Shlezinger et al., 2020). RIMSA, in the stricter sense used in later work, is a radiating metasurface aperture with reconfigurable element responses and explicit integration into the transceiver chain rather than an external environmental relay (Wei et al., 23 Jun 2025).

Several related architectures instantiate this broader concept.

Architecture	Defining characteristic	Source
DMA	Waveguide-fed radiating metasurface with analog combining/precoding	(Shlezinger et al., 2020)
RIMSA array	Parallel-fed radiating metasurface used as BS/user antenna	(Wei et al., 23 Jun 2025)
RDARS	Elements switch between reflection mode and connection mode	(Ma et al., 2023, Wang et al., 22 Apr 2026)
Transmissive RMS	Single-feed transmissive metasurface multi-antenna system	(Li et al., 2021)
RIBS	Small active array placed near RIS, forming reconfigurable base station	(Interdonato et al., 2022)

Distributed RIMSA extends these single-panel concepts to multi-aperture deployments. In the DMA formulation, multiple panels are distributed across walls, façades, or ceilings and connected to a central processing unit in a cell-free arrangement, each panel contributing few RF chains but many controllable elements (Shlezinger et al., 2020). In RIS-based index-modulated transceivers, the natural distributed extension is written explicitly as

$$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\sum_{m=1}^{M}\mathbf{H}_{r,m}\,\boldsymbol{\Theta}_m\,\mathbf{H}_{t,m}\,\mathbf{x}+\mathbf{n},$$

with per-panel reflection matrices $\boldsymbol{\Theta}_m$ and panel-specific Tx–RIS and RIS–Rx channels (Hodge et al., 2023). RDARS interprets the same distributed idea through hybrid mode synergy: some elements operate as passive reflectors while a small subset operate as fronthaul-connected distributed antennas, yielding reflection gain, distribution gain, and selection gain in a unified system model (Wang et al., 22 Apr 2026).

A recurring misconception in this literature is to treat all programmable surfaces as RIS. The cited works consistently separate at least three categories: purely passive RIS, metasurface antennas that radiate or receive directly, and hybrid architectures such as RDARS that combine passive wave control with active antenna functionality (Shlezinger et al., 2020, Wang et al., 2023, Wang et al., 22 Apr 2026). Another misconception is that “distributed” merely means multiple passive panels. In the RIMSA context, distribution can also refer to networked radiating metasurface antennas, distributed receive-only RIMSA sensors, or hybrid panels with both connected and reflective states (Wang et al., 7 Aug 2025, Ma et al., 2023).

2. Electromagnetic and signal models

At baseband, many formulations reduce to an effective channel whose structure depends on whether the metasurface is purely reflective, directly radiating, or hybrid. For a single RIS-aided link, one representative narrowband model is

$$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$

with $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$ collecting the metasurface element coefficients (Hodge et al., 2023). In distributed form, the single-RIS term becomes a sum over multiple panels, and coherent gain requires synchronization of timing, frequency, and phase references across panels (Hodge et al., 2023, Gu et al., 2022).

The electromagnetic control variable is commonly expressed through a reflection or transmission coefficient. In the electromagnetics-compliant reflect-array design for RIS-based index modulation, each meta-atom obeys

$$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$

where the tunable surface admittance $Y_s(t)$ is controlled through varactor bias and associated capacitance $C(V)$ and loss resistance $R(V)$ (Hodge et al., 2023). In transmissive and radiating metasurface formulations, analogous per-element complex coefficients $t_m=a_m e^{j\theta_m}$ or block-diagonal RIMSA phase matrices are used, with passivity or unit-modulus constraints imposed by hardware (Li et al., 2021, Wei et al., 23 Jun 2025).

For radiating metasurface antennas, the aperture field and beam pattern are explicit objects of the model. In the RIS-based reflect-array transceiver, the far-field pattern is written as

$\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$

with generalized reflection-law steering angle

$\boldsymbol{\Theta}_m$0

and practical spacing condition $\boldsymbol{\Theta}_m$1 to control grating lobes and scan loss (Hodge et al., 2023). The system-level RIS model in a LOS-dominant multi-cell setting further embeds realistic element field patterns $\boldsymbol{\Theta}_m$2 and $\boldsymbol{\Theta}_m$3, polarization handling, and angle-dependent far-field phases, emphasizing that gains depend on geometry rather than isotropic reflection assumptions (Gu et al., 2022).

DMA-based RIMSA introduces an effective analog combining matrix $\boldsymbol{\Theta}_m$4 such that the uplink effective channel becomes $\boldsymbol{\Theta}_m$5, while in downlink an analog precoder $\boldsymbol{\Theta}_m$6 is realized by the metasurface element tunings and feed phases (Shlezinger et al., 2020). In the tri-hybrid MIMO formulation, the metasurface layer is formalized as an electromagnetic precoder $\boldsymbol{\Theta}_m$7 inserted above digital and analog precoding, leading to the generic OFDM model

$\boldsymbol{\Theta}_m$8

with an architecture-dependent radiated-power constraint $\boldsymbol{\Theta}_m$9, which couples the channel and the power budget through the electromagnetic configuration (Deshpande et al., 26 Apr 2026). This suggests that in RIMSA, unlike conventional hybrid beamforming, reconfiguration reshapes not only the effective channel but also the feasible power region.

3. Architectures and hardware realizations

Waveguide-fed DMAs represent one major hardware line. They consist of multiple 1D waveguides, each connected to one RF port, with sub-wavelength-spaced metamaterial elements along the guide. Each element exhibits a tunable Lorentzian frequency response and may be made radiating or non-radiating using PIN diodes, or finely tuned using varactors (Shlezinger et al., 2020). The feed-wave phase accumulation and element responses jointly realize analog combining and precoding without corporate phase-shifter networks, thereby reducing power and cost relative to conventional phased arrays (Shlezinger et al., 2020).

Parallel-fed RIMSA arrays represent a distinct line. In the multi-user downlink RIMSA array, the base station employs $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$0 RF chains, each feeding one RIMSA aperture of $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$1 elements through a parallel coaxial feeding network, giving $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$2 total elements and a block-diagonal phase matrix

$$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$3

with unit-modulus constraints on each metasurface element (Wei et al., 23 Jun 2025). The same paper models user-side RIMSAs as receive metasurface combiners, including MU-MIMO settings in which each user has multiple RIMSA/RF links and block-diagonal analog combining matrices (Wei et al., 23 Jun 2025). A stronger hardware claim appears in the later LLM-controlled RIMSA framework, which states that each metamaterial element supports joint amplitude-phase modulation through a parallel coaxial/microstrip feed and local tuning circuits (Huang et al., 18 Aug 2025). Because that work postdates the unit-modulus downlink array paper and concerns a different architecture, the literature does not support treating independent amplitude-and-phase control as universal across all RIMSAs.

RDARS provides a hybrid element-level realization. In the original architecture, each of $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$4 panel elements can be dynamically switched between passive reflection mode and connected mode, in which the element acts as a remote distributed antenna connected to the BS through fronthaul (Ma et al., 2023). The more recent signal-processing treatment formalizes this through a diagonal mode-selection matrix $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$5, a reflection matrix $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$6, and stacked uplink observation

$$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$7

which separates passive reflected and connected-element contributions (Wang et al., 22 Apr 2026). The fabricated RDARS prototypes use $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$8 or 256-element panels, 2-bit reflection-mode phase control, and controller updates through UDP-delivered control words (Wang et al., 2023, Ma et al., 2023).

Several papers contribute concrete unit-cell data. The electromagnetics-compliant reflect-array metasurface at $$\mathbf{y}=\mathbf{H}_{\mathrm{LoS}}\mathbf{x}+\mathbf{H}_{\mathrm{rx,RIS}}\,\boldsymbol{\Theta}\,\mathbf{H}_{\mathrm{tx,RIS}}\mathbf{x}+\mathbf{n},$$9 GHz uses RT/Duroid 5880, a via-less varactor-loaded unit cell, a tunable reflection phase range of $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$0 deg at $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$1 GHz, and reflection amplitude loss $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$2 dB across tuning states (Hodge et al., 2023). The dual-functional hybrid RIS for sensing and reflection uses a $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$3 mm ($$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$4) unit cell centered around $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$5 GHz, with a high-$$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$6 sensing disc and a low-$$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$7 reflective ring sharing the same phase center, SP4T-based load tuning, and two interleaved sensing arrays with orthogonal polarization and quarter-wavelength offset (Birari et al., 23 Jan 2025). These designs are not identical architectures, but together they show that distributed RIMSA hardware spans mmWave reflect-arrays, sub-6 GHz hybrid sensing-reflection tiles, and waveguide-fed active apertures.

4. Programmability, optimization, and control

The dominant algorithmic theme is joint optimization of digital transceiver variables and metasurface states under unit-modulus, quantization, and power constraints. In MU-MISO RIMSA downlink, the sum-rate maximization problem jointly optimizes the digital precoder $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$8, BS-side RIMSA phases $$\boldsymbol{\Theta}=\mathrm{diag}(\alpha_1 e^{j\phi_1},\ldots,\alpha_N e^{j\phi_N})$$9, and user-side combiners $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$0 subject to

$$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$1

with the effective user channel $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$2 (Wei et al., 23 Jun 2025). The digital precoder is optimized by fractional programming, whereas the metasurface phase variables are updated by product manifold optimization using Riemannian gradients and retractions onto the unit-modulus manifold (Wei et al., 23 Jun 2025). In MU-MIMO, the same paper converts the sum-rate problem into weighted sum MSE minimization, yielding closed-form updates for the digital precoders and combiners and a PMO subproblem for the RIMSA configurations (Wei et al., 23 Jun 2025).

RIS-assisted index-modulated metasurface transceivers emphasize another control dimension: index selection across spatial, frequency, time-slot, and channel domains. The same actively biased reflect-array can realize M-PSK phase modulation, spatial modulation through sub-aperture partitioning, harmonic/subcarrier index modulation through time-varying $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$3, STSK/GSTSK through spatio-temporal reflection matrices, and MBM or RA-SSK through programmable radiation states (Hodge et al., 2023). A key control principle is to maximize inter-index separations $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$4 under metasurface constraints, since the pairwise error probability depends on those separations in the RIS-synthesized effective channel (Hodge et al., 2023). The distributed extension assigns spatial indices, harmonics, or activation schedules across multiple metasurfaces and requires CSI sharing, synchronization, and coordinated interference management (Hodge et al., 2023).

RDARS optimization makes the active/passive split itself a variable. In the uplink SNR-maximization problem,

$$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$5

the binary diagonal matrix $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$6 selects connection mode versus reflection mode, while $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$7 controls the passive phases (Wang et al., 22 Apr 2026). The same source lists alternating optimization, semidefinite relaxation, successive convex approximation, majorization-minimization, manifold optimization, relaxation-plus-penalty methods, branch-and-bound, greedy search, and successive rounding as solution routes (Wang et al., 22 Apr 2026). In the earlier ergodic-rate analysis of RDARS, a single-user uplink admits a closed-form optimal passive phase alignment and shows that even a small number $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$8 of connected elements can materially change the rate–cost trade-off (Ma et al., 2023).

Control architectures are correspondingly hierarchical. The FPGA-based controller in the RIS-based IM transceiver computes unit-cell phase profiles $$\Gamma(t)=\frac{Y_0-Y_s(t)}{Y_0+Y_s(t)},\quad Y_0=\frac{1}{120\pi},$$9, generates varactor bias voltages, and must meet symbol-timing synchronization constraints (Hodge et al., 2023). Distributed DMA/RIMSA deployments add a central scheduler or processing unit, panel-local analog processing, and compressed exchange of effective-channel information rather than raw element-level CSI (Shlezinger et al., 2020). The LLM-controlled RIMSA framework pushes this further by mapping pilot signals directly to metasurface configurations and precoders using a frozen GPT-2 backbone with trainable adapters, Conv1D/BiLSTM preprocessing, and spatio-temporal attention, reporting $Y_s(t)$0 M total parameters with $Y_s(t)$1 M trainable, $Y_s(t)$2 h training time, and $Y_s(t)$3 ms inference time in its tested setup (Huang et al., 18 Aug 2025). That result is architecture- and setup-specific, but it indicates an emerging trend toward learned panel control that bypasses explicit CSI estimation.

5. Communication, sensing, and anti-jamming functions

Communication performance is reported across several operating modes. The RIS-based IM transceiver shows that RIS-aided IM can outperform traditional implementations in BER, validates beam steering for a $Y_s(t)$4 metasurface at $Y_s(t)$5 and $Y_s(t)$6, and reports capacity growth with $Y_s(t)$7 from $Y_s(t)$8 to $Y_s(t)$9 under Rayleigh fading using

$C(V)$0

(Hodge et al., 2023). The DMA article reports an uplink sum-rate case study at $C(V)$1 GHz with $C(V)$2 elements, $C(V)$3 RF chains, $C(V)$4 elements per microstrip, and $C(V)$5 users within a $C(V)$6 m cell, where the DMA architecture comes closer to the $C(V)$7 sum-capacity than a conventional fully connected hybrid array with the same $C(V)$8 RF chains (Shlezinger et al., 2020). The RIMSA array downlink paper reports that its FP-PMO and WMMSE-PMO algorithms achieve significant gains over MM-Alt-Opt, AO-MO, and MO-AltMin baselines, approaching the fully digital upper bound as the number of RF chains increases (Wei et al., 23 Jun 2025).

System-level and prototype evidence also exists for distributed or hybrid deployments. In LOS-dominant multi-cell simulations with multiple RIS panels per sector, received-power gains relative to no-RIS range from approximately $C(V)$9 dB for $R(V)$0 panels of $R(V)$1 elements to approximately $R(V)$2 dB for $R(V)$3 panels of $R(V)$4 elements, while SINR gains range from approximately $R(V)$5 dB to approximately $R(V)$6 dB depending on panel count and aperture size (Gu et al., 2022). The RDARS prototype with $R(V)$7 elements reports throughput improvement over DAS and RIS-aided systems: in one measured scenario, RDARS with $R(V)$8 connected element achieves $R(V)$9 Mbps versus $t_m=a_m e^{j\theta_m}$0 Mbps for DAS and $t_m=a_m e^{j\theta_m}$1 Mbps for RIS, and in a blocked scenario $t_m=a_m e^{j\theta_m}$2 Mbps versus $t_m=a_m e^{j\theta_m}$3 Mbps and $t_m=a_m e^{j\theta_m}$4 Mbps, corresponding to additional $t_m=a_m e^{j\theta_m}$5 and $t_m=a_m e^{j\theta_m}$6 throughput improvement over DAS and RIS in the paper’s summary statement (Ma et al., 2023).

Sensing and ISAC form a second major function class. The RDARS ISAC prototype uses reflection-mode elements to assist uplink communication and connected-mode elements to collect RSSI measurements for localization, achieving reliable user localization without compromising the communication rate in a $t_m=a_m e^{j\theta_m}$7–$t_m=a_m e^{j\theta_m}$8 GHz, $t_m=a_m e^{j\theta_m}$9 GHz nominal, $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$0 MHz LTE-based setup with $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$1 elements and 2-bit phase shifters (Wang et al., 2023). The dual-functional hybrid RIS embeds two interleaved sensing arrays inside the reflective aperture so that channel parameters toward two end nodes can be sensed while the same panel provides reconfigurable reflections (Birari et al., 23 Jan 2025). Distributed anti-jamming sensing with RIMSA receivers goes further: multiple RIMSA Rx panels are deployed around a target region, each panel forms a single RF chain, MRC is applied digitally, and beamforming patterns are selected by deep reinforcement learning using a combined cross-entropy and $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$2 loss (Wang et al., 7 Aug 2025).

The anti-jamming results are especially explicit. Averaged over $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$3 trials, the reported post-processing SINR values for the proposed distributed RIMSA sensing scheme are $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$4, $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$5, and $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$6 dB at jammer powers $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$7, $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$8, and $\begin{aligned} f(\theta,\phi,t)&=\sum_{p=1}^{M}\sum_{q=1}^{N}E_{pq}(\theta,\phi)\,\Gamma_{pq}(t)\[-2pt] &\times \exp\!\Big\{j\frac{2\pi}{\lambda_c}\big[(p{-}1)d_x\sin\theta\cos\phi+(q{-}1)d_y\sin\theta\sin\phi\big]\Big\}, \end{aligned}$9 mW, compared with $\boldsymbol{\Theta}_m$00, $\boldsymbol{\Theta}_m$01, and $\boldsymbol{\Theta}_m$02 dB for ZF and $\boldsymbol{\Theta}_m$03, $\boldsymbol{\Theta}_m$04, and $\boldsymbol{\Theta}_m$05 dB with no defense (Wang et al., 7 Aug 2025). The same study states that distributed implementation overcomes environmental influences better than centralized implementation and maintains high-accuracy sensing under jamming attack (Wang et al., 7 Aug 2025). This suggests that spatial diversity and receiver-side programmable nulling are among the most distinctive advantages of distributed RIMSA compared with monolithic metasurface sensing.

6. Practical constraints, misconceptions, and research directions

The literature repeatedly emphasizes that reported gains are conditional on hardware, calibration, and synchronization assumptions. Several system-level analyses assume far-field LOS dominance and neglect mutual coupling, so densification results are upper bounds when element spacing is reduced (Gu et al., 2022). DMA and metasurface-antenna models explicitly note amplitude-phase coupling, waveguide or feed losses, nonlinear regimes at high RF amplitudes, and the need for coupling-aware calibration of the analog combining matrix $\boldsymbol{\Theta}_m$06 (Shlezinger et al., 2020). The RIS-based IM transceiver highlights bias-network parasitics, finite DAC resolution, driver jitter, calibration for manufacturing tolerances, and CSI acquisition overhead, especially when maximizing $\boldsymbol{\Theta}_m$07 by precise element-wise phase control (Hodge et al., 2023).

A second practical issue is synchronization across distributed apertures. Coherent multi-panel combining requires symbol-level timing alignment, frequency synchronization, and phase calibration across metasurfaces or RIMSA panels (Hodge et al., 2023). The system-level RIS model explicitly notes that full multi-panel coherent gains require tight synchronization and joint control of $\boldsymbol{\Theta}_m$08; otherwise, panel contributions add only partially coherently or non-coherently (Gu et al., 2022). RDARS softens this somewhat because connected elements and fronthaul facilitate centralized processing and synchronization, but multi-panel cell-free RDARS/RIMSA still poses a design challenge (Wang et al., 22 Apr 2026).

A third issue is that “independent amplitude-and-phase control” is not generic. Some architectures are strictly unit-modulus or phase-only at the optimization level, including the multi-user downlink RIMSA array (Wei et al., 23 Jun 2025). Others report amplitude-phase coupling through Lorentzian or impedance constraints, as in DMAs, parasitic arrays, and transmissive metasurfaces (Shlezinger et al., 2020, Li et al., 2021, Deshpande et al., 26 Apr 2026). Still others, such as the LLM-RIMSA hardware proposal, claim independent amplitude-and-phase control via parallel feeding and element tuning circuits (Huang et al., 18 Aug 2025). The appropriate interpretation is therefore architecture-specific rather than universal.

Open problems are comparatively consistent across sources. Wideband and OFDM operation remain difficult because metasurface responses are frequency selective and panel states may be effectively frequency flat over the coherence block (Shlezinger et al., 2020, Deshpande et al., 26 Apr 2026). Scalable channel estimation and control signaling are central bottlenecks in large distributed networks, motivating hierarchical control, sparse or codebook-based training, environment-aware codebooks, blind beamforming, and learning-based adaptation (Shlezinger et al., 2020, An et al., 2024). Near-field focusing, holographic MIMO, and spatially wideband regimes are increasingly important as apertures grow and panels move closer to users or targets (Interdonato et al., 2022, Wang et al., 22 Apr 2026). Hardware-aware optimization must also absorb finite phase resolution, insertion loss, mutual coupling, and thermal or exposure constraints (Hodge et al., 2023, Deshpande et al., 26 Apr 2026).

The most plausible synthesis of these trends is that distributed RIMSA is not a single hardware standard but an umbrella design space. Within that space, the common denominator is a distributed set of programmable metasurface apertures that participate directly in radiation, reception, or hybrid active-passive operation, with network-level coordination across digital, analog, and electromagnetic domains. The cited work indicates that this design space already spans cell-free massive MIMO, index-modulated transceivers, ISAC prototypes, hybrid distributed antennas and reflecting surfaces, and anti-jamming sensing systems, while still leaving fundamental questions of scalable calibration, synchronization, wideband modeling, and standardization unresolved (Shlezinger et al., 2020, Hodge et al., 2023, Wang et al., 2023, Wang et al., 22 Apr 2026).