---
title: 'MF-RIS: Multi-Functional Intelligent Surfaces'
url: https://www.emergentmind.com/topics/multi-functional-reconfigurable-intelligent-surfaces-mf-ris
type: topic
---

# MF-RIS: Multi-Functional Intelligent Surfaces

Multi-Functional Reconfigurable Intelligent Surfaces (MF-RIS) are reconfigurable metasurfaces that extend conventional single-functional RIS by supporting multiple electromagnetic and system functions on the same surface. Across the recent literature, the most common definition is simultaneous signal reflection, refraction or transmission, and amplification at the element level, with several works further incorporating energy harvesting, sensing, or security-oriented jamming. In this sense, MF-RIS unify passive reflecting RIS, STAR-RIS, and active RIS as special cases, while enlarging the design space from phase-only control to joint amplitude, phase, function allocation, and, in some architectures, mobility or sensing-aware adaptation [2405.16257][2410.06584][2301.13630].

## 1. From single-functional RIS to MF-RIS

Conventional single-functional RIS are typically passive and operate in one half-space only. Reflecting RIS control incident waves on one side, while transmissive surfaces serve the opposite side; both suffer from the double-fading attenuation of the cascaded BS–RIS–user link. Dual-functional architectures such as STAR-RIS remove the half-space restriction by enabling simultaneous transmission and reflection, but remain passive in the usual formulation. MF-RIS were introduced precisely to combine full-space service with active compensation of path loss through amplification [2405.16257][2410.06584].

A generic MF-RIS element is usually parameterized, for reflection and transmission sides \(p\in\{r,t\}\), by
\[
u_m^p=\sqrt{\beta_m^p}\,e^{j\theta_m^p},\quad 0\le \beta_m^p\le \beta_{\max},\quad \theta_m^p\in[0,2\pi),
\]
with the per-element split constraint
\[
\beta_m^r+\beta_m^t\le \beta_{\max},\quad \beta_{\max}\ge 1.
\]
When \(\beta_{\max}=1\), the architecture reduces to a passive transmission-reflection surface; when \(\beta_m^t=0\), it reduces to an active reflecting RIS; and when both amplification and transmission are absent, it reduces to a conventional reflecting RIS [2301.13630][2405.16257].

The term is not used in a single invariant sense. In communication-centered works, MF-RIS usually denotes simultaneous reflection, refraction, and amplification. In self-sustainable architectures, energy harvesting is added as an explicit fourth function. In secure UAV-assisted systems, the “multi-functional” aspect can instead refer to simultaneous information reflection or amplification and artificial jamming, with the overall response decomposed as
\[
\mathbf{\Theta}_n=\mathbf{\Theta}_n^R+\mathbf{\Theta}_n^J,
\]
where \(\mathbf{\Theta}_n^R\) serves legitimate users and \(\mathbf{\Theta}_n^J\) targets an eavesdropper [2310.02562][2512.19109]. A common misconception is therefore to treat MF-RIS as synonymous with a single hardware blueprint; the cited literature instead uses the term for a family of programmable surfaces whose defining property is concurrent multi-role operation.

## 2. Element architectures, impedance control, and physical realizations

A representative communication-oriented MF-RIS architecture is a three-layer structure comprising a reflective layer, an amplification layer, and a refractive layer. The reflective and refractive layers contain metallic patches, feed lines, and reconfigurable components, while the middle layer contains an amplifier and a power divider that allocate amplified power between reflection and refraction paths [2405.16257]. This produces reflection and refraction matrices of the form
\[
\Theta_r = {\rm diag}(\sqrt{\beta_1^r}e^{j\theta_1^r},\dots,\sqrt{\beta_M^r}e^{j\theta_M^r}),
\]
\[
\Theta_t = {\rm diag}(\sqrt{\beta_1^t}e^{j\theta_1^t},\dots,\sqrt{\beta_M^t}e^{j\theta_M^t}).
\]

Self-sustainable MF-RIS extend this element model by introducing harvesting and signal-relay modes. In one formulation, each element has a binary mode indicator \(\alpha_m\in\{0,1\}\), where \(\alpha_m=0\) denotes energy harvesting and \(\alpha_m=1\) denotes signal relaying; the harvested, reflected, and refracted signals are
\[
y_m^h=(1-\alpha_m)s_m,\quad
y_m^r=\alpha_m \sqrt{\beta_m^r}\, e^{j\theta_m^r} s_m,\quad
y_m^t=\alpha_m \sqrt{\beta_m^t}\, e^{j\theta_m^t} s_m.
\]
Other works relax \(\alpha_m\) to a continuous power-splitting factor, allowing each element to operate in a hybrid mode that simultaneously harvests and forwards [2310.02562][2501.11079].

At the hardware level, a foundational route to multi-functionality is continuous local impedance control. A reflective intelligent metasurface demonstrated tunable perfect absorption and tunable anomalous reflection at \(5\) GHz by embedding mixed-signal ICs in each unit cell, with IC impedance
\[
Z_{\mathrm{IC}} = R + \frac{1}{j\omega C},
\]
over the assumed ranges
\[
R \in [0, 5]~\Omega,\quad C \in [1, 5]~\text{pF}.
\]
Because both resistance and reactance are independently adjustable, the surface can move through a broad region of the complex-impedance plane rather than along a phase-only trajectory, which directly underpins amplitude-phase programmable behavior relevant to MF-RIS [1811.10082].

More recent extensions broaden the notion of reconfigurability beyond RF coefficients. A transmissive flexible intelligent metasurface allows each element to apply a phase shift and a mechanical displacement \(d_n\) along the surface normal, so that the surface shape itself becomes a design variable,
\[
\mathbf{d}=[d_1,\dots,d_N]^T,\qquad |d_n|\le d_{\max}.
\]
This does not add reflection or amplification, but it does add geometric reconfiguration as an additional degree of freedom beyond rigid metasurfaces [2510.07466].

## 3. Signal, channel, and noise models

A generic MF-RIS-aided communication model writes the effective reflected-side channel as
\[
\mathbf{h}_{\text{eff}}^H = \mathbf{h}_d^H + \mathbf{h}_r^H \mathbf{\Theta}^{\text{(ref)}} \mathbf{G},
\]
with an analogous term for transmission-side users. In the architecture of [2410.06584], a single element transforms an incident signal \(x\) into
\[
\sqrt{\alpha_r}\, g\, e^{j\theta_r} x
\]
on the reflection side and
\[
\sqrt{\alpha_t}\, g\, e^{j\theta_t} x
\]
on the refraction side, where \(\alpha_r,\alpha_t\) are power splitting coefficients and \(g\ge 1\) is the amplifier gain. At the array level, reflection and transmission responses are diagonal matrices whose entries jointly encode power splitting and amplification [2410.06584].

MF-RIS modeling departs from passive RIS in two technically important ways. First, reflection and refraction coefficients are coupled by per-element power constraints, so the two half-spaces cannot be optimized independently. Second, active circuitry introduces local thermal noise that is shaped by the same coefficients as the desired signal. In uplink channel estimation, for example, the received pilot at BS antenna \(m\) contains the term
\[
\bm{\varphi}_i^{\rm H}\mathbf{G}_m\, \mathbf{z}_i(l),
\]
where \(\mathbf{z}_i(l)\sim \mathcal{CN}(\mathbf{0}_N,\sigma_s^2 \mathbf{I}_N)\) models MF-RIS thermal noise. The equivalent covariance is
\[
\mathbf{C}_{z,m} = \sigma_s^2 \big( \mathbf{\Psi}_{m,R}\mathbf{\Psi}_{m,R}^{\rm H} + \mathbf{\Psi}_{m,T}\mathbf{\Psi}_{m,T}^{\rm H} \big) + \sigma^2 \mathbf{I}_L,
\]
which is configuration-dependent and generally non-diagonal [2501.10657].

In energy-harvesting MF-RIS, the power model is itself part of the signal model. A standard nonlinear RF-to-DC conversion law used in self-sustainable designs is
\[
P_m^{\rm A} = \frac{\Upsilon_m - Z\Omega}{1-\Omega},\qquad
\Upsilon_m = \frac{Z}{1 + e^{-a(P_m^{\rm RF}-q)}},
\]
with \(\Omega = \frac{1}{1+e^{a q}}\). The total MF-RIS output power includes both forwarded signal power and amplified thermal noise [2310.02562][2310.02564].

Secure MF-RIS variants add yet another layer to the model. In the UAV-mounted “intelligent sky mirror,” the jamming signal is generated as
\[
\mathbf{z}_n = \mathbf{\Theta}_n^{J}\mathbf{h}_n^{br}\mathbf{x}_n,
\]
and the same surface also amplifies legitimate BS signals through \(\mathbf{\Theta}_n^R\), so the MF-RIS appears simultaneously in the useful signal term, the jamming term, and the noise-amplification term [2512.19109].

## 4. Estimation, beamforming, and optimization methodologies

MF-RIS design problems are typically non-convex because they combine fractional objectives, mixed discrete-continuous variables, coupled amplitude-phase constraints, and active-noise power budgets. Channel estimation is already more involved than in passive RIS. For the uplink multi-user MF-RIS model in [2501.10657], the pilot length must satisfy
\[
L \ge N+1 \quad \text{and} \quad L \ge K,
\]
and the resulting weighted LS estimator,
\[
\hat{\mathbf{h}}^i_{m,k}
= \frac{1}{\sqrt{P_k}}
\left(\mathbf{\Theta}_{m,i}^{\rm H} \mathbf{C}_{z,m}^{-1} \mathbf{\Theta}_{m,i}\right)^{-1}
\mathbf{\Theta}_{m,i}^{\rm H}\mathbf{C}_{z,m}^{-1}\mathbf{y}_{m,k},
\]
is explicitly noise-aware. By imposing a DFT-based MF-RIS training structure, the noise covariance becomes scalar times identity, and the LS estimator achieves the CRLB under the proposed design [2501.10657].

For downlink communication design, alternating optimization is recurrent. In MF-RIS-aided NOMA, one representative approach jointly optimizes BS beamforming and MF-RIS coefficients through semidefinite relaxation, successive convex approximation, and penalty-based enforcement of rank-one structure [2301.13630]. In the self-sustainable NOMA formulation, the decision variables additionally include NOMA power allocation, operating modes, and deployment location, again solved through alternating optimization with convexified subproblems [2310.02562]. Robust energy-efficiency maximization under imperfect CSI uses the S-procedure for bounded errors and Bernstein-Type Inequality for statistical errors, then alternates between transmit beamforming and MF-RIS coefficient optimization [2412.08032].

Learning-based control appears when the environment is dynamic or the variable space is too large for repeated model-based re-optimization. In the UAV-assisted secure NOMA setting, a two-layer scheme combines SAC for trajectory, BS power allocation, amplification factors, and scheduling matrices with closed-form channel-alignment phases
\[
\theta_n^{m,g} = e^{j\angle h^{bg}_n} e^{-j\angle h^{br}_m} e^{-j\angle h^{rg}_m},
\]
so that the cascaded BS–RIS–user link is phase-aligned with the direct path [2512.19109]. In LEO and SAGIN settings, federated or hybrid multi-agent DRL methods optimize amplification, phase shifts, energy-harvesting ratios, active-element selection, and network variables such as beamforming, HAPS deployment, or user association [2501.11079][2507.16204].

A recurring system-level idea is two-timescale design. Short-term BS precoding follows instantaneous CSI, while MF-RIS coefficients are updated on a slower timescale using statistical CSI, which directly addresses the pilot and control burden created by large active surfaces [2405.16257].

## 5. Communication, sensing, security, and space-network applications

MF-RIS were first developed primarily for communication enhancement under two persistent RIS limitations: half-space coverage and double-fading attenuation. In downlink NOMA, MF-RIS jointly reflecting, transmitting, and amplifying can enlarge channel disparities for SIC, support users on both sides of the surface, and increase sum-rate beyond reflecting-only, active-only, or passive STAR-type baselines [2301.13630][2310.02562]. More broadly, the 6G-oriented MF-RIS literature explicitly positions them as enablers for unicast, multicast, and broadcast, and as channel shapers for OFDMA, NOMA, and RSMA [2410.06584].

Sensing and ISAC constitute a second major application class. MF-RIS are used to create virtual LoS sensing links, amplify echoes, form beams simultaneously in reflection and refraction spaces, and support radar-communication coexistence or dual-functional radar-communication architectures. In the surveyed DFRC example, MF-RIS outperform active RIS, STAR-RIS, and passive RIS in sum sensing SINR for a given communication QoS, with the reported ordering
\[
\text{MF-RIS} > \text{active RIS} > \text{STAR-RIS} > \text{passive RIS}
\]
as the number of elements increases [2410.06584].

Security-oriented MF-RIS exploit the same programmability for constructive and destructive propagation shaping. In the low-altitude economy scenario, the UAV-mounted MF-RIS amplifies legitimate signals and simultaneously generates artificial jamming toward an eavesdropper, with trajectory, scheduling, amplification, and phase control jointly optimized for secure energy efficiency [2512.19109]. This makes MF-RIS relevant not only to capacity enhancement but also to physical-layer security.

Space and integrated networks provide a third regime in which MF-RIS functions are strongly coupled to energy availability. In LEO networks, MF-RIS can reflect, refract, amplify, and harvest energy from RF signals, with each element controlled by phase, amplitude, and harvesting ratio. In shadow regions, harvested RF energy partially offsets the lack of solar power and supports long-term energy-efficiency optimization [2501.11079]. In space-air-ground integrated networks, MF-RIS are mounted on LEO satellites, HAPS, and BSs, with additional on/off element selection and joint optimization of communication and computing energy [2507.16204].

Deployment rules are not universal across these architectures. In self-sustainable harvesting-based MF-RIS, published results state that the surface should be deployed closer to the transmitter to maximize harvested energy and communication throughput [2310.02562][2310.02564]. In fixed amplification-budget NOMA systems, by contrast, the MF-RIS should preferably be deployed close to users [2301.13630]. In UAV-secure NOMA, the learned trajectory tends to place the aerial MF-RIS between legitimate users and the eavesdropper [2512.19109]. This suggests that placement is architecture-dependent rather than intrinsic to the MF-RIS concept.

## 6. Empirical evidence, limitations, and open research directions

Simulation evidence across scenarios consistently reports gains over passive or single-function baselines. In the downlink NOMA study, at \(P_{\max}=10\) dBm the proposed MF-RIS achieves about \(59\%\) higher sum rate than SF-RIS, about \(16\%\) over active RIS, and about \(44\%\) over STAR-RIS [2301.13630]. In the self-sustainable NOMA model, MF-RIS achieves \(23.4\%\) higher sum-rate than conventional passive RIS and \(98.8\%\) higher sum-rate than self-sustainable RIS under the reported setting [2310.02562]. In the communication-sensing tutorial scenario, performance increases with the number of RIS elements for all schemes, but with diminishing returns due to total amplification-power limits [2410.06584].

Beyond simulation, a Sub-6 GHz indoor measurement campaign implemented an MF-RIS-enabled radar and communication coexistence platform with a \(50\times 37\) dual-polarized reflective RIS at \(3.5\) GHz, an integrated \(60\) GHz MIMO radar, and local DBSCAN-plus-Kalman tracking. The reported 5G NR SU-MIMO evaluation showed a \(74\%\) reduction in throughput variance and a \(12.5\%\) sum-rate improvement within the MF-RIS near-field relative to the no-RIS setup, together with reduced delay spread and increased coherence bandwidth through a virtual-LoS path [2602.06755]. This provides direct empirical support for the broader claim that MF-RIS can harden channels rather than merely raise average power.

The main limitations identified in the literature are likewise consistent. Active amplification increases thermal noise and can degrade performance if gains are not carefully controlled [2501.10657][2512.19109]. Channel estimation and robust beamforming are substantially harder than in passive RIS because the observation noise depends on the same MF-RIS coefficients being optimized, and because reflection and refraction sides are coupled [2501.10657][2412.08032]. Hardware feasibility remains open at scale: active elements need efficient low-noise amplification, finite-resolution phase and amplitude control create quantization loss, and in impedance-programmable implementations the available phase span can remain below \(360^\circ\) because of limited capacitance range [1811.10082][2405.16257]. In robust design, it is explicitly reported that the cumulative CSI error caused by increasing the number of RIS elements is larger than that caused by increasing the number of transmit antennas [2412.08032].

Current open directions therefore center on three fronts. The first is hardware: low-noise active elements, scalable control circuitry, and integration of reflection, refraction, amplification, harvesting, and possibly sensing on a single surface. The second is modeling and estimation: near-field channels, mobile cascaded channels, robust CSI acquisition, and active-noise-aware inference. The third is control: scalable optimization for mixed discrete-continuous decisions, distributed coordination of multiple MF-RIS, and joint design across communication, sensing, security, and computing objectives. In the aggregate, the literature presents MF-RIS not as a single finished technology, but as a rapidly expanding architecture class whose defining contribution is to make the radio environment itself multi-role, power-aware, and actively programmable [2405.16257][2410.06584].

Source: https://www.emergentmind.com/topics/multi-functional-reconfigurable-intelligent-surfaces-mf-ris