Hybrid Reflecting and Sensing RISs

• Hybrid Reflecting and Sensing RISs (HRISs) are reconfigurable metasurfaces that split incident electromagnetic energy between reflection and local sensing.
• They employ tunable power-splitting and hybrid unit-cell designs—using varactor-loaded elements, PIN diodes, or interleaved sensing arrays—to optimize channel estimation and localization.
• HRISs enable integrated sensing and communication systems with improved channel recovery, secure links, and reduced pilot overhead compared to passive RISs.

Searching arXiv for recent and foundational papers on Hybrid Reflecting and Sensing RISs (HRISs). Hybrid Reflecting and Sensing RISs (HRISs) are reconfigurable intelligent surfaces whose meta-atoms or associated circuitry support both controllable reflection of impinging electromagnetic waves and local sensing or reception of a portion of those waves. In the HRIS literature, this dual functionality is introduced to overcome two limitations of purely reflective RISs: the lack of local observability for self-configuration and the difficulty of estimating the underlying individual channels when the surface is only a passive scatterer. Foundational formulations describe HRISs as metasurfaces that enable simultaneous tunable reflections and sensing, thereby supporting channel parameter estimation, localization, and potentially computationally autonomous or self-configuring metasurfaces (Alexandropoulos et al., 2021). Later estimation-theoretic treatments formalized the HRIS as a surface that can identify individual user-to-surface and surface-to-base-station channels, rather than only the cascaded channel typically accessible with passive RISs (Zhang et al., 2022).

1. Conceptual scope and taxonomy

The defining property of an HRIS is not merely that some elements are “active,” but that the surface has explicit local receive capability. In the canonical HRIS model, the impinging field is divided between a reflected branch used for propagation control and a sensed branch delivered to a small number of RF chains for local processing. This makes the surface a measurement-enabled propagation node rather than a blind programmable reflector (Zhang et al., 2022).

Within the literature, HRIS is broader than a fixed partition into reflecting-only and sensing-only subsets. Some papers model each element as simultaneously reflecting and sensing through a tunable power split, while others realize dual functionality through hybrid unit cells, interleaved sensing arrays, or a small sensing-capable subset embedded in a larger reflective aperture (Alexandropoulos et al., 2021). A conventional passive RIS appears as the special case where all incident power is assigned to the reflective branch and no sensing path is used (Zhang et al., 2022).

Several adjacent concepts are closely related but not identical. A STAR-RIS splits power into reflection and transmission or refraction toward opposite sides of the surface, whereas an HRIS splits power into reflection and reception or sensing (Zhang et al., 2022). A hybrid active/passive RIS for ISAC, where some reflecting elements include amplifiers and the sensing function remains at the BS, is also distinct: such systems improve target illumination or communication support but do not make the RIS itself a sensing receiver (Sankar et al., 2022). Likewise, hybrid relay-reflecting intelligent surfaces (HR-RISs) are important precursors because they mix passive reflecting elements with a small number of active RF-connected elements and address CSI acquisition and localization, but their active elements are modeled as amplify-and-forward relays rather than purely sensing front ends (Nguyen et al., 2021).

2. Architectural models and hardware realizations

A common analytical HRIS model assigns to each meta-atom a reflection coefficient and a sensing coefficient. In the uplink multi-user formulation of “Channel Estimation with Hybrid Reconfigurable Intelligent Metasurfaces” (Zhang et al., 2022), the $l$-th element splits the incident signal $r_l(n)$ according to

$$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$

$$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$

where $\rho_l(n)\in[0,1]$ is the power-splitting coefficient, $\psi_l(n)\in[0,2\pi)$ is the reflection phase, and $\phi_{r,l}(n)\in[0,2\pi)$ is the reception-side phase. In vector form, the reflected branch is described by a diagonal matrix $\boldsymbol{\Psi}$, whereas the sensed branch is described by an analog-combining matrix $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$ (Zhang et al., 2022). This framework accommodates both fully connected and partially connected combining. Other papers use the opposite convention, with $\rho$ or $r_l(n)$0 denoting the absorbed or sensed portion and $r_l(n)$1 or $r_l(n)$2 the reflected portion; the underlying trade-off is the same (Gavras et al., 2024).

The early hardware concept in “Hybrid Reconfigurable Intelligent Metasurfaces: Enabling Simultaneous Tunable Reflections and Sensing for 6G Wireless Communications” (Alexandropoulos et al., 2021) realizes this dual operation with varactor-loaded meta-atoms, sampling waveguides, and $r_l(n)$3 reception RF chains. There, the coupling to the sensing waveguides is kept below $r_l(n)$4 dB in full-wave simulations around $r_l(n)$5 GHz to ensure that ample power is reflected while the sensed path remains usable. The paper also notes that the sensed signal at the waveguides need not correspond one-to-one with the field at each meta-atom; instead, the metasurface-plus-waveguide structure behaves like a receiver with equivalent analog combining, and the resulting multiplexing can be pre-characterized into a sensing matrix (Alexandropoulos et al., 2021).

Subsequent hardware papers explore more specialized realizations. “Dual Dielectric Metasurface for Simultaneous Sensing and Reconfigurable Reflections” (Birari et al., 23 Jan 2025) proposes a dual-functional hybrid unit cell at $r_l(n)$6 GHz with an outer ring reflector and a central circular disc antenna that share the same phase center. The cell size is $r_l(n)$7, the central disc is miniaturized using a high-dielectric material, and two interleaved sensing arrays with orthogonal polarization and $r_l(n)$8 offset are embedded within the surface to sense channel parameters toward the two communication ends. Reflection control is implemented through SP4T-switched load-tuning matrices and lookup-table-based calibration (Birari et al., 23 Jan 2025).

At the low-complexity end, “Novel 1-bit Hybrid Reconfigurable Intelligent Surface” (Keshmiri et al., 7 Jul 2025) uses PIN-diode-loaded resonant patches, slot coupling into a parallel-plate waveguide, and only two coaxial sensing ports. Every cell participates in both functions: the patch reflects and the slot weakly couples a portion of the incident field into the PPWG for AoA sensing. The slots also support pre-coded phase randomization, with slot lengths $r_l(n)$9 mm and $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$0 mm chosen so that the four combined slot/diode states are approximately $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$1 apart at $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$2 GHz (Keshmiri et al., 7 Jul 2025).

3. Channel estimation and receiver design

The most immediate systems-level consequence of HRIS sensing is the ability to recover individual channels. In the uplink multi-user MIMO setting of (Zhang et al., 2022), the HRIS directly observes the user-to-surface channel matrix $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$3 through

$$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$4

while the BS observes the reflected product

$$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$5

This changes the identifiability structure: the HRIS estimates the individual first-hop channel $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$6, forwards $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$7 to the BS over a control link, and the BS then estimates the second-hop channel $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$8 (Zhang et al., 2022).

A central noiseless result is that exact recovery of $$y_l^{\rm RF}(n)= \rho_l\left( n \right)e^{\jmath \psi_l \left( n \right)} r_l(n),$$9 and $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$0 is possible if the total pilot length satisfies

$$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$1

For $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$2, $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$3, $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$4, and $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$5, this gives $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$6 pilots for separate recovery of $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$7 and $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$8, whereas the reflective-RIS baseline of Wang et al. requires more than $$y_{r,l}^{\rm RC} (n) = (1 - \rho_l \left( n \right))e^{\jmath \phi_{r,l} \left( n \right)} r_l(n),$$9 pilots for cascaded-channel estimation (Zhang et al., 2022). In the noisy case, the same paper derives closed-form LMMSE expressions for the MSEs of estimating $\rho_l(n)\in[0,1]$0 and $\rho_l(n)\in[0,1]$1, and uses automatic differentiation with first-order descent on a barrier-regularized objective to optimize the HRIS power-splitting and phase parameters $\rho_l(n)\in[0,1]$2 (Zhang et al., 2022).

The semi-blind line of work removes the dedicated pilot-only stage. “Semi-Blind Receivers for Hybrid Reflecting and Sensing RIS” (Magalhães et al., 2024) models the sensed and reflected uplink signals as structured third-order tensors. The HRIS-side signal obeys a double Tucker model and the BS-side signal a PARAFAC-Tucker model, enabling joint symbol detection and separate channel estimation of $\rho_l(n)\in[0,1]$3 and $\rho_l(n)\in[0,1]$4 without an a priori training stage. The paper develops iterative BALS receivers and closed-form KronF/KRF receivers, gives identifiability conditions such as

$\rho_l(n)\in[0,1]$5

for HRIS-BALS under TSTC, and shows that ambiguity-free separate channel estimation is obtained with minimal embedded pilots used only for scale normalization (Magalhães et al., 2024).

4. Sensing, localization, and integrated sensing and communications

Early HRIS papers treated sensing primarily as a means for channel parameter estimation and self-configuration, but they already emphasized localization, AoA estimation, and environment awareness. The 2021 HRIS overview explicitly argues that the sensing capability facilitates channel parameter estimation and localization, and can support self-configuring metasurfaces within smart radio environments (Alexandropoulos et al., 2021). In that sense, HRIS sensing began as communication-centric sensing rather than a full radar formulation.

A concrete localization-oriented example is “Joint Channel and Direction Estimation for Ground-to-UAV Communications Enabled by A Simultaneous Reflecting and Sensing RIS” (He et al., 2022). There, the HRIS uses global power-splitting coefficients $\rho_l(n)\in[0,1]$6 with $\rho_l(n)\in[0,1]$7, reflecting a portion of the UAV pilot toward the BS while combining the sensed portion through a single RF chain. The sensed observation is

$\rho_l(n)\in[0,1]$8

and the BS observation is

$\rho_l(n)\in[0,1]$9

Using atomic norm minimization, the system estimates the individual UAV-HRIS channel $\psi_l(n)\in[0,2\pi)$0 and HRIS-BS channel $\psi_l(n)\in[0,2\pi)$1, then applies root-MUSIC to the refined $\psi_l(n)\in[0,2\pi)$2 estimate to recover the LoS AoA. The paper also derives conditional CRLBs for the channel-estimation MSEs and shows the expected trade-off: increasing $\psi_l(n)\in[0,2\pi)$3 improves $\psi_l(n)\in[0,2\pi)$4 estimation at the BS, while increasing sensing power improves $\psi_l(n)\in[0,2\pi)$5 estimation and AoA accuracy (He et al., 2022).

Later work moves from channel estimation to true HRIS-enabled ISAC. “Simultaneous Communications and Sensing with Hybrid Reconfigurable Intelligent Surfaces” (Gavras et al., 2024) considers an XL-MIMO downlink where the HRIS reflects a fraction $\psi_l(n)\in[0,2\pi)$6 of the BS signal toward a UE and absorbs a fraction $\psi_l(n)\in[0,2\pi)$7 for passive-target localization in an area of interest. The downlink channel is

$\psi_l(n)\in[0,2\pi)$8

while the HRIS sensed signal is

$\psi_l(n)\in[0,2\pi)$9

Localization quality is measured by the PEB,

$\phi_{r,l}(n)\in[0,2\pi)$0

and the design maximizes downlink rate subject to PEB-based localization-coverage constraints over a discretized AoI (Gavras et al., 2024).

Secure ISAC extends the same principle. “Communications-Centric Secure ISAC with Hybrid Reconfigurable Intelligent Surfaces” (Gavras et al., 29 Apr 2025) uses HRIS sensing to localize both a legitimate user and an eavesdropper in a bistatic setup. The absorbed path is

$\phi_{r,l}(n)\in[0,2\pi)$1

the secrecy rate is

$\phi_{r,l}(n)\in[0,2\pi)$2

and sensing performance is quantified through a CRB-derived PEB. The formulation jointly optimizes the BS precoder, HRIS analog combiner, and reflection phases, making the absorption-versus-reflection split $\phi_{r,l}(n)\in[0,2\pi)$3 the central secure-ISAC design knob (Gavras et al., 29 Apr 2025).

5. Trade-offs, algorithms, and quantitative behavior

Across the literature, the central HRIS trade-off is consistent: more sensing or absorption improves local observability but weakens the reflected communication path, while more reflection improves communication support but starves the sensing receiver. In the channel-estimation setting of (Zhang et al., 2022), increasing $\phi_{r,l}(n)\in[0,2\pi)$4 allocates more power to reflection and less to local sensing, so the MSE of estimating $\phi_{r,l}(n)\in[0,2\pi)$5 at the HRIS rises sharply while the MSE of estimating $\phi_{r,l}(n)\in[0,2\pi)$6 at the BS initially falls and then degrades as the surface approaches the passive-RIS regime. That paper also reports that partially connected analog combining performs comparably to the fully connected architecture in the reported simulations, and that even a modest number of RF chains can be sufficient: with $\phi_{r,l}(n)\in[0,2\pi)$7, only $\phi_{r,l}(n)\in[0,2\pi)$8 receive RF chains significantly outperform the passive RIS baseline (Zhang et al., 2022).

The same balance reappears in semi-blind reception. In (Magalhães et al., 2024), when $\phi_{r,l}(n)\in[0,2\pi)$9, the HRIS behaves as mostly sensing and almost non-reflecting, yielding low HRIS SER and improved $\boldsymbol{\Psi}$0 but poor BS symbol recovery and degraded $\boldsymbol{\Psi}$1. As $\boldsymbol{\Psi}$2 increases, BS-side detection initially improves, yet excessive reflection eventually hurts the BS as well because the HRIS estimate $\boldsymbol{\Psi}$3 becomes unreliable. The paper also reports that semi-blind joint channel estimation and detection incurs about $\boldsymbol{\Psi}$4 dB penalty in t-SNR for channel estimation relative to a pilot-assisted baseline, but transmits useful data during the estimation block rather than pilots only (Magalhães et al., 2024).

The communications-and-localization trade-off in (Gavras et al., 2024) is spatial as well as energetic. For $\boldsymbol{\Psi}$5, $\boldsymbol{\Psi}$6, $\boldsymbol{\Psi}$7 dBm, and $\boldsymbol{\Psi}$8, the desired PEB threshold is satisfied over about $\boldsymbol{\Psi}$9 of the AoI. Increasing $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$0 improves localization coverage but lowers the achievable downlink rate; increasing the AoI discretization size $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$1 improves coverage robustness up to a limiting region near $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$2 in the reported setup (Gavras et al., 2024).

Hardware papers quantify a complementary trade-off between sensing capability and reflection loss. In the 1-bit PPWG-based HRIS of (Keshmiri et al., 7 Jul 2025), adding the coupling slots changes the reflection coefficient magnitude by less than $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$3 dB, while the best demonstrated CGS sensing configuration achieves $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$4 AoA accuracy with $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$5 masks at $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$6 dB SNR. The same paper shows that an MLP using only four masks remains robust over SNRs from $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$7 dB to $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$8 dB, illustrating that data-driven sensing can reduce mask overhead in strongly quantized HRIS hardware (Keshmiri et al., 7 Jul 2025).

Application-specific performance gains have also been reported. In the RISS-assisted WPCN of (Luo et al., 2023), replacing full CSI acquisition with autonomous DOA sensing at a surface having a small active subset and a large passive subset yields gains of $\boldsymbol{\Phi}\in\mathbb{C}^{N_r\times N}$9 dB and $\rho$0 dB over the conventional counterpart for $\rho$1 and $\rho$2, respectively (Luo et al., 2023).

6. Relation to neighboring paradigms and open problems

A recurring misconception is to treat all “hybrid RIS” proposals as HRISs. The literature is more specific. A hybrid active/passive RIS that assists an ISAC transmitter but does not contain sensing RF chains at the surface is not, in the strict sense, a reflecting-and-sensing RIS (Sankar et al., 2022). Conversely, the relay-reflecting lineage is highly relevant architecturally but remains distinct because its active elements are designed for amplification. The HR-RIS papers show that only a small number of enhanced-function elements can produce most of the gain and motivate fixed versus dynamic role assignment, but they do not provide a sensing-side signal model, echo processing, or local observation channel at the surface (Nguyen et al., 2021).

The main open problems identified across the HRIS literature are unusually cross-disciplinary. At the hardware level, the 2021 overview and later metasurface papers call for more physically accurate and experimentally validated electromagnetic models, fuller characterization of coupling between the reflection and sensing parameters, and scalable prototypes beyond narrowband proof-of-concept designs (Alexandropoulos et al., 2021). At the signal-processing level, open issues include broader uses of HRIS sensing beyond channel estimation, treatment of direct BS-user links, and accounting for control-link limitations between the HRIS and the BS (Zhang et al., 2022). At the receiver-design level, assumptions such as quasi-static flat fading, an error-free HRIS-BS control link, and limited modeling of practical hardware impairments remain common simplifying hypotheses (Magalhães et al., 2024).

The more recent ISAC and secure-ISAC formulations expose additional challenges rather than closing them. Joint designs over transmit precoders, HRIS reflection coefficients, analog combiners, and power-splitting parameters are highly nonconvex and currently rely on alternating optimization, semidefinite relaxation, Gaussian randomization, or first-order methods (Gavras et al., 2024). The secure-ISAC setting adds worst-case region-based sensing and secrecy constraints at significant computational cost, with SDP-based steps that scale poorly in the number of antennas, RF chains, and discretization points (Gavras et al., 29 Apr 2025). This suggests that practical HRIS research will continue to hinge on the co-design of electromagnetic hardware, sparse or tensorized signal models, and reduced-complexity optimization.

In its mature form, the HRIS concept is therefore neither a passive RIS with incidental monitoring nor merely an active/passive reflecting surface. It is a metasurface architecture in which reflection control and local observation are coupled by design. That coupling is what makes separate channel estimation, self-configuration, localization support, and communications-sensing integration possible, and it is also what defines the field’s central technical problem: how to allocate aperture, power, and algorithmic complexity between sensing and reflection without negating the gains of either function.