Pinching Antennas: Reconfigurable Waveguide Radiators

• Pinching antennas are reconfigurable guided-wave radiators that use small dielectric perturbations on low-loss waveguides to create controllable leaky radiation points.
• Their design leverages coupled-mode theory and geometric placement, with optimization algorithms like particle swarm optimization refining phase alignment and beam shaping.
• Applications include short-range communications, sensing, and physical-layer security, offering enhanced performance over conventional centralized arrays and RIS techniques.

Pinching antennas are a class of flexible, guided-wave radiators in which small dielectric perturbations are placed on a low-loss dielectric waveguide so that guided energy leaks into free space at controllable locations. In the associated pinching-antenna system, or PASS, a base station feeds one or more waveguides, while the active radiation points are created by attaching, activating, or repositioning pinching elements along those guides. The central technical idea is that waveguide transport incurs much lower loss than long-range free-space propagation, so radiation can be moved close to users or targets, creating short-distance line-of-sight links, reconfiguring aperture geometry, and introducing spatial degrees of freedom that fixed arrays do not possess (Yang et al., 18 Jan 2025, Liu et al., 30 Jan 2025, Mu et al., 23 Feb 2025).

1. Physical principle and electromagnetic basis

A pinching antenna is formed by locally “pinching” a dielectric waveguide with a small dielectric particle or perturbation. The pinch changes the local boundary conditions of the guided mode and creates a leaky radiation point, so that part of the guided power is emitted into free space. In the most general description used across the literature, each pinch behaves as a controllable radiator whose field inherits two phases: a guided-wave phase accumulated from the feed point to the pinch, and a free-space propagation phase from the pinch to the observation point (Yang et al., 18 Jan 2025, Wang et al., 14 Jul 2025).

The waveguide itself is treated as a low-loss transmission medium. One recurring model uses a guided wavelength

$\lambda_g=\frac{\lambda}{n_{\mathrm{eff}}},$

where $\lambda$ is the free-space wavelength and $n_{\mathrm{eff}}$ is the effective refractive index of the guide (Mu et al., 23 Feb 2025, Wang et al., 14 Jul 2025). In survey-level treatments, pinching antennas are contrasted with fluid and movable antennas by their ability to relocate the effective radiating point over meter-scale waveguide lengths rather than over only a few wavelengths, which permits direct mitigation of large-scale path loss and restoration of line-of-sight conditions (Yang et al., 18 Jan 2025).

A more explicit hardware model treats the pinching element as an open-ended directional coupler. In that model, the guided and pinched modes satisfy coupled-mode equations, and the power exchange is

$$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$

where $L$ is the coupling length, $\kappa$ is the coupling coefficient, and $F$ is the maximum coupling efficiency (Liu et al., 30 Jan 2025). A related physics-based treatment derives the coupled-mode fields $A(x)$ and $B(x)$ along the waveguide and shows how multiple pinches on the same guide are coupled through progressive upstream power depletion, motivating equal-power and proportional-power radiation models (Wang et al., 9 Feb 2025).

Most early PASS work models pinches as passive dielectric elements. A later variant, the electronically reconfigurable pinching antenna, replaces purely passive pinches with varactor-loaded perturbation modules whose capacitance controls the local leakage level. That design uses a low-loss rectangular dielectric waveguide, modular radiation elements, and a copper reflector to create electronically tunable radiation points at 60 GHz, thereby extending the concept from mechanically or discretely activated radiators to electronically controlled leakage (Plata-Orozco et al., 25 Jun 2026).

2. Architectures and deployment modes

The canonical PASS architecture consists of a base station feeding a long dielectric waveguide and multiple pinching antennas radiating from adjustable positions on that guide. In the multicast formulation, a single waveguide of length $L$ is deployed parallel to the $\lambda$0-axis at height $\lambda$1, with the $\lambda$2-th PA located at

$\lambda$3

subject to ordering and spacing constraints such as

$\lambda$4

with $\lambda$5 typically set to $\lambda$6 to avoid mutual coupling (Mu et al., 23 Feb 2025). The same basic geometry appears in single-user rate maximization, array-gain analysis, and blockage-aware designs (Xu et al., 18 Feb 2025, Ouyang et al., 10 Jan 2025, Xie et al., 3 Jan 2026).

Architectural classifications in the PASS literature distinguish single-waveguide and multi-waveguide systems. A single waveguide with a single activated pinch supports orthogonal access or cluster service; a single waveguide with multiple activated pinches supports simultaneous radiating points fed by one RF chain; multiple waveguides with one or more pinches per guide enable MIMO-like or hybrid beamforming behavior (Yang et al., 18 Jan 2025, Liu et al., 30 Jan 2025). One survey formalizes this split as SWMAP, MWSAP, and MWMAP, emphasizing that digital beamforming across waveguides can coexist with analog or radiative beamforming along each waveguide (Yang et al., 18 Jan 2025).

A distinctive PASS-specific concept is pinching beamforming, namely beam shaping through PA placement rather than solely through complex weights on fixed antennas. In the non-multiplexing architecture, a waveguide carries a single stream and spatial shaping is produced by pinching positions alone. In multiplexing architectures, baseband precoding is combined with pinching beamforming; proposed designs include sub-connected, fully-connected, and phase-shifter-based fully-connected schemes (Liu et al., 30 Jan 2025).

Deployment scenarios are typically distributed rather than centralized. The literature explicitly considers waveguides installed along ceilings, walls, facades, corridors, tunnels, roadsides, bridges, and building edges, so that radiators can be activated near users or sensing regions while the RF feed remains centralized (Yang et al., 18 Jan 2025, Mu et al., 23 Feb 2025). This deployment logic is also central to blockage-aware work, which treats the waveguide as infrastructure embedded in obstacle-rich indoor environments such as shopping malls, airports, offices, factories, and warehouses (Xie et al., 3 Jan 2026).

3. Channel modeling, signal models, and array behavior

PASS channel models combine a free-space spherical-wave term with an in-waveguide phase term. In multicast PASS, the in-waveguide channel from the feed point to the $\lambda$7-th PA is represented by

$\lambda$8

while the PA-to-user free-space channel is

$\lambda$9

with $n_{\mathrm{eff}}$0 and $n_{\mathrm{eff}}$1 (Mu et al., 23 Feb 2025). The corresponding user SNR is

$n_{\mathrm{eff}}$2

and the multicast rate is $n_{\mathrm{eff}}$3 (Mu et al., 23 Feb 2025).

A closely related single-user downlink model writes the $n_{\mathrm{eff}}$4-th PA channel as

$n_{\mathrm{eff}}$5

where $n_{\mathrm{eff}}$6 is the free-space distance from the PA to the user and $n_{\mathrm{eff}}$7 is the guide distance from the feed to the PA. Under equal power splitting across $n_{\mathrm{eff}}$8 active PAs, the rate becomes

$n_{\mathrm{eff}}$9

(Xu et al., 18 Feb 2025). This model makes explicit that PA placement jointly affects amplitude through $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$0 and phase through both $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$1 and $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$2.

More general survey treatments express the effective channel to user $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$3 as

$$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$4

where $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$5 is a coupling coefficient, $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$6 is a complex weight, $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$7 and $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$8 are guide attenuation and phase constants, and $$P_{\mathrm{guide}} = 1 - F \sin^2(\kappa L), \qquad P_{\mathrm{pinch}} = F \sin^2(\kappa L),$$9 is the free-space gain from pinch $L$0 to user $L$1 (Yang et al., 18 Jan 2025). For multi-waveguide systems, the effective MIMO channel is written as

$L$2

which makes clear that PASS can be modeled as a channel reconfiguration mechanism acting simultaneously on geometry, guide phase, and coupling (Yang et al., 18 Jan 2025).

Array behavior differs materially from conventional fixed-aperture arrays. In the array-gain analysis, the PASS gain is

$L$3

with waveguide-induced phase compensation

$L$4

That work derives a closed-form upper bound under half-wavelength spacing and shows that, for fixed total power, the array gain is non-monotonic in the number of pinches and vanishes asymptotically as $L$5, implying the existence of an optimal finite number of active antennas (Ouyang et al., 10 Jan 2025). When mutual coupling is incorporated through a $L$6-based coupling matrix, the same study proves that there is also an optimal inter-antenna spacing; for $L$7 and $L$8, the numerical maximizer is $L$9 (Ouyang et al., 10 Jan 2025).

4. Beamforming, placement optimization, and learning-based control

The central PASS optimization problem is usually geometric: place the PAs to maximize a communication or sensing metric. In multicast PASS, the objective is the max–min SNR

$\kappa$0

subject to ordering and spacing constraints (Mu et al., 23 Feb 2025). Because $\kappa$1 depends on nonlinear phase terms in both free-space and guide propagation, the resulting objective is non-convex and non-smooth (Mu et al., 23 Feb 2025).

A recurring algorithmic baseline is particle swarm optimization. For multicast PASS, each particle encodes a PA-position vector and is updated via

$\kappa$2

$\kappa$3

with a decreasing inertia weight $\kappa$4, boundary projection, and penalties for violating spacing constraints (Mu et al., 23 Feb 2025). The total complexity is reported as $\kappa$5, with possible reduction of penalty checking from $\kappa$6 to $\kappa$7 by nearest-neighbor checks (Mu et al., 23 Feb 2025).

For single-user downlink rate maximization, a different two-stage structure is used. First, the PAs are packed at minimum spacing $\kappa$8 and centered over the user’s $\kappa$9-coordinate to minimize large-scale path loss; second, the individual positions are refined within small wavelength-scale windows to enforce constructive phase alignment (Xu et al., 18 Feb 2025). Under the condition $F$0, with $F$1, the first-stage optimum is

$F$2

which yields a closed-form centered aperture (Xu et al., 18 Feb 2025). A cognate three-stage procedure appears in cognitive radio PASS, where coarse waveguide-level placement is followed by wavelength-level refinements that enforce constructive combining at the intended user and destructive combining at the unintended user, and then by closed-form secondary-transmit power control (Sun et al., 17 Nov 2025).

Several works formulate PASS beamforming jointly with digital precoding. In the physics-based beamforming paper, transmit-power minimization under user SINR constraints is solved by a penalty-based alternating optimization algorithm and a low-complexity zero-forcing-based algorithm (Wang et al., 9 Feb 2025). In the multicell setting, weighted sum-rate maximization is handled by an alternating-optimization framework combining fractional programming, block coordinate descent, and PSO, with digital precoders, per-pinch power allocation, and PA placement optimized in separate blocks (Chen et al., 11 Jun 2026). In ISAC, one paper uses semidefinite relaxation for digital beamforming and a pinching-beamforming subproblem solved by successive convex approximation, a penalty method, and element-wise optimization (Li et al., 27 Aug 2025), while another multi-waveguide ISAC design uses a fine-tuning approximation plus SCA to jointly optimize TPA positions and beamforming under a radar-SNR constraint (Mao et al., 30 May 2025).

Learning-based control has also been proposed. A graph-attention model, called a bipartite GAT, represents users and pinching elements as a complete bipartite graph, with user coordinates, antenna intervals, powers, and user–antenna distances as node and edge features (Xie et al., 8 Feb 2025). The BGAT readout is explicitly constructed to satisfy the power budget, spacing, and deployment-region constraints, and is trained in an unsupervised manner on the energy-efficiency objective (Xie et al., 8 Feb 2025). In blockage-aware PASS, continuous PA placement is combined with WMMSE beamforming and a DDPG agent to handle the non-smooth LoS transitions produced by obstacles (Xie et al., 3 Jan 2026).

5. Communication, security, sensing, and ISAC applications

PASS has been studied in single-user downlink, multicast, multigroup multicast, cognitive radio, multicell systems, physical-layer security, and sensing/ISAC settings. The unifying rationale is that moving the radiation point changes both large-scale path loss and phase geometry, which in turn affects communication rate, fairness, secrecy, or estimation accuracy (Yang et al., 18 Jan 2025, Liu et al., 30 Jan 2025).

In multicast communications, PASS is naturally aligned with a single common message because all PAs on a waveguide are fed by the same signal. A dedicated multicast framework optimizes PA locations to maximize the worst-user SNR and reports that PASS significantly outperforms conventional multiple-antenna transmission, especially when the number of PAs is small and short-range LoS can be created for all users (Mu et al., 23 Feb 2025). Multigroup multicast requires explicit multiple-access design. One study compares treating interference as noise, NOMA, and TDMA under pinching switching and pinching multiplexing protocols, and concludes that TDMA-PS generally achieves the best max–min fairness, while NOMA can surpass TDMA-PS in high transmit-power regimes with heterogeneous group distributions (Shan et al., 25 Feb 2026).

Single-waveguide multiuser access has also been linked to NOMA because multiple pinches on the same guide share one feed signal. Analytical work on NOMA-assisted pinching antennas derives rate expressions, shows how PA placement can create effective channel disparities that facilitate SIC, and reports superior performance over OMA in appropriate geometries (Ding et al., 2024). In cognitive radio, the same principle is used to shape constructive interference toward the intended user and destructive interference toward the unintended user, yielding nearly orthogonal co-channel transmission between primary and secondary networks through wavelength-level pinching-beamforming refinements (Sun et al., 17 Nov 2025).

Physical-layer security is an especially direct application because the active set of pre-installed pinches determines the coherent sum seen at the legitimate user and the eavesdropper. A coalitional-game formulation uses Shapley values to evaluate each pinch’s contribution to secrecy rate, and an activation algorithm merges or splits pinches until a Nash-stable active set is reached (Wang et al., 14 Jul 2025). In the reported simulations, the Shapley-based method improves secrecy rate over both a conventional ULA baseline and simpler coalition-value heuristics (Wang et al., 14 Jul 2025).

Sensing and ISAC studies exploit the same reconfigurable geometry for estimation performance rather than only for communication. In CRLB-based positioning, pinching arrays are shown to lower the position CRB relative to a conventional circular array by distributing elements across the service area and increasing Fisher information for edge and off-center users (Ding, 8 Apr 2025). For $F$3, a user-centric square arrangement yields the closed-form optimum

$F$4

revealing that positioning elements directly above the user is not always optimal for sensing accuracy (Ding, 8 Apr 2025). Another sensing architecture combines PASS transmission with LCX-based distributed reception and derives a multi-target CRB minimized by a two-stage design: PSO for PA positions and convex waveform-covariance optimization (Wang et al., 21 May 2025). In multi-waveguide ISAC, coordinated transmit and receive pinching antennas are optimized under a radar-SNR requirement, and the results highlight a distinct communication–sensing trade-off with non-smooth rate drops as sensing constraints tighten (Mao et al., 30 May 2025). A related full-duplex PASS-ISAC framework uses transmitting PAs and a receiving ULA, and reports that PASS is less affected by stringent communication constraints than conventional MIMO-ISAC because the communication links can be maintained with lower path loss (Li et al., 27 Aug 2025).

6. Comparisons, limitations, and open questions

PASS is repeatedly contrasted with centralized arrays, distributed antenna systems, RIS, fluid antennas, and movable antennas. Relative to centralized arrays and massive MIMO, PASS reduces the free-space distance before radiation and can expose many simple radiators without assigning one RF chain to each element (Mu et al., 23 Feb 2025, Yang et al., 18 Jan 2025). Relative to RIS, PASS is an actively fed distributed aperture rather than a passive reflector, and therefore avoids two-hop path loss and dependence on the quality of the incident path (Mu et al., 23 Feb 2025, Wang et al., 14 Jul 2025). Relative to fluid and movable antennas, PASS operates over much larger spatial extents and is aimed at line-of-sight creation and large-scale path-loss reduction rather than only small-scale fading mitigation (Yang et al., 18 Jan 2025).

Several misconceptions are explicitly corrected in the literature. One is that pinching antennas are merely another form of local antenna motion; the distinguishing feature is that they move the effective radiating point anywhere along a waveguide distributed near users (Yang et al., 18 Jan 2025). Another is that PASS always requires per-element active phase shifters; in many formulations, control is achieved only through geometry-induced guided and free-space phases, with equal power across active pinches and no explicit per-pinch phase shifter (Mu et al., 23 Feb 2025, Wang et al., 14 Jul 2025). A further nuance is that not all implementations are mechanically passive: electronically reconfigurable varactor-loaded E-pinching antennas constitute a later hardware extension rather than the baseline PASS assumption (Plata-Orozco et al., 25 Jun 2026).

The main limitations are also consistent across studies. Most analytical models assume LoS-dominated propagation, negligible in-waveguide attenuation, ideal calibration, and either omitted or simplified mutual coupling (Mu et al., 23 Feb 2025, Xu et al., 18 Feb 2025, Wang et al., 9 Feb 2025). Discrete activation can approximate continuous placement only if the available candidate set is sufficiently dense; one beamforming study reports that more than approximately $F$5 positions per meter are needed to match continuous activation well (Wang et al., 9 Feb 2025). Optimization can be expensive when the numbers of users, waveguides, or PAs are large, which motivates surrogate models, element-wise search, or learning-based approximations (Mu et al., 23 Feb 2025, Xie et al., 8 Feb 2025, Chen et al., 11 Jun 2026).

Blockage-aware modeling introduces an additional complication: LoS indicators change discontinuously as PAs cross obstacle tangents. Deterministic cylinder-based models have therefore been proposed for indoor environments, along with assignment, surrogate block-coordinate search, and WMMSE-DDPG methods that explicitly exploit obstacles both to preserve desired LoS and to block interference (Xie et al., 3 Jan 2026). A plausible implication is that PASS is not only a blockage-mitigation technology but also, in some layouts, an interference-shaping technology; the blockage-aware and LoS-blockage studies both report throughput gains arising from turning obstacles into spatial shields (Xie et al., 3 Jan 2026, Wang et al., 14 Jul 2025).

Open problems identified across the corpus include waveguide deployment and topology design, low-overhead channel estimation for dynamically activated nonuniform arrays, uplink processing, robustness to calibration and location uncertainty, wideband guided-wave phase modeling, joint optimization of placement and amplitude weighting, and integration with learning-based control for real-time mobility support (Yang et al., 18 Jan 2025, Mu et al., 23 Feb 2025, Wang et al., 14 Jul 2025). Taken together, these works position pinching antennas as a reconfigurable infrastructure technology whose defining research challenge is the joint treatment of electromagnetics, geometry, and communications rather than beamforming alone (Liu et al., 30 Jan 2025, Wang et al., 9 Feb 2025).