Pinching-Antenna Technology

• Pinching-antenna technology is a flexible antenna architecture that uses dielectric waveguides and movable dielectric pinches to dynamically configure transmission paths.
• It employs coupled-mode theory and multiport-network models to predict power extraction and optimize radiative performance in various system setups.
• The method enhances link quality and energy efficiency, benefiting applications from multicast beamforming to integrated sensing and communications.

Pinching-antenna technology denotes a flexible-antenna architecture in which electromagnetic energy is first guided through a low-attenuation medium—most commonly a dielectric waveguide—and is then radiated into free space at selected locations by small dielectric “pinches” or pinching antennas (PAs). Its defining feature is that the radiating point is not fixed at fabrication time: PA position becomes a system variable, so the propagation geometry itself can be reconfigured to shorten transmission distance, create or strengthen line-of-sight (LoS) links, mitigate blockage, and exploit near-field effects over meter-scale apertures rather than wavelength-scale motion alone (Yang et al., 18 Jan 2025, Liu et al., 11 Aug 2025, Liu et al., 26 Jan 2026). The concept was experimentally verified in the 60 GHz band, and the subsequent literature has developed hardware models, signal models, performance analyses, optimization methods, and application-specific designs spanning downlink communications, multicast, NOMA, sensing, integrated sensing and communications (ISAC), SWIPT, and multicell networks (Yang et al., 18 Jan 2025).

1. Physical basis and hardware realizations

In the canonical realization, a dielectric waveguide carries a guided mode, and a small dielectric particle attached to the waveguide perturbs the guided field and causes localized leakage into free space. This mechanism has been modeled both as a small dielectric scatterer and as an open-ended directional coupler or short secondary waveguide coupled to the main guide; the latter admits a coupled-mode-theory description of power exchange between the guided mode and the radiating element (Wang et al., 9 Feb 2025, Liu et al., 26 Jan 2026). In simplified communication models, the guided wavelength is typically written as

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

while an earlier principles paper also gives the dielectric-waveguide wavelength as

$\lambda_g=\frac{\lambda_0}{\sqrt{\varepsilon_r}}.$

The same literature notes a representative dielectric implementation with PTFE of relative permittivity around $2.1$ surrounded by air of relative permittivity near $1.0$ (Yang et al., 18 Jan 2025).

The hardware literature distinguishes between physically simple but idealized coupling models and more complete circuit abstractions. In the coupled-mode view, the waveguide and the PA exchange power according to modal amplitudes whose solution yields explicit radiated and residual guided powers; in the matched case, these reduce to sinusoidal power-transfer relations. The more general multiport-network model represents the PA as a three-port network with scattering parameters that incorporate reflection, transmission, and radiation efficiency, and recovers the simpler model under perfect matching (Wang et al., 9 Feb 2025, Liu et al., 11 Aug 2025, Liu et al., 26 Jan 2026).

PASS has also been generalized beyond dielectric-waveguide prototypes. The tutorial literature treats dielectric-waveguide pinching antennas, LCX-based pinching antennas, and pinching-inspired surface-wave systems within a common framework: all create, reposition, or deactivate localized radiation sites on demand along a signal-guiding medium. Activation can be mechanical, electronic, or hybrid. Mechanical activation moves or attaches a radiator; electronic activation uses switching, couplers, or engineered apertures; hybrid activation combines a guided structure with active radiation control (Xu et al., 15 Oct 2025).

2. Signal and channel models

The basic PASS channel is a two-stage channel: in-medium propagation along the guide, followed by free-space radiation from the selected PA location to the receiver. A representative single-user model writes the effective channel as

$$h=\frac{\eta^{1/2}}{\|\psi-\widetilde{\psi}\|}e^{-\alpha \tilde x} \exp\!\left(-j\left(\frac{2\pi}{\lambda}\|\psi-\widetilde{\psi}\|+\frac{2\pi}{\lambda_g}\tilde x\right)\right),$$

where $\tilde x$ is the PA position along the guide, $\alpha$ is the in-waveguide attenuation coefficient, and the phase contains both free-space and guided-wave terms (Xu et al., 15 Oct 2025). In many communication papers, waveguide loss is omitted because it is treated as negligible, which makes the in-waveguide channel phase-only and emphasizes the central design role of PA positions (Mu et al., 23 Feb 2025).

For multi-PA systems, the received field is the coherent superposition of multiple spherical-wave components. A near-field LoS model used for array-gain analysis places the $n$-th PA at $\boldsymbol\psi_n=[x_n,0,d]^T$ and a user at $\mathbf u=[x_{\rm u},0,0]^T$, giving

$\lambda_g=\frac{\lambda_0}{\sqrt{\varepsilon_r}}.$0

and, with equal power allocation,

$\lambda_g=\frac{\lambda_0}{\sqrt{\varepsilon_r}}.$1

which is the array-gain metric analyzed in the PASS literature (Ouyang et al., 10 Jan 2025).

A distinctive modeling issue for multiple PAs on a single waveguide is sequential power extraction. The hardware-oriented modeling literature therefore introduces two radiation models. Under the equal power model, antenna lengths are chosen so each PA radiates equally; under the proportional power model, all PAs have the same length and later PAs radiate a fixed proportion of the residual guided power. The latter is simpler to realize, and numerical results reported for PASS beamforming find that it performs comparably to the equal power model (Wang et al., 9 Feb 2025, Liu et al., 11 Aug 2025).

3. System architectures and structural variants

The PASS literature is organized around the number of waveguides and the number of activated PAs. A single waveguide with one activated PA behaves as a highly reconfigurable SISO-like link; a single waveguide with multiple activated PAs behaves as a linear array fed by the same signal; multiple waveguides introduce MIMO-like multiplexing and hybrid digital-analog beamforming; and later surveys add architectural variants such as segmented PASS, center-fed PASS, and multi-mode PASS (Yang et al., 18 Jan 2025, Ding et al., 2024, Liu et al., 26 Jan 2026).

Architecture	Defining feature	Representative papers
Single waveguide, single PA	One movable radiating point; often analyzed under TDMA/OMA	(Ding et al., 2024, Yang et al., 18 Jan 2025)
Single waveguide, multiple PAs	Same signal radiated from multiple positions; natural for multicast or NOMA	(Ding et al., 2024, Mu et al., 23 Feb 2025)
Multiple waveguides	One or more PAs per waveguide; supports MIMO and hybrid beamforming	(Bereyhi et al., 3 Feb 2025, Liu et al., 11 Aug 2025)
SWAN / segmented PASS	Multiple short, unconnected waveguide segments	(Liu et al., 26 Jan 2026)
C-PASS	Center feeding with forward and backward propagation	(Liu et al., 26 Jan 2026)
M-PASS	Multi-mode waveguides for mode-selective or mode-combining operation	(Liu et al., 26 Jan 2026)

A single-waveguide, multiple-PA architecture is constrained by the fact that all active PAs on that waveguide are fed with the same signal. This is the reason the literature repeatedly identifies multicast as a natural application and NOMA as a natural multiple-access strategy: one common waveform is already present in the hardware structure, and user differentiation must come from geometry, power-domain superposition, or scheduling rather than from independent RF chains on the same guide (Mu et al., 23 Feb 2025, Ding et al., 2024).

Multiple-waveguide systems, by contrast, support richer designs. Each waveguide may be connected to a dedicated RF chain, allowing digital precoding across guides while PA placement provides an analog-domain degree of freedom. This is the setting for downlink multiuser MIMO beamforming, multicell weighted sum-rate optimization, SWIPT, and ISAC (Bereyhi et al., 3 Feb 2025, Chen et al., 11 Jun 2026, Li et al., 7 Jun 2025, Mao et al., 30 May 2025). The surveys further distinguish practical transmission protocols such as waveguide switching, waveguide division, and waveguide multiplexing for multiuser operation (Liu et al., 11 Aug 2025).

4. Beamforming, resource allocation, and algorithmic design

PASS design problems are characteristically joint placement-and-resource-allocation problems. The main novelty is that antenna location itself is an optimization variable, coupled to conventional variables such as transmit power, time allocation, subcarrier allocation, beamforming vectors, SIC decoding order, and antenna activation. Because channel gains depend nonlinearly on PA locations through both distances and oscillatory phases, the resulting problems are typically highly non-convex (Zeng et al., 6 Jun 2025, Liu et al., 26 Jan 2026).

Model-based optimization spans several families. Fractional programming and block coordinate descent have been used for downlink weighted sum-rate maximization in PAS-assisted MIMO, where the digital precoder admits a regularized zero-forcing-type update and each PA position is updated by one-dimensional search (Bereyhi et al., 3 Feb 2025). WMMSE and alternating optimization are used in MIMO PASS-aided SWIPT, with beamforming updated by WMMSE and PA positions updated by a Gauss-Seidel procedure (Li et al., 7 Jun 2025). Penalty-based alternating optimization and a low-complexity ZF-based method have been proposed for joint transmit and pinching beamforming; the same work introduces continuous and discrete activation models and reports similar performance for its ZF-based and penalty-based algorithms (Wang et al., 9 Feb 2025). For multi-waveguide NOMA with discrete candidate PA positions, coalition-formation games determine waveguide assignment and antenna activation, while monotonic optimization or SCA solves the power-allocation subproblem (Wang et al., 3 May 2025). Symbol-level precoding has also been combined with PAS, using alternating optimization and projected gradient descent to exploit constructive interference by jointly designing beamforming vectors and PA positions (Pang et al., 14 Mar 2026).

Metaheuristic and stochastic methods are prominent when geometry dominates the difficulty. Particle swarm optimization (PSO) is used for PASS-enabled multicast placement (Mu et al., 23 Feb 2025), robust multi-user placement under user-position uncertainty (Zeng et al., 16 Jul 2025), discrete multi-waveguide activation problems (Wang et al., 3 May 2025), CRB-based sensing geometry design (Wang et al., 21 May 2025), and high-dimensional PA placement in multicell weighted-sum-rate maximization (Chen et al., 11 Jun 2026). SCA and stochastic SCA also appear in ISAC and two-timescale PASS designs, respectively: the ISAC literature combines a fine-tuning approximation with SCA to couple communication beamforming and sensing constraints, while two-timescale downlink designs use primal-dual decomposition, KKT-guided dual learning, and SSCA to separate short-term transmit beamforming from long-term pinching-beamforming updates (Mao et al., 30 May 2025, Zhang et al., 13 Apr 2025).

Learning-based control has emerged as a second major line. A bipartite graph attention network (BGAT) models users and antennas as a complete bipartite graph and directly outputs feasible PA positions and power allocations for energy-efficiency maximization (Xie et al., 8 Feb 2025). Tutorial and survey papers also review GNNs, transformers, KKT-guided dual learning, mixture-of-experts models, and diffusion-style methods for PASS optimization and CSI acquisition, typically because the conventional optimization problems are multimodal, large-scale, and expensive to solve online (Liu et al., 11 Aug 2025, Zeng et al., 6 Jun 2025, Liu et al., 26 Jan 2026).

5. Performance characteristics and principal applications

A central analytical result is that PASS changes large-scale propagation rather than merely reshaping small-scale fading. Early analyses show that, for a single PA on a waveguide serving TDMA users, positioning the PA close to a user yields a larger ergodic sum rate than a conventional fixed antenna, and the gain grows with deployment area size in the high-SNR regime (Ding et al., 2024). The motivation is especially strong in high-frequency systems, because one article states that LoS links are typically about 100 times stronger than NLoS links, while outdoor microcellular results at 28 GHz show that NLoS path loss is more than 20 dB higher than LoS (Yang et al., 18 Jan 2025).

For array behavior, the literature rejects a monotonic “more is better” interpretation. In the array-gain analysis of PASS, fixing the inter-antenna spacing at $\lambda_g=\frac{\lambda_0}{\sqrt{\varepsilon_r}}.$2 leads to a closed-form upper bound whose asymptotics imply

$\lambda_g=\frac{\lambda_0}{\sqrt{\varepsilon_r}}.$3

This establishes that, under equal total transmit power and centered geometry, there exists an optimal number of active PAs rather than indefinite gain from growing $\lambda_g=\frac{\lambda_0}{\sqrt{\varepsilon_r}}.$4. The same paper shows that, once mutual coupling is included, there also exists an optimal inter-antenna spacing, numerically identified in the two-antenna analysis as $\lambda_g=\frac{\lambda_0}{\sqrt{\varepsilon_r}}.$5. The same result explicitly demonstrates that more antennas are not always better and smaller spacing is not always better (Ouyang et al., 10 Jan 2025).

In beamforming and power-efficiency studies, PASS can be markedly more power-efficient than fixed arrays. A physics-based PASS beamforming paper reports that PASS reduces transmit power by over 95% compared to conventional and massive MIMO, that discrete activation causes minimal performance loss but requires a dense antenna set to match continuous activation, and that the proportional power model yields performance comparable to the equal power model (Wang et al., 9 Feb 2025). Downlink multiuser MIMO studies report higher weighted sum-rate than fixed-location antenna systems (Bereyhi et al., 3 Feb 2025), while multicell PASS with joint precoding, power allocation, and placement outperforms average-power, fixed-placement, conventional MIMO, and massive MIMO baselines across tested settings (Chen et al., 11 Jun 2026).

Application-specific studies extend these advantages. PASS-enabled multicast on a single waveguide significantly outperforms conventional multiple-antenna transmission after PA-position optimization (Mu et al., 23 Feb 2025). Multi-waveguide ISAC shows a distinct communication-sensing trade-off and near-optimality relative to exhaustive search (Mao et al., 30 May 2025). Wireless sensing via PASS, combined with LCX-based echo reception, improves the average position error bound and robustness relative to conventional MIMO and fixed-PA baselines (Wang et al., 21 May 2025). Robust downlink design under uncertain user locations minimizes total power subject to outage constraints and outperforms fixed-antenna schemes (Zeng et al., 16 Jul 2025). A later multiuser study under random LoS blockage and NLoS scattering further shows that PASS retains gains beyond idealized LoS-only models and supports globally optimal bisection-based designs for average-SNR and outage-constrained objectives (Xu et al., 4 Dec 2025).

6. Limitations, misconceptions, and open research directions

The main misconception corrected by the literature is that PASS performance is monotonic in hardware scale. The array-gain analysis shows that finite, nontrivial values of antenna count and spacing maximize performance under realistic power-splitting and mutual-coupling considerations (Ouyang et al., 10 Jan 2025). A second misconception is that flexibility alone guarantees gains: multiple papers note that PASS with non-optimized or poorly optimized pinching positions can underperform conventional fixed antennas, so placement design is fundamental rather than cosmetic (Ouyang et al., 10 Jan 2025, Mu et al., 23 Feb 2025).

Open problems are correspondingly broad. Surveys and tutorials emphasize waveguide deployment and topology optimization, accurate electromagnetic modeling, channel estimation under compressed observations, uplink design, reciprocity and uplink reradiation, wideband and OFDM modeling with frequency-selective waveguide effects, robustness to user-position uncertainty, and real-time low-complexity control when mechanical repositioning is slow or costly (Yang et al., 18 Jan 2025, Zeng et al., 6 Jun 2025, Liu et al., 11 Aug 2025, Liu et al., 26 Jan 2026, Xu et al., 15 Oct 2025). The same sources repeatedly observe that much of the present analysis assumes negligible waveguide attenuation and LoS-dominant propagation; this suggests that bridging idealized PASS models with blockage-rich, frequency-selective, and fabrication-constrained deployments remains a central research task.

Future work is also structural. PASS variants such as SWAN, C-PASS, and M-PASS require new channel models and optimization frameworks; generalized implementations based on LCX, surface-wave guides, radio stripes, and other media expand the design space but also the burden of hardware characterization (Liu et al., 26 Jan 2026, Xu et al., 15 Oct 2025). Machine-learning-based operation is promising for scalability, yet the surveys identify unresolved questions in generalization, interpretability, and robustness (Zeng et al., 6 Jun 2025, Liu et al., 11 Aug 2025). Across this literature, the common conclusion is that pinching-antenna technology is not merely a new array geometry: it is a redefinition of antenna placement, radiation-site activation, and propagation shaping as joint physical-layer resources.