Flexible Pinching-Antenna Overview

• Flexible pinching-antennas are reconfigurable radiating elements formed by introducing a dielectric perturbation on a guided-wave structure.
• They enable dynamic control of radiation patterns and propagation geometry by creating, removing, or repositioning pinching points along waveguide routes.
• Optimization challenges include nonconvex design, mutual coupling effects, and trade-offs between array gain and positioning accuracy for communications, sensing, and power transfer.

A flexible pinching-antenna is a reconfigurable radiating point formed on a guided-wave structure—most commonly a dielectric waveguide—by locally perturbing the guide with a small dielectric particle, coupler, or analogous “pinch” so that part of the guided energy leaks into free space. In the broader pinching-antenna system (PASS) framework, the essential flexibility lies not only in radiation-pattern control but in the ability to create, remove, and reposition effective antenna ports over meter-scale or tens-of-meters-scale waveguide routes, thereby changing both large-scale path loss and propagation geometry while preserving a low-loss guided feed architecture (Yang et al., 18 Jan 2025, Liu et al., 26 Jan 2026).

1. Physical realization and the meaning of flexibility

In the baseline PASS formulation, RF energy is launched into a dielectric waveguide and radiates only where a pinching element is attached. The physical implementations described in the literature include small dielectric particles applied on or near the waveguide, short secondary waveguides operating as directional couplers, and mechanically movable or pre-deployed activatable pinching points (Xu et al., 18 Feb 2025, Liu et al., 26 Jan 2026). In a generalized low-frequency realization, the same principle is implemented with leaky coaxial cables (LCX), where controllable slots serve as pre-installed pinching points whose activation or deactivation determines where radiation occurs (Wang et al., 4 Dec 2025).

The flexibility of the concept has several distinct meanings. First, the waveguide itself is mechanically flexible in deployment: it can be routed along ceilings, walls, building facades, tunnels, or roadside infrastructure over macroscopic distances (Zhang et al., 11 Aug 2025, Mu et al., 23 Feb 2025). Second, the radiating ports are reconfigurable in space: pinching points can be created or removed, activated one at a time or in subsets, or approximated in practice by selecting among many pre-configured candidate positions (Zhang et al., 11 Aug 2025, Wang et al., 3 May 2025). Third, the number of effective antenna elements is itself reconfigurable, which distinguishes PASS from conventional arrays with fixed element counts and from fluid or movable antennas whose movement range is typically limited to a few wavelengths (Yang et al., 18 Jan 2025).

A recurring distinction in the literature is that PASS can alter large-scale propagation conditions rather than merely exploit small-scale fading. Existing flexible-antenna systems were characterized in early PASS work as being designed mainly for non-line-of-sight fading environments, whereas pinching antennas can place the radiating point close to the user and establish line-of-sight links that are described as typically 100 times stronger than NLoS links (Yang et al., 18 Jan 2025). This does not imply that all PASS deployments are automatically line-of-sight; rather, the architecture supplies a geometric degree of freedom that conventional fixed-position arrays do not possess.

2. Electromagnetic and geometric channel models

The canonical PASS channel model separates guided propagation inside the waveguide from free-space propagation between a pinching antenna and a receiver. A concise baseline expression writes the end-to-end scalar channel for a single PA as

$$h=\frac{\eta e^{-j 2\pi R/\lambda}}{R}\,\rho(\theta,\phi)\,\kappa\,e^{-j 2\pi L/\lambda_g},$$

where $R$ is the free-space PA–user distance, $L$ is the feed-to-PA distance along the waveguide, $\lambda_g$ is the guided wavelength, and $\rho(\theta,\phi)\kappa$ captures radiation and coupling (Liu et al., 26 Jan 2026). In single-user downlink formulations, the same structure appears as a coherent superposition of per-PA terms with distance-dependent amplitude $$1/\|\boldsymbol{\psi}_m-\widetilde{\boldsymbol{\psi}}_n\|$$, free-space phase, and in-waveguide phase $$\theta_n= \frac{2\pi}{\lambda_g}\|\widetilde{\boldsymbol{\psi}}_0-\widetilde{\boldsymbol{\psi}}_n\|$$ (Xu et al., 18 Feb 2025).

The resulting channel is inherently near-field and geometry dependent. In multicast PASS, the received SNR depends on

$$\gamma_k = \frac{P\eta}{M\sigma_0^2}\left|\sum_{m=1}^{M}\frac{e^{-j\frac{2\pi}{\lambda}d_{k,m}}\,e^{-j\frac{2\pi}{\lambda_g}(x_m+\frac{L}{2})}}{d_{k,m}}\right|^2,$$

so moving a PA changes both the amplitude and the phase of its field contribution through the free-space distance $d_{k,m}$ and the guided-path phase offset (Mu et al., 23 Feb 2025). In the indoor-positioning model, this same dual structure is made explicit through a deterministic AP–PA–user channel in which the AP–PA segment is guided and the PA–user segment is modeled as LoS with free-space loss and phase (Zhang et al., 11 Aug 2025).

A notable feature of PASS analysis is that more elements are not automatically better. For array gain, a closed-form upper bound under fixed half-wavelength spacing shows that the array gain tends to zero as $N\to\infty$, establishing the existence of an optimal finite number of pinching antennas (Ouyang et al., 10 Jan 2025). The same letter proves that there also exists an optimal inter-antenna spacing once mutual coupling is incorporated; for the two-antenna case it derives a coupling-aware approximation

$R$0

and reports an optimal normalized spacing $R$1 for $R$2 (Ouyang et al., 10 Jan 2025). This directly contradicts the common simplification that PASS performance should improve monotonically with denser or more numerous pinching points.

3. System architectures and PASS variants

PASS has evolved from the basic single-waveguide abstraction into several architectural families that trade flexibility, multiplexing capability, and implementation complexity (Liu et al., 26 Jan 2026).

Variant	Main mechanism	Reported role
Segmented PASS / SWAN	Multiple short segments, typically one active PA per segment	Avoids uplink IAR, improves maintainability
Center-fed PASS (C-PASS)	Center feed creates forward and backward propagation	Two DoFs per waveguide
Multi-mode PASS (M-PASS)	Multiple guided modes with mode-selective or mode-combining PAs	Mode-domain multiplexing
Generalized LCX-based PASS	Controllable LCX slots as generalized pinching points	Practical low-frequency implementation

Segmented PASS, also described as SWAN in the survey literature, replaces a single long waveguide with multiple short segments and can operate in segment-selection, segment-aggregation, or segment-multiplexing modes (Liu et al., 26 Jan 2026). This architecture is motivated by uplink inter-antenna radiation, waveguide-loss accumulation, and maintainability issues associated with very long continuous guides.

Center-fed PASS addresses the one-stream-per-waveguide limitation of conventional end-fed PASS by feeding the guide at its center and splitting energy into forward and backward directions. The survey characterizes this as yielding two degrees of freedom per waveguide and discusses power-splitting, direction-switching, and time-switching protocols for allocating resources between the two propagation directions (Liu et al., 26 Jan 2026).

Multi-mode PASS generalizes the architecture further by exciting multiple guided modes on the same waveguide and employing either mode-selective or mode-combining PAs. The survey describes this as enabling per-waveguide multiuser multiplexing in the mode domain, especially when guided modes such as TE$R$3 and TM$R$4 can be separated or intentionally combined (Liu et al., 26 Jan 2026).

A separate practical line of work treats continuous PA motion as an idealization and instead equips each waveguide with many discrete candidate pinching sites, activated through binary variables $R$5 (Wang et al., 3 May 2025). This discrete formulation is especially prominent in multi-waveguide NOMA models, where waveguide assignment, antenna activation, decoding order, and power allocation are optimized jointly. By contrast, the multicast model emphasizes that one waveguide carries one RF signal and is therefore naturally suited to a common-message service, with PA positions optimized for minimum-user SNR or multicast rate (Mu et al., 23 Feb 2025).

4. Communication-theoretic optimization and algorithmic design

Because PA positions affect both path losses and guided/free-space phases, PASS design problems are typically high-dimensional and nonconvex. This pattern appears across single-user rate maximization, multicast, NOMA, symbol-level precoding, multi-cell weighted sum-rate design, and energy-efficiency optimization (Xu et al., 18 Feb 2025, Chen et al., 11 Jun 2026).

In single-user downlink rate maximization, a relaxed formulation separates large-scale path-loss minimization from phase alignment. The resulting two-stage algorithm first centers a tightly packed array around the user projection and then refines positions on a wavelength scale to improve constructive combining; the paper reports nearly the same performance as the highly complex exhaustive search-based benchmark (Xu et al., 18 Feb 2025). In multicast, PA positions are optimized by particle swarm optimization (PSO) to maximize the minimum user SNR, and numerical results show that PASS can significantly outperform the conventional multiple-antenna transmission baseline (Mu et al., 23 Feb 2025).

For NOMA, one line of work studies a practical discrete multi-waveguide model in which waveguide assignment and antenna activation are cast as coalition-formation games, while power allocation is solved by monotonic optimization or successive convex approximation (SCA) (Wang et al., 3 May 2025). Another studies a two-user downlink QoS-constrained NOMA scenario on a single waveguide and proposes a block coordinate descent plus SCA method, while also deriving a closed-form global optimum for the single-PA special case (Xu et al., 18 Apr 2025). These works treat PASS not merely as a beamforming device but as a channel-engineering mechanism for controlling both gain disparity and interference structure.

More recent algorithmic developments integrate PASS with optimization-aware learning. A bipartite graph attention network (BGAT) models users and pinching antennas as a bipartite graph and jointly optimizes placement and power allocation for energy-efficiency maximization; the paper reports advantages in optimality, scalability, and computational efficiency (Xie et al., 8 Feb 2025). In two-timescale beamforming, pinching positions are optimized on a long timescale by stochastic successive convex approximation, while short-term transmit beamforming is learned by a KKT-guided dual-learning approach (Zhang et al., 13 Apr 2025). In symbol-level precoding, alternating optimization decouples conventional beamforming from PA position design, and a projected gradient descent procedure is used to update each PA position within a feasible movable interval, reducing transmit power under constructive-interference constraints (Pang et al., 14 Mar 2026). In multi-cell systems, weighted sum-rate maximization is handled by alternating optimization, fractional programming, block coordinate descent, and PSO, and the proposed scheme is reported to outperform average power allocation, fixed antenna placement, conventional MIMO, and massive MIMO baselines (Chen et al., 11 Jun 2026).

A consistent methodological theme is that PASS introduces a third design layer beyond conventional digital and analog beamforming: the geometry of the radiating points themselves. This suggests that many classical formulations in wireless communications remain relevant, but their variable sets must be extended to include waveguide-level and PA-level spatial design.

5. Localization, sensing, and wireless power transfer

PASS has been applied not only to communications but also to indoor positioning, wireless sensing, and SWIPT, where its geometric determinism and extended aperture are particularly valuable (Zhang et al., 11 Aug 2025, Wang et al., 21 May 2025, Zhang et al., 4 Sep 2025).

In indoor positioning, a single AP sequentially activates one PA per time slot, measures received power, converts each RSSI sample into a distance estimate, and then estimates the two-dimensional user position through a PASS-based weighted least-squares algorithm (Zhang et al., 11 Aug 2025). The paper reports three central observations: more PAs improve positioning accuracy and robustness; once the number of PAs exceeds a threshold the performance gain becomes marginal; and users located between and near PAs achieve superior positioning accuracy (Zhang et al., 11 Aug 2025). In the reported simulation setting, the room is $R$6 m, $R$7 GHz, and the improvement beyond about seven PAs becomes small.

For sensing, PASS has been combined with LCX reception to create a distributed transmit-receive architecture in which PAs illuminate targets and LCX cables collect echoes over a wide area (Wang et al., 21 May 2025). The paper derives the multi-target Cramér–Rao bound, formulates a joint optimization of waveform covariance and PA positions, and proposes a two-stage PSO-based algorithm. Numerical results are reported to show significant gains in sensing accuracy and robustness over conventional sensing systems (Wang et al., 21 May 2025). The sensing interpretation is straightforward: moving PAs changes target-dependent distances and phases, which directly reshapes the Fisher information.

In SWIPT, a single flexible PA is positioned along one of three waveguide layouts—edge deployment, center deployment, or diagonal deployment—and the resulting average harvested energy and achievable rate are characterized under a hybrid time-switching and power-splitting protocol (Zhang et al., 4 Sep 2025). The paper derives closed-form expressions for linear energy harvesting and an upper bound for a nonlinear logistic model. It also reports that, in a square region, diagonal deployment can yield the largest rate–energy region, whereas in rectangular settings the relative ordering depends on geometry (Zhang et al., 4 Sep 2025). This supports a broader PASS design principle: waveguide layout is itself a system variable, not merely an installation detail.

6. Limitations, misconceptions, and open research problems

A common misconception is that PASS is simply another distributed phased array. The literature repeatedly distinguishes it from fixed arrays, RIS, and wavelength-scale movable antennas by emphasizing that PASS reconfigures physical radiation locations, can scale the number of radiators by adding or removing pinching points, and can transport energy through a low-loss guided medium before radiating it near the user or target (Yang et al., 18 Jan 2025, Liu et al., 26 Jan 2026). Another misconception is that increasing the number of PAs always improves performance; array-gain analysis and positioning experiments both show threshold effects and non-monotonic behavior (Ouyang et al., 10 Jan 2025, Zhang et al., 11 Aug 2025).

At the same time, much of the current PASS literature relies on simplifying assumptions. Many communication papers assume perfect CSI, ideal or negligible in-waveguide loss, static users during optimization, deterministic LoS propagation, equal power splitting among active PAs, and simplified handling of mutual coupling via minimum spacing constraints (Mu et al., 23 Feb 2025, Xu et al., 18 Feb 2025, Wang et al., 3 May 2025). Practical deployment papers explicitly note that arbitrary continuous PA placement is difficult to realize, motivating discrete candidate positions, on/off activation models, or slower-timescale reconfiguration (Wang et al., 3 May 2025, Zhang et al., 13 Apr 2025). The survey additionally identifies reflections, reciprocity, inter-antenna radiation in uplink, bandwidth limitations, dispersion, and actuator speed as central implementation issues (Liu et al., 26 Jan 2026).

Open problems identified across the literature include waveguide deployment and topology optimization, channel estimation and beam training for geometry-changing arrays, wideband and dispersive modeling, robust design under imperfect CSI, multi-cell coordination, hardware-aware modeling of coupling and losses, and scalable optimization or learning for very large PASS deployments (Yang et al., 18 Jan 2025, Liu et al., 26 Jan 2026). The emergence of generalized LCX-based PASS at low frequencies broadens the hardware scope of the field, but it also raises new questions about slot control, interference localization, and frequency-dependent propagation in guided media (Wang et al., 4 Dec 2025).

Flexible pinching-antennas are therefore best understood as a reconfigurable guided-wave antenna paradigm in which the primary optimization object is not merely a complex weight vector, but the spatial realization of the radiating interface itself. The central research challenge is to convert that geometric flexibility into robust, hardware-realizable gains in communications, sensing, and wireless power transfer without relying on idealized assumptions that erase the very implementation difficulties the architecture introduces.