Pinching Antenna Systems (PASSs)

• Pinching Antenna Systems (PASSs) are flexible antenna architectures that use dielectric waveguides with locally perturbative radiating elements to implement reconfigurable beamforming.
• They employ pinching beamforming by dynamically adjusting the positions and activation patterns of pinching antennas, thus altering the effective aperture without traditional phase shifters.
• PASSs find applications in MIMO, multicast, indoor positioning, sensing, and ISAC, with research demonstrating optimal PA spacing and enhanced performance over conventional arrays.

Pinching Antenna Systems (PASSs) are flexible-antenna architectures in which RF signals are conveyed by dielectric waveguides and radiated into free space by pinching antennas (PAs), namely dielectric particles or local perturbations attached along the waveguide. Their defining mechanism is pinching beamforming: the effective aperture is reconfigured by changing PA positions and activation patterns, rather than relying only on fixed-array weights or phase-shifter networks. Recent work studies PASSs as single-waveguide and multi-waveguide systems for uplink, downlink, multicast, MIMO, positioning, sensing, integrated sensing and communications (ISAC), UAV support, and over-the-air federated learning, typically exploiting short-distance, strong line-of-sight links and near-field geometry (Liu et al., 30 Jan 2025, Bereyhi et al., 5 Mar 2025).

1. Physical basis and channel geometry

A PA is formed on a dielectric waveguide by locally perturbing it with a small dielectric particle or mechanical deformation. The waveguide carries the signal over a long, low-loss path; each pinching point couples part of the guided mode into free space and behaves as a radiating element. In the basic PASS abstraction, the contribution of the $n$-th PA to a user is written as

$$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$

where $r_n$ is the free-space distance from PA to user, $d_n$ is the guided distance from the feed point to the PA, and $n_{\mathrm{eff}}$ is the effective refractive index of the guided mode. This form makes the PASS channel explicitly geometry-dependent in both amplitude and phase (Liu et al., 30 Jan 2025).

A more detailed MIMO-PASS model writes the effective scalar channel from waveguide $m$ to user $k$ as a coherent sum over pinching elements,

$$g_{k,m}(l_m)=\xi \alpha_k \sum_{n=1}^{N}\frac{\exp\!\left(-j\kappa\big(D_{k,m}(\ell_{m,n})+i_{\mathrm{ref}}\ell_{m,n}\big)\right)}{\sqrt{N}\,D_{k,m}(\ell_{m,n})},$$

where $D_{k,m}(\ell)$ is the distance between the user and the element at location $\ell$, and the term $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$0 captures guided-wave phase accumulation. This is a near-field, geometry-dependent channel in which moving $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$1 changes both path loss and phase (Bereyhi et al., 5 Mar 2025).

Two modeling conventions coexist in the literature. Many communication papers adopt negligible in-waveguide attenuation and retain only phase accumulation, which is justified by the low attenuation of dielectric waveguides relative to free-space propagation. By contrast, PASS-based indoor positioning explicitly models waveguide attenuation through

$$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$2

so that the AP–PA path contributes both amplitude decay and phase rotation before free-space propagation between PA and user (Mu et al., 23 Feb 2025, Zhang et al., 11 Aug 2025). This contrast is not a contradiction; it reflects different operating assumptions and performance metrics.

The physical significance of PASS lies in the replacement of a long free-space hop by a long guided hop plus a short radiating hop. The architecture survey frames this as wireless communications over the “last meter,” while other works emphasize “meter-scale reconfigurability,” “macroscopic port mobility,” and the ability to create short-distance, strong line-of-sight links that are not achievable with wavelength-scale rigid arrays (Liu et al., 30 Jan 2025, Zhang et al., 11 Aug 2025).

2. Architectural families and transmission structures

PASS research distinguishes several architectural families according to how many waveguides are used, how many RF chains feed them, and whether the design supports one stream or multiple streams.

Family	Characterization	Representative use
Non-multiplexing architecture	Simple baseband signal processing; relies on pinching beamforming only	Coverage-focused links (Liu et al., 30 Jan 2025)
Multiplexing architecture	Joint baseband and pinching beamforming	Multiuser transmission (Liu et al., 30 Jan 2025)
Sub-connected	Each RF chain feeds one waveguide	Hybrid MIMO-like PASS (Liu et al., 30 Jan 2025)
Fully-connected	Each RF chain connects to all waveguides	Enhanced spatial DoF (Liu et al., 30 Jan 2025)
PS-based fully-connected	Fully-connected plus phase shifters	Tri-hybrid beamforming (Liu et al., 30 Jan 2025)
Waveguide multiplexing (WM)	All waveguides jointly serve all groups	Multicast-oriented (Zhao et al., 20 Aug 2025)
Waveguide division (WD)	One stream per waveguide	Lower-complexity multi-group transmission (Zhao et al., 20 Aug 2025)
Waveguide switching (WS)	Time-domain orthogonalization across groups	Unicast-oriented (Zhao et al., 20 Aug 2025)

In single-waveguide settings, all PAs on a waveguide naturally radiate the same signal stream. This makes multicast a particularly natural fit, while multiple independent data streams may require multiple waveguides or more sophisticated RF feeding structures (Mu et al., 23 Feb 2025). The architecture survey accordingly separates non-multiplexing designs, where a waveguide is primarily a physically reconfigurable one-stream aperture, from multiplexing designs, where PASS is combined with digital beamforming and, in the PS-based fully-connected variant, analog phase shifting (Liu et al., 30 Jan 2025).

Multi-waveguide PASS generalizes the aperture in a second spatial dimension. In MIMO-PASS, multiple parallel waveguides are fed at one end and host movable pinching elements subject to waveguide length and minimum-spacing constraints. In PASS-enabled multi-user communications, the WM, WD, and WS structures formalize three different ways to map data streams, power allocation, and PA positions onto the same physical infrastructure; the reported design conclusion is that WS and WM are suitable for unicast and multicast communications, respectively, while the performance gap between WD and WM can be significantly alleviated when the users are geographically isolated (Bereyhi et al., 5 Mar 2025, Zhao et al., 20 Aug 2025).

Separated transmit/receive architectures also appear. PASS-ISAC uses two waveguides, one for the information-bearing ISAC transmission and the other for receiving reflected echoes, while PASS-based sensing combines dielectric-waveguide transmission with LCX-based reception. In both cases, the architectural point is that the waveguide does not merely replace a feed network; it is the medium through which the aperture is physically redistributed across space (Zhang et al., 10 Apr 2025, Wang et al., 21 May 2025).

3. Pinching beamforming and optimization methodologies

“Pinching beamforming” denotes the design of PA positions and activation patterns, optionally together with digital beamforming variables, power allocation, scheduling, or waveform design. The resulting problems are typically continuous-discrete, geometry-coupled, and highly non-convex because the objective depends on distances, guided phases, and free-space phases simultaneously.

In multicast communications over a single waveguide, PASS-enabled multicast communication optimizes PA positions to maximize the minimum user SNR. One formulation uses particle swarm optimization (PSO) directly on the PA position vector under waveguide-span and minimum-separation constraints (Mu et al., 23 Feb 2025). A later multicast treatment derives a closed-form single-PA solution for linearly distributed users, then proposes an element-wise alternating optimization algorithm for multiple PAs and arbitrary user distributions; in the multiple-waveguide case, it combines majorization-minimization (MM) with second-order cone programming (SOCP) for transmit beamforming and sequential refinement for pinching beamforming (Shan et al., 31 May 2025).

MIMO-PASS formulates weighted sum-rate maximization in both downlink and uplink. The downlink method combines fractional programming, an outer block coordinate descent between digital precoding and PA locations, and an inner Gauss–Seidel update over individual pinching-element positions; a lower-complexity zero-forcing alternative is also given. The uplink counterpart uses MMSE detection with location optimization, again via Gauss–Seidel and one-dimensional search over feasible positions (Bereyhi et al., 5 Mar 2025).

For multi-group multicast with the WM, WD, and WS structures, PASS-enabled multi-user communications uses penalty dual decomposition (PDD). The method introduces auxiliary variables to decouple the complex exponential and fractional couplings, applies an augmented Lagrangian relaxation, and then alternates over beamforming, PA-position, and auxiliary-variable blocks; in the WS unicast case, a lower-complexity method exploits phase alignment and path-loss minimization (Zhao et al., 20 Aug 2025).

Beyond data transmission, optimization formulations diversify. PASS-based indoor positioning estimates user–PA distances from RSSI and then applies a weighted least squares (WLS) solver for two-dimensional coordinates (Zhang et al., 11 Aug 2025). Wireless sensing via PASS derives the Cramér–Rao bound (CRB) for multi-target sensing and minimizes it through a two-stage PSO-based algorithm together with a convex waveform-design stage (Wang et al., 21 May 2025). PASS-ISAC uses a penalty-based alternating optimization algorithm to maximize target illumination power while satisfying a communication quality-of-service constraint (Zhang et al., 10 Apr 2025). PASS-enabled UAV delivery employs a double-layer optimization structure with hierarchical alternating optimization (HAO) for delivery-sequence planning and either Branch-and-Bound (BnB) or incremental search and local refinement (ISLR) for PA activation (Lv et al., 30 Sep 2025). Energy-efficient OTA-FL via PASS jointly tunes PASS parameters and device scheduling for minimal energy consumption during over-the-air aggregation (Asaad et al., 15 Feb 2026). In blockage-aware multicast, the non-LoS formulation is handled by MM, with convex surrogate subproblems solved by either a candidate search method (CSM) or a bisection search method (BSM) (Hanif et al., 7 Feb 2026).

Across these formulations, the recurring pattern is that PASS adds a slow-timescale geometric control layer to the usual fast-timescale precoding layer. The literature treats that geometric layer as a first-class optimization variable, not as a secondary implementation detail.

4. Communication-theoretic properties and empirical trends

A central finding is that PASS behavior does not follow the monotonic rules familiar from conventional fixed arrays. For array gain, a closed-form upper bound under half-wavelength spacing shows that, as the number of antennas grows without bound, the array gain tends to zero: $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$3 The same work proves the existence of an optimal number of antennas and an optimal inter-antenna spacing, and in one numerical example at $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$4 GHz and $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$5 m the optimal number is $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$6, while in the two-antenna coupled case the optimal spacing is $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$7 (Ouyang et al., 10 Jan 2025). This directly counters the common intuition that denser or larger PASS apertures are always better.

Uplink analysis reaches a related conclusion from a different angle. “On the Performance of Uplink Pinching Antenna Systems (PASS)” studies three scenarios—multiple PAs for a single user (MPSU), a single PA for a single user (SPSU), and a single PA for multiple users (SPMU)—and shows that the proposed PASS significantly outperforms conventional Multiple-input Single-output networks, that the PA distribution follows an asymmetric non-uniform distribution in the MPSU scenario, and that optimizing PA positions significantly enhances the ergodic sum rate performance (Hou et al., 17 Feb 2025). In the far zone, the optimal structure approaches uniform spacing; in the near zone, it is asymmetric and non-uniform.

For multicast, two distinct results are notable. First, the single-waveguide analysis proves that, in the high-SNR regime, the average multicast rate achieved by PASS employing a single PA is strictly higher than that of a conventional fixed-location antenna system, and that the corresponding multicast rate gain increases monotonically with $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$8 (Shan et al., 31 May 2025). Second, PASS-enabled multicast communication reports that PASS can significantly outperform the conventional multiple-antenna transmission when multiple PAs on a single waveguide radiate the broadcast signal to multiple users (Mu et al., 23 Feb 2025).

In multiuser MIMO, MIMO-PASS reports weighted-sum-rate gains greater than $$y_n = \frac{\beta_n\sqrt{P_n}}{r_n}\,e^{-j\frac{2\pi}{\lambda}\left(r_n+n_{\mathrm{eff}}d_n\right)}x,$$9 over massive MIMO and greater than $r_n$0 over small-scale MIMO in the evaluated downlink settings, together with the qualification that when $r_n$1 the weighted sum-rate of PASS decays fast and fully digital massive MIMO can eventually outperform PASS (Bereyhi et al., 5 Mar 2025). PASS-enabled multi-user communications gives a complementary max–min fairness view: for a multicast setup with WM, $r_n$2, $r_n$3, and $r_n$4 dBm, the reported minimum rate is approximately $r_n$5 bps/Hz for PASS, versus approximately $r_n$6 bps/Hz for conventional fully-digital MIMO and approximately $r_n$7 bps/Hz for hybrid MIMO; the same study concludes that WS is preferable for unicast, WM for multicast, and WD becomes competitive when user groups are geographically isolated (Zhao et al., 20 Aug 2025).

Recent work also weakens the pure-LoS assumption. “Multicasting Pinching Antenna Systems With LoS Blockage” models a Bernoulli LoS indicator with distance-dependent LoS probability, optimizes PA positions under this blockage model, and reports superior performance of multicasting PASS in non-LoS environments compared to conventional antenna systems. It also shows that, for $r_n$8 PAs and $r_n$9 users, the execution time with the CSM is approximately $d_n$0 times that with the BSM (Hanif et al., 7 Feb 2026). The significance is methodological as much as numerical: PASS optimization is moving from idealized LoS settings to blockage-aware environments.

5. Localization, sensing, ISAC, and related functions

PASS has been extended from communications to positioning and sensing largely because the architecture exposes a deterministic geometry. PASS-based indoor positioning formulates an uplink model with one PA active per time slot, estimates user–PA distances from received signal strength indication, and recovers user coordinates through a PASS-based weighted least squares algorithm. The reported observations are that more PAs on the waveguide improve the positioning accuracy and robustness, that the performance gain becomes marginal when the number of PAs exceeds a threshold, and that user locations between and near PAs yield superior positioning accuracy; in one Monte Carlo comparison, $d_n$1 gives mean error approximately $d_n$2 m and variance approximately $d_n$3, versus mean error approximately $d_n$4 m and variance approximately $d_n$5 for $d_n$6 (Zhang et al., 11 Aug 2025).

Wireless sensing via PASS proposes a hybrid architecture with dielectric-waveguide transmission and LCX-based reception. The paper derives the CRB for multi-target sensing, then minimizes it through the joint optimization of the transmit waveform and PA positions via a two-stage PSO-based algorithm. The reported outcome is significant gains in sensing accuracy and robustness over conventional sensing systems, attributed to the combination of optimized PASS transmission and uniform echo collection by LCX cables over a wide area (Wang et al., 21 May 2025).

PASS-ISAC develops a separated ISAC design for the two-waveguide PASS, where one waveguide emits the information-bearing ISAC signal and the other receives the reflected echoes. Its optimization objective is to maximize illumination power while ensuring a communication quality-of-service requirement, and the reported numerical result is that the proposed PASS-ISAC scheme outperforms the conventional antenna scheme (Zhang et al., 10 Apr 2025). The broader implication is that PASS naturally supports functional separation at the waveguide level, not only at the signal-processing level.

Other extensions make the same point in application-specific language. PASS-enabled UAV delivery jointly optimizes the UAV delivery sequence and the PA activation vector to minimize communication energy consumption in one cycle; the abstract reports that PASS outperforms the conventional multi-antenna systems, especially with higher communication rate requirements (Lv et al., 30 Sep 2025). Energy-efficient over-the-air federated learning via PASS studies a PASS-assisted server for OTA-FL and reports that, with a single-waveguide PASS in a moderately sized area, the required energy for model aggregation is drastically reduced relative to a fully-digital MIMO server (Asaad et al., 15 Feb 2026). These use cases differ in algorithmic detail, but they share the same architectural advantage: guided transport plus spatially reconfigurable radiation points.

6. Limitations, misconceptions, and open problems

The literature repeatedly cautions against several oversimplified readings of PASS. The first is the idea that “more PAs are always better.” Array-gain analysis proves the existence of an optimal number of antennas and an optimal inter-antenna spacing, and indoor positioning shows explicit saturation: more PAs improve accuracy and robustness at first, but beyond about $d_n$7 to $d_n$8 PAs in the simulated room the additional gain is marginal (Ouyang et al., 10 Jan 2025, Zhang et al., 11 Aug 2025). The second is that PASS is merely a low-cost substitute for massive MIMO. In some multiuser MIMO regimes, especially when the number of users exceeds the number of RF chains or waveguides, fully digital massive MIMO can outperform PASS (Bereyhi et al., 5 Mar 2025).

A second cluster of limitations concerns modeling assumptions. Multiple works assume LoS-dominant propagation, perfect CSI or perfect user-location knowledge, negligible waveguide attenuation, narrowband signaling, and minimum-spacing constraints as a stand-in for a full mutual-coupling model (Hou et al., 17 Feb 2025, Liu et al., 30 Jan 2025). Indoor positioning explicitly notes that the current model neglects significant multipath or NLoS effects common indoors, and sensing formulations rely on idealized propagation and calibration assumptions (Zhang et al., 11 Aug 2025, Wang et al., 21 May 2025). The recent blockage-aware multicast study shows that the field is beginning to relax the LoS-only abstraction, but that extension is not yet universal (Hanif et al., 7 Feb 2026).

A third issue is implementation. The architecture survey, MIMO-PASS, and several application papers all identify hardware and control challenges: mechanical sliders, detachable clips, or micro-actuators for moving pinching elements; activation and reconfiguration latency; waveguide deployment and alignment; practical dielectric loss and bending constraints; frequency selectivity; and the control overhead required for large numbers of movable or activatable elements (Liu et al., 30 Jan 2025, Bereyhi et al., 5 Mar 2025). This suggests that PASS should be understood less as a drop-in antenna replacement than as a new class of reconfigurable infrastructure.

The open-problem list is correspondingly broad. Explicit directions appearing across the papers include robust design under imperfect CSI, wider study of NLoS and multipath-rich environments, dynamic user motion, joint optimization of PA locations and waveguide layouts, wideband and OFDM modeling, multi-waveguide and multi-user scheduling, hardware prototyping, and machine-learning-assisted configuration design (Liu et al., 30 Jan 2025, Zhang et al., 11 Aug 2025, Bereyhi et al., 5 Mar 2025). A plausible implication is that the long-term importance of PASS will depend not only on raw rate or accuracy gains, but on how effectively these geometric degrees of freedom can be integrated with estimation, control, and practical waveguide hardware.