PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey of Pinching-Antenna Systems (PASS)

Yuanwei Liu, Hao Jiang, Xu Gan, Xiaoxia Xu, Jia Guo, Zhaolin Wang, Chongjun Ouyang, Xidong Mu, Zhiguo Ding, Arumugam Nallanathan, Octavia A. Dobre, George K. Karagiannidis, Robert Schober

arXiv:2601.18927v2eess.SP

TL;DR

Next-generation wireless networks need flexible architectures that address propagation conditions, scalability, and sensing-communication demands. This survey organizes PASS foundations, variants, sensing applications, performance analysis, optimization, and machine learning, while identifying practical modeling and implementation challenges. It concludes that PASS offers large-scale channel reconfiguration and related communication and sensing benefits, but hardware-aware modeling and optimization remain open issues.

  • Problem

    Existing flexible-antenna approaches face limited wavelength-scale flexibility or high deployment costs, while PASS introduces distinct propagation and hardware challenges requiring systematic study.

  • Method

    The paper surveys PASS hardware and signal foundations, emerging designs, sensing applications, communication and sensing performance analysis, optimization, and machine learning.

  • Results

    The survey reports PASS benefits including channel-gain enhancement, reduced inter-user interference, improved achievable rate and outage performance, and learning-based optimization with lower complexity or millisecond-level response in studied settings.

  • Takeaways & Limitations

    PASS provides a flexible architecture for large- and small-scale channel reconfiguration, LoS creation, near-field benefits, and scalable implementation across communication and sensing applications.

  • Takeaways & Limitations

    Practical PASS models and optimizations must account for reflections, mutual coupling, calibration dependence, finite positioning resolution, spacing, coupling ranges, errors, and reconfiguration time.

Abstract

from arXiv · show

The pinching-antenna system (PASS), recently proposed as a flexible-antenna technology, has been regarded as a promising solution for several challenges in next-generation wireless networks. It provides large-scale antenna reconfiguration, establishes stable line-of-sight links, mitigates signal blockage, and exploits near-field advantages through its distinctive architecture. This article aims to present a comprehensive overview of the state of the art in PASS. The fundamental principles of PASS are first discussed, including its hardware architecture, circuit and physical models, and signal models. Several emerging PASS designs, such as segmented waveguide-enabled PASS (SWAN), center-fed PASS (C-PASS), and multi-mode PASS (M-PASS), are subsequently introduced, and their design features are discussed. In addition, the properties and promising applications of PASS for wireless sensing are reviewed. On this basis, recent progress in the performance analysis of PASS for both communications and sensing is surveyed, and the performance gains achieved by PASS are highlighted. Existing research contributions in optimization and machine learning are also summarized, with the practical challenges of beamforming and resource allocation being identified in relation to the unique transmission structure and propagation characteristics of PASS. Finally, several variants of PASS are presented, and key implementation challenges that remain open for future study are discussed.

I. INTRODUCTION

PASS extends flexible-antenna concepts from wavelength-scale repositioning toward meter-scale channel reconfiguration, using low-attenuation waveguides and repositionable pinching antennas. The introduction situates PASS within MIMO’s evolution toward larger, denser, and more flexible architectures while emphasizing its potential for LoS creation, near-field benefits, and scalable implementation.

  • PASS architecture: PASS uses tens-of-meters low-attenuation waveguides with repositionable pinching antennas to proactively reconfigure wireless channels.Its architecture supports meter-scale PA positioning and integrates guided propagation with short-range free-space transmission.
  • PASS benefits: PASS is positioned as an outer layer of tri-hybrid beamforming that provides LoS creation, near-field benefits, and low-cost, scalable implementation.Low-cost waveguides and attachable or detachable PAs support scalability compared with other flexible-antenna forms.
  • MIMO trends: MIMO is evolving toward larger apertures, denser antenna placement, and more flexible antenna architectures to meet growing wireless-network demands.Larger arrays improve beamforming and multiplexing, while denser apertures enable continuous source-current control and near-field focusing.
  • Implementation challenges: Practical implementation of larger and denser arrays remains challenging because optimizing many beamforming variables can impose substantial signal-processing overhead.Flexible antennas are introduced partly to provide channel reconfigurability without relying solely on increasingly large numbers of electronically controlled elements.
  • Flexible antennas: Conventional flexible antennas typically reposition elements over only a few wavelengths, limiting their ability to manipulate large-scale fading in LoS-dominated high-frequency channels.In such channels, spatial diversity may be unavailable because channel conditions remain relatively flat across the array.
  • PASS benefits: PASS addresses deployment limitations of flexible antennas by supporting large- and small-scale channel reconfiguration while allowing antenna additions or removals at lower cost.This contrasts with systems whose large-scale flexibility is costly or whose deployed antenna and port changes are difficult to realize.

C. Motivation and Contribution

This survey provides an integrated review of PASS, covering its fundamentals, advanced architectures, sensing, performance analysis, optimization, and machine learning. It also positions the work against existing PASS literature and outlines the paper’s organization.

  • Scope and positioning: The paper presents a comprehensive review of PASS and compares its coverage with existing magazine, survey, and tutorial papers.Table II provides the comparison, using symbols for high-level overviews, detailed tutorials, and topics mentioned without detailed consideration.
  • Fundamentals: The survey develops PASS fundamentals through implementation principles, directional-coupler waveguide modeling, multiport theory, and PASS signal modeling.These foundations support characterization of guided-signal electromagnetic behavior and the subsequent signal-model discussion.
  • Advanced architectures: It introduces advanced PASS architectures, including SWAN, C-PASS, and M-PASS, with distinct waveguide segmentation, signal-routing, and multimode propagation features.SWAN uses shorter segments and dedicated feeds; C-PASS uses a T-junction splitter; M-PASS supports multiple modes.
  • Sensing: The survey reviews PASS-assisted sensing across sensing, localization, user tracking, and the interaction between sensing and communication.It covers both sensing-aided communications and communication for sensing.
  • Performance analysis: It categorizes PASS performance analyses by achievable rate and outage probability for communications, and by sensing metrics for sensing.The survey organizes existing studies according to the performance measures they employ.
  • Optimization and machine learning: It surveys structure-based, population-based, and game-theoretic optimization, while highlighting machine learning as an approach for PASS optimization.The optimization discussion addresses mathematical tools for PASS’s intrinsic challenges.

A. PASS Signal and Channel Models

PASS models transmission as a two-stage process: signal propagation inside the waveguide followed by free-space propagation from each PA to the user. PA positions jointly determine in-waveguide and free-space distances, enabling phase alignment and coherent signal combining.

  • Signal propagation: PASS decomposes signal propagation into in-waveguide propagation and free-space propagation linked by the PA.The waveguide stage includes attenuation and phase rotation before the PA radiates toward the user.
  • Multiple-PA channel model: For multiple PAs, cascaded coupling accounts for each PA’s extraction and the signal transmitted through preceding PAs.The effective coupling factor combines the n-th PA coupling with cumulative transmission along the waveguide.
  • Single-PA channel model: The end-to-end channel includes free-space path loss, PA radiation, coupling, and in-waveguide phase shift.The radiation pattern ρ(θ, φ) and coupling factor κ characterize the PA contribution.
  • Received signal: The received signal is the superposition of individual PA channel contributions plus additive noise.Each contribution depends on the user’s direction relative to its PA.
  • Pinching beamforming: PA positions control both free-space distances R_n and in-waveguide distances L_n, allowing phase alignment for coherent summation and strong overall channel gain.This position-based strategy is referred to as pinching beamforming.

B. PA Implementation and Hardware Models

PA implementations extract and radiate guided energy through interactions with the waveguide’s evanescent field, with models ranging from simple dielectric scattering to coupled-waveguide designs. Multiport network models simplify system analysis but sacrifice detailed physical insight into PA behavior.

  • Physical principle: A PA extracts guided energy by introducing a secondary dielectric element into the main waveguide’s evanescent-field region and radiating the coupled wave.The radiation and coupling factors depend on the PA’s physical implementation.
  • Small Dielectric Scatterer (SDS) Model: The small dielectric scatterer model is mechanically simple and inexpensive but offers low power efficiency and limited controllability.Its electrically small scatterer interacts with only a small fraction of the evanescent field, producing inherently low coupling.
  • Dielectric-Coupled-Waveguide (DCW) Model: The dielectric-coupled-waveguide model uses a secondary short waveguide to enable stronger and more controllable power extraction through coupled-mode interaction.Energy can either leak directly from the secondary element or be guided to a dedicated radiator.
  • Multiport network model: Multiport network theory abstracts a PA as a lumped three-port network using measurable or simulatable scattering parameters.S21 represents remaining main-waveguide transmission, S31 represents radiation and coupling, and S11 represents reflection.
  • Multiport network model: The multiport model supports system-level analysis but provides limited guidance about coupling strength, radiation pattern, and polarization.Its simplicity comes at the cost of physical insight into the PA.

D. Discussion and Outlook

PASS research has progressed from foundational modeling toward segmented architectures, but practical deployment remains constrained by hardware fidelity, wideband effects, actuator limits, and non-convex optimization. SWAN addresses long-waveguide loss and maintainability while supporting more tractable uplink modeling.

  • Discussion and Outlook: Large-scale PASS deployment still requires hardware characterization, calibrated channel modeling, and system-level multi-user validation beyond existing prototype evidence.
  • Hardware Limitations: SDS coupling is typically inefficient and difficult to calibrate because scattered fields depend strongly on scatterer geometry, materials, and contact conditions.
  • Hardware Limitations: Physical PA movement introduces latency, positioning error, mechanical wear, and tolerance constraints that may limit tracking of fast user movement.
  • Modeling Limitations: Common system models neglect reflections, mutual coupling, and inter-PA radiation, making hardware-aware modeling dependent on full-wave simulation, calibration, or multiport characterization.
  • Optimization Limitations: PASS optimization is highly non-convex because PA positions and coupling states interact with propagation, reflections, coupling, actuators, and deployment constraints.
  • SWAN Outlook: SWAN uses short, independently fed waveguide segments to reduce path loss, localize failures, and support tractable multi-PA uplink modeling.

2) Operating Protocols of SWAN:

SWAN offers three operating protocols that trade hardware complexity against beamforming capability, and it can extend these choices through tri-hybrid digital, analog, and EM processing. The protocols range from low-complexity segment selection to multi-RF-chain processing that approaches the SWAN performance upper bound.

  • Operating Protocols: Segment selection connects one segment to an RF chain at a time, providing straightforward implementation with very low hardware complexity.
  • Operating Protocols: Segment aggregation connects all feed points through a power splitter, allowing simultaneous segment contribution but requiring more complex RF hardware.
  • Operating Protocols: Segment multiplexing assigns each segment a dedicated RF chain, enabling MRC and MRT while requiring substantially more RF chains and baseband processing.
  • Tri-Hybrid Beamforming Structures: Tri-hybrid beamforming combines digital, analog, and EM processing, jointly optimizing their weights to exploit SWAN’s spatial degrees of freedom.
  • Tri-Hybrid Beamforming Structures: Phase-shifter and switching architectures support different complexity-flexibility trade-offs, while dynamic designs select active segments before refining their analog weights.
  • Tri-Hybrid Beamforming Structures: With multiple RF chains, segments may connect to every chain for maximum spatial flexibility or be partitioned into sub-segment-arrays to reduce complexity.

B. Center-Fed PASS (C-PASS)

Conventional end-fed PASS is limited to one degree of freedom because its radiated signals are phase-shifted replicas from a single source. C-PASS addresses this through bidirectional center feeding, while its protocols trade flexibility and multiplexing capability against implementation complexity and synchronization demands.

  • Motivation: Conventional end-fed PASS restricts communication to DoF = 1 and prevents independent multi-stream transmission.
  • C-PASS Architecture: C-PASS bifurcates signals into forward and backward propagation, generating two independent data streams and extending to M degrees of freedom with distributed center feeding.
  • Operating Protocols: C-PASS supports power splitting, direction switching, and time switching as practical operating protocols.
  • Operating Protocols: Power splitting simultaneously serves both directions and offers the highest design DoF, but requires precise tuning at every splitter.
  • Operating Protocols: Direction switching assigns feed ports to forward or backward propagation using binary splitting states, reducing degrees of freedom while simplifying implementation.
  • Operating Protocols: Time switching decouples forward and backward beamforming and lowers design complexity, but requires stringent synchronization between direction-allocated periods.

C. Multi-Mode PASS (M-PASS)

M-PASS exploits multiple guided modes in a dielectric waveguide to increase spatial degrees of freedom and support multi-user transmission over a single waveguide. Its mode-selection and mode-combining protocols trade implementation simplicity for coupling flexibility and adaptive beam shaping.

  • M-PASS architecture: M-PASS uses multiple guided modes with distinct propagation constants to create a full-rank mode-domain channel over one waveguide.The resulting modal diversity increases per-waveguide spatial degrees of freedom for multi-user communications.
  • M-PASS architecture: A coupled-mode theory model characterizes radiation from multiple guided modes into single-mode pinching antennas.The model is represented by M + 1 coupled-mode equations under the stated implementation assumption.
  • Operating protocols: Mode selection phase-matches each pinching antenna to one guided mode, suppressing nonselected modes through phase mismatch.Grouping pinching antennas by selected mode provides a practical implementation, with leakage depending on the operating regime.
  • Operating protocols: Mode combining continuously tunes each pinching antenna's propagation constant so it can extract power from multiple guided modes.This makes each antenna a programmable multi-mode coupler, but requires joint optimization of beamforming, antenna positions, and propagation constants.
  • Operating protocols: Mode-selective structures suit separable modes, static environments, and sparse users, whereas mode-combining structures suit dense or rapidly changing deployments.The latter supports dynamic modal configuration and adaptive beam shaping but has greater optimization demands.

D. Discussion and Outlook

PASS variants broaden deployment and sensing possibilities, but realizing their benefits requires new physical models, optimization frameworks, and deployment protocols. Environmental obstructions and complex sensing models remain important practical boundaries.

  • Open design challenges: New PASS variants require physics-based in-waveguide channel models and optimization frameworks tailored to their architectures.These developments are identified as necessary for accurately characterizing and exploiting variant designs.
  • Deployment outlook: Wireless links between base stations and waveguide feed points could improve deployment flexibility when buildings or terrain obstruct wired connections.The associated protocol design and joint communication-sensing implications remain future research directions.
  • Sensing outlook: PASS-assisted sensing is motivated by underexplored sensing capability alongside established communication benefits.The discussion highlights sensing designs and their interaction with communication functionality.

A. Motivation of PASS-Assisted Sensing

PASS-assisted sensing combines large-aperture near-field effects, improved sensing resolution, and line-of-sight creation. The survey reviews pure sensing, localization, tracking, and integrated communication-sensing applications while identifying unresolved modeling and compatibility issues.

  • Sensing benefits: PASS-assisted sensing offers near-field benefits, enhanced sensing resolution, and line-of-sight creation.Its tens-of-meters aperture produces useful spherical-wave curvature while requiring only a handful of pinching antennas.
  • Sensing benefits: Large PASS apertures expose polar-domain target coordinates and velocities through spherical-wave effects and can enlarge the near-field region with fewer antennas.The near-field boundary is given by 2D^2/λ.
  • Sensing applications: Pure sensing studies evaluate sensing SNR and Cramér–Rao lower bounds as measures of received target power and estimation accuracy.PASS sensing work includes CRLB derivation and particle-swarm-based CRLB minimization.
  • Sensing applications: PASS supports localization through spatially distributed pinching antennas and tracking through Doppler-based extraction of full-dimensional target mobility.Near-field sensing can capture polar-domain positions and velocities for moving users.
  • Integrated sensing and communications: Integrated sensing and communications includes sensing-aided communication, where sensing results inform later transmission design, and communication for sensing, where communication signals sense targets.The two paradigms respectively emphasize environmental information for communications and shared radio resources for sensing.
  • Open challenges: Conventional PASS sensing models can become intractable because received signals are aggregated in the waveguide before reradiation.The survey identifies SWAN architectures and cross-architecture sensing-communication compatibility as directions for addressing this issue.

2) Machine Learning for Sensing:

PASS performance research has developed rate and outage analyses across increasingly varied deployments, while machine learning is proposed to address the complexity of sensing optimization. The surveyed communication studies collectively report gains in channel gain, interference reduction, rate, and outage performance.

  • Machine learning for sensing: Echo signals create a highly multimodal sensing optimization space that makes convex optimization difficult.Existing element-wise searches for pinching-antenna positions also have high computational complexity, motivating machine-learning approaches.
  • Communication performance analysis: Communication-performance analyses commonly use average achievable rate and outage probability under random user distributions.The surveyed work applies stochastic geometry and considers single-user, multi-user, multi-waveguide, and multicell settings.
  • Achievable-rate analysis: Rate studies extend from optimized single-PA placements to discrete activation positions, multiple PAs, multiple waveguides, multi-user access, and blockage-aware placement.The survey includes models with propagation loss, mobility or spatial randomness, hybrid beamforming, and interference management.
  • Outage-probability analysis: Outage-probability studies progress from lossless and lossy single-waveguide models to wireless-powered, multi-user, WDMA, and multicell PASS.These works derive closed-form or integral-form performance expressions under different coverage and interference configurations.
  • Summary: Collectively, the surveyed communication studies show that PASS can enhance user channel gain and reduce inter-user interference.These effects are associated with improved achievable rate, outage performance, network coverage, and throughput.

B. Sensing Performance

PASS sensing research uses CRLB-based analysis to assess sensing accuracy, while Bayesian formulations incorporate prior target distributions. The section also identifies sensing–communication trade-offs and optimization challenges arising from PA placement and spherical-wave channels.

  • Performance Analysis: CRLB is the most frequently used metric for analyzing PASS sensing accuracy, with Bayesian CRLB averaging performance over prior sensing-parameter distributions.The Bayesian formulation reduces dependence on evaluating unknown sensing-parameter values.
  • Performance Analysis: The sensing centroid can differ from the distribution centroid, so placing PAs directly above the target is not generally optimal for sensing.The distribution centroid is the center of the target prior, whereas the sensing centroid minimizes BCRLB.
  • Sensing–Communication Trade-off: Communication and sensing can prefer different PA placements: target-aligned PAs may reduce path loss and improve communication rate while missing the sensing-sensitive region.This motivates studying sensing–communication trade-offs in PASS-enabled ISAC systems.
  • Optimization Challenges: PASS optimization is difficult because PA locations jointly affect signal phase and path loss, producing highly non-convex, oscillatory, and coupled design problems.The coupling extends across PA placement, beamforming, and power allocation as system dimensions grow.

1) Structure-based Optimization Methods:

PASS optimization methods span structure-based, population-based, and game-theoretic approaches. Their suitability depends on model structure, system scale, and whether continuous or discrete decisions must be optimized in real time.

  • Structure-based Methods: Structure-based methods exploit phase alignment, KKT conditions, block decomposition, or surrogate bounds to derive iterative updates with convergence guarantees.These methods use the mathematical structure of PASS optimization problems directly.
  • Structure-based Methods: Globally optimal branch-and-bound methods can guarantee optimal PA placement and beamforming solutions, but their complexity remains affordable only for small-scale systems.McCormick relaxation accelerates exhaustive search without removing the high computational burden.
  • Structure-based Methods: Suboptimal structure-based methods use analytical phase alignment, alternating updates, and successive convex approximation to obtain scalable solutions for throughput and power-related objectives.These methods separately or iteratively update transmission variables and PA positions.
  • Population-based Methods: Population-based methods, including PSO and differential evolution, explore non-convex PASS design spaces stochastically and have been applied to sensing, covert communication, and M-PASS optimization.PSO-based designs can jointly optimize waveform, PA positions, activation, or beamforming depending on the application.
  • Game-theoretic Methods: Game-theoretic methods model interactions among PAs, waveguides, and users through matching or coalition games to obtain stable, lower-complexity allocation and activation designs.Matching methods target discrete activation, while coalition methods can jointly handle assignment, activation, decoding order, and power allocation.
  • Motivation for Learning-based Methods: Large-scale PASS optimization remains challenging because spherical-wave channels are non-convex and exhaustive search is computationally expensive.Deep learning is introduced to learn PA positions, activation patterns, and power allocations for lower-complexity real-time implementation.

1) Learning for Communications:

Learning-based PASS communication methods use GNNs, DNNs, unfolding networks, and Transformers to jointly learn PA positions and transmission strategies. Reported studies achieve near-optimal or higher spectral efficiency with lower complexity, while discrete decisions and fast channel variation remain constraints.

  • GNN-based Learning: GNN-based methods learn PA activation or jointly learn PA positions and beamforming while exploiting permutation properties to improve scalability and generalization.Reported simulations show robustness to localization uncertainty and near-optimal joint policies with low training complexity.
  • DNN-based Learning: Multi-block DNNs jointly learn antenna positioning and precoding for multi-waveguide IoT systems and can achieve near-optimal spectral efficiency in simulations.Separate blocks handle feature extraction, antenna positioning, and precoding design.
  • Model-based Learning: Gradient-input unfolding networks embed objective and constraint gradients to jointly learn antenna positions and beamforming for weighted-sum-rate maximization.The model-based structure simplifies the mappings learned by the DNN.
  • Transformer-based Learning: KDL-Transformer learns a small set of dual variables by leveraging the KKT beamforming structure, achieving higher spectral efficiency than an optimization-based method with millisecond-level response.The approach targets multi-waveguide, multi-PA joint optimization.
  • Learning for ISAC: Learning-based ISAC designs jointly optimize PA positions and beamforming for sensing and communication, achieving performance close to or better than optimization-based methods with much lower computational complexity.GNNs exploit permutation properties and optimal beamforming structure in multi-target, multi-user settings.
  • Limitations and Design Considerations: Learning methods reduce inference complexity after training, but continuous relaxations may lose optimality for discrete decisions and short coherence times constrain fast-timescale applications.Slow variables can be learned offline, whereas fast variables such as baseband beamforming may require lightweight model-based optimization.

VII. CONCLUSIONS

The survey consolidates PASS foundations, variants, sensing, performance analysis, and optimization tools, then identifies implementation challenges. Open issues include channel acquisition, PA reciprocity, EM-to-system modeling, and deployment flexibility.

  • Survey Scope: The survey integrates PASS physical foundations, variants such as SWAN, C-PASS, and M-PASS, sensing applications, performance analysis, and optimization methods from conventional algorithms to machine learning.It also discusses sensing–communication interplay and future research directions.
  • Channel Estimation: PASS channel estimation is constrained by single-RF receiver architectures that aggregate distributed PA signals into compressed, rank-deficient observations.A general acquisition framework for hardware-coupled coefficients and dimension-compressed observations remains open.
  • Channel Estimation: Future channel-estimation methods should exploit PASS geometry, propagation, and hardware structure while addressing calibration errors, positioning uncertainty, mutual coupling, and waveguide impairments.Integrating sensing and localization with channel acquisition is identified as a promising direction.
  • PA Reciprocity: Reciprocal PAs support bidirectional transmission and full-duplex functions but can cause inter-PA reradiation that degrades received signal strength and complicates CSI acquisition.Non-reciprocal PAs resolve uplink inter-PA radiation but support only half-duplex operation.
  • PA Reciprocity: The system-level trade-off among uplink reception, full-duplex or ISAC functions, and hardware complexity remains insufficiently understood for reciprocal and non-reciprocal PA architectures.Reciprocity-aware models and potentially hybrid or controllable directionality are proposed as future directions.
  • Hardware Implementation: PASS EM performance depends strongly on the PA–waveguide interface, including contact shape, dimensions, and mounting positions that determine radiation intensity, directionality, and 3dB beamwidth.A calibrated bridge from EM device behavior to tractable system-level optimization parameters is still lacking.
  • Deployment: Outdoor deployment remains an open direction because wired waveguide connections limit flexibility, motivating wireless-connected architectures such as Wi-PASS.PASS is positioned for outdoor use through large-scale channel reconfiguration and line-of-sight link creation.
Loading 2601.18927v2…