---
title: Pinching Antenna Systems (PASS)
url: https://www.emergentmind.com/topics/pinching-antenna-systems-pass-8ca4c7e9-3195-4038-8ef1-735431796da8
type: topic
---

# Pinching Antenna Systems (PASS)

Pinching Antenna Systems, more commonly denoted **PASS** in the recent literature, are waveguide-based flexible antenna architectures in which radio signals are first transported through a low-attenuation dielectric waveguide and then intentionally radiated into free space at selected locations by attaching **pinching antennas (PAs)** to the waveguide. Their distinctive capability is that the effective radiation location becomes a design variable over meter-scale or tens-of-meters-scale structures, so beamforming is achieved not only by complex weights but also by physically changing where radiation occurs. This principle has been described as **pinching beamforming** and is closely associated with “last meter” communication, stable line-of-sight (LoS) links, blockage mitigation, and near-field beam focusing [2601.18927, 2501.18409].

## 1. Terminology, concept, and physical motivation

In the papers considered here, the official notation is overwhelmingly **PASS**, while **PA** denotes a single pinching antenna. The shorter form **PAS** appears as an informal variation, but the system-level acronym used in the formal models is **PASS**. A PA is created by bringing a secondary dielectric element into contact with, or into the evanescent-field region of, a dielectric waveguide; if the element is removed, the radiation disappears and the waveguide returns to acting primarily as a transmission medium. This attach–detach–move capability is the defining architectural departure from conventional fixed arrays [2502.16624, 2601.18927].

The central motivation for PASS is that conventional arrays keep the radiating aperture fixed at deployment, while movable or fluid antennas typically reconfigure only over wavelength-scale regions. PASS instead enables **large-scale antenna reconfiguration** along pre-deployed waveguides, so the over-the-air segment can be shortened and shaped in response to geometry. This is the physical basis of the “last meter” viewpoint: most of the propagation is transferred to a low-loss guide, while only the final local segment is wireless. The architecture has therefore been proposed for ceilings, walls, facades, rooftops, roadsides, and similar surfaces where a long waveguide can be embedded and PAs can be activated near users or targets [2501.18409, 2601.18927].

This also clarifies a common misconception. PASS is not merely a conventional array with unusual packaging. The literature treats the **location of the radiating point itself** as a communication and sensing variable, so geometry, phase accumulation inside the waveguide, and free-space propagation are jointly designed rather than separated into “hardware” and “beamforming” layers [2501.18409, 2601.18927].

## 2. Electromagnetic principle and channel modeling

A physics-based PASS model treats a pinching antenna as an **open-ended directional waveguide coupler**. If the main dielectric waveguide field is written as
\[
\mathbf{E}_{\mathrm{guide}}(x,y,z)=\mathbf{D}_{\mathrm{guide}}(y,z)e^{-j\beta_{\mathrm g}x}s_{\mathrm p},
\]
and the pinching element field as
\[
\mathbf{E}_{\mathrm{pinch}}(x,y,z)=\mathbf{D}_{\mathrm{pinch}}(y,z)e^{-j\beta_{\mathrm p}x}s_{\mathrm p},
\]
then coupled-mode theory yields
\[
\frac{dA(x)}{dx}=-j\kappa B(x)e^{-j\Delta\beta x},\qquad
\frac{dB(x)}{dx}=-j\kappa A(x)e^{j\Delta\beta x},
\]
with \(\Delta\beta=\beta_{\mathrm p}-\beta_{\mathrm g}\). The corresponding power exchange is
\[
P_{\mathrm{guide}}(x)=|A(x)|^2,\qquad
P_{\mathrm{pinch}}(x)=|B(x)|^2,
\]
so the radiated power is directly controlled by coupling length, coupling strength, and phase matching. Under \(\Delta\beta=0\), the idealized exchange simplifies to \(A(x)=\cos(\kappa x)\) and \(B(x)=-j\sin(\kappa x)\), which underlies several simplified communication models [2502.05917].

At system level, PASS channels are usually modeled as a cascade of **in-waveguide propagation**, **coupling/radiation**, and **free-space propagation**. For a lossless guided segment of length \(L\), one survey writes the end-to-end coefficient as
\[
h=
\frac{\eta \exp\left(-j \tfrac{2\pi}{\lambda} R\right)}{R}
\times
\rho(\theta,\phi)\,\kappa
\times
\exp\left(-j \frac{2\pi}{\lambda_g} L\right),
\]
where \(R\) is the PA-to-user distance, \(\lambda_g\) is the guided wavelength, \(\rho(\theta,\phi)\) is the PA radiation pattern, and \(\kappa\) is the coupling factor. For multiple serial PAs on one waveguide, the \(n\)-th PA inherits the residual guided power left after the previous \(n-1\) couplers, yielding the cascaded coefficient
\[
\widetilde{\kappa}_n=\kappa_n\prod_{i=1}^{n-1}\sqrt{1-\kappa_i^2},
\]
which is one reason PASS does not behave like an independently fed phased array [2601.18927, 2502.05917].

This sequential extraction motivates two widely used simplified power models. In the **equal power model**, different coupling lengths are chosen so that each PA radiates the same amount of power. In the **proportional power model**, all PAs use identical coupling lengths, so each extracts the same fraction of the remaining guided power; radiated power therefore decays along the waveguide. The former is analytically clean, while the latter is easier to manufacture [2502.05917, 2501.18409].

The literature is not uniform on waveguide loss. Many communication papers assume negligible in-waveguide attenuation and retain only phase accumulation, whereas the indoor positioning model explicitly uses a lossy guided factor
\[
l(y_{li})=e^{-(\alpha+j\beta)y_{li}},
\]
with \(\gamma=\alpha+j\beta\) derived from dielectric parameters. This distinction matters: “negligible waveguide loss” is a modeling assumption in several PASS communication works, not a universal constitutive law of all PASS deployments [2508.08185, 2502.16624, 2601.18927].

## 3. Pinching beamforming, array behavior, and placement laws

The canonical PASS beamforming mechanism is geometric. One architecture paper writes the \(n\)-th PA contribution as
\[
y_n = \frac{\beta_n \sqrt{P_n}}{r_n} e^{-j \frac{2 \pi}{\lambda} \left( r_n + n_{\mathrm{eff}} d_n \right)} x,
\]
so moving a PA changes both the free-space distance \(r_n\) and the in-waveguide distance \(d_n\). Pinching beamforming is therefore the deliberate alignment of these geometry-induced amplitudes and phases by changing PA positions, not only by changing electronic weights [2501.18409].

The first systematic array-gain analysis establishes that, even under idealized LoS spherical-wave propagation and fixed inter-antenna spacing \(\lambda/2\), the normalized array gain
\[
a=\frac{|\mathbf{h}^{\mathsf T}\boldsymbol{\phi}|^2}{N}
\]
is **not monotonic** in the number of PAs. For symmetric half-wavelength placement around the user, the paper proves
\[
\lim_{N\rightarrow\infty} a = 0.
\]
Hence there exists an optimal finite number of activated PAs rather than an unlimited monotonic scaling law. In the numerical setup with \(f_c=28\) GHz, \(d=3\) m, and \(n_{\rm eff}=1.4\), the reported optimum is
\[
N^\star = 923,
\]
with required waveguide length
\[
\frac{\lambda}{2}N^\star = 4.94\ \text{m}.
\]
The same paper also shows, for a two-antenna model with mutual coupling, that tighter spacing is not always better; the reported optimal spacing is
\[
\Delta^\star = 0.715\,\lambda.
\]
These results directly refute the common intuition that “more PAs” or “smaller spacing” must always increase gain [2501.05657].

Uplink analyses sharpen this geometric view. In the **multiple PAs for a single user (MPSU)** setting, optimized phase-aligned PA positions satisfy
\[
x_n=\frac{n\lambda(2d_0+n\lambda)}{2(d_0+n\lambda)},\qquad n=-N,\ldots,N,
\]
with \(d_0=\sqrt{y_{ui}^2+h^2}\). The resulting PA arrangement is **asymmetric and non-uniform** in the near zone, but approaches
\[
x_n\approx n\lambda
\]
in the far zone. The same work derives closed-form analytical, asymptotic, and approximated ergodic-rate expressions for optimized uplink PASS and concludes that optimizing PA positions significantly enhances ergodic sum rate [2502.12365].

A plausible implication is that PASS should be interpreted less as a dense-array technology and more as a **geometry-allocation technology**. The dominant design question is often not “how many elements can be packed,” but “which radiating points should exist, where, and with what extraction pattern.”

## 4. Architectural classes and communication modes

The architecture literature distinguishes between basic PASS transmission classes and more advanced variants.

| Variant | Defining feature | Design implication |
|---|---|---|
| Non-multiplexing PASS | One data stream per waveguide; relies on pinching beamforming only | Simple baseband processing |
| Multiplexing PASS | Joint baseband and pinching beamforming | Supports richer multiuser transmission |
| Segmented PASS / SWAN | Multiple short waveguide segments; one PA per segment | Reduces in-waveguide loss and uplink IAR issues |
| Center-fed PASS (C-PASS) | Center feed with forward and backward propagation | \(2\) DoFs instead of \(1\) for end-fed PASS |
| Multi-mode PASS (M-PASS) | Multiple guided modes in one waveguide | Enables a full-rank effective channel on one waveguide |

The non-multiplexing and multiplexing categories, including **sub-connected**, **fully-connected**, and **phase-shifter-based fully-connected** multiplexing forms, were introduced as practical transmission architectures. Later survey work added **segmented PASS (S-PASS/SWAN)**, **center-fed PASS (C-PASS)**, and **multi-mode PASS (M-PASS)** to address uplink reradiation, in-waveguide loss, and the one-waveguide degree-of-freedom bottleneck [2501.18409, 2601.18927].

These architectural ideas have been instantiated in several communication problems. For multicast, one paper studies a **single dielectric waveguide** carrying a **single common signal** to multiple users and optimizes PA positions to maximize the **worst-user** multicast SNR via PSO; the setting is particularly natural because a single waveguide “can only be fed with the same signal,” making PASS structurally aligned with broadcast or multicast services [2502.16624]. A blockage-aware extension replaces deterministic LoS with Bernoulli LoS indicators and optimizes the **minimum average SNR** through a provably convergent MM procedure; the paper reports that with **8 PAs and 25 users**, the execution time of **CSM is approximately 2.5 times** that of **BSM**, illustrating the growing importance of scalable inner solvers as PASS size increases [2602.07421].

Multiuser MIMO formulations generalize this to hybrid beamforming. In **MIMO-PASS**, the access point uses \(M\) waveguides and \(N\) movable pinching elements per waveguide; in downlink the digital precoder \(W\) and positions \(L\) are jointly optimized via fractional programming and Gauss-Seidel updates, while uplink uses an iterative hybrid multiuser detection design. The reported numerical results show weighted sum-rate gains over conventional MIMO, classical hybrid analog-digital MIMO, and, in some settings, fully digital massive MIMO [2503.03117]. For **over-the-air computation**, PASS adds PA positions as variables in a joint mean-squared-error minimization over receive combining, user powers, and PA locations, solved by alternating optimization with Gauss-Seidel position updates [2505.07559]. For **secure multicast**, the literature combines digital transmit beamforming with pinching beamforming and develops SDR, Dinkelbach-ADMM, MM, and SOCP formulations for single-group and multi-group secrecy-rate maximization, with PASS consistently outperforming fixed-location architectures in the reported settings [2509.16045].

## 5. Sensing, positioning, and integrated sensing and communications

PASS has rapidly expanded from communication-focused studies into positioning, sensing, and ISAC.

For **indoor positioning**, a single-waveguide PASS uplink model uses RSSI-based ranging and a PASS-specific weighted least squares estimator. The ranging law explicitly includes waveguide attenuation,
\[
\hat d_{ik} = \frac{c e^{-\alpha y_{li}}}{\sqrt{P_{ik}/P_k}\,4\pi f_c},
\]
and the WLS stage estimates the 2D user position from collinear PA anchors. The reported observations are: more PAs improve positioning accuracy and robustness; the gain becomes marginal when the number of PAs exceeds roughly **7**; and users located **between** and **near** PAs obtain superior accuracy. In the cited Monte Carlo results, increasing the number of PAs from \(I=3\) to \(I=9\) reduces mean error and variance from approximately \(0.74\) m and \(0.75\) to about \(0.30\) m and \(0.29\), respectively [2508.08185].

For **wireless sensing**, one architecture combines PASS transmission with **leaky coaxial (LCX) cables** for reception. The transmit side uses dielectric waveguides with movable PAs; the receive side uses LCX cables to collect echoes over a wide area. The paper derives a multi-target Fisher information matrix and a CRB for target positions, then minimizes \(\mathrm{Tr}(\mathrm{CRB}(\boldsymbol{\theta}))\) by jointly designing PA positions and waveform covariance. The proposed solution is a two-stage PSO-based method for PA placement followed by convex waveform optimization [2505.15430].

ISAC studies follow two main directions. A separated two-waveguide design uses one waveguide for transmitting information-bearing signals and another for receiving reflected echoes. On the transmit guide, the \(N\) PAs are partitioned into a communication subarray of size \(N_1\) and a sensing subarray of size \(N_2=N-N_1\); the sensing metric is illumination power,
\[
P(\phi_{\mathrm{s}},r_{\mathrm{s}})
=
\mathbf{h}_{1,\mathrm{s}}^H\mathbf{w}\mathbf{w}^H\mathbf{h}_{1,\mathrm{s}}
+
\mathbf{h}_{2,\mathrm{s}}^H\mathbf{R}_{\mathrm{s}}\mathbf{h}_{2,\mathrm{s}},
\]
subject to a minimum communication rate. The beamforming subproblem admits a tight SDR, and the paper reports that equal power allocation performs nearly as well as an ideal optimal power allocation in the presented setting [2504.07709]. A second ISAC formulation uses a full-duplex BS with transmitting PASS waveguides and a receiving ULA, and minimizes the target localization CRB trace under communication QoS, power-budget, and PA-deployment constraints via AO, SDR, SCA, penalty methods, and element-wise position optimization. In the reported simulations, the PASS-assisted ISAC framework is less affected by stringent communication constraints than conventional MIMO-ISAC and improves further as the number of waveguides and PAs per waveguide increases [2508.19540].

Taken together, these works show that PASS geometry is not only a communication resource. It is also an **estimation-theoretic resource**: PA placement directly changes Fisher information, CRB, illumination power, and the conditioning of localization equations.

## 6. Optimization methods, assumptions, and open problems

PASS optimization is structurally difficult because PA positions change both amplitude and phase through nonlinear spherical-wave geometry and guided-wave propagation. The literature therefore uses a broad algorithmic toolbox: **particle swarm optimization**, **penalty-based alternating optimization**, **penalty dual decomposition**, **fractional programming**, **Gauss-Seidel coordinate updates**, **minorization-maximization**, **successive convex approximation**, **semidefinite relaxation**, **ADMM**, **SOCP**, **matching/game-theoretic designs**, and increasingly **machine learning**. Surveyed learning models include **GNNs**, **deep unfolding**, and **Transformer-based architectures** for joint PA-position and beamforming control [2502.05917, 2503.03117, 2509.16045, 2601.18927].

Some algorithmic results are already quite specific. The blockage-aware multicast MM framework compares a **candidate search method (CSM)** with a **bisection search method (BSM)** and reports identical objective values with substantially better scalability for BSM as the number of users grows [2602.07421]. The secure multicast formulations report complexity orders such as \(\mathcal O(M^6)\) for single-group SDR and \(\mathcal O(I_{\rm iter}M^3)\) for single-group Dinkelbach-ADMM, while the multi-group SOCP method is proposed precisely because MM-SDR becomes expensive [2509.16045]. The general trend is clear: once PASS is scaled beyond small proof-of-concept geometries, **position optimization dominates** both modeling and runtime.

At the same time, the current literature rests on strong assumptions. Common ones include pure or dominant LoS propagation, negligible waveguide attenuation, matched ports, negligible reflections, equal power splitting, single-antenna users, perfect or sufficiently accurate channel and location information, and one-dimensional PA movement along predefined waveguides. Several papers explicitly note additional omissions, such as hardware nonidealities, switching loss, blockage uncertainty, clutter, dynamic mobility, self-interference, and wideband dispersion [2502.05917, 2508.08185, 2508.19540, 2601.18927].

A second misconception is therefore worth correcting. PASS does not automatically eliminate the difficulties of high-frequency wireless propagation. Rather, it **moves** part of the challenge from free-space path loss toward waveguide design, coupling control, actuation, calibration, reciprocity management, and channel estimation. The survey literature highlights unresolved issues in uplink modeling, inter-antenna reradiation, multi-mode and center-fed hardware realization, wideband operation, reflection and matching control, real-time PA actuation, and scalable learning-assisted control [2601.18927].

The present state of the field suggests a precise interpretation. PASS is best viewed as a **physically reconfigurable waveguide-fed antenna paradigm** in which the radiating aperture is no longer fixed. Its mature core is the insight that **antenna location itself is an optimization variable**; its unsettled frontier is how to realize that insight under realistic loss, calibration, mobility, and network-scale constraints.

Source: https://www.emergentmind.com/topics/pinching-antenna-systems-pass-8ca4c7e9-3195-4038-8ef1-735431796da8