Optimizing Pinching Antenna Placement

Updated 6 July 2026

Pinching antenna placement is the process of selecting antenna positions along dielectric waveguides to optimize communication, sensing, or joint performance objectives.
It considers geometric and electromagnetic factors that affect free-space path loss, guided-wave phase, and interference management in multi-antenna configurations.
Optimization objectives include throughput, fairness, energy efficiency, and robustness, addressing both fixed and mobile scenarios under real-world physical constraints.

Pinching antenna placement is the design problem of selecting where along one or more dielectric waveguides pinching antennas are activated, moved, or pre-installed so that the effective radiating points of a wireless system optimize communication, sensing, or joint objectives. In pinching-antenna systems (PASs) and pinching-antenna systems (PASSs), placement is a primary spatial degree of freedom because changing the longitudinal coordinate of a pinching element changes free-space path loss, guided-wave phase, and, in multi-antenna settings, the coherent superposition across antennas. Depending on the architecture, placement may mean continuous sliding of a pinching point along a waveguide, sequential per-user repositioning, or selective activation of pre-installed positions (Xie et al., 20 Feb 2025, Feng et al., 10 Apr 2026, Chen et al., 11 Jun 2026).

1. Geometric and electromagnetic foundations

In the canonical geometry, a dielectric waveguide is deployed along the $x$ -axis at a fixed height $d$ , while users lie on the plane $z=0$ . A pinching antenna on the waveguide has position $(x,0,d)$ in the single-waveguide case, or $(x_{z,l}^k,y_z,d)$ in multi-waveguide systems. The placement variable is therefore typically a continuous scalar coordinate along each waveguide, subject to physical bounds defined by waveguide length (Feng et al., 10 Apr 2026, Chen et al., 11 Jun 2026).

Placement enters the channel model through both amplitude and phase. In single-PA models, the received power depends on the Euclidean distance from the pinching point to the user, so the channel magnitude reduces to a free-space term proportional to $1/\|\mathbf u-\tilde{\mathbf v}\|^2$ . In multi-PA models, each antenna contributes a term whose amplitude depends on free-space distance and whose phase contains both the free-space propagation phase and the guided-wave phase accumulated from the feed point to the pinching location. Consequently, moving one PA changes not only its own path loss but also the constructive or destructive superposition with other PAs (Feng et al., 10 Apr 2026, Zhang et al., 21 Dec 2025).

This dependence becomes more pronounced in near-field formulations. In multi-cell and mmWave models, the effective channel is explicitly a nonlinear function of PA positions through spherical-wave propagation, distance-dependent phase curvature, and guided-wave phase. Several works therefore treat placement as geometry control of the effective channel itself rather than as a secondary refinement of beamforming (Chen et al., 11 Jun 2026, Zhou et al., 12 Apr 2025).

A related distinction is architectural. Some papers assume truly movable or continuously placeable pinching points along a waveguide. Others assume multiple pre-installed pinching antennas at fixed discrete locations and perform runtime antenna activation. In the latter case, “effective placement” is realized by selecting which pre-installed antenna radiates, often jointly with user assignment or waveguide assignment (Wang et al., 14 Jul 2025).

2. Optimization objectives and physical constraints

Across the literature, pinching antenna placement appears inside a wide range of optimization problems. The objective may be the sum rate, weighted sum rate (WSR), max–min multicast rate, energy efficiency (EE), outage probability, Bayesian Cramér–Rao bound (BCRB), or a hybrid communication–computation metric. The choice of objective strongly changes the placement structure: fairness-oriented formulations tend to favor balanced geometric locations, while throughput-oriented or NOMA formulations can bias the placement toward users with intrinsically stronger channels (Ding et al., 17 Jul 2025, Shan et al., 18 Feb 2026).

Setting	Objective	Placement variables
Single-/multi-user PAS	Sum-rate maximization or low-complexity placement design	Single-PA $x^{\rm Pin}$ , or per-user multi-PA offsets (Xie et al., 20 Feb 2025)
Robust PAS under bounded uncertainty	Minimize total power under worst-case QoS	Single-PA $v$ or multi-PA vector $\mathbf v$ (Feng et al., 10 Apr 2026)
Multi-waveguide PASS	WSR maximization	$\{x_{z,l}^k\}$ across cells and waveguides (Chen et al., 11 Jun 2026)
ISAC with Tx/Rx PASS	Communications-centric, sensing-centric, Pareto-optimal design	$d$ 0 (Shan et al., 18 Feb 2026)
Mobility-aware PASS	Long-term SE or average sum rate	Time-varying PA positions and beamforming (Zhao et al., 28 Feb 2026, Amhaz et al., 8 May 2026)

The physical constraints are remarkably consistent. PAs must remain on the waveguide, so each coordinate is bounded by the waveguide length. Adjacent PAs are often required to satisfy a minimum spacing, typically $d$ 1, to avoid strong mutual coupling or to preserve the assumptions used in the channel model. In mobile settings, additional movement constraints appear, such as a maximum per-slot displacement. When placement is coupled to QoS, outage, or SIC feasibility, these constraints become indirect geometric restrictions because a given placement must preserve minimum SNR, rate, or sensing accuracy under the induced channel (Feng et al., 10 Apr 2026, Zhang et al., 21 Dec 2025, Zhao et al., 28 Feb 2026).

3. Single-pinching-antenna placement regimes

The single-PA case has yielded the clearest analytical results. In the low-complexity TDMA design of a single-pinching-antenna system, the optimal placement for user $d$ 2 is the point on the waveguide directly above that user’s $d$ 3-coordinate, namely $d$ 4. This is the globally optimal solution of the single-user TDMA subproblem because it minimizes the user–antenna distance subject to the waveguide constraint (Xie et al., 20 Feb 2025).

Fairness-oriented OMA leads to a different geometry. In “Analytical Optimization for Antenna Placement in Pinching-Antenna Systems,” the optimal single-PA location for user-fairness-oriented OMA is

$d$ 5

and the paper explicitly shows that the users’ distances to the waveguide have no impact on the location selection. This establishes a strong separation between horizontal balancing along the waveguide and vertical or lateral offsets away from it (Ding et al., 17 Jul 2025).

Throughput-oriented formulations do not preserve that symmetry. In the same paper, the greedy-allocation-based OMA analysis shows, under a high-SNR approximation, that the optimal antenna location is in close proximity to the user who is nearest to the waveguide. For two-user NOMA with fairness, the closed-form solution

$d$ 6

shows that even a fairness-oriented NOMA objective places the antenna near the user positioned closest to the waveguide, not at a point that benefits all users equally (Ding et al., 17 Jul 2025).

Single-PA placement also interacts with deployment geometry. In a circular indoor IoT setting, the PA is placed at the point on the waveguide closest to the device, which reduces to $d$ 7 under full coverage and $d$ 8 under partial coverage. When propagation loss along the waveguide is included, the system exhibits a non-monotonic trend with respect to the waveguide length, and the optimal length decreases as the attenuation coefficient increases. A plausible implication is that single-PA placement cannot be separated from waveguide coverage design when guided-wave attenuation is non-negligible (Zhang et al., 5 Sep 2025).

4. Multi-PA and multi-waveguide placement

Once multiple PAs are present, placement becomes a coherent array design problem on a constrained one-dimensional manifold. The central difficulty is that each position controls both amplitude and phase, so the objective landscape is highly non-convex. In robust multi-PA systems under bounded user-location uncertainty, the worst-case channel gain

$d$ 9

has no simple closed form, and placement is optimized by a multi-start coordinate descent algorithm after a numerical worst-case gain evaluation based on boundary sampling and local refinement (Feng et al., 10 Apr 2026).

A different line of work addresses the same difficulty by transforming the communication objective rather than the uncertainty set. In “Multi-Waveguide Pinching Antenna Placement Optimization for Rate Maximization,” the average per-user data-rate maximization problem is converted by fractional programming into a tractable form, and a projected gradient ascent algorithm is constructed for continuous placement under a stringent physical minimum spacing constraint. A notable technical contribution is an exact nearest-point projection onto the non-convex feasible set induced by the spacing constraint. The reported gains over geometric baselines are large: at $z=0$ 0 and $z=0$ 1 m, the improvements over Closest-to-User Placement, Uniform Pre-Placement with Closest Selection, and Random Pre-Placement with Closest Selection are about $z=0$ 2, $z=0$ 3, and $z=0$ 4, respectively (Zhang et al., 21 Dec 2025).

In multi-cell systems, placement couples desired links and inter-cell interference across all cells. “Pinching-Antenna Enabled Multicell Wireless Systems” formulates a joint WSR maximization over precoding, PA-level power allocation, and antenna placement, using alternating optimization with fractional programming for precoding and power, and particle swarm optimization (PSO) for the high-dimensional PA placement problem. In that model, placement is a set of $z=0$ 5 continuous coordinates, and PSO is used because the landscape is multimodal and analytically difficult (Chen et al., 11 Jun 2026).

Placement also serves interference shaping. In “Antenna Placement Design for Interference Exploitation in Pinching-Antenna Systems,” PAs are jointly optimized with symbol-level precoding (SLP). The placement subproblem is decomposed into scalar updates for PA position coefficients, and a projected gradient descent method is applied within feasible movable regions. This formulation treats placement as a way to increase constructive interference margins rather than merely as path-loss minimization (Pang et al., 14 Mar 2026).

5. Robustness, mobility, blockages, and CSI acquisition

Robust placement has emerged because perfect user locations are unrealistic. Under bounded uncertainty regions, the single-PA problem can be converted into a convex semidefinite program via the S-procedure, yielding a globally optimal robust position that guarantees QoS for all user locations in the uncertainty set. In the multi-PA case, robustness is more difficult because user perturbations alter the interference pattern among PAs; the resulting design therefore combines numerical worst-case gain evaluation, closed-form power allocation, and block coordinate descent for placement (Feng et al., 10 Apr 2026).

A complementary uncertainty model assumes Gaussian-distributed localization errors. In that setting, the outage probability constraint is expressed through the first-order Marcum $z=0$ 6-function, the minimum user power is derived analytically for a fixed antenna position, and PSO is then used to search for the antenna position that maximizes EE. This formulation makes placement sensitive to the distance between the antenna projection and the mean of the user-location distribution rather than to a deterministic user point (Feng et al., 27 Jan 2026).

User mobility transforms placement into a sequential control problem. In urban micro networks, one line of work uses a bilevel design: a soft actor-critic (SAC) agent controls continuous antenna positions in the outer loop, while zero-forcing precoding is updated analytically in the inner loop. Under the reported setup, the SAC-based policy achieves a Mean Sum SE of $z=0$ 7 versus $z=0$ 8 for fixed placement, corresponding to a $z=0$ 9 improvement over the fixed-PA baseline (Zhao et al., 28 Feb 2026). A related study with probabilistic blockage and random waypoint mobility uses deep deterministic policy gradient (DDPG) to jointly optimize beamforming and time-varying pinching locations, emphasizing that the performance gains of PASS critically rely on the ability to track or predict user trajectories in real time (Amhaz et al., 8 May 2026).

Placement also interacts with blockage geometry. In systems with LoS blockages and pre-installed PAs, runtime activation becomes the effective placement variable. The joint waveguide-assignment and antenna-activation problem is then solved by a matching-based algorithm using two distinct preference designs, with the explicit aim of establishing LoS links for desired users and NLoS links for eliminating inter-user interference (Wang et al., 14 Jul 2025).

Finally, accurate placement requires accurate CSI at arbitrary waveguide positions. “Channel Estimation for mmWave Pinching-Antenna Systems” shows that activating only a near-end subarray and a far-end subarray can emulate a large-aperture array, estimate multipath parameters, and reconstruct CSI at all candidate positions. The paper reports that in mixed LoS/NLoS the refined estimation scheme can achieve a rate within about $(x,0,d)$ 0 of the perfect-CSI rate at high pilot power, while using only a small number of activated pinching antennas during training (Zhou et al., 12 Apr 2025).

6. Comparative findings, recurring misconceptions, and research directions

Several recurring findings cut across the literature. First, optimized pinching antenna placement consistently outperforms fixed-antenna baselines in multicast rate, WSR, EE, SE, sensing accuracy, or hybrid metrics, especially when the service area is large or when users are non-uniformly distributed (Xie et al., 20 Feb 2025, Shan et al., 18 Feb 2026). Second, “closest to the user” is not a universal rule. It is exactly optimal in some single-user or per-slot TDMA settings, but it becomes suboptimal in fairness-oriented OMA, in NOMA with SIC structure, and in interference-limited multi-user systems where global phase and interference effects dominate (Ding et al., 17 Jul 2025, Zhang et al., 21 Dec 2025).

A further misconception is that adding more PAs always helps. In robust multi-PA design with fixed positions, the total power can increase with the number of antennas because destructive superposition can make some worst-case channel realizations extremely weak. This suggests that placement optimization is essential once coherent multi-PA effects matter (Feng et al., 10 Apr 2026). Likewise, full waveguide coverage is not automatically preferable to partial coverage: in circular indoor IoT systems with propagation loss, full coverage can be worse than partial coverage because attenuation can dominate the geometric benefit of longer reach (Zhang et al., 5 Sep 2025).

The literature remains constrained by several simplifying assumptions: perfect CSI, narrowband channels, single-cell operation, one-dimensional waveguide geometry, negligible inter-waveguide coupling, or idealized movement mechanisms. Many papers explicitly identify extensions toward multi-waveguide two-dimensional deployments, multi-cell coordination, wideband and frequency-selective models, more general uncertainty sets, and learning-based approximations of optimization-based placement policies (Chen et al., 11 Jun 2026, Feng et al., 10 Apr 2026, Feng et al., 27 Jan 2026). A plausible implication is that pinching antenna placement is evolving from a purely geometric design variable into a coupled control variable spanning communications, sensing, interference management, uncertainty mitigation, and real-time adaptation.