Center-Fed Pinching Antenna System

• C-PASS is an advanced antenna architecture that uses a center-fed dielectric waveguide with reconfigurable pinching antennas to generate counter-propagating modes and two-stream MIMO channels.
• It leverages rigorous waveguide theory and multiport network modeling with optimization algorithms like PSO and alternating optimization to optimize PA placement and beamforming.
• The system enhances spatial multiplexing and reduces losses, achieving significant capacity gains (e.g., a 3.59 dB improvement) over conventional end-fed architectures in both communications and sensing.

The Center-Fed Pinching Antenna System (C-PASS) is an advanced architecture developed to achieve high spatial flexibility, low-loss beamforming, and enhanced spatial-multiplexing gain in modern wireless communications and sensing scenarios. Unlike end-fed Pinching Antenna Systems (PASS), C-PASS excites a dielectric waveguide from its physical center, generating counter-propagating guided modes that radiate via movable, reconfigurable pinching antennas. This enables a two-stream MIMO channel, effectively doubling the degrees of freedom and enabling new capacity scaling regimes. C-PASS combines rigorous waveguide theory, multiport network modeling, and algorithmic geometry optimization to address the highly constrained design problem arising from practical waveguide and stub physics, geometric constraints, and performance objectives across communication, localization, and sensing applications (Gan et al., 14 Dec 2025, Zhou et al., 2 Sep 2025, Wang et al., 9 Feb 2025, Wang et al., 21 May 2025, Wang et al., 6 Sep 2025).

1. Physical and System Architecture

C-PASS consists of a dielectric waveguide, center-fed via a T-junction splitter or power divider, supporting two bidirectional guided waves. Along both sides of the center feed, sets of dielectric “pinching” antennas (PAs) are positioned at variable locations. Each PA, modeled as an open-ended directional coupler or a multiport scattering element, taps a controllable fraction of the guided power, radiating into free space. The waveguide may be extended with co-located leaky coaxial (LCX) cables for wide-area, low-loss reception in wireless sensing configurations (Gan et al., 14 Dec 2025, Wang et al., 21 May 2025, Wang et al., 9 Feb 2025, Wang et al., 6 Sep 2025).

The defining feature is the center-fed geometry: the input signal is launched at the midpoint, creating a “forward” sub-array (PAs to the right) and a “backward” sub-array (to the left), each addressable via relative phase and power-split coefficients. The guided mode in the waveguide propagates with $\beta_g = 2\pi n_g / \lambda$; stubs or couplers produce out-coupling with controllable amplitude and phase according to model-specific S-matrix entries or coupled-mode theory (Wang et al., 9 Feb 2025, Wang et al., 6 Sep 2025).

Co-location of PAs and dense PA positioning (potentially sub-wavelength, with constraints determined by the minimum spacing for mutual coupling, typically $\Delta x \ge 0.3\lambda$) allows “programming” of the radiation centers over large arrays greater than 30 meters in length or, in mmWave regimes, sub-centimeter granularity (Zhou et al., 2 Sep 2025, Gan et al., 14 Dec 2025, Wang et al., 21 May 2025).

2. Signal and Channel Models

The end-to-end channel from each feed port (forward or backward) to user or target consists of three segments: guided propagation from the center feed to each PA location, guided-to-radiated coupling (with amplitude and phase described by directional-coupler models), and free-space or multipath LoS/NLoS propagation to multiple receive antennas or scattering points.

For a given scalar input $x$ of total power $P$, the received signal at the $m$-th antenna from all $|\mathcal{P}|$ activated PAs is:

$$y_m = \sqrt{\tfrac{P}{|\mathcal{P}|}}\sum_{n\in\mathcal{P}} h_{m,n} x + w_m$$

with per-path channel coefficients

$$h_{m,n} = \sqrt{\eta} \frac{1}{\sqrt{(x̃_m - ℓ_n)^2 + d^2}}\exp\left\{-j\left[\frac{2\pi}{\lambda}\sqrt{(x̃_m - ℓ_n)^2 + d^2} + \frac{2\pi}{\lambda_g}(ℓ_n-ℓ_f)\right]\right\}$$

for user antennas at $(x̃_m, d)$ above the waveguide.

In multi-user or multi-target sensing, the receive model incorporates the spatial multiplexing of the forward/backward subarrays, the S-parameter transfer function for each PA, and the path loss of each free-space segment, often under the narrowband and far-field approximations (Zhou et al., 2 Sep 2025, Wang et al., 9 Feb 2025, Wang et al., 21 May 2025, Gan et al., 14 Dec 2025).

3. Theoretical Capacity, Spatial-Multiplexing Gain, and DoF

A major breakthrough of C-PASS is its inherent spatial-multiplexing capability on a single waveguide. In mathematical terms, the DoF for an end-fed PASS is one (i.e., the MIMO channel is of rank at most one), whereas C-PASS supports two spatial streams:

$$\mathrm{DoF}_{\mathrm{C\!-\!PASS}} = 2, \quad \mathrm{DoF}_{\mathrm{end\!-\!fed}} = 1$$

via the two independent counter-propagating arrays (Gan et al., 14 Dec 2025).

The instantaneous capacity of C-PASS, when phase-aligned via optimal PA positioning, follows:

$$C = \log_2\det\left(\mathbf{I}_2 + \frac{P_T}{2N_0}\mathbf{H}^{\mathrm{eff}}(\mathbf{H}^{\mathrm{eff}})^H\right)$$

with an array gain $$G^{\mathrm{A}} \sim \mathcal{O}\left(\frac{\ln^2 N}{N}\right)$$ and a multiplexing gain $$G^{\mathrm{M}} \sim \mathcal{O}\left(P_T \frac{\ln^4 N}{N^2}\right)$$. The ergodic capacity slope versus $\log_2 P_T$ matches the rank (factor of two increase). A 3.59 dB enhancement at $N = 50$ and $P_T = 30$ dBm was observed over end-fed, with scaling laws matching analytic predictions (Gan et al., 14 Dec 2025).

This is summarized in the following table:

Architecture	DoF	Array Gain	Multiplexing Gain
Center-Fed	2	$\mathcal{O}(\ln^2N/N)$	$\mathcal{O}(P_T\ln^4N/N^2)$
End-Fed	1	$\mathcal{O}(\ln^2N/N)$	0

4. Placement and Beamforming Optimization Algorithms

The highly nonconvex, constrained optimization problem of PA placement and beamforming in C-PASS has attracted several algorithmic solutions:

• Two-Layer Centered Placement and Compressed Phase-Alignment: The first layer selects the center PA position to minimize aggregate path-loss to all $M$ user antennas (unique solution under concavity, or 1-D search). The second layer “grows” PAs outward, at each step using a linearized phase alignment and a compressed span-minimization search to identify the next best PA position for joint phase alignment (Zhou et al., 2 Sep 2025).
• Penalty-Based Alternating Optimization: Alternates beamformer and PA position updates, using channel decomposition and quadratic penalties to enforce stackable free-space-plus-coupling constraints; per-position optimization is reduced to a one-dimensional search (Wang et al., 9 Feb 2025).
• Zero-Forcing (ZF) Low-Complexity Search: For scenarios where strict nulling/cancellation is viable, the beamformer is constructed via matrix pseudo-inverse, with transmit power minimization at each PA position.
• Multiport-Network Alternating Search: In multiport network models, practical directional-coupler PAs (with amplitude-only or amplitude-constrained phase control) require alternately maximizing over coupling coefficients (quasi-Newton/BFGS) and PA locations (grid search), each step guaranteed to increase the objective until convergence (Wang et al., 6 Sep 2025).
• Particle Swarm Optimization (PSO) for Sensing: For joint waveform and PA-geometry optimization in C-PASS sensing, a two-stage PSO solver (PA geometry first, then waveform covariance via CVX) reduces the Cramér–Rao Bound (CRB) for multi-target localization (Wang et al., 21 May 2025).

5. Performance Metrics, Comparative Results, and Sensing Applications

C-PASS achieves significant performance gains over end-fed and conventional MIMO architectures, both in communications and sensing:

• Communications: In short-range/dense-PA scenarios, with $N = 32, M = 2$ at $P_T = 20$ dB, C-PASS achieves ~12 bps/Hz compared to ~8 bps/Hz for the single-PA baseline (a 4 bps/Hz gain) (Zhou et al., 2 Sep 2025).
• Sensing: At $N = 5$ waveguides, $K = 2$ targets, optimized C-PASS (center-fed, PSO-optimized) achieves RMSE$_x$ = 0.27 m versus 0.48 m for a conventional $10 \times 10$ UPA. At high percentiles, C-PASS halves estimation errors and is less sensitive to calibration (Wang et al., 21 May 2025).

A summary table from (Wang et al., 21 May 2025) is as follows:

System	RMSE$_x$ (m)	RMSE$_y$ (m)
Conventional MIMO $5\times5$	0.62	0.58
Conventional MIMO $10\times10$	0.48	0.45
PASS (fixed PAs)	0.41	0.39
PASS (optimized PAs)	0.27	0.25

System strengths include hardware-free beam steering (no phase shifters/RF chains), sub-wavelength granularity, robust NLoS/quasi-LoS performance, and practical complexity ($O(NMZ \log MZ)$). However, C-PASS is sensitive to geometry miscalibration, assumes mostly LoS/narrowband operation, and may incur reduced benefit under rapid mobility or strong waveguide/dielectric loss at high frequencies (Zhou et al., 2 Sep 2025, Gan et al., 14 Dec 2025).

6. Multiport Network Modeling and Reconfigurability

The C-PASS architecture leverages both ideal (fully reconfigurable, arbitrary S-matrix) and practical (directional-coupler based) PA models. Under ideal PA assumptions, both amplitude and phase of the radiated field can be controlled independently, yielding globally optimal performance for aligned clustering of PAs close to the target location. For DC-based practical PAs, amplitude-only or amplitude-constrained phase control suffices to approach ideality, provided PA spacing is dense and respective coupler geometries are tuned (Wang et al., 6 Sep 2025).

Reconfigurability is crucial: amplitude control dominates the performance gain when PA positions can be freely optimized; phase control becomes essential in fixed or sparse deployments. Energy conservation and coupling physics (e.g., $$|\widetilde{\Theta}_1|^2 + |\widetilde{\Theta}_2|^2 = 1$$ for return and radiating ports of the DC-based PA) set hard constraints on achievable S-parameters.

7. Practical Implementation, Guidelines, and Future Directions

Implementation of C-PASS involves key choices in waveguide and PA design:

• Power dividers (Wilkinson/T-junction) for symmetric, phase-matched center feeding;
• Subwavelength PA placement for fine beam resolution (mmWave feasibly allows $\Delta \le 0.01$ m);
• Real-time tuners for power splitting, e.g., gyromagnetic or mechanical septa;
• Dense PA arrays to enable near-continuous activation (≥300 positions/m closes the gap with analog);
• Use of LCX cables for dense, distributed echo reception in sensing;
• Calibration procedures for the network’s S-parameters and structural geometry (Gan et al., 14 Dec 2025, Wang et al., 9 Feb 2025, Wang et al., 21 May 2025, Wang et al., 6 Sep 2025).

Anticipated deployment scenarios include mmWave fronthaul/backhaul, indoor hotspot coverage (blockage-prone), multi-user MIMO, UAV communications, and precision wireless sensing. Future research directions encompass multi-band operation, adaptive feed ratios for hybrid multiplexing-beamforming, $K$-input extensions ($\mathrm{DoF} = K$), and integration with reconfigurable intelligent surfaces (RIS) (Gan et al., 14 Dec 2025).

---

References:

(Gan et al., 14 Dec 2025, Zhou et al., 2 Sep 2025, Wang et al., 9 Feb 2025, Wang et al., 21 May 2025, Wang et al., 6 Sep 2025)