Integrated Sensing and Communication for Segmented Waveguide-Enabled Pinching Antenna Systems

In this paper, an integrated sensing and communication (ISAC) design for segmented waveguide-enabled pinching-antenna array (SWAN) systems is proposed to improve the performance of systems by leveraging the low in-waveguide propagation loss of segmented waveguides. The hybrid segment selection and multiplexing (HSSM) protocol is implemented to provide favorable performance with less hardware cost. To achieve this, a joint transmit beamforming optimization, segment selection, and pinching antenna positioning problem is formulated to maximize the sum communication rate with the constraints of sensing performance. To solve the maximization problem, we propose a segment hysteresis based reinforcement learning (SHRL) algorithm to learn segment selection and pinching antenna positions in different progress to explore better strategies. Simulation results demonstrate that 1) the proposed SWAN-ISAC scheme outperforms the other baseline schemes, and 2) the proposed HARL algorithm achieves better performance compared to conventional RL algorithms.

💡 Research Summary

**
This paper proposes an integrated sensing and communication (ISAC) framework that leverages segmented waveguide‑enabled pinching‑antenna arrays (SWAN) to address the stringent requirements of emerging 6G networks. Traditional phased‑array or reconfigurable antennas suffer from high cost, complex feeding networks, and limited flexibility, while pinching antennas—created by physically “pinching” a waveguide—offer low‑complexity radiation points but have been implemented on long, single waveguides, leading to excessive propagation loss and maintenance overhead.

The authors therefore divide each waveguide into multiple electrically isolated segments, each connected to a dedicated RF chain. This segmented architecture (SWAN) enables selective activation of waveguide sections, dramatically reducing in‑waveguide loss and allowing fine‑grained spatial resource allocation. To exploit this capability, a Hybrid Segment Selection and Multiplexing (HSSM) protocol is introduced: each segment can be independently turned on or off, and the active segments are multiplexed across the available RF chains, achieving hardware cost savings while preserving performance.

A joint optimization problem is formulated over three coupled design variables: (i) the transmit beamforming matrix W, (ii) binary segment‑selection variables φ, and (iii) the three‑dimensional positions ψ of the pinching antennas along each waveguide. The objective maximizes the sum communication rate across all downlink users, subject to (a) a minimum illumination power Γ̃ for each sensing target (ensuring sufficient radar‑SNR), (b) total transmit power constraints, (c) binary bounds on segment activation, and (d) a minimum spacing constraint between any two pinching antennas (to avoid mutual coupling). The resulting problem is highly non‑convex due to exponential phase terms, absolute‑value operations, and the mixed discrete‑continuous nature of the variables.

To solve this challenging problem, the paper introduces a Segment‑Hysteresis Reinforcement Learning (SHRL) algorithm. The system is modeled as a Markov Decision Process (MDP) where the state includes current communication and sensing channel state information (CSI), as well as the previous antenna positions and segment selections—information essential for the hysteresis mechanism. The action space comprises continuous updates to W, ψ, and raw segment logits φ. A probabilistic hysteresis gate is applied to the raw logits: with probability p_update the new segment decision is accepted; otherwise the previous decision is retained. This gate prevents abrupt, frequent re‑mapping of antennas across segments, thereby stabilizing the learning process and reducing non‑stationarity.

The learning backbone is an Advantage Actor‑Critic (A2C) scheme. The actor outputs both the continuous beamforming/position parameters and the raw segment logits; the critic estimates the state‑value function to compute the advantage. The reward function is defined as the sum of user rates minus a large penalty whenever any target’s illumination falls below the required threshold, encouraging the agent to satisfy sensing constraints while maximizing throughput.

Simulation settings reflect realistic mmWave parameters: carrier frequency 28 GHz, waveguide height 5 m, total transmit power 100 W, each segment containing N = 10 pinching antennas, and a maximum of M = 3 segments. Six communication users and one sensing target are randomly placed within a 50 m × 60 m ground area. Two deployment scenarios are examined: a “sparse” case where users and the target are well separated, and a “dense” case where they are clustered. The proposed SWAN‑HSSM architecture is compared against a conventional PASS scheme (fixed, non‑adaptive antenna layout). Four reinforcement‑learning baselines are also evaluated: (i) SPRL (segment selection at fixed intervals), (ii) standard A2C, (iii) Proximal Policy Optimization (PPO), and (iv) a random‑action policy.

Results show that (1) SWAN‑HSSM consistently outperforms PASS in both communication rate and sensing illumination, with the advantage being most pronounced in the sparse scenario; (2) the SHRL algorithm achieves the highest cumulative reward, converges faster, and exhibits the most stable learning curve among all RL baselines; (3) SPRL converges more slowly due to its reduced exploration flexibility; (4) PPO and A2C attain respectable performance but display larger variance and require more episodes to reach comparable rewards; (5) the random policy performs worst, confirming the necessity of informed decision making. A further study varying the total waveguide length demonstrates that SHRL maintains superior ISAC performance across all lengths, underscoring its robustness to hardware scaling.

In summary, the paper makes three key contributions: (i) it introduces a segmented waveguide‑based pinching‑antenna architecture that mitigates propagation loss and reduces hardware complexity; (ii) it proposes the HSSM protocol that efficiently multiplexes selected segments, achieving better power and spectrum utilization; and (iii) it develops the SHRL algorithm, which integrates a hysteresis mechanism into reinforcement learning to handle the non‑convex, mixed‑integer ISAC optimization problem with improved stability and performance. The findings suggest that combining flexible segmented hardware with hysteresis‑aware learning offers a promising pathway for practical, high‑performance ISAC deployments in future wireless networks. Future work may explore hardware prototyping, multi‑target multi‑user extensions, and lightweight online RL schemes for real‑time adaptation.