⚡ electrical engineering

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

This paper proposes an Integrated Sensing and Communication (ISAC) framework for segmented waveguide-enabled pinching antenna (SWAN) systems that utilizes a hybrid segment selection and multiplexing protocol and a segment hysteresis-based reinforcement learning algorithm to jointly optimize beamforming, segment selection, and antenna positioning, thereby maximizing communication rates while satisfying sensing constraints with reduced hardware costs.

Original authors: Qian Gao, Ruikang Zhong, Hyundong Shin, Yuanwei Liu

Published 2026-01-29

📖 4 min read☕ Coffee break read

CC0 1.0

Original authors: Qian Gao, Ruikang Zhong, Hyundong Shin, Yuanwei Liu

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Smart, Flexible "Light Show" for Data and Radar

Imagine you are trying to do two things at once in a large, dark room:

Talk to friends (Communication): You need to send clear messages to specific people scattered around the room.
Find lost items (Sensing): You need to shine a light to see where objects or people are hiding.

In the world of 6G wireless networks, doing both of these efficiently is a huge challenge. This paper proposes a new way to build the "spotlights" (antennas) that handle this job. They call their system SWAN-ISAC.

The Problem: The "Long Hose" Issue

Traditionally, these antennas are like one giant, continuous hose (a waveguide) that runs for tens of meters. To send a signal, you have to pump energy all the way down this long hose.

The Flaw: Just like water losing pressure as it travels through a very long pipe, the signal loses strength (energy) as it travels down a long waveguide.
The Rigidity: If you want to change where the light shines, you often have to move heavy mechanical parts or use expensive, complex electronics. It's like trying to rearrange a giant, heavy chandelier just to light up a different corner of the room.

The Solution: The "Modular Train" (Segmented Waveguides)

The authors suggest breaking that giant hose into smaller, independent segments. Think of it like a train where each car is a separate spotlight.

Pinching Antennas: Instead of moving the whole train, you can "pinch" (activate) specific spots on any of the train cars to create a beam of light. It's like having a flexible ruler where you can create a light source at any point along the line.
The Hybrid Strategy (HSSM): The system uses a smart protocol called "Hybrid Segment Selection and Multiplexing." Imagine a conductor who decides: "Today, we only need cars 1, 3, and 5 to run. Let's turn off cars 2 and 4 to save fuel." This saves money and energy while keeping the signal strong because the light doesn't have to travel as far through the "pipe."

The Brain: The "Hysteresis" Coach (SHRL)

The hardest part is deciding which train cars to use and where to pinch the light, especially when people (users) and objects (targets) are moving. This is a massive math puzzle that is too hard for a standard computer to solve quickly.

The authors created a new AI coach called SHRL (Segment Hysteresis Reinforcement Learning).

How it learns: The AI tries different combinations of train cars and light positions. If a combination works well (people hear clearly, and objects are seen), it gets a "reward."
The Secret Sauce (Hysteresis): This is the paper's main innovation. Imagine a coach who is a bit stubborn. If the team is doing okay, the coach doesn't immediately panic and change the whole lineup just because of one small mistake.
- Without Hysteresis: A standard AI might flip-flop wildly, changing the train configuration every second, which confuses the system.
- With Hysteresis: The SHRL coach says, "Wait, the current setup is still good enough. Let's stick with it unless the new idea is significantly better." This prevents the system from shaking and jittering, allowing it to learn faster and more stably.

The Results: Who Won the Race?

The authors ran simulations (computer tests) to see how their new system compared to old methods and other AI algorithms.

Better than the Old Way: The new "Modular Train" system (SWAN) performed much better than the old "Giant Hose" system (PASS). It sent more data and found targets more accurately.
The Best Coach: The SHRL algorithm (the stubborn coach) beat other popular AI coaches like A2C, PPO, and Random guessing.
- In tests where users and targets were spread out (Sparse case), SHRL was the clear winner.
- Even when the room was crowded and chaotic (Dense case), SHRL still managed to keep the signal strong and the radar working, while others struggled.
Handling Length: When they made the "hoses" very long (100 meters), the signal naturally got weaker for everyone. However, SHRL was the most robust, handling the loss better than the others.

Summary

This paper introduces a smarter, more flexible way to build 6G antennas. Instead of one long, inefficient wire, they use a modular system of segments that can be turned on and off like train cars. To control this complex system, they invented a new AI method that is "stubborn" enough to avoid making unnecessary changes, resulting in faster learning and better performance for both talking to phones and sensing the environment.

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

1. Problem Statement

The paper addresses the challenges of designing Integrated Sensing and Communication (ISAC) systems for sixth-generation (6G) networks, which require simultaneous high-rate data transmission and fine-grained environmental awareness. While traditional phased arrays and reconfigurable antennas offer beam-steering capabilities, they often suffer from high hardware costs, complex feeding networks, and limited flexibility.

Recent advancements introduced "pinching antennas" (PAs), which create radiation points by pinching a waveguide, offering high spatial flexibility and lower complexity. However, existing PA designs typically rely on single or long continuous waveguides (tens of meters), which degrade performance over large regions due to propagation loss and incur high maintenance costs. To address these limitations, the paper investigates Segmented Waveguide-Enabled Pinching Antenna (SWAN) systems. The core challenge is to jointly optimize transmit beamforming, segment selection, and pinching antenna positioning to maximize the sum communication rate while satisfying sensing performance constraints, a problem characterized as non-convex and nonlinear.

2. Methodology

System Model

The authors propose a downlink SWAN-ISAC system where a base station is equipped with $M$ segmented waveguides, each containing $N$ pinching antennas. The system serves $K_c$ communication users and detects $K_s$ sensing targets.

Signal Model: The system employs a Hybrid Segment Selection and Multiplexing (HSSM) protocol. Each waveguide segment is connected to a dedicated Radio Frequency (RF) chain. Signals are fed into the waveguides, and the PAs radiate phase-shifted versions of the signal, accounting for in-waveguide propagation loss.
Metrics: The objective is to maximize the sum communication rate ( $R_{kc}$ ) subject to a minimum illumination power threshold ( $\tilde{\Gamma}$ ) for sensing targets. Constraints include total transmit power limits and minimum spacing between adjacent pinching antennas.

Proposed Solution: Segment Hysteresis Reinforcement Learning (SHRL)

To solve the non-convex joint optimization problem, the authors propose a Segment Hysteresis based Reinforcement Learning (SHRL) algorithm.

MDP Formulation: The problem is modeled as a Markov Decision Process (MDP). The state space includes channel state information (CSI), previous antenna positions, and previous segment assignments. The action space involves continuous updates for antenna positions, segment selection logits, and beamforming weights.
Hysteresis Mechanism: Unlike conventional Reinforcement Learning (RL) that updates segment selections independently at every step, SHRL introduces a probabilistic hysteresis gate. This mechanism prevents abrupt changes in the antenna-segment mapping. New segment decisions are only accepted if they provide sufficient improvement or exceed a specific threshold; otherwise, previous selections are retained. This reduces structural variance and mitigates non-stationarity caused by rapid switching.
Training: The algorithm utilizes the Advantage Actor-Critic (A2C) framework. The actor outputs beamforming parameters and raw segment logits, while the hysteresis gate refines the segment selection before execution. The critic evaluates state values to guide policy updates.

3. Key Contributions

SWAN-ISAC Framework: The paper proposes an ISAC design leveraging segmented waveguides to reduce in-waveguide propagation loss and enhance spatial resource allocation flexibility compared to continuous waveguide designs.
HSSM Protocol: Implementation of a hybrid segment selection and multiplexing protocol that connects segments to dedicated RF chains, balancing performance with reduced hardware costs.
SHRL Algorithm: Development of a novel RL algorithm that incorporates a hysteresis-based update mechanism. This specifically addresses the stability issues in optimizing discrete segment selections within a continuous control environment, ensuring temporal consistency in policy exploration.
Joint Optimization: Formulation and solution of a complex joint optimization problem involving transmit beamforming, segment selection, and physical antenna positioning under sensing constraints.

4. Simulation Results

The authors conducted simulations in a 50m $\times$ 60m area with 6 users and 1 target, operating at 28 GHz.

Protocol Comparison: The proposed SWAN-HSSM scheme significantly outperformed the SWAN-SM (single multiplexing) and conventional PASS (Periodic Antenna Selection Scheme) baselines in both sparse and dense user/target scenarios. SWAN-HSSM achieved the highest rewards by effectively selecting active segments to reduce power spreading and mitigate interference.
Algorithm Performance: SHRL demonstrated superior convergence stability and achieved the highest final reward compared to SPRL (periodic selection), A2C, PPO, and Random schemes. The hysteresis mechanism prevented unnecessary segment switching, leading to more stable learning.
Configuration Sensitivity:
- Waveguide Length: As total waveguide length increased (up to 80m), performance improved due to larger spatial aperture. However, at very long lengths (100m), propagation loss caused degradation across all algorithms, with SHRL remaining the most robust.
- Segment Number: Increasing the number of segments ( $M$ ) improved sensing SNR (due to lower per-segment attenuation) but reduced communication rates (due to fewer antennas per segment and reduced array gain). SHRL consistently maintained the best trade-off between these competing metrics.

5. Significance and Claims

The paper claims that the proposed SWAN-ISAC architecture, combined with the SHRL algorithm, offers a viable solution for high-performance, power-limited ISAC deployments. The significance lies in:

Hardware Efficiency: Utilizing segmented waveguides and pinching antennas reduces hardware complexity and maintenance costs compared to traditional phased arrays while offering greater flexibility than single continuous waveguides.
Stable Optimization: The introduction of the hysteresis mechanism in RL provides a method to stabilize the learning process in systems with discrete structural changes (segment selection), a challenge often faced by standard RL algorithms like A2C or PPO.
Performance Gains: The system achieves a superior balance between communication throughput and sensing accuracy, maintaining sensing SNR near operational thresholds while significantly boosting data rates compared to baseline schemes.

The authors conclude that segmented-waveguide antenna systems are a promising enabler for future 6G ISAC applications, particularly where flexible beamforming and efficient separation of communication and sensing functions are required.