Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Channel Estimation for Pinching-Antenna Systems (PASS)

Jian Xiao, Ji Wang, Yuanwei Liu

TL;DR

Channel estimation in PASS is challenging because the in-waveguide channel $\mathbf{g}$ couples with the wireless channel $\mathbf{h}_k$, making the problem ill-posed from low-dimensional pilot observations. The authors introduce two DL-based estimators: PAMoE, a mixture-of-experts network that uses dynamic padding and Fourier positional embeddings to leverage PA positions; and PAformer, a Transformer-based architecture that predicts per-antenna channel coefficients and scales to arbitrary PA counts with self-attention. They build a large offline dataset and demonstrate that both estimators outperform conventional LS/LMMSE and DL baselines, with zero-shot generalization to unseen PA configurations and reduced pilot overhead. The results indicate real-time feasibility (microsecond-scale inference) and practical applicability for PASS deployments with flexible PA configurations.

Abstract

Pinching Antennas (PAs) represent a revolutionary flexible antenna technology that leverages dielectric waveguides and electromagnetic coupling to mitigate large-scale path loss. This letter is the first to explore channel estimation for Pinching-Antenna SyStems (PASS), addressing their uniquely ill-conditioned and underdetermined channel characteristics. In particular, two efficient deep learning-based channel estimators are proposed. 1) PAMoE: This estimator incorporates dynamic padding, feature embedding, fusion, and mixture of experts (MoE) modules, which effectively leverage the positional information of PAs and exploit expert diversity. 2) PAformer: This Transformer-style estimator employs the self-attention mechanism to predict channel coefficients in a per-antenna manner, which offers more flexibility to adaptively deal with dynamic numbers of PAs in practical deployment. Numerical results demonstrate that 1) the proposed deep learning-based channel estimators outperform conventional methods and exhibit excellent zero-shot learning capabilities, and 2) PAMoE delivers higher channel estimation accuracy via MoE specialization, while PAformer natively handles an arbitrary number of PAs, trading self-attention complexity for superior scalability.

Channel Estimation for Pinching-Antenna Systems (PASS)

TL;DR

Channel estimation in PASS is challenging because the in-waveguide channel couples with the wireless channel , making the problem ill-posed from low-dimensional pilot observations. The authors introduce two DL-based estimators: PAMoE, a mixture-of-experts network that uses dynamic padding and Fourier positional embeddings to leverage PA positions; and PAformer, a Transformer-based architecture that predicts per-antenna channel coefficients and scales to arbitrary PA counts with self-attention. They build a large offline dataset and demonstrate that both estimators outperform conventional LS/LMMSE and DL baselines, with zero-shot generalization to unseen PA configurations and reduced pilot overhead. The results indicate real-time feasibility (microsecond-scale inference) and practical applicability for PASS deployments with flexible PA configurations.

Abstract

Pinching Antennas (PAs) represent a revolutionary flexible antenna technology that leverages dielectric waveguides and electromagnetic coupling to mitigate large-scale path loss. This letter is the first to explore channel estimation for Pinching-Antenna SyStems (PASS), addressing their uniquely ill-conditioned and underdetermined channel characteristics. In particular, two efficient deep learning-based channel estimators are proposed. 1) PAMoE: This estimator incorporates dynamic padding, feature embedding, fusion, and mixture of experts (MoE) modules, which effectively leverage the positional information of PAs and exploit expert diversity. 2) PAformer: This Transformer-style estimator employs the self-attention mechanism to predict channel coefficients in a per-antenna manner, which offers more flexibility to adaptively deal with dynamic numbers of PAs in practical deployment. Numerical results demonstrate that 1) the proposed deep learning-based channel estimators outperform conventional methods and exhibit excellent zero-shot learning capabilities, and 2) PAMoE delivers higher channel estimation accuracy via MoE specialization, while PAformer natively handles an arbitrary number of PAs, trading self-attention complexity for superior scalability.

Paper Structure

This paper contains 14 sections, 11 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. Deep Learning Based Channel Estimation
  4. Dataset Construction
  5. PAMoE: Mixture of Experts-Based Channel Estimation
  6. Dynamic Padding
  7. Feature Embedding
  8. Feature Fusion
  9. MoE With Gating Network
  10. PAformer: Transformer-Based Channel Estimation
  11. Input Embedding
  12. Transformer Encoder
  13. Numerical Results
  14. Conclusions

Figures (5)

  • Figure 1: Pinching antennas assisted multi-user systems.
  • Figure 2: Proposed PAMoE model for channel estimation.
  • Figure 3: Proposed PAformer model for channel estimation.
  • Figure 4: NMSE vs. SNR for different algorithms.
  • Figure 5: NMSE vs. number of PAs $N$ for different algorithms.