Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transformer-Based Sparse CSI Estimation for Non-Stationary Channels

Muhammad Ahmed Mohsin, Muhammad Umer, Ahsan Bilal, Hassan Rizwan, Sagnik Bhattacharya, Muhammad Ali Jamshed, John M. Cioffi

TL;DR

This work tackles CSI estimation under non-stationary, high-m mobility wireless channels by introducing a pilot-aided Flash-Attention Transformer that unifies pilot acquisition with data-driven CSI reconstruction. Through patch-wise self-attention and a physics-aware composite loss, the model learns long-range time–frequency dependencies while conditioning on the pilot mask to adapt to Doppler and multipath dynamics. In standardized 3GPP NR MIMO-OFDM scenarios, it outperforms LMMSE, LSTM, and LDAMP baselines by about $13~\mathrm{dB}$ in phase-invariant NMSE and achieves markedly lower BER while reducing pilot overhead by $16\times$, approaching the oracle dense-pilot performance. This approach significantly enhances spectral efficiency and link reliability in non-stationary 5G and beyond-5G networks by enabling high-fidelity CSI reconstruction from sparse pilots.

Abstract

Accurate and efficient estimation of Channel State Information (CSI) is critical for next-generation wireless systems operating under non-stationary conditions, where user mobility, Doppler spread, and multipath dynamics rapidly alter channel statistics. Conventional pilot aided estimators incur substantial overhead, while deep learning approaches degrade under dynamic pilot patterns and time varying fading. This paper presents a pilot-aided Flash-Attention Transformer framework that unifies model-driven pilot acquisition with data driven CSI reconstruction through patch-wise self-attention and a physics aware composite loss function enforcing phase alignment, correlation consistency, and time frequency smoothness. Under a standardized 3GPP NR configuration, the proposed framework outperforms LMMSE and LSTM baselines by approximately 13 dB in phase invariant normalized mean-square error (NMSE) with markedly lower bit-error rate (BER), while reducing pilot overhead by 16 times. These results demonstrate that attention based architectures enable reliable CSI recovery and enhanced spectral efficiency without compromising link quality, addressing a fundamental bottleneck in adaptive, low-overhead channel estimation for non-stationary 5G and beyond-5G networks.

Transformer-Based Sparse CSI Estimation for Non-Stationary Channels

TL;DR

This work tackles CSI estimation under non-stationary, high-m mobility wireless channels by introducing a pilot-aided Flash-Attention Transformer that unifies pilot acquisition with data-driven CSI reconstruction. Through patch-wise self-attention and a physics-aware composite loss, the model learns long-range time–frequency dependencies while conditioning on the pilot mask to adapt to Doppler and multipath dynamics. In standardized 3GPP NR MIMO-OFDM scenarios, it outperforms LMMSE, LSTM, and LDAMP baselines by about in phase-invariant NMSE and achieves markedly lower BER while reducing pilot overhead by , approaching the oracle dense-pilot performance. This approach significantly enhances spectral efficiency and link reliability in non-stationary 5G and beyond-5G networks by enabling high-fidelity CSI reconstruction from sparse pilots.

Abstract

Accurate and efficient estimation of Channel State Information (CSI) is critical for next-generation wireless systems operating under non-stationary conditions, where user mobility, Doppler spread, and multipath dynamics rapidly alter channel statistics. Conventional pilot aided estimators incur substantial overhead, while deep learning approaches degrade under dynamic pilot patterns and time varying fading. This paper presents a pilot-aided Flash-Attention Transformer framework that unifies model-driven pilot acquisition with data driven CSI reconstruction through patch-wise self-attention and a physics aware composite loss function enforcing phase alignment, correlation consistency, and time frequency smoothness. Under a standardized 3GPP NR configuration, the proposed framework outperforms LMMSE and LSTM baselines by approximately 13 dB in phase invariant normalized mean-square error (NMSE) with markedly lower bit-error rate (BER), while reducing pilot overhead by 16 times. These results demonstrate that attention based architectures enable reliable CSI recovery and enhanced spectral efficiency without compromising link quality, addressing a fundamental bottleneck in adaptive, low-overhead channel estimation for non-stationary 5G and beyond-5G networks.

Paper Structure

This paper contains 14 sections, 1 theorem, 18 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Model
  3. Channel Model and Pilot Configurations
  4. Pilot Overhead and Achievable Rate
  5. Methodology
  6. Dataset Preparation
  7. Model Architecture
  8. Channel Estimation via Spatio-Temporal Interpolation
  9. Loss Formulation
  10. Evaluation Flow
  11. Simulation Results
  12. Baselines and Metrics
  13. Results
  14. Conclusion

Key Result

Proposition 1

Consider a block-fading channel with coherence block length $T_c$ REs, pilot fraction $\alpha=\tau_p/T_c\in(0,1)$, per-RE SNR $\rho$, Rayleigh fading, and LMMSE channel estimation from $\tau_p$ orthogonal pilots. The standard training-based achievable rate per RE is If a reconstruction mechanism enables a reduced pilot fraction $\alpha_1<\alpha_0$ while maintaining or improving the effective SNR,

Figures (3)

  • Figure 1: Proposed pilot-aided Transformer framework for CSI reconstruction. Sparse pilots are first interpolated to form an initial estimate, which is then refined by the Transformer to recover high-fidelity CSI, followed by performance evaluation.
  • Figure 2: NMSE performance and subcarrier-wise accuracy of the Transformer estimator versus baselines across SNRs.
  • Figure 3: Time–frequency magnitude heatmaps of CSI: (a) noisy and interpolated sparse-pilot estimate, (b) Transformer-reconstructed CSI, and (c) oracle dense-pilot reference.

Theorems & Definitions (2)

  • Proposition 1: Rate Gain from Pilot Reduction under Reliable Reconstruction
  • proof