Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hybrid CNN-Transformer Based Sparse Channel Prediction for High-Mobility OTFS Systems

Zhaowei Guan, Wenkun Wen, Peiran Wu, Chen Wang, Minghua Xia

TL;DR

This work tackles channel prediction for high-mobility OTFS systems in URLLC scenarios by leveraging DD-domain sparsity and temporal dynamics. It introduces LDformer, a compact CNN–Transformer architecture with CNN-based downsampling and causal-masked Transformer layers to predict $\hat{\bm{H}}_t$ from history $\{\bm{H}_{t-L},\dots,\bm{H}_{t-1}\}$, reducing the effective dimensionality from $MN \times MN$ to a latent size $R$ before reconstruction. In simulations at $500$ km/h, LDformer achieves RMSE $=0.0215$ and MAE $=0.00261$, improving RMSE by $12.2\%$ and MAE by $9.4\%$ over the strongest baselines, while maintaining a lightweight model (~$25.6$M parameters) and fast inference (~$10.6$ ms per sample). The results highlight the benefits of DD-domain representations for robust prediction in rapidly varying channels and demonstrate a practical, scalable approach for URLLC-enabled OTFS systems. Future work targets real-time deployment, edge-model compression, and extensions to multi-antenna OTFS.

Abstract

High-mobility scenarios in next-generation wireless networks, such as those involving vehicular communications, require ultra-reliable and low-latency communications (URLLC). However, rapidly time-varying channels pose significant challenges to traditional OFDM-based systems due to the Doppler effect and channel aging. Orthogonal time frequency space (OTFS) modulation offers resilience by representing channels in the quasi-static delay-Doppler (DD) domain. This letter proposes a novel channel prediction framework for OTFS systems using a hybrid convolutional neural network and transformer (CNN-Transformer) architecture. The CNN extracts compact features that exploit the DD-domain sparsity of the channel matrices, while the transformer models temporal dependencies with causal masking for consistency. Simulation experiments under extreme $500$ \si{km/h} mobility conditions demonstrate that the proposed method outperforms state-of-the-art baselines, reducing the root mean square error and mean absolute error by $12.2\%$ and $9.4\%$, respectively. These results demonstrate the effectiveness of DD-domain representations and the proposed model in accurately predicting channels in high-mobility scenarios, thereby supporting the stringent URLLC requirements in future wireless systems.

Hybrid CNN-Transformer Based Sparse Channel Prediction for High-Mobility OTFS Systems

TL;DR

This work tackles channel prediction for high-mobility OTFS systems in URLLC scenarios by leveraging DD-domain sparsity and temporal dynamics. It introduces LDformer, a compact CNN–Transformer architecture with CNN-based downsampling and causal-masked Transformer layers to predict from history , reducing the effective dimensionality from to a latent size before reconstruction. In simulations at km/h, LDformer achieves RMSE and MAE , improving RMSE by and MAE by over the strongest baselines, while maintaining a lightweight model (~M parameters) and fast inference (~ ms per sample). The results highlight the benefits of DD-domain representations for robust prediction in rapidly varying channels and demonstrate a practical, scalable approach for URLLC-enabled OTFS systems. Future work targets real-time deployment, edge-model compression, and extensions to multi-antenna OTFS.

Abstract

High-mobility scenarios in next-generation wireless networks, such as those involving vehicular communications, require ultra-reliable and low-latency communications (URLLC). However, rapidly time-varying channels pose significant challenges to traditional OFDM-based systems due to the Doppler effect and channel aging. Orthogonal time frequency space (OTFS) modulation offers resilience by representing channels in the quasi-static delay-Doppler (DD) domain. This letter proposes a novel channel prediction framework for OTFS systems using a hybrid convolutional neural network and transformer (CNN-Transformer) architecture. The CNN extracts compact features that exploit the DD-domain sparsity of the channel matrices, while the transformer models temporal dependencies with causal masking for consistency. Simulation experiments under extreme \si{km/h} mobility conditions demonstrate that the proposed method outperforms state-of-the-art baselines, reducing the root mean square error and mean absolute error by and , respectively. These results demonstrate the effectiveness of DD-domain representations and the proposed model in accurately predicting channels in high-mobility scenarios, thereby supporting the stringent URLLC requirements in future wireless systems.
Paper Structure (9 sections, 6 equations, 9 figures, 2 tables, 1 algorithm)

This paper contains 9 sections, 6 equations, 9 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. A Hybrid CNN-Transformer Architecture
  4. Input Preprocessing and Data Representation
  5. Feature Extraction via Convolutional Downsampling
  6. Temporal Dependency Modeling with Transformer Architecture
  7. Feature Reconstruction and Channel Prediction
  8. Numerical Results and Discussions
  9. Conclusions

Figures (9)

  • Figure 1: OTFS modulation system structure.
  • Figure 2: Hybrid CNN-Transformer architecture based on compact features that exploit DD-domain sparsity.
  • Figure 3: Channel variation in TF and DD domains at the speed of $500$ km/h.
  • Figure 4: Downsampling convolution block and upsampling transposed convolution block.
  • Figure 5: Temporal Transformer block structure.
  • ...and 4 more figures