Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Conditional Variational Framework for Channel Prediction in High-Mobility 6G OTFS Networks

Mohsen Kazemian, Jürgen Jasperneite

TL;DR

The paper tackles rapid channel variations in high-mobility OTFS by proposing a conditional variational autoencoder (CVAE) to learn $p(\mathbf{X}\mid \mathbf{e})$, where $\mathbf{X}$ encodes channel coefficients in the delay-Doppler domain and $\mathbf{e}$ collects system parameters such as Doppler spread, SNR, and grid sizes. To capture non-Gaussian and multimodal channel statistics, the authors augment the CVAE with normalizing flows, transforming $z_0\sim \mathcal{N}(\mathbf{0},\mathbf{I})$ through a sequence $z_k=f_k(z_{k-1})$ conditioned on $\mathbf{e}$. Training uses the conditional ELBO $\mathcal{L}(\theta,\phi)$ and evaluation shows substantial NMSE gains over a competitive RNN baseline, particularly at high Doppler frequencies and long prediction horizons. This probabilistic, conditioning-based framework enables future channel prediction without instantaneous CSI, reducing pilot overhead and enhancing spectral efficiency in 6G OTFS networks.

Abstract

This paper proposes a machine learning (ML) based method for channel prediction in high mobility orthogonal time frequency space (OTFS) channels. In these scenarios, rapid variations caused by Doppler spread and time varying multipath propagation lead to fast channel decorrelation, making conventional pilot based channel estimation methods prone to outdated channel state information (CSI) and excessive overhead. Therefore, reliable channel prediction methods become essential to support robust detection and decoding in OTFS systems. In this paper, we propose conditional variational autoencoder for channel prediction (CVAE4CP) method, which learns the conditional distribution of OTFS delay Doppler channel coefficients given physical system and mobility parameters. By incorporating these parameters as conditioning information, the proposed method enables the prediction of future channel coefficients before their actual realization, while accounting for inherent channel uncertainty through a low dimensional latent representation. The proposed framework is evaluated through extensive simulations under high mobility conditions. Numerical results demonstrate that CVAE4CP consistently outperforms a competing learning based baseline in terms of normalized mean squared error (NMSE), particularly at high Doppler frequencies and extended prediction horizons. These results confirm the effectiveness and robustness of the proposed approach for channel prediction in rapidly time varying OTFS systems.

A Conditional Variational Framework for Channel Prediction in High-Mobility 6G OTFS Networks

TL;DR

The paper tackles rapid channel variations in high-mobility OTFS by proposing a conditional variational autoencoder (CVAE) to learn , where encodes channel coefficients in the delay-Doppler domain and collects system parameters such as Doppler spread, SNR, and grid sizes. To capture non-Gaussian and multimodal channel statistics, the authors augment the CVAE with normalizing flows, transforming through a sequence conditioned on . Training uses the conditional ELBO and evaluation shows substantial NMSE gains over a competitive RNN baseline, particularly at high Doppler frequencies and long prediction horizons. This probabilistic, conditioning-based framework enables future channel prediction without instantaneous CSI, reducing pilot overhead and enhancing spectral efficiency in 6G OTFS networks.

Abstract

This paper proposes a machine learning (ML) based method for channel prediction in high mobility orthogonal time frequency space (OTFS) channels. In these scenarios, rapid variations caused by Doppler spread and time varying multipath propagation lead to fast channel decorrelation, making conventional pilot based channel estimation methods prone to outdated channel state information (CSI) and excessive overhead. Therefore, reliable channel prediction methods become essential to support robust detection and decoding in OTFS systems. In this paper, we propose conditional variational autoencoder for channel prediction (CVAE4CP) method, which learns the conditional distribution of OTFS delay Doppler channel coefficients given physical system and mobility parameters. By incorporating these parameters as conditioning information, the proposed method enables the prediction of future channel coefficients before their actual realization, while accounting for inherent channel uncertainty through a low dimensional latent representation. The proposed framework is evaluated through extensive simulations under high mobility conditions. Numerical results demonstrate that CVAE4CP consistently outperforms a competing learning based baseline in terms of normalized mean squared error (NMSE), particularly at high Doppler frequencies and extended prediction horizons. These results confirm the effectiveness and robustness of the proposed approach for channel prediction in rapidly time varying OTFS systems.
Paper Structure (8 sections, 3 equations, 3 figures, 1 algorithm)

This paper contains 8 sections, 3 equations, 3 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. Proposed Method
  4. Problem Formulation
  5. Proposed CVAE-Based OTFS Channel Modeling
  6. Latent Space Enhancement via Normalizing Flows
  7. Simulation Results
  8. Conclusion

Figures (3)

  • Figure 1: Wireless communication system with receiver-side CSI prediction.
  • Figure 2: NMSE versus maximum Doppler frequency for OTFS channel prediction, comparing the proposed CVAE4CP with the RNN-based method in 2aidriven.
  • Figure 3: NMSE versus prediction horizon (number of OTFS frames ahead), comparing the proposed CVAE$4$CP with the RNN-based method in 2aidriven.