JEPA · 2026-10-12
JEPA-kMMVAE: CSI Prediction for Time-Varying Wireless Channels
Efstathios Chatziloizos, Konstantinos Vandikas, Aneta Vulgarakis Feljan, Zheng Chen, Nikolaos Pappas
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
JEPA-kMMVAE predicts the next clean multi-antenna CSI snapshot from a finite history of noisy CSI. It combines a joint-embedding predictive architecture with a k-memory Markov variational autoencoder for temporally structured latent representations. The authors provide direct and residual decoding variants (residual uses a denoising backbone and residual references). Simulations on 3GPP UMa channels show residual JEPA-kMMVAE is best at moderate/high input SNRs, while sequence baselines win under stronger noise.
Why it matters
CSI prediction is central for using timely, task-relevant channel information in time-varying wireless systems. This work targets the specific issue of separating temporal channel evolution from measurement noise, by forecasting in a temporally structured latent space and using residual correction relative to the latest latent and noisy CSI observation.
Method
- Model causal one-step CSI prediction: map noisy history Y_t to next clean snapshot x̂_{t+1}, supervised with clean future CSI only during training/evaluation.
- Use a pretrained k-MMVAE encoder with k-memory Markov latent dynamics; freeze the backbone and propagate posterior means for prediction.
- Predict next latent via a GRU + linear residual increment, then decode future CSI with either direct decoding or residual denoising correction using the latest latent and noisy CSI reference.
Limitation
The authors’ evaluation is simulation-based on QuaDRiGa 3GPP 38.901 UMa channels (per subcarrier narrowband OFDM, without accounting for inter-subcarrier correlation).
Abstract (from arXiv)
This paper considers channel state information (CSI) prediction using neural networks for time-varying multi-antenna channels from noisy historical observations. The main challenge is to exploit temporal correlation while distinguishing channel evolution from observation noise. We propose JEPA-kMMVAE, which combines a joint-embedding predictive architecture with a k-memory Markov variational autoencoder to predict future CSI through temporally structured latent representations. The framework supports direct prediction and residual corrections relative to the latest latent state and noisy CSI observation. Simulation results show that the residual configuration outperforms the evaluated baselines at moderate and high input CSI signal-to-noise ratios, while sequence baselines remain more accurate under stronger noise.
Related papers
- JEPA-Inspired Latent Field Prediction for Sparse-Pilot MIMO-OFDM Channel Reconstruction
- Does Joint-Embedding Predictive Architecture Pretraining Help Time Series Forecasting?
- S2S-JEPA: Predicting the Predictable at Subseasonal-to-Seasonal Timescales
- A Stochastic Latent Variable Generalization of Joint-Embedding Predictive Architectures
- Self-Supervised Latent Action-Conditioned World Modeling for Time Series
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-09 · Indexed by Pub-AI: 2026-10-12 UTC
JEPA-kMMVAE: CSI Prediction for Time-Varying Wireless Channels
Preprint posted: 2026-10-09 · Original record ↗
- arxiv
- 2610.13315
Explicit source versions (1)
- v1 · 2026-10-09
- Preprint available
Source date: 2026-10-09 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
This paper considers channel state information (CSI) prediction using neural networks for time-varying multi-antenna channels from noisy historical observations. The main challenge is to exploit temporal correlation while distinguishing channel evolution from observation noise. We propose JEPA-kMMVAE, which combines a joint-embedding predictive architecture with a k-memory Markov variational autoencoder to predict future CSI through temporally structured latent representations. The framework supports direct prediction and residual corrections relative to the latest latent state and noisy CSI observation. Simulation results show that the residual configuration outperforms the evaluated baselines at moderate and high input CSI signal-to-noise ratios, while sequence baselines remain more accurate under stronger noise.
