Reducing Pilots in Channel Estimation With Predictive Foundation Models

TL;DR

This paper tackles the challenge of acquiring accurate channel state information with reduced pilot overhead in next-generation wireless systems featuring large antenna arrays. It introduces a predictive foundation model (PFM) that forecasts future CSI from historical data and fuses these predictive priors with a ViT-based pilot-processing network, enabling robust CSI reconstruction under sparse pilots and diverse environments. Through a two-phase training strategy and extensive simulations, the approach demonstrates substantial NMSE and BER improvements and notable zero-shot generalization to unseen speeds, antennas, and channel profiles. The results suggest that PFMs can serve as a scalable, generalizable backbone for intelligent, low-overhead channel acquisition in future wireless networks.

Abstract

Accurate channel state information (CSI) acquisition is essential for modern wireless systems, which becomes increasingly difficult under large antenna arrays, strict pilot overhead constraints, and diverse deployment environments. Existing artificial intelligence-based solutions often lack robustness and fail to generalize across scenarios. To address this limitation, this paper introduces a predictive-foundation-model-based channel estimation framework that enables accurate, low-overhead, and generalizable CSI acquisition. The proposed framework employs a predictive foundation model trained on large-scale cross-domain CSI data to extract universal channel representations and provide predictive priors with strong cross-scenario transferability. A pilot processing network based on a vision transformer architecture is further designed to capture spatial, temporal, and frequency correlations from pilot observations. An efficient fusion mechanism integrates predictive priors with real-time measurements, enabling reliable CSI reconstruction even under sparse or noisy conditions. Extensive evaluations across diverse configurations demonstrate that the proposed estimator significantly outperforms both classical and data-driven baselines in accuracy, robustness, and generalization capability.

Figures (14)

• Figure 1: Overall framework and slot-based workflow of the proposed PFM-aided channel estimator.
• Figure 2: Architecture of the PFM tailored to channel estimation. The model consists of data preprocessing, an input residual block, positional encoding, and a causal transformer backbone, followed by an output residual block.
• Figure 3: Architecture of the proposed ViT-based pilot processing network.
• Figure 4: Illustration of the proposed two-phase training strategy.
• Figure 5: Channel prediction NMSE performance comparison across varying user speeds. The training and evaluation datasets are adopted from liu2024llm4cp, with historical and future sequence lengths set to 16 and 4, respectively.

Theorems & Definitions (1)

• Remark 1