WiFo-E: A Scalable Wireless Foundation Model for End-to-End FDD Precoding in Communication Networks

TL;DR

WiFo-E addresses the scalability challenge of end-to-end precoding for FDD massive MIMO by framing heterogeneous configuration settings as a multi-task learning problem and training a wireless foundation model with a shared sparse MoE Transformer backbone. The approach enables knowledge sharing across configurations via multi-task pretraining and rapid adaptation to unseen setups through freezing the trunk and fine-tuning task-specific heads. Empirical results show that WiFo-E outperforms per-task training and modular baselines in spectral efficiency, while substantially reducing parameter count and improving generalization to unseen configurations and domain shifts. The work demonstrates a pathway toward robust, scalable foundation models for adaptive wireless precoding with practical implications for future 5G/6G deployments.

Abstract

Accurate precoding in massive multiple-input multiple-output (MIMO) frequency-division duplexing (FDD) systems relies on efficient channel state information (CSI) acquisition. End-to-end learning frameworks improve performance by jointly optimizing this process, but they lack scalability and fail to generalize across different system configurations, such as varying numbers of antennas and users. To overcome this limitation, we introduce WiFo-E, a wireless foundation model designed for scalable end-to-end precoding. WiFo-E employs multi-task pretraining on a diverse set of configurations to learn transferable representations of underlying wireless principles. Central to the model is a sparse Mixture-of-Experts (MoE) Transformer architecture, which mitigates task interference and enhances training efficiency by activating specialized parameter subsets adaptively. Extensive simulations demonstrate that WiFo-E outperforms conventional per-configuration training and shows strong generalization to unseen system configurations, providing a flexible and efficient foundation for adaptive massive MIMO precoding.

Figures (5)

• Figure 1: An overview of proposed multi-task pretraining framework.
• Figure 2: MoE-Transformer block.
• Figure 3: Comparison of the spectral efficiency of the proposed scheme and the baselines under different SNRs.
• Figure 4: Fine-tuning performance on Group 2 tasks.
• Figure 5: Model scaling performance.