Generative Decoding of Compressed CSI for MIMO Precoding Design

2025-11-11

AI Summary

To address the high channel state information (CSI) feedback overhead and deployment challenges of machine learning models in massive MIMO systems, this paper proposes a decoder-side generative CSI feedback scheme. At the user equipment (UE), standardized lightweight encoding is employed; at the base station (BS), an environment-aware decoder-only generative network is deployed. Leveraging site-specific digital twin-synthesized data and a target-oriented loss function, the model achieves efficient CSI reconstruction under low-data regimes via two-stage training. The key innovation lies in abandoning the conventional encoder-decoder co-design paradigm and introducing the novel “standardized encoding + generative decoding” framework—ensuring backward compatibility and practical deployability. Experiments demonstrate substantial improvements in CSI reconstruction accuracy across multiple bit-rate constraints (PSNR gain: 3.2–5.8 dB), enabling high-performance precoding while reducing feedback overhead by up to 60%.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Massive MIMO systems can enhance spectral and energy efficiency, but they require accurate channel state information (CSI), which becomes costly as the number of antennas increases. While machine learning (ML) autoencoders show promise for CSI reconstruction and reducing feedback overhead, they introduce new challenges with standardization, interoperability, and backward compatibility. Also, the significant data collection needed for training makes real-world deployment difficult. To overcome these drawbacks, we propose an ML-based, decoder-only solution for compressed CSI. Our approach uses a standardized encoder for CSI compression on the user side and a site-specific generative decoder at the base station to refine the compressed CSI using environmental knowledge. We introduce two training schemes for the generative decoder: An end-to-end method and a two-stage method, both utilizing a goal-oriented loss function. Furthermore, we reduce the data collection overhead by using a site-specific digital twin to generate synthetic CSI data for training. Our simulations highlight the effectiveness of this solution across various feedback overhead regimes.

Problem

• Reducing CSI feedback overhead in massive MIMO systems
• Addressing standardization and compatibility issues in ML-based CSI reconstruction
• Minimizing real-world data collection requirements for CSI training

Innovation

• Decoder-only generative CSI refinement using environmental knowledge
• Goal-oriented loss function with two training schemes
• Site-specific digital twin generates synthetic training data