[Paper Review] Hybrid Knowledge-Data Driven Channel Semantic Acquisition and Beamforming for Cell-Free Massive MIMO

Zhen Gao, Shicong Liu+3 more|arXiv (Cornell University)|Jul 6, 2023

Advanced MIMO Systems OptimizationEngineering3 citations

TL;DR

This paper proposes a hybrid knowledge-data driven framework for channel semantic acquisition and beamforming in cell-free massive MIMO systems, combining a data-driven MLP-Mixer autoencoder for CSI feedback with a knowledge-driven deep-unfolding beamformer for robust, low-complexity multi-user transmission. The method achieves ~96% of converged spectral efficiency in just three iterations, significantly improving performance under imperfect CSI.

ABSTRACT

This paper focuses on advancing outdoor wireless systems to better support ubiquitous extended reality (XR) applications, and close the gap with current indoor wireless transmission capabilities. We propose a hybrid knowledge-data driven method for channel semantic acquisition and multi-user beamforming in cell-free massive multiple-input multiple-output (MIMO) systems. Specifically, we firstly propose a data-driven multiple layer perceptron (MLP)-Mixer-based auto-encoder for channel semantic acquisition, where the pilot signals, CSI quantizer for channel semantic embedding, and CSI reconstruction for channel semantic extraction are jointly optimized in an end-to-end manner. Moreover, based on the acquired channel semantic, we further propose a knowledge-driven deep-unfolding multi-user beamformer, which is capable of achieving good spectral efficiency with robustness to imperfect CSI in outdoor XR scenarios. By unfolding conventional successive over-relaxation (SOR)-based linear beamforming scheme with deep learning, the proposed beamforming scheme is capable of adaptively learning the optimal parameters to accelerate convergence and improve the robustness to imperfect CSI. The proposed deep unfolding beamforming scheme can be used for access points (APs) with fully-digital array and APs with hybrid analog-digital array. Simulation results demonstrate the effectiveness of our proposed scheme in improving the accuracy of channel acquisition, as well as reducing complexity in both CSI acquisition and beamformer design. The proposed beamforming method achieves approximately 96% of the converged spectrum efficiency performance after only three iterations in downlink transmission, demonstrating its efficacy and potential to improve outdoor XR applications.

Motivation & Objective

• Address the challenge of high pilot overhead and imperfect CSI in outdoor cell-free massive MIMO systems for extended reality (XR) applications.
• Reduce CSI feedback overhead and improve reconstruction accuracy compared to conventional compressive sensing and learning-based methods.
• Develop a robust, low-complexity beamforming scheme that adapts to imperfect CSI and accelerates convergence.
• Extend the proposed framework to both fully-digital and hybrid analog-digital array architectures for practical deployment.
• Achieve high spectral efficiency with minimal training and inference time while maintaining robustness in dynamic outdoor XR environments.

Proposed method

• Proposes an end-to-end optimized MLP-Mixer-based autoencoder for channel semantic acquisition, jointly learning pilot signals, CSI quantization, and reconstruction.
• Introduces a deep-unfolding beamforming architecture based on successive over-relaxation (SOR), where iterative steps are modeled as layers in a neural network to learn optimal parameters.
• Adapts the beamformer to both fully-digital and hybrid analog-digital array APs, enabling practical deployment in real-world systems.
• Employs a channel state information feedback scheme using a CSF (channel semantic feedback) loss function to improve semantic representation and reconstruction accuracy.
• Uses a learned regularization factor and convergence factor via training, replacing analytical solutions that are suboptimal in practical scenarios.
• Applies distributed and centralized processing paradigms, with separate feedback schemes (MixerS for centralized, MixerUL for distributed) to reconstruct local CSI at each AP.

Experimental results

Research questions

• RQ1 Can a data-driven autoencoder with joint optimization of pilots, quantization, and reconstruction reduce CSI feedback overhead while improving accuracy?
• RQ2 How does deep unfolding of a conventional SOR-based beamformer improve convergence speed and robustness to imperfect CSI?
• RQ3 To what extent can the proposed beamforming scheme maintain high spectral efficiency with only a few iterations, especially under imperfect CSI?
• RQ4 Does the learned parameter adaptation (e.g., regularization and convergence factors) outperform analytically derived values in practical outdoor XR scenarios?
• RQ5 Can the proposed framework be effectively extended to hybrid analog-digital array APs without sacrificing performance?

Key findings

• The proposed hybrid feedback scheme achieves significantly lower training and inference time compared to state-of-the-art learning-based CSI feedback methods.
• The deep-unfolding beamformer attains approximately 96% of the converged spectral efficiency after only three iterations in downlink transmission, demonstrating fast convergence.
• The learned convergence factor outperforms manually selected values, especially at low transmit power, and maintains superior performance across all power levels.
• The performance gap between centralized and distributed processing paradigms is minimal, indicating strong feasibility of distributed beamforming in practice.
• The proposed beamformer shows robustness to imperfect CSI, with lower performance degradation than conventional iterative schemes.
• The learned regularization factor consistently improves spectral efficiency, particularly under low SNR conditions, where analytical solutions fail to deliver optimal performance.