Deep learning for 1-bit compressed sensing-based superimposed CSI feedback.

Area of Science:

• Wireless Communications
• Signal Processing
• Machine Learning

Background:

• Frequency-division duplexing (FDD) massive multiple-input multiple-output (MIMO) systems utilize 1-bit compressed sensing (CS) for channel state information (CSI) feedback.
• Existing 1-bit CS CSI feedback methods face challenges including low downlink CSI recovery accuracy and significant processing delays.

Purpose of the Study:

• To propose a deep learning (DL) scheme to enhance 1-bit CS-based superimposed CSI feedback in FDD massive MIMO systems.
• To address the limitations of low accuracy and high latency in current CSI feedback mechanisms.

Main Methods:

• Downlink CSI is compressed using 1-bit CS, superimposed on uplink user data sequences (UL-US), and transmitted to the base station (BS).
• A model-driven multi-task detection network at the BS jointly detects UL-US and downlink CSI, aided by superimposition-interference cancellation.
• A lightweight reconstruction scheme, combining simplified traditional methods and a single hidden layer network, reconstructs downlink CSI with reduced delay.

Main Results:

• The proposed DL scheme significantly improves the recovery accuracy of both UL-US and downlink CSI compared to traditional 1-bit CS methods.
• The scheme achieves lower processing delays in CSI reconstruction.
• The proposed method demonstrates robustness against variations in system parameters.

Conclusions:

• The developed deep learning scheme effectively enhances 1-bit CS superimposed CSI feedback in FDD massive MIMO.
• This approach offers a superior trade-off between CSI recovery accuracy and processing delay.
• The method provides a practical solution for improving feedback efficiency in advanced wireless systems.