Neural CSI Compression for Accurate Low-Overhead Feedback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for channel state information (CSI) feedback in wireless communication networks are compute-intensive and result in suboptimal quantized feedback due to structured PMI codebooks, leading to inefficient use of UE processing resources.

Innovation Solution

Implementing neural-network (NN) based channel compression and reconstruction techniques using a structured payload with interpretable and uninterpretable portions, where the uninterpretable portion carries size-reduced CSI encoded by a CSI encoder, and utilizing RS-based side information for decoding.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If existing quantized feedback methods with structured PMI codebooks are used, then feedback can be provided with standardized formats, but the processing becomes compute-intensive and results in suboptimal feedback accuracy

Engineering Contradiction:
Improvefeedback accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces traditional quantized feedback methods with neural network-based continuous feedback. The neural network model (encoder-decoder architecture) substitutes the mechanical quantization process, transforming discrete codebook-based feedback into continuous channel state information feedback. This substitution eliminates the compute-intensive evaluation of structured PMI codebooks while improving feedback accuracy through learned optimal representations.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental parameter representation from discrete quantized values (PMI, RI, CQI) to continuous channel state information. By transforming the feedback parameter space from a limited codebook structure to continuous values optimized by neural networks, the system achieves both higher accuracy and reduced processing complexity through learned parameter transformations.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If full channel state information is transmitted, then reconstruction accuracy is maximized, but payload size and transmission overhead increase significantly

Engineering Contradiction:
Improvechannel reconstruction accuracyVSAvoidpayload size
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential channel state information features through neural network encoding, separating critical information from redundant data. The encoder identifies and extracts salient channel characteristics, transmitting only these extracted features rather than complete channel matrices. This extraction process maintains reconstruction accuracy while dramatically reducing payload size.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent uses neural network-based compression to create a compressed representation (copy) of the channel state information. Instead of transmitting the original full-dimensional channel data, the system creates a compact encoded copy that preserves essential information. The decoder reconstructs the full channel state from this compressed copy, achieving high fidelity with minimal transmission overhead.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4111609B1Channel state information feedback using channel compression and reconstruction
Publication Date: 2025.07.02 QUALCOMM INC
  • EP4111609B1 patent drawingFigure 1
  • EP4111609B1 patent drawingFigure 2
  • EP4111609B1 patent drawingFigure 3

AI summary

Systems and methods facilitating feedback of robust channel state information (CSI), such as to provide full CSI feedback or otherwise providing CSI feedback, are described. CSI encoders and/or decoders used by network nodes may implement channel compression/reconstruction based upon neural-network (NN) training of collected channels. A structured payload having an interpretable payload portion and an uninterpretable payload portion may utilized with respect to CSI feedback. The channel compression provided according to some aspects of the disclosure supports feedback of robust CSI, in some instances including full CSI, as determined by a particular network node. Other aspects and features are also claimed and described.