CSI Compression Encoder Using Differential Convolution Steps

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Massive multiple-input multiple-output (m-MIMO) antenna systems require accurate channel state information (CSI) for efficient operation, but existing methods face challenges in effectively compressing and encoding CSI data while maintaining accuracy and reducing complexity.

Innovation Solution

The proposed solution involves an encoding method for CSI performed by user equipment (UE) using a first target CSI compression encoder, which includes N composite convolution layers and one fully-connected layer. Each composite convolution layer has a delay-domain convolution step smaller than the angle-domain convolution step, allowing for differential compression based on correlation strengths in the delay and angle domains.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If CSI compression encoding is performed using conventional methods, then the complexity of CSI feedback is reduced, but the compression performance and accuracy of CSI data deteriorate

Engineering Contradiction:
ImproveCSI feedback complexityVSAvoidCSI data accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent changes the parameter of convolution step size in different domains. Specifically, it uses a first convolution step size in the delay domain and a second convolution step size in the angle domain, where the ratio between them matches the ratio of correlation strengths. This parameter adjustment enables differential compression that adapts to the correlation characteristics of each domain, improving compression performance while maintaining CSI accuracy.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent applies local quality by treating the delay domain and angle domain differently based on their distinct correlation properties. It performs compression encoding with different convolution step sizes tailored to each domain's characteristics - using larger steps in high-correlation dimensions and smaller steps in low-correlation dimensions. This localized approach optimizes compression for each specific domain while preserving overall CSI accuracy.

Inventive Principle:
Principle #3Local quality

2Ease of manufacture

If uniform compression is applied to all CSI dimensions, then the encoding process is simplified, but the compression efficiency deteriorates due to ignoring differential correlation strengths

Engineering Contradiction:
Improveencoding process simplicityVSAvoidcompression efficiency
Core Design Contradiction:
Ease of manufactureVSProductivity

Solution Approach 1:

The patent introduces different convolution step size parameters for different domains based on their correlation strengths. By setting the ratio of first to second convolution step sizes equal to the ratio of delay domain to angle domain correlation strengths, the method achieves adaptive compression that improves efficiency while maintaining reasonable encoding complexity through a systematic parameter relationship.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS12278681B2Encoding method and decoding method for channel state information and communication device
Publication Date: 2025.04.15 BEIJING XIAOMI MOBILE SOFTWARE CO LTD
  • US12278681B2 patent drawing
  • US12278681B2 patent drawing
  • US12278681B2 patent drawing

AI summary

An encoding method for channel state information (CSI), performed by a user equipment (UE), includes: encoding, based on a first target CSI compression encoder, a target CSI matrix in a delay domain and an angle domain, to generate a compressed encoded value, in which the first target CSI compression encoder includes N composite convolution layers and one fully-connected layer, each composite convolution layer includes a delay-domain convolution step and an angle-domain convolution step, and the delay-domain convolution step of the first composite convolution layer in the N composite convolution layers is smaller than the angle-domain convolution step of the first composite convolution layer in the N composite convolution layers, where N is a positive integer.