AI/ML CSI Feedback with Decoder Output Restrictions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing wireless communication systems face challenges in reducing downlink interference and efficiently collecting data for artificial intelligence and machine learning models, particularly due to limitations in codebook subset restriction and data collection configurations.

Innovation Solution

The proposed solution involves providing user equipment (UE) with restriction information related to the channel, enabling it to ensure that the precoder determined using a decoder model is orthogonal to the restricted sub-space. This is achieved by configuring the UE with a restricted sub-space and decoder outputs, allowing it to generate channel state information (CSI) feedback that avoids interference. Additionally, the system enables UE to establish data collection sessions, request specific data types, and declare supported data collection configurations, optimizing data collection processes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If codebook subset restriction is implemented by setting basis vector coefficients to zero, then downlink interference is reduced, but the system cannot determine restricted PMIs in AI/ML-based CSI generation where no basis vector exists

Engineering Contradiction:

Improvedownlink interferenceVSAvoidcompatibility with AI/ML-based CSI reporting

Core Design Contradiction:

Object-affected harmful factorsVSAdaptability or versatility

Solution Approach 1:

The patent changes the parameter representation from basis vector coefficients (traditional linear algebra approach) to decoder output coefficients (AI/ML approach). By modifying how restriction information is represented and applied - using decoder outputs instead of basis vectors - the system maintains codebook subset restriction functionality while adapting to AI/ML-based CSI generation methods where traditional basis vectors do not exist

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

Instead of restricting PMIs by setting basis vector coefficients to zero (traditional approach), the patent inverts the approach by providing restriction information that directly constrains decoder outputs. The base station generates restriction information based on desired downlink interference reduction, and the UE applies this to the AI/ML decoder outputs, achieving restriction through the inverse path of traditional codebook subset restriction

Inventive Principle:

Principle #13The other way round (Inversion)

2Reliability

If UE is configured with CSI report configuration for data collection, then channel measurements can be performed, but UE computes and reports unnecessary CSI types consuming excessive processing resources

Engineering Contradiction:

Improvedata collection capabilityVSAvoidprocessing resource efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent extracts only the necessary data collection functionality from the full CSI report configuration. Instead of configuring UE with complete CSI reporting capabilities, the system selectively enables only the specific data types and measurement configurations needed for AI/ML model training and inference, removing unnecessary CSI computation and reporting components

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies local quality by making data collection configuration specific to each UE's AI/ML needs. Different UEs receive customized data collection configurations based on their local model training requirements, enabling each UE to collect only the specific data types (e.g., channel matrices, CSI feedback, target CSI) needed for its particular AI/ML applications rather than uniform CSI reporting

Inventive Principle:

Principle #3Local quality

3Reliability

If UE collects data for AI/ML model training, then model performance improves, but UE wastes resources preparing uplink channel for transmitting unnecessary CSI

Engineering Contradiction:

Improvemodel training data qualityVSAvoiduplink transmission resource waste

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent implements feedback mechanisms where the base station receives information about UE's AI/ML data collection needs and capabilities, then provides targeted data collection configurations. This feedback loop ensures that uplink resources are allocated efficiently - the base station knows exactly what data the UE needs to collect and can configure uplink transmissions accordingly, preventing waste on unnecessary CSI reports

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20250055534A1Data collection and precoding restrictions for artificial intelligence and machine learning based reporting of communication channel information

Publication Date: 2025.02.13 SAMSUNG ELECTRONICS CO LTD

AI summary

An apparatus may include a receiver configured to receive a reference signal using a channel, and receive restriction information relating to the channel, a processing circuit configured to determine channel information based on the reference signal, and generate, using a machine learning model, a representation based on the channel information and the restriction information, a transmitter configured to transmit the representation.