Joint CSI Training and PMI Feedback via Compressed-Uncompressed Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional CSI feedback mechanisms in AI/ML-enabled networks are limited by compressed CSI feedback, which restricts the model's capability to design an optimal compression scheme, and specifying new AI/ML-based reporting configurations involves significant specification impact.
Innovation Solution
A CSI feedback mechanism that includes reporting a subset of CSI in a compressed domain for reduced overhead and another subset in an uncompressed domain for AI/ML model training, with the network configuring the UE on CSI compression types based on AI/ML inference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If compressed CSI feedback is used, then feedback overhead is reduced, but the AI/ML model's capability to design optimal compression scheme is restricted
Solution Approach 1:
The patent segments the CSI feedback into two distinct parts: compressed CSI feedback for reduced overhead and uncompressed CSI feedback for AI/ML model training. This segmentation allows each part to serve its specific purpose optimally - the compressed part reduces overhead while the uncompressed part provides full information for model training and compression scheme design.
Solution Approach 2:
The uncompressed CSI feedback serves multiple functions: it provides training data for AI/ML models, enables the model to design optimal compression schemes, and maintains full information integrity. This multi-functionality resolves the contradiction by ensuring that despite compression in one part, the other part preserves full capability.
2Adaptability or versatility
If new AI/ML-based reporting configurations are specified, then AI/ML model training capability is enhanced, but specification impact increases significantly
Solution Approach 1:
The patent specifies preliminary action by pre-defining the uncompressed CSI feedback mechanism and its structure in the specification. This allows AI/ML model training capability to be enhanced without requiring complex new specification changes, as the basic framework is already established and can be utilized directly.
Solution Approach 2:
The patent leverages existing CSI feedback structures and mechanisms, copying and adapting them for AI/ML training purposes rather than creating entirely new specification frameworks. This approach enhances training capability while minimizing specification impact by reusing established protocols.
3Loss of substance
If CSI is reported in compressed domain, then feedback overhead is reduced, but measurement precision for AI/ML training is compromised
Solution Approach 1:
The patent segments CSI feedback into compressed and uncompressed domains, allowing the compressed domain to reduce overhead for standard communication while the uncompressed domain maintains full measurement precision for AI/ML training purposes. Each segment serves its specific function optimally without compromising the other.
Solution Approach 2:
The uncompressed CSI feedback acts as an intermediary that bridges the gap between compressed feedback efficiency and training precision requirements. It provides the full-information reference needed for accurate AI/ML training while the compressed feedback handles the overhead reduction for actual communication.
Data Source
AI summary
Various aspects of the present disclosure relate to techniques for joint CSI training and PMI feedback for AI-enabled networks. A user equipment (UE)s configured to receive, based on a CSI reporting setting, a set of RSs for a channel measurement corresponding to an NZP CSI-RS resource, generate a CSI report based on the channel measurement, the CSI report comprising precoding matrix information for a precoding matrix and a set of coefficients that correspond to at least one dimension of the precoding matrix, and transmit the CSI report to a network.