CSI Pre-Processing with Eigenvector Polarization Separation for ML Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CSI reporting frameworks in wireless communication face challenges such as poor correlation between estimated CSI samples due to cross-polarization, leading to inefficient training of ML models and increased computational burden, particularly in CNN- and transformer-based models.
Innovation Solution
Implementing enhanced sample-invariant pre-processing techniques like phase discontinuity compensation (PDC), one-step and two-step polarization separation, and position-based re-ordering to extract eigenvectors (EVs) from raw CSI, transforming them into a beam-frequency or beam-delay domain, thereby improving sparsity and reducing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sample-variant pre-processing is used to adapt to different CSI samples, then the ML model can capture sample-specific features better, but the feedback overhead increases due to additional information that must be transmitted
Solution Approach 1:
The patent transforms CSI parameters from the antenna domain to the beam domain using pre-defined beamforming vectors, changing the representation parameters without altering the underlying information. This transformation enables sample-invariant pre-processing while maintaining estimation accuracy, as the beam-domain representation captures essential spatial characteristics without requiring sample-specific adaptation parameters.
Solution Approach 2:
The patent segments the CSI processing into distinct domains: the pre-processing stage operates on beam-domain representations using sample-invariant operations, while sample-specific features are implicitly captured by the ML model in the post-processing stage. This segmentation allows different processing strategies for different aspects of CSI, reducing overall feedback overhead.
2Device complexity
If cross-polarization components are processed together, then the processing is simpler, but the correlation between CSI samples deteriorates due to phase discontinuities
Solution Approach 1:
The patent segments cross-polarization components into separate processing streams, handling each polarization independently through the beam domain transformation. This segmentation prevents phase discontinuities from different polarizations from interfering with each other, maintaining sample correlation while keeping individual processing streams relatively simple.
Solution Approach 2:
The patent introduces a new dimension (beam domain) as an intermediate representation space between the antenna domain and the processing domain. By transforming CSI to the beam domain before processing, the patent creates an additional processing dimension that separates cross-polarization components effectively, improving correlation without significantly increasing overall complexity.
3Measurement precision
If detailed pre-processing is applied to maintain spatial information, then the ML model performance improves, but the computational burden increases
Solution Approach 1:
The patent performs preliminary transformation of CSI to the beam domain using pre-defined beamforming vectors before the ML model processing. This preliminary action organizes the spatial information in a more efficient representation, allowing the ML model to work with pre-processed data that requires less computational effort while maintaining spatial information integrity.
Solution Approach 2:
The patent changes the parameter representation from antenna-domain CSI to beam-domain CSI through a linear transformation. This parameter change reorganizes the data structure to better suit ML processing, preserving spatial information while reducing the computational complexity of subsequent model operations through the inherent structure of beam-domain representations.
Data Source
AI summary
Techniques pertaining to channeling state information (CSI) pre-processing are described. A user equipment (UE) that is in wireless communication with a base station node extracts eigenvectors (EVs) from CSI acquired by the UE. The UE generates pre-processed CSI for compression by a machine-learning (ML)-based encoder of the UE into CSI feedback for the base station node by at least performing one or more of a phase discontinuity compensation (PDC), a one-step polarization separation with re-ordering, or a two-step polarization separation that includes separation based on polarization type and separation by position on the EVs.