CSI Feedback Compression Using Differential Frequency Transforms

The patent extracts and reports only the most significant components of CSI information. Specifically, it identifies and reports dominant eigenvalues and selected eigenvectors that capture the essential channel characteristics, while omitting redundant or less significant components. This extraction approach reduces reporting overhead while preserving the most critical channel state information needed for effective precoding and resource allocation.

The patent transforms the representation of CSI by changing parameters from reporting full channel matrices to reporting processed features such as eigenvalues, eigenvectors, and differential values. It applies transformations like singular value decomposition (SVD) to convert channel matrices into compact spectral representations, and uses differential encoding to report only changes in channel conditions rather than absolute values, thereby reducing overhead while maintaining measurement precision.

The patent applies partial action by reporting only a subset of CSI components that are sufficient for effective precoding. Instead of reporting all channel matrix elements, it selectively reports dominant eigenvalues and key eigenvectors that capture the essential spatial characteristics of the channel. This partial reporting approach achieves adequate precoding accuracy while significantly improving resource utilization by reducing feedback overhead.

The patent segments the CSI reporting into distinct components with different levels of detail and importance. It divides the channel information into dominant and non-dominant components, reports dominant eigenvalues and selected eigenvectors with higher precision, and either omits or coarsely quantizes less significant components. This segmentation allows the system to allocate reporting resources efficiently, focusing on the most critical parameters for precoding accuracy.

The patent extracts the essential characteristics of the channel matrix through eigenvalue and eigenvector decomposition. Instead of transmitting the complete channel matrix H, it extracts and reports only the dominant eigenvalues λ₁, λ₂, ... and selected eigenvectors v₁, v₂, ... that capture the principal modes of channel variation. This extraction reduces the quantity of feedback data from O(N²) elements to O(K) components where K ≪ N, while preserving the most significant channel state information.

The patent changes the parameter representation from raw channel matrix elements to transformed spectral parameters (eigenvalues and eigenvectors). It applies mathematical transformations like SVD to convert the channel matrix into a compact spectral form, and further transforms these spectral parameters into differential representations that report only changes from previous measurements. This parameter transformation dramatically reduces feedback overhead while maintaining channel state accuracy.