CSI Class-Based Compression for Lower MIMO Feedback Overhead

The patent segments CSI elements into multiple classes based on channel characteristics (e.g., delay spread, Doppler spread, angle of arrival). Each class is associated with a specialized encoder that processes only the relevant CSI parameters for that class. This segmentation allows the system to transmit only necessary CSI information for each channel type, reducing overall feedback overhead while maintaining accuracy for each specific class.

Different encoder configurations are applied to different CSI classes based on their specific characteristics. For example, classes with high delay spread use encoders optimized for time-selective fading, while classes with wide angular spread use encoders optimized for spatial diversity. This local optimization ensures that each CSI element is processed with the most appropriate encoding strategy, improving precision without uniformly increasing resource consumption across all CSI types.

The patent performs preliminary classification of CSI elements into distinct classes before encoding. By pre-categorizing CSI based on channel characteristics (such as identifying whether the channel is rich scattering, line-of-sight, static, or mobile), the system can select the appropriate encoder configuration in advance. This preliminary action avoids the complexity of dynamically switching between multiple encoders during real-time operation, as the classification is done once and then the corresponding encoder is used consistently for that class.

The patent changes encoding parameters based on CSI class characteristics. Instead of using multiple complex encoder architectures, the system uses a single encoder framework with configurable parameters that are adjusted according to the CSI class. For example, the encoder may change the transformation basis, the number of feedback bits, or the quantization strategy based on the identified channel class, achieving optimized performance without the complexity of multiple dedicated encoders.

The patent extracts and transmits only the most relevant CSI parameters for each CSI class rather than compressing all CSI elements uniformly. For example, for classes with dominant line-of-sight components, only the angle of arrival and delay of the dominant path are extracted and fed back. For classes with rich scattering, only statistical parameters like correlation coefficients are extracted. This selective extraction reduces feedback overhead significantly while preserving the essential information needed for each channel type.

The patent applies partial compression by transmitting full-precision CSI for certain critical parameters while using compressed representations for less critical parameters. For example, the rank indicator or precoding matrix indicator may be transmitted with higher precision than the channel coefficients themselves. This partial action approach ensures that the most important information is preserved with minimal loss while achieving overall compression of the CSI feedback.