AI Channel Estimation Reducing Pilot Overhead

The patent segments the channel estimation process into two distinct feature extraction paths: linear feature extraction from pilot signals and nonlinear feature extraction from data signals. This segmentation allows each path to be optimized independently, reducing the reliance on extensive pilot overhead while maintaining estimation accuracy through complementary feature fusion.

The patent makes the data signal serve multiple functions: it is used for both nonlinear feature extraction and as a reference for channel estimation. By extracting nonlinear features (such as amplitude information) from data signals and fusing them with linear features from pilot signals, the system reduces pilot overhead while maintaining channel estimation performance.

The patent introduces nonlinear features extracted from data signals as an intermediary that bridges the gap between reduced pilot overhead and maintained estimation performance. These nonlinear features (particularly amplitude information) act as a mediator that compensates for the reduced pilot density, enabling accurate channel estimation even with sparse DM-RS placement.

The patent changes the parameter representation by transforming data signals into nonlinear features (such as amplitude envelope information) that provide complementary channel state information. This parameter transformation allows the system to extract useful channel characteristics from data signals without requiring additional pilot overhead, thereby maintaining reliability with reduced pilot density.