Adaptive Filter Branch Sharing for Faster MIMO DPD Convergence

The patent segments the DPD algorithm into multiple parallel branches, each handling a subset of antennae. This segmentation allows the system to manage massive MIMO configurations by dividing the overall computation into smaller, more manageable tasks that can be processed simultaneously, thereby reducing the computational burden on individual processing units while maintaining the enhanced beamforming capability provided by the large number of antennae.

The patent implements a feedback mechanism where gradient-based information is transferred and shared among all parallel branches of the adaptive filter system. This feedback loop allows each branch to benefit from the learning progress of other branches, enabling faster convergence despite the increased number of parallel processing units. The gradient information exchange ensures that all branches converge toward the optimal solution more efficiently than if they operated independently.

The patent extracts only the essential gradient-based information needed for parameter updates and transfers it among the parallel branches. By taking out only the critical gradient data rather than sharing all intermediate computations or raw data, the system achieves faster convergence while minimizing the communication overhead and information loss associated with data exchange between branches.