AI-Based Coherent Uplink MIMO for Flexible UE Antennas

The patent replaces the traditional mechanical/codebook-based MIMO transmission system with an AI-based neural network system. The neural network encoder processes channel state information and generates precoding matrices dynamically, eliminating the need for predefined codebooks and enabling flexible adaptation to various UE antenna configurations and channel conditions while maintaining system efficiency.

The patent changes the fundamental parameters of UL MIMO transmission by introducing AI-based dynamic precoding instead of static codebook-based approaches. The neural network model processes channel state information and outputs optimized precoding matrices in real-time, allowing the system to adapt to different transmission scenarios without requiring complex reconfiguration of the underlying transmission framework.

The patent applies preliminary action by pre-training the neural network encoder and decoder models before actual communication occurs. The encoder is trained at the UE side and the decoder at the base station side using offline training data, so that during real-time communication, only lightweight inference and decoding operations are needed, significantly reducing the computational overhead and complexity during active transmission.

The patent uses copying by transferring the pre-trained neural network encoder model from the UE to the base station via the PUSCH channel. This allows the base station to replicate the UE's channel characterization capabilities, enabling efficient coherent MIMO transmission without requiring the base station to perform complex local channel analysis, thereby reducing overall system complexity.