Full-Duplex Ethernet Link Training with Concurrent Equalizer Adaptation

The patent implements concurrent training where both PHY transceivers perform equalizer and echo canceler training simultaneously using independent clocks, eliminating idle periods and continuous coordination overhead. This allows useful training action to continue without interruption in both directions, reducing total training time while maintaining manageable complexity through standardized training sequences.

The patent uses preliminary training sequences transmitted before actual data communication to pre-train equalizers and echo cancelers. These preliminary actions include sending known training patterns that allow receivers to adapt their parameters in advance, ensuring ready-to-use trained parameters when full-duplex communication begins, thus reducing operational delays.

The patent segments the full-duplex training process into independent clock domains for each direction. Each PHY transceiver uses its own independent clock for generating and processing training sequences, allowing separate optimization of each direction's data rate. This segmentation eliminates the need for complex bidirectional clock synchronization while maintaining flexibility.

The patent introduces an intermediary clock recovery and resampling mechanism that bridges independent clocks. The receiver recovers the transmitter's clock from incoming training sequences and uses it for resampling and processing, acting as an intermediary that enables independent clock operation while ensuring proper synchronization for echo cancellation and equalization.

The patent implements periodic training sequences with distinct patterns for equalizer training and echo canceler training. By alternating or interleaving these periodic training signals in time, the system maintains concurrent operation (improving power efficiency) while ensuring that each training function receives dedicated signal periods free from interference (maintaining accuracy).

The patent applies different local signal characteristics to different training functions. Equalizer training uses specific signal patterns optimized for channel equalization, while echo canceler training uses different patterns optimized for echo path identification. This local differentiation of signal quality allows concurrent operation without mutual degradation of training accuracy.