Abstract

This paper presents a Long Short-Term Memory (LSTM) Neural Equalizer. The LSTM is integrated as a terminal in the system where the spoiled signal series from the transmission line are directly processed by the LSTM neural block. The feedback information is achieved by the long-short memories states. The LSTM neural equalizer shows significant improvement in the eye width, height, and jitter compared to the active Feed-Forward Equalizer and Decision Feedback Equalizer (FFE-DFE) approach.

Key Words: LSTM, Neural Equalizer, Signal Processing, Feed-Forward Equalizer, Decision Feedback Equalizer

Main points

• Proposes a Long Short-Term Memory (LSTM) Neural Equalizer.

• The LSTM is integrated as a terminal in the system.

• Shows significant improvement in the eye width, height, and jitter compared to the active Feed-Forward Equalizer and Decision Feedback Equalizer (FFE-DFE) approach.

Citation

@ARTICLE{10038624,
  author={Wang, Zihao and Xu, Zhifei and He, Jiayi and Delingette, Hervé and Fan, Jun},
  journal={IEEE Transactions on Signal and Power Integrity}, 
  title={Long Short-Term Memory Neural Equalizer}, 
  year={2023},
  volume={2},
  number={},
  pages={13-22},
  doi={10.1109/TSIPI.2023.3242855}}