High-Speed Channel Equalization Applying Parallel Bayesian Machine Learning

Abstract

Recovering attenuated signals caused by different issues including connections between connectors and chips of the devices, channel loss, and crosstalk is a challenging problem. Equalization is the most popular way to restore distorted signals. Different optimization algorithms are used to find the best tap coefficients for each equalization that improves eye opening and decreases the bit error rate (BER). Nowadays algorithms of equalization mainly operate to reduce the difference between input and output signals. In turn, this will increase eye height indirectly, but direct maximization of the eye height will restore the signal even better. The paper proposes a new efficient optimization of joint Feed-Forward Equalization (FFE) and Decision Feedback Equalization (DFE) for binary as well as multi-level signals. Unlike to above-mentioned method, which is a linear optimization problem, direct maximization of the eye height applying joint FFE and DFE is a non-linear problem and cannot be solved analytically. Paper proposes the black-box function optimization using the regular as well as parallel Bayesian machine learning to find the best tap coefficients for joint FFE and DFE equalization. The efficiency of the parallel Bayesian algorithm is shown for binary NRZ (nonreturn-to-zero) and PAM4 (Pulse Amplitude Modulation) signals.