Bayesian Optimization of First-Order Continuous-Time Linear Equalization in High-Speed Links Including Crosstalk

Abstract

In this work, the Bayesian optimization algorithm is utilized to find optimal coefficients for a first-order Continuous-Time Linear Equalizer (CTLE). The goal function for the optimization is derived from the single-bit response in the time domain. The performance of the optimized CTLE is compared to a frequency domain-based method for predicting the CTLE coefficients and a grid search over all possible combinations. The analysis is presented for two different PCB-based high-speed channels and with a 32 Gbps data rate. As a novel contribution, far-end crosstalk is included in the optimization. The resulting noisy behavior of the single-bit response is learned during the optimization as part of the Gaussian process model.