Stochastic quasi-Newton method for minimum bit error rate nonlinear equalizers online training

A sliding window hybrid quasi Newton algorithm is proposed in this paper for minimum bit error rate nonlinear equalizers online training. Switching between sliding window stochastic gradient algorithm and sliding window quasi Newton algorithm makes the new algorithm be stable and converge fast. Moreover, by modifying the quasi Newton method, the new algorithm can be applied to high-dimensional parameters. In simulations, the new algorithm is used for training nonlinear equalizers in direct sequence spread spectrum communications and its high efficiency is proved by simulation results.