Escape from stable equilibria in blind adaptive equalizers

The authors apply diffusion modeling and large deviations theory to obtain information about the behavior of blind adaptive equalizers when leaving local equilibria. In particular, they obtain estimates of the asymptotic behavior of the expected value of the exit time, and also about which trajectories are likely when an exit occurs. The conclusions drawn are that the residence times for local equilibria are (exponentially) much shorter than for the global minimum, the principal factors dictating this exponent being the step size of the adaptive process and prediction error variance at the equilibrium point. Some simulation results are included to support the theory