Channel equalization with rapid convergence based on ε-support vector machines

In this paper, the problem of adaptive channel equalization is considered. We extend the previous work in a new direction with periodic training sequence and formulate the original channel equalization problem into the equivalent linear regression in which ϵ-support vector regression machines are proposed. Slack variables are introduced in our approach to solve the problem of quadratic programming derived from ϵ-support vector regression machines. Simulation results show that the bit error rate of the proposed equalizer based on ϵ-support vector regression machines comes near to that of the optimal equalizer and is better than that of wavelet neural network equalizer. Simulation results also support the rapid convergence of the proposed approach, which can meet the realtime requirement of time-varying channels effectively in modern digital communication systems.