A novel kernel filtering algorithm based on the generalized half-quadratic criterion

In this work we combine the kernel method and the generalized half-quadratic criterion, and a kernel adaptive filtering algorithm is proposed based on the generalized half-quadratic criterion (KLGHQC). The generalized half-quadratic criterion (GHQC) guarantees the stability of the algorithm under the environment of the stable distribution noise, and the shape of the GHQC performance surface is determined by a constant, which improves the rate of convergence of the algorithm. Finally, the simulated results in two environments, Mackey-Glass sequence prediction and non-linear system identification. The outcome demonstrates that the KLGHQC algorithm proposed in this research outperforms other kernel filtering algorithms in the filtering accuracy and error magnitude.