Abstract: Kernel Least Mean Square(KLMS) algorithm has a good covergence performance in nonlinear systems.But its mean square error gradient is estimated by the instantaneous gradient that results in larger randomness.However,the block adaptive filtering theory can reduce the steady-state error of KLMS algorithm by estimating the mean square error gradient with the multiple input-output errors.For this purpose,the block adaptive filtering theory is applied to the KLMS algorithm,and the Kernel Block Least Mean Square(KBLMS) algorithm is proposed.Based on the basic idea of the steepest descent algorithm,the weight vector update equation of KBLMS is derived.Then the filter output expression is calculated by utilizing the kernel method,and the computational complexity is reduced by using parallel processing.Simulations results show that KBLMS effectively improves the steady-state performance of KLMS and has lower Bit Error Rate(BER) than Block Least mean Square(BLMS) algorithin.
Kernael Least Mean Square(KLMS) algorithm,
block adaptive filtering,
steepest descent algorithm,
nonlinear channel equalization