Active Noise Control based on Kernel Least-Mean-Square algorithm

Active Noise Control (ANC) is an important and efficient method to attenuate acoustic noise signals, especially those in low frequency range. Typical ANC system utilizes Filtered-x LMS algorithm (FXLMS), which shows low complexity and high attenuation under linear system assumptions. However, nonlinearity, existing in a real system from noise source to canceling points through electrical and acoustic paths, degrades the attenuation performance of the linear ANC methods. We take into account the possible system nonlinearity and introduce the Kernel LMS algorithm for mapping the data from low dimensional input space to high dimensional ¿feature space¿, where linear operations are applicable. Experimental results are presented for nonlinear primary path transfer function and chaotic nonlinear noise source. Comparison of KLMS and conventional LMS is also shown.