Gaussian Kernel Least Mean Square Algorithm With Improved Novelty Criterion

Abstract

The kernel-based adaptive algorithm has been widely applied to noise cancellation, but the computational complexity of kernel function causes it can't perform well in embedded real-time control system. This paper proposes a design of the Gaussian kernel least mean square algorithm with improved novelty criterion, which aims to reduce computational load with universal approximation and fast convergence speed. The methodology slows down the growth rate of the network by multiple operations on the training set. It finds out the filter parameter from the collection data by improved novelty criterion, so the filter network has a much smaller scale and computation complexity, which allow it to be used in the embedded real-time control system.