Skip to Main Content
This paper examines a method to apply to channel equalization problem by model selection. The selection process is based on finding a subset model to approximate the response of the full two weighted neural network model for the current input vector, and not for the entire input space. When the channel equalization problem is nonstationary, the requirement to update all the kernel weights locations is removed, and its complexity is reduced. Using computer simulations, we show that the number of kernel weights can be greatly reduced without compromising classification performance.