ABSTRACT
Many neural network models possess two significant properties that often allow them to outperform more conventional techniques in signal processing applications. Their ability to adapt continuously to new data allows them to track changes in a system over time, and their ability to learn arbitrary, nonlinear transfer functions permits them to solve problems that cannot be handled adequately with more conventional adaptive linear techniques. However, linear methods generally converge to a solution much faster than neural networks, and they currently have a stronger theoretical foundation for predicting their behavior. This section shows how neural networks can be used for channel equalization, signal prediction, and noise canceling tasks.
