A simple predictor based on delay-induced negative group delay

Abstract

A very simple linear signal predictor that uses past predicted values rather than past signal values for prediction is presented. Man-made or natural systems utilizing this predictor would not require a memory of input signal values but only of already predicted, internalized states. This delay-induced negative group delay (DINGD) predictor affords real-time prediction of signals without the need for a specific signal model. Its properties are derived analytically and are numerically tested on various types of broadband input data.