Sequential decision making for automatic modulation classification

Abstract

In this paper, an algorithm is introduced to use deep learning to perform automatic modulation classification in a sequential manner. At each time step, a decision is made whether to request more data or to return a classification decision. This allows for the data, and therefore time, needed to make a decision to be minimized while maintaining a high degree of accuracy. The performance of this algorithm is studied using multiple strategies and lists of modulations to be classified.