In a preceding paper (Part I) the solution to the same problem has been reviewed using decision-theoretic algorithms. In this paper the artificial neural network (ANN) approach for the modulation recognition is studied in some detail. Three categories of ANN-based modulation recognition algorithms, using the same datasets as for the corresponding algorithms utilising the decision-theoretic approach of Part I, are presented. It is found that normalising the datasets corresponding to the maximum value of each key feature reduces the training time. Generally, the ANN removes the need for separate threshold determination, removes the active choice of the time-ordering of threshold testing, and often offers a better success rate than the decision-theoretic algorithms.
Read full abstract