CNN-Based Automatic Modulation Classification in OFDM Systems

Abstract

Convolutional neural network (CNN)-based modulation classification schemes for orthogonal frequency division multiplexing (OFDM) signals have recently been reported. In this paper, we examine the effect of hyperparameters in a CNN model on classification performance and present improved performance of automatic modulation classification for OFDM signals. To do this, we first set a baseline CNN model for OFDM signal modulation classification and then conduct experiments by varying the hyperparameters, such as the size and number of convolution kernels, and the number of fully connected neurons, through computer simulations. We show that the kernel size has a dominant effect on the classification accuracy and should be large enough within an appropriate range to achieve high classification accuracy for a given in-phase and quadrature data set. Finally, we show that the tuned model outperforms the conventional work in terms of classification accuracy.