Deep Learning Based Automatic Modulation Recognition Using GELU Activation Function

Abstract

One promising technique for detecting signal modulation schemes in cognitive radio networks is automatic modulation recognition (AMR). AMR based on highperformance deep learning (DL) techniques have been made easier recently by the growing research on DL. But, as DL is evolving daily, AMR approaches must perform better, and new approaches must be developed. This research presents a new DL based technique for AMR used in modern communication systems' cognitive radio networks. To simultaneously learn the spatio-temporal signal correlations with Gaussian Error Linear Unit (GELU) activation function, the network architecture is built with multiple distinct convolutional blocks. The suggested technique achieves an overall 6-modulation classification rate of 80% at 20 dB SNR in the simulations performed with the generated dataset.