Low Complex Hybrid Deep Learning Model for Automatic Modulation Classification
Bandaru Bhavana, Samrat Lagnajeet Sabat, Swetha Namburu, Trilochan Panigrahi · 2023
The manually extracted features based automatic modulation classification algorithms have high computational complexity. In this paper, we introduce a low complex deep neural network (DNN) architecture for automatic modulation classification of the radar and radio signal. The proposed DNN architecture is a hybrid of Inception and long short-term memory network (LSTM), named the Inception_LSTM hybrid model. It combines the advantages of a faster convergence rate of Inception model and higher classification accuracy of the LSTM model. The proposed hybrid architecture has lower computational complexity with improved classification accuracy in classifying different modulated signals of radar modulated and radio signals. We also evaluate the classification accuracy of the model by reducing the size of the training dataset using a compressive sampling technique, resulting in less training time and complexity while incurring a minimal loss of classification accuracy.