Automatic Modulation Classification using amalgam CNN-LSTM

The accurate classification of received wireless signals through modulation is vital for both military and civilian uses. The recent evolution in Deep Learning (DL) have led to a growing interest among wireless researchers to apply DL algorithms to modulation classification. To support this, the RadioML2016.10a dataset has been created using the GNU Radio framework, simulating real-world communication channel imperfections. This work proposes the use of two DL models - standalone RNN and combination of Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM). The amalgam CNN-LSTM model showed the highest accuracy that is 95 % for 18 Signal to Noise Ratio (SNR) value. The models have been trained and tested using the RadioML2016.10b dataset.