Automatic Modulation Recognition Using Multi Channel Network and Deep Neural Network with Talking Heads Attention

In recent days, the rapid growth of modern communication system has led to an increased demand for the efficient and consistent techniques. Additionally, the Orthogonal Frequency Division Multiplexing (OFDM) has developed as a best choice for high-speed data transmission. However, the increasing complexity of OFDM-based systems has created a need for Automatic Modulation Recognition (AMR) techniques to identify the modulation type in real-time, lead to noise robustness and modulation similarity. Hence, this research proposes a hybrid Multi-Channel Network (MCL) and Deep Neural Network (DNN) with Talking-Heads Attention (THA) used for signal processing of AMR in OFDM-based communication system. This procedure begins with the collection of modulation signals from RadioML2016.10a dataset, as it is a source of single carrier-based modulation signals. Next, the collected data is preprocessed with Recursive Least Square (RLS) technique, as it updates filter coefficients to minimize error. Then, Higher Order Statistics (HOS) is utilized to extract the optimal features from preprocessed signals. Finally, MCLDNN with THA is introduced to classify the modulation signals in communication systems. From the results, the proposed MCLDNN-THA model offered outperforming results in accuracy with 95.9% when compared with the existing Residual Network (ResNet) model.