Deep Learning-Based Robust Automatic Modulation Classification Using Higher Order Cumulant Features

The Internet of Things (IoT) represents one of the pivotal technologies in our daily lives. Within these systems, wireless communication plays an indispensable role in connecting IoT devices. Consequently, signal modulation emerges as a technique within wireless communication systems, enabling the transmission of baseband signals at higher frequencies. Given the array of modulation types, modulation classification comes into play to differentiate the modulation type of the received signal. Due to the presence of noise and attenuation, automatic modulation classification (AMC) stands as the primary method for such classification. AMC methods can operate without requiring any prior knowledge regarding the received signal. Multiple research groups have extended the depth of deep learning (DL) models or augmented the neuron count in each layer to enhance performance, albeit at the expense of increased model complexity. To tackle this challenge, we present model that merges the gated recurrent unit (GRU) and long short-term memory (LSTM) architectures. Furthermore, data pre-processing techniques, specifically Min-Max normalization and fourth-order cumulant (FOC), have been employed to enhance classification accuracy. The outcomes underscore that our proposed model surpasses both LSTM and GRU models. Moreover, the classification accuracy of normalized data outperforms that of data without normalization.