Multichannel Deep Learning Network-Based Automatic Modulation Classification in Presence of Symmetric \(\alpha \) Stable Distributed Impulsive Noise

The development of Automatic Modulation Classification (AMC) was driven by the realization of its paramount importance in various military sectors like electronic warfare, intelligence, surveillance, and threat analysis. In numerous civilian and military applications, AMC serves as a pivotal stage bridging signal identification and demodulation. To match with practical wireless channel conditions, this work considered both symmetrical \(\alpha \) stable distributed impulsive noise and additive white Gaussian noise (AWGN) as a generalized noise model along with Rayleigh multipath frequency selective channel. Most of the existing methods utilize feature-based approaches which require human intervention. To create a completely automated process, deep learning (DL) framework is used. DL model accepts raw signal as input in in-phase/quadrature (I/Q) complex form. DL models are capable of extracting spatial and temporal features from raw signals. The neural network utilizes the extracted features for the purpose of classification. This literature utilizes state-of-the-art (SoA) DL models which is integration of convolutional layers, long-short term memory layers, and deep neural network into a single model (CLDNN) and its multichannel structures (MCLDNN) with convolutional neural network (CNN) as baseline model for AMC. The study proves that MCLDNN and CLDNN both achieve significant accuracy of about 5–6% higher compared to CNN which is considered a baseline model.