ResCNNGRU: A Hybrid Deep Learning Based Algorithm for Modulation Classification in NOMA Systems

Automatic modulation classification (AMC) is pivotal in non-orthogonal multiple access (NOMA) systems, as it dynamically adjusts modulation schemes to maximize spectral efficiency and minimize inter-user interference. We introduce a novel deep learning (DL) based architecture that integrates a residual convolutional neural network (CNN) with a gated recurrent unit (GRU) to classify the modulation formats of the interference users in NOMA downlink scenarios. This hybrid model harnesses the strengths of both spatial and temporal patterns extracted from the superposed in-phase and quadrature-phase (IQ) signals. Through comprehensive experimentation, we observe that for power ratio 4 and 200 symbol points our model demonstrates an accuracy of more than $99 \%$ at signal-to-noise (SNR) levels of 6 dB and above. This approach not only surpasses existing methods in classification accuracy but also proposes a computationally efficient modulation classifier.