OFDM sinyallerinin alttaşıyıcıları için derin öğrenme temelli modülasyon sınıflandırma

Automatic modulation classification (AMC) is automatically identifying and classifying the modulation schemes employed in digital communications. By accurately identifying the modulation scheme, AMC enables communication systems to adapt their parameters, optimizing efficiency, spectral utilization, and overall performance. The majority of the literature on AMC focuses on the single-carrier communications systems. This thesis addresses the gap between the AMC and multi-carrier communications systems. Two architectures are proposed. Both employ a filter bank-convolutional neural network (CNN) complex. The first architecture uses raw features and a maximum operation to perform classification, whereas the second architecture learns feature patterns by employing a fully connected neural network (FNN). It is observed that the raw features are not sufficiently informative for theoretical and practical purposes. It is further observed that putting together the raw features and allowing the transformations on the combinations of the raw features, effectively forming a decision context, improves the performance significantly. The performances of both architectures are analyzed through the accuracy metric and confusion matrices. Finally, the thesis is concluded by summarizing the experiments, results, and implications and mentioning the possible future work.