Time-Series Classification Using a CNN-LSTM Model with Attention, Autoencoder, and K-Fold Cross-Validation

Here, we propose a hybrid Deep Learning (DL) framework consisting of a Denoising Autoencoder (DAE), Convolutional Neural Network (CNN), Bidirectional LSTM (BiLSTM), and a custom Attention mechanism based on these classic models, capable of resolving sixth generation signal identification and localization. The model deals with both real and imaginary parts of the signals to achieve robustness, adopting advanced Data Augmentation (DA) and regularization techniques. When tested with k-fold cross-validation and learning rate optimization, it showed constant accuracy. This framework enables inductive learning to handle complicated 6G signals and thus can help enhance AI techniques in next-generation Wireless Communication Systems (WCS).