Deep Learning-based Signal Identification in Wireless Communication Systems: A Comparative Analysis on 3G, LTE, and 5G Standards

Efficient signal identification in wireless communication systems is critical for optimal service provision. However, the complexity of contemporary criteria and factors such as noise and fading make it hard to do so. To address this problem, convolutional neural networks (CNNs) are used to classify signals using 3G, LTE, and 5G standards. This approach involves creating a range of datasets with different Signal-to-Noise Ratios (SNR) and introducing Rayleigh fading to represent real-world environments. Two CNN architectures for dependable assessment, VGG19 and ResNet18, with robust 5-fold cross-validation, are employed. To test model resilience, the dataset includes Poisson noise and Thermal noise. Despite noise and fading in the system, VGG19 and ResNet18 show high accuracies across all standards. However, ResNet18 demonstrates relatively better performance, especially under Poisson noise conditions. Both models also have good signal detection from among noises generated by Poisson thermal or Rayleigh distribution. ResNet18 demonstrates a commendable average accuracy of 99.52%, while VGG19 Net demonstrates 97.14%. CNNs effectively identify signals amidst noise scenarios and contribute to advancing deep learning techniques in signal processing, enhancing the reliability of wireless communication systems.