Comparative Analysis of CNN and LSTM for Channel Distribution Learning and Classification in Beyond 5G

Abstract

In next generation communication systems, accurate channel distribution information is critical for developing adaptive communication strategies. In this study, we aim to accurately classify the channel distribution in wireless communication systems using artificial intelligence. In this study, a large data set of BPSK modulated signals is generated based on Rayleigh, Rician and Nakagami channel models. The resulting I-Q components are processed using CNN and LSTM architectures and the performance of both models in channel classification is compared. Experimental results show that both models achieve high accuracy rates, with the CNN-based model slightly outperforming the LSTM, especially for the Rician and Nakagami channels. The training process was performed in Google Colab environment using Python-based TensorFlow and Keras libraries, Tesla P100 GPU and high memory configuration. The results show that deep learning-based channel classification methods offer significant potential for effective modeling and discrimination of dynamic channel conditions, and are expected to contribute to improving the efficiency of next generation wireless communication systems.