Saidov Behruz Badridinovich, Isfandiyori Sadullo, Huseynzoda Nasimjon Burikhon

Annotation: 

This study presents the modeling and investigation of artificial neural networks (ANNs) aimed at predicting the parameters of radio signals with different spectral densities. The architectures ADALINE, Focused Linear Network (FLN), and Nonlinear Autoregressive Network (NARX), implemented in the MATLAB Neural Network Toolbox environment, were used. A detailed analysis of one-step and multi-step forecasting was carried out using error functionals (RMS, MAE, MPE). Special attention is given to the application of the obtained results in intelligent data processing for fifth-generation (5G) wireless communication systems, where radio signal forecasting enhances adaptive resource allocation efficiency, reduces transmission delay, and optimizes modulation parameters. The simulation results show that the accuracy and stability of neural network forecasting depend on the spectral complexity of the signal and the parameters of the learning algorithm. The FLN architecture demonstrated the best balance between learning speed and prediction accuracy, while the ADALINE model showed high stability for narrowband signals. The findings can be applied to the prediction of radio channel parameters in intelligent 5G communication systems, where neural network algorithms can improve adaptive resource allocation, reduce latency, and enhance channel robustness under dynamic propagation conditions.

Keywords: artificial neural networks, signal forecasting, ADALINE, 5G, radio channel.