Application of artificial neural network modeling techniques to signal strength computation

Area of Science:

• Electromagnetics and Signal Propagation
• Computational Intelligence
• Atmospheric Science

Background:

• Received signal strength (RSS) prediction is crucial for broadcast engineering.
• Traditional models like ITU-R P. 526 have limitations in accurately predicting signal strength.
• Atmospheric parameters significantly influence radio wave propagation.

Purpose of the Study:

• To develop and evaluate artificial neural network (ANN) models for computing VHF RSS.
• To assess the impact of different ANN architectures and training parameters on model performance.
• To compare the accuracy of ANN models against established diffraction models.

Main Methods:

• Development of ANN models using the Levenberg-Marquardt back-propagation (LMBP) algorithm.
• Systematic evaluation of activation functions, hidden layer neuron counts, and data normalization techniques.
• Training and testing the models with measured atmospheric and signal strength data.
• Comparison of ANN predictions with measured data and ITU-R P. 526 diffraction model results.

Main Results:

• ANN models demonstrated a good fit between computed and measured signal strength values.
• Low mean square error (MSE) values ranging from 0.0027 to 0.0043 were achieved during training.
• The trained ANN model showed high accuracy on independent datasets with MSE values of 0.0069 and 0.0040.
• A strong correlation was observed between measured field strength and ANN-computed signals, unlike diffraction models.

Conclusions:

• ANN models are effective tools for computing VHF RSS based on atmospheric parameters.
• ANN models offer superior accuracy compared to the ITU-R P. 526 diffraction model for the studied scenario.
• The developed ANN approach provides a reliable method for signal strength prediction in broadcasting.