Modelling of propagation path loss using adaptive hybrid artificial neural network approach for outdoor environments.

Abstract

Prediction of signal power loss between transmitter and receiver with minimal error is an important issue in telecommunication network planning and optimization process. Some of the basic available conventional models in literature for signal power loss prediction includes the Free space, Lee, COST 234 Hata, Hata, Walficsh- Bertoni, Walficsh-Ikegami, dominant path and ITU models. But, due to poor prediction accuracy and lack of computational efficiency of these traditional models with propagated signal data in different cellular network environments, many researchers have shifted their focus to the domain of Artificial Neural Networks (ANNs) models. Different neural network architectures and models exist in literature, but the most popular one among them is the Multi-Layer Perceptron (MLP) ANN which can be attributed to its superb architecture and comparably clear algorithm. Though standard MLP networks have been employed to model and predict different signal data, they suffer due to the following fundamental drawbacks. Firstly, conventional MLP networks perform poorly in handling noisy data. Also, MLP networks lack capabilities in dealing with incoherence datasets which contracts with smoothness. Firstly, in this work, an adaptive neural network predictor which combines MLP and Adaptive Linear Element (ADALINE) is developed for enhanced signal power prediction. This is followed with a resourceful predictive model, built on MLP network with vector order statistic filter based pre-processing technique for improved prediction of measured signal power loss in different micro-cellular urban environments. The prediction accuracy of the proposed hybrid adaptive neural network predictor has been tested and evaluated using experimental field strength data acquired from Long Term Evolution (LTE) radio network environment with mixed residential, commercial and cluttered building structures. By means of first order statistical performance evaluation metrics using Correlation Coefficient, Root Mean Squared Error, Standard Deviation and Mean Absolute Error, the proposed adaptive hybrid approach provides a better prediction accuracy compared to the conventional MLP ANN prediction approach. The superior performance of the hybrid neural predictor can be attributed to its capability to learn, adaptively respond and predict the fluctuating patterns of the reference propagation loss data during training.