Wavelet Generalized Regression Neural Network Approach for Robust Field Strength Prediction

Abstract

Accurate predictive field strength and coverage modelling during and after cellular network planning process is one key factor that contribute to a successful and robust wireless communication network performance. Accurate field strength coverage prediction will provide realistic idea about the level of field strength and link quality in the entire coverage service areas. It will also assist in close-fitting fringe areas that are likely to be imparted negatively by interference, and cell edge/contour areas with poor signal coverage. Therefore, opting for a suitable predictive field strength system model that will enable superb cellular network planning environment will be of a great succor to the radio network planner and stakeholders, including the network end users as well. This work presents spatial electric field strength prediction engaging hybrid wavelet-neural modelling approach. The proposed is called Wavelet-GRNN. To accomplish this task, the spatial field strength data is first routed through a wavelet-based decomposition process employing three decomposition levels. The decomposed field strength constituents are then utilised as input data to GRNN neural network model where relevant extracted information is captured and trained for robust predictive learning. In the third phase of the model, the outputs from the GRNN predictor are combined with wavelet coefficients to form the final predicted output. The degree of prediction accuracy using the Wavelet-GRNN model over other prediction techniques are also statistically quantified and provided using six different first order statistics.