Mobile radio propagation path loss prediction using Artificial Neural Networks with optimal input information for urban environments

Abstract

The propagation of radio waves in a built-up area is of great importance for the design of a mobile communication network and the Path Loss (PL) of the transmitted power is one of the characteristic parameters of the space channels. Traditional methods applied for the estimation of the PL are theoretical and empirical models proposed and well known in the literature. The Artificial Neural Network (ANN) methodology has been introduced as an alternative method for the PL prediction and has been proved effective in finding, via a stochastic evolutionary procedure, the influence of the configuration of the built up urban environment to the signal attenuation during its propagation through it. In most of the ANNs applied for this purpose general parameters, as the mean values of geometrical characteristics of roads and built blocks are used. In this paper the research has focused on the synthesis of ANNs which could obtain PL prediction of sufficient accuracy, using small amount but of proper kind input data. Analytical results are presented, which compared with respective ones received via theoretical methods, prove that the proposed ANNs technique with the optimal input information, is effective in estimating the power PL of the transmitted signals.

Introduction

The prediction of radiowave propagation in a wireless communication environment, as in every type of radio access technology, is crucial for the planning of a mobile network being the physical layer of its operation. It is also a difficult task for several physical mechanisms, as the reflection, the diffraction and generally the scattering of the waves as well as the multipath phenomenon contribute to the transmission of the signal power. Taking into account the continuous movement of the users, the profile of the radio channel changes in real time and the parameters of the propagation process become random variables. The quantities, the values of which, have to be predicted are (a) the reduction of the mean power level of the signal versus the distance from the transmitter (Trx), termed as the ‘large scale’ attenuation and known as Path Loss (PL) prediction and (b) the usual rapid variation of the signal power inside a small area around any point of the path, known as ‘small scale’ variation. What makes, the a priori estimation of both type transmission characteristics, be complicated, is that the propagation occurs in random manmade environment. So, the synthesis of a global model suitable for every built-up area profile, is very difficult. For all that, several PL prediction models have been proposed in the literature. These models, used extensively in communication network planning and signal interference studies, can broadly be classified [1], [2], [3], [4], [5], [6], [7], [8], [9] as (a) empirical and (b) deterministic. The models of the former class are easier to implement and require less computational effort but are less sensitive to the environment's physical and geometrical configuration. Those of the latter category have a certain physical basis and are more accurate but at the cost of more computation effort and the necessity of more detailed information about the coverage area. All the aforementioned methods, being flexible and reliable tools for the PL estimation, can effectively replace the realization of measurements, an also efficient process, which, however requires the installation of a complete system of suitable equipment.

The work at hand proposes Artificial Neural Network (ANN) models for the PL prediction in urban environments. The ANN technique has been introduced in several articles of the literature, for the solution of this problem [10], [11], [12], [13], [14], [15], [16], [17], [18], [19]. The majority of these works use, as input to the ANNs, information which concern general parameters of the manmade terrain as, the mean road width, the mean height and length of built blocks, the percentage coverage of the built area, etc. These formulations lead to easy to handle simple ANNs, because small amount and easy to be gathered input information is required. However this kind of data give to the ANNs, substantially very little information for the relief of the environment and lead inevitably to results of decreased accuracy for points at which the built-up characteristics diverge significantly from their mean values over the entire area. On the other hand there have been proposed ANN techniques of large ANNs [16], which use detailed information for the environment thus obtaining accurate prediction but at the cost of the requirement of a large amount of information data to be selected for the area under consideration. The present work focuses on finding the appropriate kind and amount of input information in a way that, on the one hand this information to include details for the terrain of the environment and on the other hand the amount of data given to the ANNs, namely the number of its input nodes, to be small. The research was made at 900 MHz and the pre-condition in finding these optimal input data, was the ensuring, as far as possible, of the accuracy of the prediction