This paper presents an alternative procedure for the prediction of propagation path loss in urban environments. It is based on Neural Network (NN) algorithms and uses the detailed environment proflle instead of the mean values of its structural parameters. The general performance of the NN shows its efiectiveness to yield results with satisfactory accuracy in short time. The received results are compared to the respective ones yielded by the Ray-Tracing model and exhibit satisfactory accuracy either for uniform or for non-uniform distribution of the manmade structured environment. 1. INTRODUCTION The prediction of electromagnetic wave propagation is of great importance in the design and plan- ning of a cellular-network both for mobile and flxed wireless-access systems. A prediction, based on theoretical models, is really valuable since it ofiers the capability of determining optimum base locations, in order to obtain suitable data rates, to estimate their coverage and evaluate the quality of the wireless network without the need of expensive and time consuming measurements. The theoretical models used for the estimation of the path-loss in various urban or suburban areas, even within buildings, are grouped in two categories (1): the empirical or statistical models (e.g., the COST-231-Walflsch-Ikegami model, the Hata model, etc.) and the site-speciflc or de- terministic ones (e.g., the Ray Tracing technique, the Image Method, the FDTD or the Moment Method, etc.). The models of the former category are easier to implement and require less compu- tational efiort but are less sensitive to the environment's physical and geometrical structure. Those of the latter category have a certain physical basis and are more accurate but at the cost of more computations and at the necessity of more detailed information about the coverage area. In the present work, a prediction model based on Neural Network (NN)-architectures is pro- posed. Published works have introduced the NN-methodologies as e-cient techniques for indoor and outdoor estimation of path-loss propagation. They have given solutions, using measured or theoretically produced data and feeding the input of the NN by the values of some of the geometry parameters of the environment, e.g., the mean height and mean dimensions of the buildings and the mean width of the roads (2{6). In the work at hand a Multiple Layer Perceptron (MLP) neural network was composed, and the collections of data, by which it was trained, include the detailed environment proflle. These data were produced using the Ray Tracing technique. Although the calculation for the training collections were made for simple and uniform distribution of the man- made structures, the appropriate grid modeling of the built-up area as well as the way by which the input data were presented to the NN made it e-cient to give, in the generalization phase, results for arbitrary environments, if their proflle is provided. 2. FORMULATION The propagation of radio waves in built-up areas is strongly in∞uenced by the nature of the en- vironment, in particular by the size and the density of the buildings. Urban areas are dominated by tall building blocks with high density and non uniform distribution. Empirical prediction mod- els use mean values for the parameters of the manmade terrain (meanly the mean values of the roads' widths or of the buildings' height). Many proposed NN models use also these parameters as information for the NN. The present work suggests an alternative approach for the prediction of path loss based on a NN-methodology which uses detailed description of the entire coverage area. The used Neural Network was composed via a Multiple Layer Perceptron (MLP) architecture. The input layer consists of a large number of nodes that accept analytical information for the structure of the built-up environment as well as for the coordinates of the position at which the path-loss is going to be estimated. A single node output layer gives the value of this path loss.
Search from 250 M+ research papersSearch from 250 M+ research papers
Open Access
Dec 16, 2023
Seokwon Jang + 4
