Study and Comparison of Macro cell Path Loss Prediction Models for Efficient Radio Network Planning

Abstract

Radio network planning needs proper channel characterization and hence approximation of path loss propagation is its key aspect. Path loss is a phenomenon that happens due to the attenuation suffered by the signal as the propagation over the transmitting medium. Wireless systems deployment is expensive; therefore, path loss models propagate dare necessary for the predicting signal strength received(RSS) at the transmitter from a given distance; estimation of radio coverage areas of Base Transceiver Stations (BTS); frequency tests; interference analysis, without conducting huge propagation measurements which are time taking and costly. Thus, predicting the path loss is critical to properly address the lack of signal strength at various places within the coverage region. Empirical path loss prediction models such as Okumura-Hata, COST-231-Hata are less sensitive to change in physical geometry like environmental structures; local terrain profiles; and weather conditions of the site. Artificial Intelligence came to the fore in understanding the past data to make future predictions and is suggested as an alternative approach in this field. In this work, propagation quantities like transmitting to receiving antennas distance, transmitted power and elevation of terrain, clutter height, latitude and longitude are used as inputs to develop path loss model based on AI techniques such as Multi-Layer Perceptron (MLP), Radial Basis Function (RBF)and Adaptive Neuro-Fuzzy Inference Systems (ANIFS)to predict the path loss in macro cell environment. The performance of these AI based models is compared with widely followed HATA model. The experimental results are provided on the recently released dataset containing the RSS values of GSM network working at 900 MHz and 1800 MHz. Mean Square Error (MSE) and R-square fit is achieved between the reading values and the outputs of model as a performance measure. ANFIS model’s performed better compared to other AI models in the prediction of path loss as it is the hybrid model which utilize the advantages of neuro-fuzzy.