Received Power Prediction for Suburban Environment based on Neural Network

Accurate received power prediction is important to wireless network planning and optimization, and appropriate channel modeling approach is highly demanded. The existing channel modeling approaches mainly include deterministic models and stochastic models. The deterministic models usually require detailed three-dimensional (3D) environment model including geometry and material information. The stochastic models are based on mathematical expressions which are difficult in describing complex environments. For mountaineous and suburban scenarios, where the environment is complex and difficult to reconstructed, both channel modeling approaches are limited. In this work, back propagation (BP) neural network which is the most widely used artificial neural network (ANN), is employed to accurately predict received power in the suburban scenario. An environmental feature is defined, which can describe the propagation environment only by using limited environmental types instead of complex 3D environment reconstruction. Furthermore, the low-dimensional environmental feature is generated by principal component analysis (PCA). From the measurements, the information of base station (BS) and the receiver (Rx), including 3D locations, frequency, transmitted power, angle information of antenna, and the received power of all the locations are obtained. The information of BS and Rx is combined with environmental features to form datasets for ANN training and testing. The training samples are randomly selected with different percentage from the formed datasets. The mean error, standard deviation and the prediction accuracy of BP networks are explored in the work, which will help researchers to carry out network planning and communication system design.