Radio Map Estimation Using a Generative Adversarial Network and Related Business Aspects

Abstract

Radio maps are widely used in the network resource management of 5G communication systems. However, a frequent radio map update is expensive and inefficient in practice because of the need to collect measurements from many devices. In this paper, we propose a highly accurate deep learning technique to predict the propagation path loss from any point on a planar domain with respect to the transmitter. Our proposal adopts a generative adversarial network to yield precise path loss estimations which are very close to ray tracing simulations but are computationally more efficient.The use of our proposal offers a three-fold benefits to a firm in terms of commercialization: Dynamic estimation of radio map corresponding to the changes in the environment, zero-touch automation of radio map generation, and more importantly, the development of end-to-end AI based network planning solution.