Improvement of source localization via cellular network using machine learning approach

Hyperbolic source localization and Taylor-series estimations are widely used and standardized by applying the time difference of a received signal between sensors. However, due to the different installation positions of sensors, data must be transmitted via a wide area and local network to arrive at a processing center. These are essential parameters in finding the correct position of a wave source. Therefore, machine learning-based prediction methods have drawn significant attention due to their outstanding execution and robust modeling potential. This paper aims to improve the source localization process, including the transmission rate occurring in a public network, through a combination of graphical modes, a Taylor-series expansion, and machine learning regression by developing so-called machine learning-based cross-correlation (ML-CCR) algorithms. The actual FM radio signals from three broadcast stations were used to train the model, and the regression algorithm was performed with another dataset containing untrained data. The experimental results indicated that the ML-CCR algorithm with random forest and boost regression provided more beneficial outcomes for range difference determinations and Taylor-series estimations than the standard cross-correlation technique. Moreover, our developed ML-CCR algorithm can improve the source localization error due to unwanted delay in a wide area and local network.