Enhancing cellular network-based localisation using CNN-CellImage method

Location-based services (LBS) are essential in daily applications. However, GPS-based localisation often suffers from high power consumption, added hardware costs, and unreliable performance in weak or jammed signal environments. Cellular network-based localisation offers a practical alternative by exploiting existing mobile infrastructure. In our earlier work, we developed methods to collect and process cell information datasets using received signal strength (RSS) and evaluated traditional approaches such as centroid, weighted centroid, linear regression, support vector regression, multilayer perceptron, and fingerprinting. In this study, we propose CNN-CellImage, a novel technique that converts RSS and geographical cell relationships into image data for convolutional neural network processing. Using a dataset of 21,155 measurements collected over multiple days and environments, our method achieved a mean localisation error of 119.7 metres, significantly outperforming Cell-ID, centroid, and other machine learning approaches. These results highlight the effectiveness of CNN-CellImage for accurate and robust cellular-based localisation.