During the past decade, with the rapid development and spread of the Internet of Things (IoT), cloud computing, and intelligent terminals, the application of Location-Based Services (LBS) has attracted wide attention from both academia and industry. In the outdoor environment, satellite-based positioning technologies (like GPS) can provide convenient location services for people, to support applications such as vehicle navigation and cargo tracking. However, in indoor environments and the dense urban areas, due to severe object occlusion and multipath effects of signal propagation, the accuracy of satellite-based positioning technologies decreases and cannot meet the applications’ demands. Accurate and real-time positioning is highly demanded by location-based services and can be beneficial for radio resource management in 5G networks currently being deployed to achieve significant performance improvement over existing cellular networks. Many new technologies, for example, massive Multiple Input Multiple Output (MIMO), millimeter Wave (mmWave) communication, Ultra-Dense Network (UDN), and Device-to-Device (D2D) communication, are introduced in 5G networks, not only to enhance communication performance but also to provide the opportunity to improve positioning accuracy significantly. It is envisioned that 5G networks will be capable of locating a User Equipment (UE) with an accuracy of sub-meter and with high network utility. In the year 2021, the research team made several steps forward in outdoor and indoor positioning by creating Machine Learning-based positioning solutions. A novel ML-based mobile positioning framework was developed which can utilize several ML methods (Decision Tree, Random Forest, Extra Trees, AdaBoost, Neural Networks) combined with timing measurements to determine the position based on the radio network measurements. The novel method can be used in every (indoor and outdoor) environment and contains various parameters that can be fine-tuned for every environment. After it was tested in various environments, such as rural, suburban, and dense urban environments in Japan, rural environments in Germany, and urban environments in the USA, it was clear that it outperforms the legacy AECID mobile positioning method and has similar results as our previously developed fingerprinting-based mobile positioning solution. The new ML-based solution needs more testing and a scalability analysis to replace our previous solution which was deployed at the end of last year. Our novel ML-based method was further improved with measurement augmentation and propagation model-based measurement generation. These new features are in test phase and show great potential not just in positioning accuracy but in model robustness. Hopefully, all these new features will be deployed in 2022.
Dr. Alija Pašić, Ferenc Mogyorósi, Péter Revisnyei, Gergely Dobreff
2021-09-08
Támogató: Ericsson Magyarország Kft.