A Comparative Performance Evaluation of Machine Learning Algorithms for Fingerprinting Based Localization in DM-MIMO Wireless Systems Relying on Big Data Techniques

Mobile terminal (MT) localization based on the fingerprint approach is a strong contender solution for utilization in microcells urban environments and indoor settings that suffer from severe multipath and signal degradation. In this paper, we investigate and evaluate the performance of thirteen machine learning (ML) algorithms (including multi-target algorithms) employed in conjunction with fingerprint based MT localization for distributed massive multiple input multiple-output (DM-MIMO) wireless systems configurations. The fingerprints will rely solely on the received signal strengths (RSS) from the single-antenna MT collected at each of the receive antenna elements of the massive MIMO base station. The performance is evaluated through numerical simulations incorporating practical millimeter-wave signal propagation models suited for 5G wireless systems in combination with ray-tracing techniques, and in conjunction with the 3D OpenStreetMap to replicate real-life environments. In addition, the ML computational platform, and implementation of the proposed framework was selected with a focus on efficiently handling the anticipated big data that could be generated from a typical 5G network with expected large subscriber cell density (1 million/km(2)). To that end, an Apache Spark based ML platform is proposed and employed. Several DM-MIMO system topologies and configuration parameters combinations affecting MT localization were investigated to analyze performance. Numerical simulation results demonstrated that the location of a MT could be effectively predicted by means of a subset of the collection of considered ML algorithms. The obtained results of MT localization performance evaluation metrics served to identify an optimum ML algorithm and methodology for employment in DM-MIMO systems.