A Machine Learning Method for Positioning in the 5G Cellular Networks Open Access
Downloadable ContentDownload PDF
Technologies in the developing of 5G-communication system provide interesting prospects, which are beneficial for positioning. The substantial increase in the number of mobile users (MUs) and their demand on higher data rates and volumes, create challenges on the efficient distribution of limited resources in future 5G networks. Indoor and outdoor positioning of users could be of benefit in spectral balancing, power efficiency and geographic routing by base stations (BS). Although existing positioning techniques can mostly overcome problems caused by path loss, background noise and Doppler effects, multiple paths in complex indoor or outdoor environments present additional challenges. In addition, dynamic environments in the outdoor will also bring some troubles in the localization of cellular network.In the first part of this dissertation, we are trying to introduce some new technologies employed in the 5G systems, which will be added in our localization system design. The emerging technology of flexible massive multiple-input multiple-output (MIMO) through multiple antennas reduce the interference while transmitting more information from many more antennas at the same time. We will employ the advantages of MIMO to transmit and receive more signals of mobile users and thereby increase the efficiency of localization. Also several beamforming designs have been proposed in existing literature for channel communications that could be collaborated with the conventional MIMO systems. Different beamforming techniques can be flexibly employed in adapting the distribution of wireless bandwidth.In the second part of the dissertation, in order to address the detailed problems, we propose a BeamMaP positioning system to locate users and steer the beams efficiently in a distributed massive MIMO system. To simulate a realistic environment, we evaluate the positioning accuracy with channel fingerprints from uplink received signal strength (RSS) data, including line-of-sight (LoS) and Non-line-of-sight (NLoS), in the training data sets. Analytical models prove the better performance compared with conventional positioning system.In an effort to improve the flexibility in the outdoor environment, we propose design an improved adaptive BeamMaP that can instantaneously locate users in dynamic environment urban in the third part of dissertation. In addition, based on the adaptive beamforming, we employ the Rice distribution to sample the current mobile users locations in the testing data sets. Our simulation results achieve reduced root-mean-squared estimation error (RMSE) performance with increasing volume of training data. The results demonstrate the effectiveness of the adaptive beamforming model in the test process.Finally, we discuss the inadequate consideration of our proposed methods and expand our motivation to improve the better performance for our future works in the last chapter.