Fingerprint Positioning for Massive MIMO Systems Based on Machine Learning

Abstract

Fingerprint-based positioning is promising for mobile cellular systems in dense scattering environments. Massive multiple-input multiple-output (MIMO) has significant advantages in providing accurate positioning due to the high angular resolution. This paper investigates the fingerprint-based positioning problem in massive MIMO systems utilizing machine learning. Adopting a improved spatial beam-based channel model, we exploit a novel beam domain channel amplitude matrix as the location-related fingerprint. We transform the positioning problem into a pattern recognition problem through fingerprint information. The base station can independently classify and distinguish the position fingerprints of different mobile user terminals via machine learning. Then, we propose a categorization-based positioning method by using the neural network. The performance of the proposed machine learning based fingerprint positioning method is evaluated with a geometry-based stochastic channel model. Simulation results demonstrate that the proposed positioning method can perform better than conventional approaches.