Machine-Learning Assisted Outdoor Localization via Sector-based Fog Massive MIMO

Gains in area spectral efficiency have been recently demonstrated in networks of large-antenna arrays by means of fog massive MIMO operation and virtual sector-based processing. In this paper, we adopt such sector-based processing and operation to localize users in the network. We investigate the viability of some widely used supervised-learning methods in estimating user locations by observing across the fog massive MIMO network signals transmitted by the users. In particular, we evaluate linear regression (LR), weighted K-nearest neighbors (WKNN), and neural networks (NN) in the context of a network of massive-antenna remote-radio heads (RRHs) using simulations based on a spatially consistent channel model. As our simulations reveal, NN-based location estimators trained with user-sector channel gains could be a viable approach for providing user location information at the network side.