Fingerprint-Based 3D Hierarchical Localization for Cell-Free Massive MIMO Systems

In this article, we address the fingerprint-based three-dimensional (3D) user positioning issue for wide-area cell-free massive multiple-input multiple-output (MIMO) systems. To adequately capture the multipath channel features, we propose a novel angle-space channel power matrix (ASCPM) fingerprint, which is shown to be closely related to the geographical position. Through further analyzing the distribution of fingerprint sparsity, we reveal that each sub-region possesses a unique fingerprint-position mapping relationship, which is intractable for the bulky whole-region-enabled positioning neural networks to obtain. Therefore, a handcrafted hierarchical localization framework is designed for wide low-altitude stereo coverage scenes. To be specific, we exploit the weighted K-nearest neighbor (WKNN) algorithm based on the coarse-grained received signal strength (RSS) to quickly identify the sub-region to which the user belongs. Subsequently, an adaptive three-stage residual convolutional neural network (ATResNet) with fine-grained ASCPM fingerprint as the input is adopted to determine accurate 3D position estimation. By intensive simulations in a ray-tracing signal propagation scenario based on the Wireless Insite (WI) software, the proposed hierarchical framework is demonstrated to achieve high 3D positioning accuracy with low computational complexity and storage overhead.