Enhancing 6G Positioning Accuracy through AI-Driven Localization Algorithms

Sixth-generation (6G) wireless networks target ultra-high data rates, sub-millisecond latencies, and centimeterto-millimeter-level positioning accuracy. This unprecedented localization precision is essential for missioncritical applications such as autonomous transportation, smart grids, extended reality (XR), and industrial automation. However, operating at high frequencies (millimeter-wave and Terahertz bands) exposes 6G signals to severities including rapid path loss, signal blockages, and complex multipath interference. Traditional geometric and range-based methods face fundamental performance ceilings under these non-line-of-sight (NLoS) and dynamic channel conditions. This paper proposes a hybrid, artificial intelligence-assisted localization framework to solve these challenges. We design a multi-layer feedforward backpropagation Artificial Neural Network (ANN) trained on multi-path Received Signal Strength Indicator (RSSI) profiles. To tackle high-velocity user equipment (UE) tracking in dense urban centers, we deploy a self-tuning Kalman Filter (KF) configured directly over the raw neural network predictions. We evaluate our design in a simulated100m equilateral 6G base station grid. Under dynamic shadowing conditions (1.5 dB) and high velocity (35 m/s or 126 km/h), the proposed hybrid model successfully reduces the mean physical positioning error from 7.970 meters (baseline static ANN) to 4.380 meters—achieving a substantial 45.04% improvement in tracking accuracy and a 41.83% improvement in the 90th percentile bounds (down to 7.793 meters). These empirical results validate the robustness of blending low-overhead AI models with temporal kinetic filters, supporting high-fidelity location services in next-generation networks.