مكان‌يابي داخلي مبتني بر سطوح هوشمند قابل‌پيكربندي در ارتباطات امواج ميلي‌متري با استفاده از شبكه‌هاي عصبي گرافي

With the ever-increasing deman‎d for communication services an‎d location-based applications in wireless networks, accurate user localization, especially in indoor environments, has become one of the primary challenges for the next generation of telecommunications networks. Since global satellite navigation systems (GNSS) face significant accuracy degradation in indoor spaces, there is a growing need for alternative methods based on local infrastructures. In this regard, using wireless access points based on Wi-Fi technology in new frequency ban‎ds, especially millimeter waves, is considered a suitable option due to its broad coverage an‎d ability to integrate with existing networks. However, challenges such as signal blockage, non-line-of-sight (NLOS) paths, an‎d power loss at high frequencies make the adoption of novel technologies essential. One effective solution to address these limitations is the use of reconfigurable intelligent surfaces (RIS), which, by providing controlled signal reflections, enable improvements in positioning accuracy. In this study, a three-dimensional indoor localization system is designed an‎d simulated in a 5*100*100 meter environment. This system includes three Wi-Fi-based access points operating at $60$GHz, five reconfigurable intelligent surfaces, an‎d a user with an unknown position. The localization process occurs in two steps: first, an initial estimation using time of arrival (TOA) extraction through inverse Fourier transform (IFFT) an‎d maximum likelihood estimation (MLE), followed by an initial position estimate using a least-squares (LS) method based on time difference of arrival (TDOA). This initial estimate is then fed into a four-layer graph attention network (GAT) to improve the accuracy of the user’s initial position using the network’s geometric structure. Simulation results show that the proposed model improves the accuracy of the initial estimate by approximately 98% These results demonstrate the effectiveness of combining Wi-Fi-based access points, RISs, an‎d graph neural networks in overcoming the challenges of indoor localization, achieving an average test-set accuracy on the order of 2 cm.

محمدجواد اميدي , فروغ السادات طباطباء

حميد سعيدي سورك

احسان يزديان , نغمه سادات مويديان