Machine Learning over Riemannian Manifolds for Activity Detection in Grant-Free Random Access
Conference
ElSakka, OM, Ibrahim, AS, Ismail, MH. (2025). Machine Learning over Riemannian Manifolds for Activity Detection in Grant-Free Random Access
. 2015 IEEE GLOBAL COMMUNICATIONS CONFERENCE (GLOBECOM), 6490-6495. 10.1109/GLOBECOM59602.2025.11432106
ElSakka, OM, Ibrahim, AS, Ismail, MH. (2025). Machine Learning over Riemannian Manifolds for Activity Detection in Grant-Free Random Access
. 2015 IEEE GLOBAL COMMUNICATIONS CONFERENCE (GLOBECOM), 6490-6495. 10.1109/GLOBECOM59602.2025.11432106
Grant-Free Random Access (GFRA) is a key enabler for massive Machine-Type Communications (mMTC) and ultrareliable low-latency communication (URLLC) in 5G and beyond. Traditional methods for user activity detection often rely on prior knowledge of Large Scale Fading Coefficients (LSFC) and the number of active users, which limits their practical applicability. In this paper, we propose a novel manifold-based activity detection method that utilizes the covariance structure of received signals to efficiently identify active users without requiring LSFC or prior knowledge of the number of active users. The proposed method integrates Riemannian distance metrics with machine learning techniques, including logistic regression, Gaussian modeling, and k-means clustering, to enhance detection accuracy. We evaluate three variants: (1) a threshold-based approach trained using logistic regression, (2) an adaptive Gaussian modeling technique for real-time threshold determination, and (3) a k-means clustering method, which proves to be the most effective among the three. Simulation results demonstrate that the manifold-based method outperforms the Multiple Measurement Vector (MMV) approach in detection accuracy while maintaining significantly lower computational complexity. Our findings establish the proposed manifold-based framework as a promising solution for scalable, low-latency activity detection in next-generation wireless networks.
