Empirical Data-Driven Multiple User Detection for Dynamic Indoor Factory Settings

The future of the manufacturing industry is envisioned to be built with a massive number of connected devices, which necessitates robust machine-type connectivity. Grant-free access protocols leveraging non-orthogonal multiple access have emerged as a promising technique to minimize signaling overhead and enhance resource utilization in such scenarios. A major challenge in this regard is that the base station needs to reliably detect active devices in the absence of sparsity and channel state information during the initial access phase. To address this problem, we evaluate a novel multi-user detection algorithm under realistic conditions, integrating empirical data collected with mobile automated guided vehicles in a production plant. By restructuring the original problem as a binary relevance problem, we exploit deep learning techniques and propose a low-complex deep neural network for precise and scalable detection. Comprehensive numerical evaluations validate the proposed model’s effectiveness under realistic conditions and emphasize its potential for implementation in real-life scenarios. Our proposed model detects closer to 99% under higher sparsity scenarios, laying the groundwork for intelligent and adaptive enablers in next-generation industrial ecosystems.