Random access (RA) or preamble collision is one of the crucial problems in massive internet-of-things (IoT) at the network entry stage. Since a massive number of IoT nodes simultaneously attempt RAs on the same physical random access channel (PRACH), preambles may be selected by multiple nodes, incurring preamble collisions at the first step of the RA procedure. However, conventional RA models are limited to binary preamble detections which poses severe RA performance loss in the massive IoT environment. In this paper, we propose a deep learning (DL)-based end-to-end RA framework which has detection and resolution abilities for the collided preambles. In particular, advanced preamble classification and timing advance (TA) classifications are performed using deep neural networks (DNNs) for improving the probability of RA success while reducing the delay of the entire RA procedure. The effectiveness of the proposed DNN-based preamble and TA classifiers are demonstrated through extensive simulations. We further evaluate the system-level performance of the proposed DL-based RA model. It shows a significantly higher probability of instant RA success, which makes every node succeed in RA with very limited reattempts, and also maintains a significantly lower RA delay in massive IoT environment.

中文翻译：

---

从眼底图像进行糖尿病视网膜病变分级的斑块级和图像级注释的稳健协作学习


随机接入（RA）或前导码冲突是大规模物联网（IoT）网络进入阶段的关键问题之一。由于大量 IoT 节点同时在同一物理随机接入信道 (PRACH) 上尝试 RA，因此多个节点可能会选择前导码，从而在 RA 过程的第一步中引发前导码冲突。然而，传统的 RA 模型仅限于二进制前导码检测，这在大规模物联网环境中造成了严重的 RA 性能损失。在本文中，我们提出了一种基于深度学习（DL）的端到端RA框架，该框架具有对冲突前导码的检测和解决能力。特别是，使用深度神经网络 (DNN) 执行高级前导码分类和定时提前 (TA) 分类，以提高 RA 成功的概率，同时减少整个 RA 过程的延迟。通过广泛的模拟证明了所提出的基于 DNN 的前导码和 TA 分类器的有效性。我们进一步评估了所提出的基于深度学习的 RA 模型的系统级性能。它表现出显着更高的即时 RA 成功概率，这使得每个节点在重试次数非常有限的情况下在 RA 中成功，并且在大规模物联网环境中保持显着较低的 RA 延迟。