Scholarlink

<p>The integration of the low Earth orbit (LEO) satellite and the terrestrial networks has extended the coverage of the Internet of Things (IoT) from densely populated areas to the entire globe. Random access plays an important role in LEO satellite-based IoT (SIoT) since many sensors on the ground need to send the data back to the LEO satellites with a stringent delay requirement. Due to the significant difference in inherent characteristics between the LEO satellite-based systems and the terrestrial networks, the factors of consideration for random access are quite different. First and foremost, a LEO satellite has limited resources, and the coverage is rather dynamic. Secondly, the services provided require scalability and differentiated quality of service (QoS). Thirdly, the received packets are sporadic and sparse at the satellites. In this paper, we propose using a deep neural network box (DNNB) to resolve collisions for resource reservation in the SIoT. An active sensor node sends a reservation packet, which contains a randomly generated ticket number and a password with a checksum. The former is converted into a signature by the mapping of the Finite Projective Plane (FPP). The resource allocator (RA) at the LEO satellite uses the output of the DNNB to determine the active sensor nodes of the reservation packet and assign resources accordingly. The confirmation of resource reservation is doubly checked by the integrity of passwords, placed independently and sequentially in the password section. Through such a dual checking system, the RA at the LEO satellite-based system can take either a conservative policy, an aggressive policy, or a hybrid policy in allocating resources. The reservation-based random access with the assistance of machine learning (ML) can provide high throughput, high scalability, differentiated QoS, and age of information (AoI). In the performance evaluation, we analyze the expected throughput and mean delay for the reservation-based system, and compare the proposed DNNB with CRDSA and IRSA. Lastly, we provide the design of a multi-class QoS mechanism.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning Assisted Random Access in LEO Satellite-Based Internet of Things

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Tai-Kuo Woo,
  • Chen-Hua Fu,
  • Yichun Wu

摘要

The integration of the low Earth orbit (LEO) satellite and the terrestrial networks has extended the coverage of the Internet of Things (IoT) from densely populated areas to the entire globe. Random access plays an important role in LEO satellite-based IoT (SIoT) since many sensors on the ground need to send the data back to the LEO satellites with a stringent delay requirement. Due to the significant difference in inherent characteristics between the LEO satellite-based systems and the terrestrial networks, the factors of consideration for random access are quite different. First and foremost, a LEO satellite has limited resources, and the coverage is rather dynamic. Secondly, the services provided require scalability and differentiated quality of service (QoS). Thirdly, the received packets are sporadic and sparse at the satellites. In this paper, we propose using a deep neural network box (DNNB) to resolve collisions for resource reservation in the SIoT. An active sensor node sends a reservation packet, which contains a randomly generated ticket number and a password with a checksum. The former is converted into a signature by the mapping of the Finite Projective Plane (FPP). The resource allocator (RA) at the LEO satellite uses the output of the DNNB to determine the active sensor nodes of the reservation packet and assign resources accordingly. The confirmation of resource reservation is doubly checked by the integrity of passwords, placed independently and sequentially in the password section. Through such a dual checking system, the RA at the LEO satellite-based system can take either a conservative policy, an aggressive policy, or a hybrid policy in allocating resources. The reservation-based random access with the assistance of machine learning (ML) can provide high throughput, high scalability, differentiated QoS, and age of information (AoI). In the performance evaluation, we analyze the expected throughput and mean delay for the reservation-based system, and compare the proposed DNNB with CRDSA and IRSA. Lastly, we provide the design of a multi-class QoS mechanism.