Acesso aleatório não-ortogonal baseado em Q-learning para redes IoT-satélite auxiliadas por relays

Abstract

In a scenario with multiple IoT devices, spread over a very large area, the biggest difficulty is providing access, in view of the great competition for resources available in a central station that provides connectivity. Here, the use of LEO satellites is an alternative. However, IoT devices do not have the transmission power needed to stablish connection with the satellite, só the use of intermediate relays that capture this information is crucial. With this scenario in mind, this work proposes a random access protocol, based on non-orthogonal multiple access (NOMA) with access control through machine learning with 𝑄-Learning. The proposed QL-NOMA algorithm decides, from a 𝑄-Table, the communication channel and the timeslot in which each device will transmit in connection with a relay that retransmits to a satellite, aiming to improve the throughput in the device-relay link. Still, even if there are collisions, relays apply successive interference cancellation (SIC), seeking to decode signals from different devices transmitted on the same timeslot-channel. From the successful transmission of each device, the 𝑄-Table of each one is updated. As a comparison, the Slotted-ALOHA random access protocol was chosen, which chooses the channel and timeslot at random. Numerical results show that the proposed method significantly improves performance compared to Slotted-ALOHA with NOMA, despite being a decentralized strategy. Furthermore, it is possible to conclude that there is an optimal number of relays and channels depending on the number of IoT devices trying to access the network.