Open Access

ARTICLE

Deep Reinforcement Learning-Based Long Short-Term Memory for Satellite IoT Channel Allocation

S. Lakshmi Durga¹, Ch. Rajeshwari¹, Khalid Hamed Allehaibi², Nishu Gupta^3,*, Nasser Nammas Albaqami⁴, Isha Bharti⁵, Ahmad Hoirul Basori⁶

1 Electronics and Communication Engineering Department, Vaagdevi College of Engineering, Warangal, 506005, India
2 Department of Computer Science, Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah, 21589, Saudi Arabia
3 Electronics and Communication Engineering Department, Chandigarh University, Mohali, 160036, India
4 Department of Information Technology, Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah, 21589, Saudi Arabia
5 Senior Business Analyst & Solution Architect, SAP Technology & Innovation, Capgemini America Inc., 75039, USA
6 Department of Information Technology, Faculty of Computing and Information Technology in Rabigh, King Abdulaziz University, 21589, Saudi Arabia

* Corresponding Author: Nishu Gupta. Email:

Intelligent Automation & Soft Computing 2022, 33(1), 1-19. https://doi.org/10.32604/iasc.2022.022536

Received 10 August 2021; Accepted 26 September 2021; Issue published 05 January 2022

Abstract

In recent years, the demand for smart wireless communication technology has increased tremendously, and it urges to extend internet services globally with high reliability, less cost and minimal delay. In this connection, low earth orbit (LEO) satellites have played prominent role by reducing the terrestrial infrastructure facilities and providing global coverage all over the earth with the help of satellite internet of things (SIoT). LEO satellites provide wide coverage area to dynamically accessing network with limited resources. Presently, most resource allocation schemes are designed only for geostationary earth orbit (GEO) satellites. For LEO satellites, resource allocation is challenging due to limited availability of resources. Moreover, due to uneven distribution of users on the ground, the satellite remains unaware of the users in each beam and therefore cannot adapt to changing state of users among the beams. In this paper, long short-term memory (LSTM) neural network has been implemented for efficient allocation of channels with the help of deep reinforcement learning (DRL) model. We name this model as DRL-LSTM scheme. Depending on the pool of resources available to the satellite, a channel allocation method based on the user density in each beam is designed. To make the satellite aware of the number of users in each beam, previous information related to the user density is provided to LSTM. It stores the information and allocates channels depending upon the requirement. Extensive simulations have been carried out which have shown that the DRL-LSTM scheme performs better as compared to the traditional and recently proposed schemes.

---

Keywords

---