Energy-Efficient Data Gathering and Computing in LEO Satellite-Assisted Marine IoT Networks

Satellite communication has emerged as a promising technology for achieving a wide range of communication coverage and providing a variety of services in the marine Internet of Things (IoT) network. However, it has to cope with the energy efficiency and the scarcity of radio resources. This paper thus investigates how to improve the system-wide energy efficiency through the efficient data gathering and computing in the low earth orbit (LEO) satellite-assisted marine IoT network. To be specific, in different time slots, a group of sensing devices (SDs) deployed in the offshore area utilize non-orthogonal multiple access (NOMA) to upload their sensor data to the LEO satellite passing over the feasible communication area. The LEO satellite begins to process its received sensor data from the next time slot when a SD completes its data uploading. To ensure the energy-efficient data gathering and computing, we aim to minimize the overall energy consumption required for the data gathering and computing by jointly optimizing the data gathering time, the amount of gathering data, and the computation resource allocation subject to the minimum gathering-plus-computing latency. Due to its non-convex nature, we obtain the near-optimal solution to the proposed optimization problem by alternatively solving the following two sub-problems with the variable replacement: a subproblem optimizing the data gathering time in the top layer and a subproblem optimizing the amount of gathering data in the bottom layer. Therefore, we first propose a successive convex approximation-based (SCA-based) algorithm to obtain the optimal data sizes to transmit to the LEO satellite under the given data gathering time in the bottom layer. Then, a cross entropy-based learning (CEL) algorithm is proposed to optimize the time of gathering sensor data in the top layer. Finally, we present numerical results to evaluate the performance of our proposed scheme and algorithms. The simulation results demonstrate that our proposed algorithms outperform comparative benchmarks.