Many of the next generation of satellites will be equipped with numerous degrees of freedom in power and bandwidth allocation capabilities, making manual resource allocation impractical. Therefore, it is desirable to automate the operation of these highly flexible satellites. This paper presents a novel approach based on Deep Reinforcement Learning to allocate power in multibeam satellite systems. The proposed architecture represents the problem as continuous state and action spaces. We make use of the Proximal Policy Optimization algorithm to optimize the allocation policy for minimum unmet system demand and power consumption. Finally, the performance of the algorithm is analyzed through simulations of a multibeam satellite system. The analysis shows promising results for Deep Reinforcement Learning to be used as a dynamic resource allocation algorithm.

Automating resource management strategies is a key priority in the satellite communications industry. The future landscape of the market will be changed by a substantial increase of data demand and the introduction of highly flexible communications payloads able to operate and reconfigure hundreds or even thousands of beams in orbit. This increase in dimensionality and complexity puts the spotlight on Artificial Intelligence-based dynamic algorithms to optimally make resource allocation decisions, as opposed to previous fixed policies. Although multiple approaches have been proposed in the recent years, most of the analyses have been conducted under assumptions that do not entirely reflect operation scenarios. Furthermore, little work has been done in thoroughly comparing the performance of different algorithms.In this paper we compare some of the recently proposed dynamic resource allocation algorithms under realistic operational assumptions, addressing a specific problem in which power needs to be assigned to each beam in a multibeam High Throughput Satellite (HTS). We focus on Genetic Algorithms, Simulated Annealing, Particle Swarm Optimization, Deep Reinforcement Learning, and hybrid approaches. Our multibeam operation scenario uses demand data provided by a satellite operator, a full radio-frequency chain model, and a set of hardware and time constraints present during the operation of a HTS. We compare these algorithms focusing on the following characteristics: time convergence, continuous operability, scalability, and robustness. We evaluate the performance of the algorithms against different test cases and make recommendations on the approaches that are likely to work better in each context.

In the recent years, communications satellites' payloads have been evolving from static subsystems to highly flexible components. Modern satellites are able to provide four orders of magnitude higher throughput than their predecessors fourty years ago, going from a few Mbps to several hundreds of Gbps. This enhancement in performance is aligned with an increasing highly-variable demand. In order to dynamically and efficiently manage the satellite's resources, an automatic tool is needed.This work presents an implementation of a new metaheuristic algorithm based on Particle Swarm Optimization (PSO) to solve the joint power and bandwidth allocation problem. We formulate this problem as a multi-objective approach that considers the different constraints of a communication satellite system. The evaluation function corresponds to a full-RF link budget model that accounts for adaptive coding and modulation techniques as well as multiple types of losses. We benchmark the algorithm using a realistic traffic model provided by a satellite communications operator and under time restrictions present in an operations environment.

No abstract

No abstract