Learning to Optimize: Balancing Two Conflict Metrics in MB-HTS Networks

TL;DR

This paper introduces novel power allocation strategies for MB-HTS satellite networks that balance throughput and user satisfaction, employing both model-based and supervised learning solutions to improve performance under congestion.

Contribution

It formulates a new multi-objective optimization problem and proposes two effective solutions, one model-based and one learning-based, to address congestion in satellite networks.

Findings

Solutions effectively handle congestion and improve user satisfaction.

Proposed methods outperform previous approaches in data throughput.

Supervised learning reduces computational complexity.

Abstract

For multi-beam high throughput (MB-HTS) geostationary (GEO) satellite networks, the congestion appears when user's demands cannot be fully satisfied. This paper boosts the system performance by formulating and solving the power allocation strategies under the congestion control to admit users. A new multi-objective optimization is formulated to balance the sum data throughput and the satisfied user set. After that, we come up with two different solutions, which efficiently tackle the multi-objective maximization problem: The model-based solution utilizes the weighted sum method to enhance the number of demand-satisfied users, whilst the supervised learning solution offers a low-computational complexity design by inheriting optimization structures as continuous mappings. Simulation results verify that our solutions effectively copes with the congestion and outperforms the data throughput demand than the other previous works.