In the rapidly evolving landscape of satellite technologies, a novel concept known as space computing power networks (Space-CPN) is emerging as a potential solution to address the growing demands for efficient data processing and transmission in space-based applications. A recent article published in Engineering titled "Space Computing Power Networks: Fundamentals and Techniques" delves into the intricacies and potential of Space-CPN, highlighting its ability to integrate communication and computation capabilities across various types of satellites.
Over the past few decades, satellite technologies have made significant strides, enabling seamless global connectivity and extensive data collection through remote sensing. However, traditional methods of downloading raw data from satellites for further processing on the ground face several challenges, including short visible windows for satellite–ground communication, high-frequency band limitations, and privacy concerns. To overcome these hurdles, Space-CPN has been proposed as a promising architecture that leverages advanced edge computing techniques and onboard computing payloads to empower fast and trustworthy onboard information processing.
Space-CPN expands the concept of computing power networks by incorporating innovations in satellite networks. It integrates the communication and computation capabilities of low-Earth-orbit (LEO), medium-Earth-orbit (MEO), and geostationary-Earth-orbit (GEO) satellites, allowing for flexible scheduling of computing power to achieve secure, fast, and accurate onboard intelligent data processing. For instance, GEO/MEO/LEO satellites can serve as space computing centers for emergency scenarios, while ground stations act as terrestrial computing centers for high-demand computing tasks.
The development of Space-CPN faces several key challenges. One lies in the design of communication principles that can support specific computation tasks, shifting from traditional data-oriented transmission to task-oriented communication. The article introduces the robust information bottleneck (RIB) principle, which aims to maximize the mutual information between the result and the label of the data sample for accuracy, while minimizing the mutual information between the feature and the input sample for compression. This approach enhances communication robustness without additional overhead.
Another challenge is the onboard computing architecture, where limited energy resources necessitate low-power solutions. Neuromorphic computing, which mimics the human brain's integrated memory and processing capabilities, is highlighted as a potential solution. The article discusses the application of spiking neural networks (SNNs) and proposes satellite federated and decentralized neuromorphic learning network architectures to enable efficient onboard training.
Resource allocation in Space-CPN is also a critical issue, given the dynamic and uncertain nature of space networks. The article proposes robust optimization methods, including robust reinforcement learning for satellite microservice deployment and distributionally robust optimization for satellite task scheduling. These methods aim to quantify the relationship between computation task requirements and network resources, ensuring efficient utilization in highly dynamic environments.
Space-CPN represents a significant step forward in the integration of communication and computation in space networks. By addressing the challenges of task-oriented communication, energy-efficient computing architectures, and robust resource allocation, Space-CPN has the potential to revolutionize how data is processed and transmitted in space-based applications, paving the way for more efficient and intelligent satellite systems.
The paper "Space Computing Power Networks: Fundamentals and Techniques," is authored by Linling Kuang, Yuanming Shi, Kai Liu, Chunxiao Jiang. Full text of the open access paper: https://doi.org/10.1016/j.eng.2025.06.026
