In order to promote exchanges and interactions among various subdivisions of commercial aerospace, NiHaoSpace has launched an expert column. Invite industry leaders such as founders, technical leaders, and senior experts from each segment to write articles and share topics such as industry progress, technology paths, and market prospects.<\/p>\n
The author of this issue is Cang Jirong, co-founder of Star Test Future, sharing with you: “The core of the next generation constellation competition-satellite intelligence”.<\/p>\n
As reusable rockets and mega-constellations gradually become the norm, the next focus of competition in commercial aerospace has focused on the intelligence level of the satellites themselves. The traditional “data downlink, ground processing” model is no longer able to meet the minute-level response needs of real-time remote sensing, tactical reconnaissance and other scenarios. Satellites must learn to “think” in orbit and transform from “porters” of data to “decision-makers” of information. On-board intelligence is evolving from a cutting-edge technology to a core competitiveness that determines the performance of next-generation constellations. <\/p>\n
01<\/p>\n
Satellite Intelligence Trend<\/p>\n
With the acceleration of the implementation of real-time remote sensing, satellite Internet and other applications, single-satellite intelligence and constellation collaborative mission intelligent scenarios have emerged, and spaceborne intelligence has become a rigid need and standard configuration of satellites. In the field of environmental monitoring, real-time analysis of satellite images can provide valuable information support for disaster prevention and emergency rescue. For example, Orbital Sidekick’s energy pipeline monitoring constellation and OroraTech and Google’s wildfire monitoring constellation have already been put into operation. In the defense field, satellite on-orbit intelligent processing is the core technology of tactical intelligence, surveillance and reconnaissance (ISR) and battle management and command and control (BMC3). For example, Palantir’s planetary time-sensitive intelligence system Meta-Constellation and the “Proliferated Warfighter Space Architecture” (PWSA) being built by the U.S. Space Force have formed a space-based intelligence network. In the field of communications, spaceborne intelligence can provide communication link optimization, constellation management and operation, autonomous obstacle avoidance and other functions, empowering giant constellations integrating communications and remote control. For example, SpaceX’s Starlink and StarShield carrying military payloads are being deployed in batches. In the field of space situational awareness, spaceborne intelligence assists people in drawing space traffic maps, assisting in garbage cleanup and on-orbit services. For example, LeoLabs, a company that provides space traffic management services, and Clearspace-1, a space debris cleanup project funded by the European Space Agency, are about to launch a verification mission. Space computing centers are an emerging satellite application model in recent years, aiming to use space computing to deploy cloud computing facilities. For example, the American startup Starcloud plans to use NVIDIA H100 to build a data center. The original on-board computing power was mainly designed for a single simple task, using a relatively old computing architecture and backward aerospace-grade process component selection. The performance of a single processing chip was weak, and a large number of stacked processing chips were needed to meet the needs of specific computing tasks. There were problems such as high system complexity, strong coupling of software and hardware, and slow technology iteration. The above application scenarios put forward higher requirements for spaceborne intelligence, which requires making full use of limited satellite resources to provide highly reliable and high-performance computing power to truly serve users’ flexible and changing mission needs. <\/p>\n
02<\/p>\n
Computing power requirements of smart satellites<\/p>\n
The current demand for spaceborne intelligence for efficient aerospace information services can be divided into two categories in terms of computing functions: data processing and multi-satellite collaboration. Taking remote sensing scenarios as an example, data processing mainly includes on-orbit preprocessing of remote sensing data and intelligence interpretation tasks. Multi-satellite collaboration is mainly used for collaborative observation task scheduling of multiple remote sensing satellites to enhance the intelligence value of real-time remote sensing. In addition, spaceborne intelligence also has corresponding mission requirements in scenarios such as satellite communications, space science, and situational awareness. In terms of computing power, we currently use space-based distributed computing, with computing power of 1,000 TOPS and below, serving single-star edge computing needs. The space computing center is a kind of space-based centralized computing, which may be comparable to or even exceed ground data centers (100 EOPS level) in the future. As mentioned above, the data center planned by Starcloud has reached a scale of 1,000 EOPS, and can serve cloud computing needs in space and on the ground. Satellite resources are limited and on-board missions are diverse, so it is necessary to match the appropriate computing power form based on demand. Based on the accumulated research experience of star testing for many years in the future, several typical computing scenario algorithms and computing power requirements are proposed, as shown in the figure below. Image preprocessing and task planning requirements mostly use traditional algorithms, which are characterized by clear rules and efficient calculations, and are suitable for general-purpose computing chips; target detection, image segmentation and other requirements use AI algorithms, which are characterized by data-driven, good at complex tasks, and are suitable for AI chips such as GPU\/NPU. <\/p>\n
03<\/p>\n
The challenge of on-planet computing power<\/p>\n
In order to achieve the goal of bringing computing power to the sky, solve the limitations of existing on-board computing power and meet the growing demand for satellite intelligence, it is necessary to solve the core problems of on-board computing one by one and form a systematic solution. Problem 1: Onboard computing is complex and timeliness requirements are high – software and hardware collaborative optimization Hardware layer optimization: Optimize the design based on the underlying computing architecture, use advanced COTS devices to achieve low cost and high energy efficiency ratio, and use multiple chip types to form a hyper-heterogeneous computing architecture, which can flexibly schedule multiple types of computing resources to achieve comprehensive performance optimization; Software layer optimization: Focus on typical business scenarios, carry out algorithm lightweight and accelerated deployment designs for typical processing algorithms such as signal processing and deep learning to improve hardware resource utilization efficiency. Problem 2: Business scenarios are changing and intelligence is urgently needed – intelligent software architecture with software and hardware decoupling. Layer-decoupled spaceborne intelligent software design: build a multi-layer decoupled software architecture to support flexible orchestration and unified scheduling of business programs; cooperate with OTA upgrades to achieve sustainable updates of business scenarios. Problem 3: Harsh space environment and high reliability requirements – system-level fault-tolerant reinforcement design for COTS devices. System-level radiation-resistant fault-tolerant reinforcement design: Focusing on high-performance COTS devices, fault-tolerant reinforcement design is carried out from various aspects such as structure, hardware, and software to reduce costs while improving system-level reliability. Analogous to the intelligent development of automobiles, with the optimization of computing architecture and upgrading of computing power, satellites will also have a higher level of intelligence and release greater application potential. <\/p>\n
04<\/p>\n
The future of satellite + AI<\/p>\n
Satellite intelligence is becoming a key force in promoting industry change. By deeply integrating AI capabilities into satellite platforms, full-link upgrades from data collection to application services can be achieved. Empowering space-based computing power to unleash the potential of satellites: Based on the satellite-borne intelligent platform, satellites can complete data processing and compressed transmission in orbit. Whether it is minute-level response to key intelligence or joint observation of space science, space-based computing power allows satellites to be upgraded from “data collectors” to “intelligent decision-makers”. Open ecosystem co-construction, lowering the threshold of intelligence: Based on the open intelligent platform, it provides standardized development tools and algorithm libraries to support customers to quickly customize AI applications. It can support global partners to jointly build an aerospace information service ecosystem, allowing various applications to obtain satellite wisdom with minimal investment. Deep integration of multiple scenarios creates sustainable value: from real-time remote sensing to satellite Internet, from space traffic management to deep space exploration, spaceborne intelligence is reshaping industry application models. Through continuous algorithm updates and service subscriptions, satellites in orbit can continue to evolve, providing accurate and reliable intelligent services for major areas such as aerospace infrastructure and people’s livelihood applications. Satellite intelligence is becoming another new opportunity for commercial aerospace after reusable rockets and low-orbit communication constellations. It is not only about improving the capabilities of single satellites, but also reshaping satellite and constellation operation models and industry competition. In the future, whoever can establish advantages in on-board computing power and application ecology will be able to take the initiative in the aerospace information service market. <\/p>\n
About the author<\/p>\n
Cang Jirong, co-founder and CEO of Star Testing Future, holds a bachelor’s degree and a Ph.D. in the Department of Engineering and Physics of Tsinghua University, a postdoctoral fellow in the Department of Astronomy, a senior engineer, a rising entrepreneurial star in Beijing, a Zhongguancun U30 winner, and the deputy director of the “Space Particle Information and Space Astronomical Detection” Center of the Hebei Tsinghua Development Research Institute. He has long been engaged in research in the fields of particle and astrophysics detector electronics and high-performance heterogeneous computing. He is currently focusing on the deep integration of “AI + aerospace” and has led the team to develop and put in orbit more than 20 internationally leading on-board intelligent processing payloads.<\/p>\n","protected":false},"excerpt":{"rendered":"
In order to promote exchanges and interactions among various subdivisions of commercial aerospace, NiHaoSpace has launched an expert column. Invite industry leaders such as founders, technical leaders, and senior experts from each segment to write articles and share topics such as industry progress, technology paths, and market prospects. The author of this issue is Cang […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":"","_links_to":"","_links_to_target":""},"categories":[2],"tags":[],"class_list":["post-25202","post","type-post","status-publish","format-standard","hentry","category-news"],"acf":[],"_links":{"self":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/25202"}],"collection":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/comments?post=25202"}],"version-history":[{"count":0,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/25202\/revisions"}],"wp:attachment":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/media?parent=25202"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/categories?post=25202"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/tags?post=25202"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}