Session
Session I: Advanced Technologies 1 - Research & Academia
Location
Salt Palace Convention Center, Salt Lake City, UT
Abstract
Utah State University has been developing hardware accelerators and machine learning algorithms for real-time detection and attribution of space weather events on small satellites. This research, funded under the Low-power Array for CubeSat Edge Computing Architecture, Algorithms, and Applications (LACE-C3A), is a NASA STMD-funded effort as part of the University Smallsat Technology Partnerships (USTP). In this paper, we present an FPGA implementation of an AI/ML algorithm for intelligent decision-making, enabling real-time space weather monitoring and now-casting of Equatorial Plasma Bubbles (EPBs). EPBs are low-density plasma structures that form at low latitudes within the Earth’s ionosphere, rising along magnetic field lines after sunset and persisting throughout the night. These structures cause severe scintillation in radio signals, degrading the performance of GPS and other satellite communication systems. The objective of LACE-C3A is to develop an FPGA-based edge computing platform and AI/ML processing algorithms at the CubeSat scale for a variety of detection and attribution problems. Given the global reliance on GPS for navigation, aviation, and communications, real-time detection of plasma bubbles is critical for mitigating their impacts. To achieve real-time detection, data from in-situ sensors such as Langmuir probes and impedance probes onboard CubeSats can be used to identify EPBs. Utah State University developed a suite of plasma sensors, Space Weather Probes (SWP), which flew and collected data on EPBs during the Scintillation Prediction and Observation Research Task (SPORT) mission. An XGBoost machine learning model was trained on SPORT mission electron density data from SWP to detect plasma bubbles using a desktop computer. This paper presents our research extending that work by optimizing and implementing XGBoost on a low-power flash-based FPGA for real-time inference using a custom decision tree accelerator and FFT-based feature vector creation. The FPGA processes sensor data onboard in real time, generating timestamped plasma bubble detections and attributions, which can be reported via an inter-satellite link. This FPGA-accelerated machine learning implementation will be demonstrated on the upcoming ITA-SAT2 mission. As part of the ITA-SAT2 mission, the LACE-C3A hardware and algorithms will be integrated into a constellation of three 16U CubeSats. This presentation will focus on the implementation of XGBoost on FPGA hardware, the adaptation of machine learning for space-based detection of plasma bubbles, and the expected benefits of real-time, autonomous detection and attribution from CubeSat constellations.
Document Type
Event
Real-Time Space Weather Event Detection and Attribution Using FPGA-Based Machine Learning on CubeSats
Salt Palace Convention Center, Salt Lake City, UT
Utah State University has been developing hardware accelerators and machine learning algorithms for real-time detection and attribution of space weather events on small satellites. This research, funded under the Low-power Array for CubeSat Edge Computing Architecture, Algorithms, and Applications (LACE-C3A), is a NASA STMD-funded effort as part of the University Smallsat Technology Partnerships (USTP). In this paper, we present an FPGA implementation of an AI/ML algorithm for intelligent decision-making, enabling real-time space weather monitoring and now-casting of Equatorial Plasma Bubbles (EPBs). EPBs are low-density plasma structures that form at low latitudes within the Earth’s ionosphere, rising along magnetic field lines after sunset and persisting throughout the night. These structures cause severe scintillation in radio signals, degrading the performance of GPS and other satellite communication systems. The objective of LACE-C3A is to develop an FPGA-based edge computing platform and AI/ML processing algorithms at the CubeSat scale for a variety of detection and attribution problems. Given the global reliance on GPS for navigation, aviation, and communications, real-time detection of plasma bubbles is critical for mitigating their impacts. To achieve real-time detection, data from in-situ sensors such as Langmuir probes and impedance probes onboard CubeSats can be used to identify EPBs. Utah State University developed a suite of plasma sensors, Space Weather Probes (SWP), which flew and collected data on EPBs during the Scintillation Prediction and Observation Research Task (SPORT) mission. An XGBoost machine learning model was trained on SPORT mission electron density data from SWP to detect plasma bubbles using a desktop computer. This paper presents our research extending that work by optimizing and implementing XGBoost on a low-power flash-based FPGA for real-time inference using a custom decision tree accelerator and FFT-based feature vector creation. The FPGA processes sensor data onboard in real time, generating timestamped plasma bubble detections and attributions, which can be reported via an inter-satellite link. This FPGA-accelerated machine learning implementation will be demonstrated on the upcoming ITA-SAT2 mission. As part of the ITA-SAT2 mission, the LACE-C3A hardware and algorithms will be integrated into a constellation of three 16U CubeSats. This presentation will focus on the implementation of XGBoost on FPGA hardware, the adaptation of machine learning for space-based detection of plasma bubbles, and the expected benefits of real-time, autonomous detection and attribution from CubeSat constellations.
