Intersatellite Imaging Data Transfer Under Bandwidth Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Nanosatellites in low-Earth orbit collect more imaging data than can be transmitted to ground stations due to bandwidth constraints, leading to data discard and inefficient use of onboard storage, while existing ML training methods overlook hardware limitations and unrealistic data distribution assumptions, resulting in low model performance and slow convergence.
Innovation Solution
Implement onboard computing to preprocess imaging data, utilize intersatellite links for data transfer and training, and schedule communication based on satellite status and link topology to optimize data sharing and model aggregation among satellites.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If nanosatellites transmit all collected imaging data to ground stations, then complete data utilization is achieved, but bandwidth constraints and transmission time prevent this from being feasible
Solution Approach 1:
The patent extracts only the essential model training data from the collected imaging data and transfers it to ground stations, rather than transmitting all raw imaging data. This selective extraction of critical information reduces transmission time and bandwidth requirements while maintaining effective model training capabilities.
Solution Approach 2:
The nanosatellites perform preliminary ML model training onboard before data transmission, completing essential processing in advance. This preliminary action allows the satellites to train local models using collected data, then only transfer the trained models or selected data subsets to ground stations, significantly reducing transmission time while preserving data utility.
2Loss of information
If nanosatellites store all collected imaging data onboard, then complete data retention is achieved, but limited onboard storage capacity causes data discard
Solution Approach 1:
The system extracts only the most valuable data subsets for model training from the complete imaging dataset, identifying and retaining critical training samples while discarding redundant data. This selective extraction enables effective model training with minimal onboard storage requirements.
Solution Approach 2:
The patent changes the parameter of data representation by transforming raw imaging data into extracted features, descriptors, or model parameters that occupy significantly less storage space. This parameter transformation maintains the essential information needed for ML training while reducing storage requirements by orders of magnitude.
3Reliability
If existing ML training methods are used without considering hardware limitations, then theoretical model performance is achieved, but practical implementation results in low model performance and slow convergence
Solution Approach 1:
The patent implements local quality by tailoring the ML training approach to the specific hardware constraints of each nanosatellite. Different satellites can use different data subsets, model architectures, or training parameters optimized for their individual storage, compute, and power capabilities, rather than applying a uniform approach that ignores hardware variations.
Solution Approach 2:
The system dynamically adapts the ML training process based on real-time satellite status, available resources, and mission requirements. The training parameters, data selection criteria, and model complexity are dynamically adjusted to match current hardware conditions, enabling effective operation across diverse nanosatellite platforms with varying capabilities.
Data Source
AI summary
A computing device including a processor configured to receive satellite status data from satellites included in a satellite constellation. The processor is further configured to determine a link topology of the satellites. Based at least in part on the satellite status data and the link topology, the processor is further configured to identify a first satellite constellation subset including one or more selected satellite pairs. Identifying the one or more selected satellite pairs includes computing respective link utility values associated with a plurality of candidate pairs of satellites included in the satellite constellation based at least in part on the satellite status data and the link topology. The one or more selected satellite pairs are selected based at least in part on the link utility values. The processor is further configured to transmit, to the satellites included in the first satellite constellation subset, instructions to perform intersatellite imaging data transfer.