Session

Session 1: Big Data From Small Satellites 1

Abstract

Tool sets, algorithms and technologies developed to create value from the availability of big data have potential not only to justify and reward the collection of sensor data from space but also to improve the quality of sensor data collection. The 2017 HawkEye360 Pathfinder mission will demonstrate the use of tight interactions between constrained, space-based compute platforms with sensor hardware and an approach to ground-segment data processing typical of cloud-based, Big Data analysis to maximize the performance of payload hardware on-orbit. We present specific examples related to the improvement of time- and frequency-of-arrival (TOA and FOA) estimation for AIS transmissions due to specific-emitter characterization on-orbit made feasible by the application of machine learning to take place on the ground. Using a small corpus of raw AIS data captured from commodity hardware on planes over the Chesapeake Bay, we investigate early prototype machine-learning models and test hypotheses as to on-orbit collection improvements. Providing a description of the compute resources available as part of the HawkEye Pathfinder payload, we discuss system design considerations and practical approaches to deploying payload sensor data collection enhancement as part of an automated system for smallsat data collection, ingestion and enhancement. Limitations facing the application of techniques derived from Big Data analytics to the problem of enhanced payload data collection via specific-emitter characterization arise as part of the system design discussion. The HawkEye Pathfinder power budget and payload processing resources will not support constant execution for the most effective methods to enhance TOA and FOA estimation on-orbit, and sensor connectivity to the ground system lags most terrestrial Big Data processing systems in most aspects. We describe the HawkEye Pathfinder analytic software stack, focusing on how it leverages code and concepts developed to enable Big Data processing and how those concepts extend to facilitate improved sensor data collection as part of a mutual feedback system between space and ground processing components. Typical Big Data business models involving power-sensitive commodity hardware sensors at the periphery of a system serviced by a backbone of cloud compute resources have evolved a number of effective open-source and academic software resources amenable to the smallsat use case. We posit ideas for mitigating these factors through the application of predictive analytics.

nick_mccarthy.pdf (3335 kB)
Presentation

Share

COinS
 
Aug 7th, 2:45 PM

Eating Your Own Big-Data Dogfood: Exquisite Collection with Non-Exquisite Hardware

Tool sets, algorithms and technologies developed to create value from the availability of big data have potential not only to justify and reward the collection of sensor data from space but also to improve the quality of sensor data collection. The 2017 HawkEye360 Pathfinder mission will demonstrate the use of tight interactions between constrained, space-based compute platforms with sensor hardware and an approach to ground-segment data processing typical of cloud-based, Big Data analysis to maximize the performance of payload hardware on-orbit. We present specific examples related to the improvement of time- and frequency-of-arrival (TOA and FOA) estimation for AIS transmissions due to specific-emitter characterization on-orbit made feasible by the application of machine learning to take place on the ground. Using a small corpus of raw AIS data captured from commodity hardware on planes over the Chesapeake Bay, we investigate early prototype machine-learning models and test hypotheses as to on-orbit collection improvements. Providing a description of the compute resources available as part of the HawkEye Pathfinder payload, we discuss system design considerations and practical approaches to deploying payload sensor data collection enhancement as part of an automated system for smallsat data collection, ingestion and enhancement. Limitations facing the application of techniques derived from Big Data analytics to the problem of enhanced payload data collection via specific-emitter characterization arise as part of the system design discussion. The HawkEye Pathfinder power budget and payload processing resources will not support constant execution for the most effective methods to enhance TOA and FOA estimation on-orbit, and sensor connectivity to the ground system lags most terrestrial Big Data processing systems in most aspects. We describe the HawkEye Pathfinder analytic software stack, focusing on how it leverages code and concepts developed to enable Big Data processing and how those concepts extend to facilitate improved sensor data collection as part of a mutual feedback system between space and ground processing components. Typical Big Data business models involving power-sensitive commodity hardware sensors at the periphery of a system serviced by a backbone of cloud compute resources have evolved a number of effective open-source and academic software resources amenable to the smallsat use case. We posit ideas for mitigating these factors through the application of predictive analytics.