Session
Weekday Session 11: Advanced Technologies II
Location
Utah State University, Logan, UT
Abstract
Satellites with synthetic aperture radar (SAR) payloads are growing in popularity, with a number of new institutional missions and commercial constellations launched or in planning. As an active instrument operating in the microwave region of the electromagnetic spectrum, SAR provides a number of unique advantages over passive optical instruments, in that it can image in all weather conditions and at night. This allows dense time-series to be built up over areas of interest, that are useful in a variety of Earth observation applications. The polarisation and phase information that can be captured also allows for unique applications not possible in optical frequencies.
The data volume of SAR captures is growing due to developments in modern high-resolution multi-modal SAR. Instruments with higher spatial resolution, wider swaths, multiple beams, multiple frequencies and more polarization channels are being launched. Miniaturization and the deployment of SAR constellations is bringing improved revisit times. All of these developments drive an increase in the operational cost due to the increase in data downlink required. These factors will make on-board data compression more crucial to overall system performance, especially in large scale constellations.
The current deployed state-of-the-art of on-board compression in SAR space-borne payloads is Block Adaptive Quantization (BAQ) and variations such as Flexible BAQ, Entropy Constrained BAQ and Flexible Dynamic BAQ. Craft Prospect is working on an evolution of these techniques where machine learning will be used to identify signals based on dynamics and features of the received signal, with this edge processing allowing the tagging of raw data. These tags can then be used to better adjust the compression parameters to fit the local optimum in the acquired data.
We present the results of a survey of available raw SAR data which was used to inform a selection of applications and frequencies for further study. Following this, we present a comparison of a number of SAR compression algorithms downselected using trade-off metrics such as the bands/applications they can be applied to and various complexity measures. We then show an assessment of AI/ML feasibility and capabilities, with the improvements assessed on mission examples characterised by the SAR modes and architecture for specific SAR applications. Finally, future hardware feasibility and capability is assessed, targeting a Smallsat SAR mission, with a high level roadmap developed to progress the concept toward this goal.
Adaptive On-Board Signal Compression for SAR Using Machine Learning Methods
Utah State University, Logan, UT
Satellites with synthetic aperture radar (SAR) payloads are growing in popularity, with a number of new institutional missions and commercial constellations launched or in planning. As an active instrument operating in the microwave region of the electromagnetic spectrum, SAR provides a number of unique advantages over passive optical instruments, in that it can image in all weather conditions and at night. This allows dense time-series to be built up over areas of interest, that are useful in a variety of Earth observation applications. The polarisation and phase information that can be captured also allows for unique applications not possible in optical frequencies.
The data volume of SAR captures is growing due to developments in modern high-resolution multi-modal SAR. Instruments with higher spatial resolution, wider swaths, multiple beams, multiple frequencies and more polarization channels are being launched. Miniaturization and the deployment of SAR constellations is bringing improved revisit times. All of these developments drive an increase in the operational cost due to the increase in data downlink required. These factors will make on-board data compression more crucial to overall system performance, especially in large scale constellations.
The current deployed state-of-the-art of on-board compression in SAR space-borne payloads is Block Adaptive Quantization (BAQ) and variations such as Flexible BAQ, Entropy Constrained BAQ and Flexible Dynamic BAQ. Craft Prospect is working on an evolution of these techniques where machine learning will be used to identify signals based on dynamics and features of the received signal, with this edge processing allowing the tagging of raw data. These tags can then be used to better adjust the compression parameters to fit the local optimum in the acquired data.
We present the results of a survey of available raw SAR data which was used to inform a selection of applications and frequencies for further study. Following this, we present a comparison of a number of SAR compression algorithms downselected using trade-off metrics such as the bands/applications they can be applied to and various complexity measures. We then show an assessment of AI/ML feasibility and capabilities, with the improvements assessed on mission examples characterised by the SAR modes and architecture for specific SAR applications. Finally, future hardware feasibility and capability is assessed, targeting a Smallsat SAR mission, with a high level roadmap developed to progress the concept toward this goal.
