Next-generation SAR systems will be capable of performing high-resolution, wide-swath acquisitions at frequent revisit times. The overcoming of conventional SAR limitations will also lead to the generation of very large volumes of onboard data which need to be stored and managed by the system and downlinked to the ground. This poses severe constraints in terms of onboard memory requirements and downlink capacity and, in this challenging scenario, the onboard quantization of SAR raw data represents a crucial aspect, acting as a trade-off between the achievable product quality and the resulting onboard volume of data. State-of-the-art quantization schemes allow for enhanced data rate allocation, however, the optimization is directly performed on raw data, without targeting a desired performance on the final higher-level SAR/InSAR product. In this paper, we investigate the use of artificial intelligence (AI), and in particular of deep learning (DL), for developing a flexible onboard SAR raw data quantization method, with the aim of deriving an optimized and fully adaptive data rate allocation given a set of desired performance metrics and requirements in the resulting focused SAR and InSAR products, without relying on a priori information on the acquired scene. Different performance parameters are considered, such as the signal-to-quantization noise ratio (SQNR), the phase errors, the InSAR coherence loss as well as the resulting noise equivalent sigma zero (NESZ), extending the capabilities of the architecture to provide multiple bitrate estimations for a single input scene at the same time, depending on the desired application case. We use experimental TanDEM-X bistatic SAR data, both for the training of the DL model as well as for the validation and demonstration of the suitability of the proposed method. In view of a potential onboard implementation, a possible hardware architecture for the proposed compression scheme is investigated as well.

Present and next-generation synthetic aperture radar (SAR) missions require an increasing volume of onboard data, due to the employment of large bandwidths, multiple channels and polarizations, and large swath widths acquired by bi- and multi-static sensor configurations. This leads to stringent requirements in terms of onboard memory and downlink capacity, hence making the proper quantization of the SAR raw data represents an task of utmost importance, as it affects the amount of data but also the quality of the SAR and InSAR products. This paper presents novel methods for efficient SAR raw data compression, which make use of artificial intelligence for the joint optimization of bitrate allocation and the resulting performance and exploit the potential of transform and predictive coding schemes for data volume reduction in the context of multi-azimuth channel (MAC) SAR. Simulations and analyses on real data are presented, showing the suitability of the proposed methods.

This paper presents the MLAB project, a research and development activity funded by ESA General Support Technology Programme under the lead of Airbus Defence and Space GmbH, with the goal of developing a machine learning application benchmark for space applications. First, the need for a benchmark dedicated to machine learning applications in spacecraft is explained, and examples of applications are described including their design challenges. Then the benchmark design is presented, including the rules of the metrics, guidelines and scenarios for references. These scenarios include a description of the reference workloads that have been selected during the activity as representative for spacecraft applications. Lastly, the submission concept is introduced.

Artificial intelligence has found its way into space, and similar to the situation on ground demands powerful hardware to unfold its full potential. With the heterogeneous compute platform that is offered by the space-grade variant of the Versal, AMD Xilinx presents a system that is particularly targeted at accelerating AI inference in space. This paper investigates the design flow and the achievable performance of this novel device. We present benchmark results in terms of concrete figures and measurements, i.e., throughput, latency, and power consumption, achieved by a predesigned hardware accelerator realized on the system, and compare them to a previous generation platform.

Higher autonomy in satellite operation is seen as the key game changer for the space systems market in the next decade, with a considerable amount of agencies and startups focusing on bringing machine learning to space.

Compute in space, e.g., in miniaturized satellites, requires dealing with special physical and boundary constraints, including the limited energy budget. These constraints impose strict operational conditions on the on-board data processing system and its capability in dealing with sophisticated workloads suchlike Machine Learning (ML). In the meantime, the breakthroughs in ML based on Deep Neural Networks (DNNs) in the last decade promise innovative solutions to expand the functional capabilities of on-board data processing and to drive the space industry forward. Therefore, due to the aforementioned special requirements, performance- and power-efficient, and novel solutions and architectures for deploying ML via, e.g., FPGA-enabled SoC, particularly Commercial-Off-The-Shelf (COTS) solutions, are gaining significant interest in the space industry. Therefore it is essential to conduct extensive benchmarking and feasibility and efficiency analyses in different aspects: such analyses would require the investigation of options for programming and deployment as well as the investigation of various real-world models and datasets. To this end, a research and development activity is funded by the European Space Agency (ESA) General Support Technology Programme and is led by Airbus Defence and Space GmbH with the goal of developing an ML Application Benchmark (MLAB) that covers benchmarking aspects mentioned above.