On-Board Data Compression for Future SAR Systems: An Overview of the Research Activities at the DLR Microwaves and Radar Institute

Abstract

Synthetic aperture radar (SAR) represents nowadays a well-established technique for a broad variety of remote sensing applications, being able to acquire high-resolution images of the Earth's surface, independently of daylight and weather conditions. In the last decades, innovative spaceborne radar techniques have been proposed to overcome the limitations which typically constrain the capabilities of conventional SAR for the imaging of wide swaths and, at the same time, of fine spatial resolutions. In addition to that, present and future spaceborne SAR missions are characterized by the employment of multi-static satellite architectures, large bandwidths, multiple polarizations, and fine temporal sampling. This inevitably leads to the acquisition of an increasing volume of on-board data, which poses hard requirements in terms of on-board memory and downlink capacity of the SAR system. This paper presents an overview of the efficient raw data quantization and data volume reduction methods which have been developed at the Microwaves and Radar Institute of DLR in the last years. In particular, we focus our attention on the exploitation of the use of artificial intelligence (AI), and in particular of deep learning (DL), with the goal of deriving an optimized and fully adaptive bitrate allocation to be used for raw data quantization, depending on a set of desired performance metric and requirements in the resulting focused SAR/InSAR products, without relying on a priori information on the acquired scene. The derived bitrate allocation maps (BRMs) is employed for adapting a state-of-the-art block-adaptive quantizer (BAQ) to the local characteristics of the input raw data and to the desired performance, and the results obtained on experimental TanDEM-X interferometric data demonstrate the potentials of the proposed method as a helpful tool for performance budget definition and data rate optimization of present and future SAR missions. In the second part of the paper, we investigate and discuss potentials for data volume reduction in multi-channel SAR. These systems allow for high-resolution imaging of a wide swath but, on the other hand, require for their operation the acquisition and downlink of a huge amount of data: together with the intrinsic requirement related to resolution and swath width, this is due to the fact that the effective pulse repetition frequency (PRF) generated by the multiple channels is typically higher than the processed Doppler bandwidth, which introduces a certain oversampling in the azimuth raw data. Therefore, convenient data volume reduction strategies can be proposed, based on Doppler-based transform coding (TC) or linear predictive coding (LPC), which aim at exploiting the existing correlation between subsequent azimuth samples. We consider realistic multi-channel SAR system architectures, and simulate multi-channel raw data using synthetic as well as real backscatter data from TanDEM-X. We analyze the statistical properties (such as autocorrelation and Doppler power spectrum) exhibited by the multi-channel raw signal and discuss the impact of relevant system parameters, highlighting potentials and limitations of the proposed approaches in terms of achievable data volume reduction. DLR is also member of the Consultative Committee for Space Data Systems (CCSDS) and the authors currently support the Data Compression Working Group with the main objective of defining and standardizing a data compression method for SAR systems. For this purpose, we carried out dedicated simulations and performance assessment on test (simulated as well as real SAR) data, which are also summarized in the last section of the paper. Finally, conclusions and outlook are provided.