Hardware Implementation Design for An Efficient On-board Imaging Processing System in Spaceborne SAR

‍ ‍On-board Synthetic Aperture Radar （SAR） imaging technology is a key tool for time-sensitive remote sensing applications such as disaster monitoring and military reconnaissance. The typical construction of an on-board SAR imaging system using a Field-Programmable Gate Array （FPGA） coupled with a Digital Signal Processor （DSP） as the processing core is well-regarded for its heterogeneous computing power， energy efficiency， and flexibility. However， current FPGA+DSP systems are still under-researched in terms of algorithm support， large-grain processing， on-chip parallel acceleration， and complex matrix transposition， with significant room for performance enhancement. This paper analyses the Nonlinear Chirp Scaling （NCS） algorithm， suitable for large squint， high-resolution imaging， and partitions the algorithm into main and auxiliary paths based on computational complexity and type. A heterogeneous mapping scheme for the NCS algorithm within the FPGA+DSP system is subsequently proposed. To address the transformation of data storage formats post multi-channel Fast Fourier Transform （FFT） processing， which complicates pipelined processing， this work introduces a multi-channel FFT collaborative method based on time-frequency extraction switching to ensure efficient parallel FFT operations. Additionally， to handle the complexity of transpose requirements across various granularities and parallelisms， a universal cross transpose solution using X-Direct Memory Access （XDMA） and on-chip segmented transposition is developed. Employing two VX690T FPGAs and two FT 6678 DSPs as core processors， this paper presents the development of an on-board SAR imaging card that implements the proposed system design. Moreover， a validation environment using simulated sources plus ground testing is established， processing simulation array data for strip/scan/spotlight/sliding spotlight/TOPS modes and actual data for strip/sliding spotlight modes. The array data demonstrates a two-dimensional peak sidelobe ratio of about -13.2 dB and an integrated sidelobe ratio of around -10.1 dB， indicative of good imaging quality. For instance， in strip mode with an image size of 32K×16K， the average imaging time using the NCS algorithm is 7.81 seconds， markedly accelerating imaging speed compared to existing approaches.