On SAR Imaging Core Algorithm's Configurable Software Enhancement Technology Research

Hardware-implemented fault tolerance technology, was costly for spaceborne applications in terms of energy, area and complex circuits overhead. So software-implemented technology was developed, for its good flexibility, configurability and higher fault coverage. This kind of technology met development needs of onboard software, and was used for reliability enhance-ment without modifying hardware.With continuous development of software-implemented fault tolerance technology, there were massively good performence algorithms which formed plenty of ingenious and strong error detection capability algorithms both in control flow and data flow enhancement aspects. However, they also had the compromise selective problem between computation cost and enhancement strength. Therefore, ABFT(Algorithm Based Fault Tolerance) which aimed at characteristics of specified algorithms, and CBFT(Configurable Based Fault Tolerance) configur-able enhancement technology, had became especially popular.This paper draw on experiences of software-implementedly classical algorithms, and aimed at characteristics of RDA algorithm, proposes a multi-level enhancement technology which combines computing features, simple yet meets the needs of computational performance. In this paper, we mainly focus on several research areas as following:1. Aimed at RDA's imaging core component as FFT, operations were proposed base on weighed data values&checksum fault tolerance algorithm——WCFT. Fault detection capability, performance overhead, design characteristics and advantages of the algorithm are analyzed.2. This paper presents ABFT(Algorithm Based Fault Tolerance) of RDA program, which integrates checksum methods with WCFT algorithm, be an effective fault detection&recovery mechanisms. For calculation accuracy and performance overhead weighing, we make double precision arithmetic and trigonometric functions' tabulation. For bit-reverse function of FFT as easily error part, we use bit-reverse tabulation method. For others, we use incremental computation fault detection algorithm, redundantly check method to global varieties, and so on.3. For parallel operation under conditions of multi-processors, we analyze problems of redundant multi-threaded execution and memory access contention. Incremental twiddle factor error problems, and related technologies such as improved methods for twiddle factor error detection are also analyzed.4. Inputing multiple rows/columns of data simultaneously in one time FFT parallelled computation, also called SIMD(Single Instruction Multiple Data), its performance had been verified to improve nearly double to the original program. And basing on this conclusion, we propose recovery mechanism which corresponding to fault detection features. Saving four matrix transpose operations with this method, memory overhead of data migration and check are reduced greatly.5. According to different applications of demand levels, this paper proposes CBFT for SAR's program, which has two levels of enhancement strength, and makes use of redundant algorithm for fault detection and recovery implementation.