Real-time Parallel Synthetic Aperture Radar Data Processing

Synthetic Aperture Radar is an active microwave imaging radar with high resolution obtained through pulse compression and synthesize technique. Because of the characteristics of all-weather, high resolution and wide coverage, SAR is used widely in the field of Military applications, Earth Remote Sensing, Environmental Protection and Disaster Examination etc.Along with the technical development of SAR, higher resolution and shorter processing time especially real time processing are needed. The real-time SAR data processing is characterized the theme as processing a continuous stream of data collected from radar sensors. The rate at which data samples flow from the sensor(s) to the computational platform is typically very high– often on the order of tens or hundreds of millions of samples per second and even higher. Furthermore, the number of calculations to be performed on each sample is typically at least 100 FLOPs (floating point operations), which amounts to an overall computational throughput requirement ranging from at least one to ten billion FLOPs (and often much higher).Approaches capable of providing a computational platform that could achieve these types of computational throughput rates typically involved multi-processor parallel computing such as a"pipeline of interconnected processors"and a"data exchange interconnected processors"style of architecture. Such approaches could be valid and effective architectures in some cases. However, situations often arose in which the throughput requirements dictated that 100 or more processors were required. In many situations, the associated level of space and power requirement for the computational platform alone posed a severe problem, because of the strict space and power budgets available on UAVs (unmanned aerial vehicles) and satellites where these systems are deployed.In this thesis, we showed how a 32-node IBM PC Cluster system could be optimally configured to achieve a near real time SAR data processing. We showed that by carefully choose the total number of processing nodes, the maximum throughput could be achieved. Also in the situation with fixed number of processors, the ratio of the arranged number of processors used for range processing to the number of processors for azimuth processing is an important factor to the performance of SAR processing, such that a ratio of 1:2 can give us the best performance which implies that an optimal configuration can be derived based on application of mathematical programming techniques. Moreover, in this thesis, we showed by comparing the results (speedup, efficiency and throughputs) obtained in these two styles of architectures (pipelined and data exchanging),"pipeline of interconnected processors"is super than"data exchange of interconnected processors".The achievements in this thesis including three aspects: firstly, a C-coded MPI based parallel SAR data processing software was developed and implemented a 32-node IBM PC Cluster system. Then, two processing algorithms were implemented in this software, i.e., pipeline based and data exchanged based processing algorithms. Finally, an optimal configuration (maximum throughput) was derived to achieve near real-time SAR data processing based on application of mathematical programming techniques.