The Study Of Cluster-based Synthetic Aperture Radiometer Signal-level Simulation

Synthetic aperture microwave radiometer is a passive imaging tool, in which manysmall antennas are used to form an antenna array. With the array reception andinterferometric techniques, it can break through the restrictions of aperture size andimprove the resolution performance. Moreover, its’ abilities to work under all weatherconditions and at any desired time make it has a wide range of applications in satelliteremote sensing, target detection, environmental monitoring, et al. But the actual hardwaresystem is highly complex and difficult to conduct accurate analysis, while using computersimulation for error analysis, model and algorithm validation, and performance evaluationcan significantly reduce the developing risk and demonstration period, which for theconduct of synthetic aperture imaging of microwave radiation has a significant meaning.Firstly, we represent the state-of-art of the simulation for the synthetic aperturemicrowave radiometer and the parallel computation technology. And then for morerealistic simulation of the entire microwave radiometer detection process, this articleestablishes the whole chain signal level model from the synthetic aperture radiometerimaging method and principle, and divides the simulation system into four sub-modules,and then analyzes the computational complexity and the data scale and its evolutionprograms throughout every module of the simulation system, from which it can beconcluded that signal level simulation system faces the problems of large computation anddata. Since simulation of microwave radiation detection process itself has a very highdegree of parallelism, this study researches on the paralleling implementation method ofevery sub-module in details, and presents a paralleling and distributed solution, coveringmodule-level parallel task size and scheduling strategy. Finally, on small Linux cluster andthe Shanghai Supercomputer Center’s Shuguang5000A magic platform, simulations ofdifferent scene and scale are conducted and results all prove that the parallel solutionworks correctly, and solves the computational complexity and data issues effectively,which has good acceleration and application prospects compared to the serial algorithm.