Optimization of Fast Factorized Backbrojection execution performance

Abstract

Real-time signal processing often requires high computational performance from the signal processing system. In order to increase performance computer systems have moved from the traditional one-core CPU to multi-core systems. This requires, however, parallel software to use all the available performance. It is therefore not only important to have efficient algorithms but also efficient parallel implementations of them. The signal processing for a low-frequency synthetic aperture radar system, used to create high-resolution radar maps of the ground, is studied in this master's thesis. The datasets used to create the maps are often very large and therefore the computational burden is high. The efficient Fast factorized backprojection algorithm is used to create the images but still the images cannot be produced in real-time on a single core system. This thesis describes how the Fast factorized backprojection is optimized and parallelized. OpenMP and vector instructions are used to reach real-time performance on a multi-core platform, for small and medium sized images.