Parallel implementation of image reconstruction for the CARABAS long-wave penetrating radar

Abstract

CARABAS (Coherent All RAdio BAnd Sensing) is a new type of radar that has the unique property of being able to penetrate through vegetation, and to some extent into upper levels of soil depending on water content. This can be done by using long radar waves in the range 3–15 meters, and new algorithms for image reconstruction from information in reflected radar waves. These algorithms are related to methods used for computer tomography, and are very computationally expensive. Two classes of algorithms for image reconstruction are direct Fourier methods and filtered backprojection. Even though filtered backprojection is more computationally demanding, we chose that method since it is easier to parallelize, it has better real-time properties, and it is easier to compensate for disturbances and achieve good image quality.

In this paper we report results from the first parallel implementation of the CARABAS algorithms. The benchmarking was done on a Parsytec PowerGC MIMD computer with 128 PowerPC 601 processors. We come close to achieving the real-time requirement for significant parts of the computation.

This work was supported by Ericsson Microwave AB, and FOA, the Swedish National Defense Research Establishment.