Spaceborne Synthetic Aperture Radar Data Focusing on Multicore-Based Architectures

Abstract

This paper describes a general-purpose parallel scheme for efficiently focusing synthetic aperture radar (SAR) data on multicore-based shared-memory architectures. The rationale of the proposed tiling-based parallel focusing model is first discussed, and then, its implementation structure is illustrated. The adopted parallel solution, which is based on a canonical processing pattern, exploits a segmented-block-based approach and works successfully on data acquired by different spaceborne SAR platforms. Insofar as a significant portion of the focusing algorithm is amenable to tiling, our approach decomposes the problem into simpler subproblems of the same type, also providing a suitable mechanism to explicitly control the granularity of computation through the proper specification of the tiling at the different stages of the algorithm itself. Relevant implementation makes use of multithreading and high-performance libraries. Achievable performances are then experimentally investigated by quantifying the benefit of the parallelism incorporated into the prototype solution, thus demonstrating the validity of our approach. Accordingly, canonical performance metrics have been evaluated, and the pertinent scalability has been examined on different multicore architectures. Furthermore, in order to emphasize the practical ability of the proposed parallel model implementation to efficiently deal with data of different SAR sensors, a performance analysis has been carried out in different realistic scenarios including data acquired by the Envisat/ASAR, RADARSAT-1, and COSMO-SkyMed platforms.