Multiparadigm Space Processing for Hyperspectral Imaging

Projected demands for future space missions, where on-board sensor processing and autonomous control rapidly expand computational requirements, are outpacing technologies and trends in conventional embedded microprocessors. To achieve higher levels of performance as well as relative performance versus power consumption, new processing technologies are of increasing interest for space systems. Technologies such as reconfigurable computing based upon FPGAs and vector processing based upon SIMD processor extensions, often in tandem with conventional software processors in the form of multiparadigm computing, offer a compelling solution. This paper will explore design strategies and mappings of a hyperspectral imaging (HSI) classification algorithm for a mix of processing paradigms on an advanced space computing system, featuring MPI-based parallel processing with multiple PowerPC microprocessors each coupled with kernel acceleration via FPGA and/or AltiVec resources. Design of key components of HSI including autocorrelation matrix calculation, weight computation, and target detection will be discussed, and hardware/software performance tradeoffs evaluated. Additionally, several parallel-partitioning strategies will be considered for extending single-node performance to a clustered architecture. Performance factors in terms of execution time and parallel efficiency will be examined on an experimental testbed. Power consumption will be investigated, and tradeoffs between performance and power consumption analyzed. This work is part of the Dependable Multiprocessor (DM) project at Honeywell and the University of Florida, one of the four experiments in the Space Technology 8 (ST-8) mission of NASA´s New Millennium Program.