Gaussian Belief Propagation on a Field-Programmable Gate Array for Solving Linear Equation Systems

Solving Linear Equation System (LESs) is a common problem in numerous fields of science. Even though the well studied and powerful solvers are available nowadays, solving LES still bottleneck many numerical applications concerning computation time. This issue especially pertains to mobile robotics constrained by real-time requirements, where on-top power consumption weight play an important role. paper provides general framework approximately solve large LESs Gaussian Belief Propagation (GaBP), which extremely suitable for parallelization implementation hardware on Field-Programmable Gate Array (FPGA). We derive simple update rules Message Passing Algorithm GaBP show how implement approach efficiently Programmable Chip (SoPC). In particular, multiple dedicated co-processors take care recurring computations GaBP. Exploiting Direct Memory Access (DMA) controllers scatter-gather mode arithmetic logic slices calculations accelerate algorithm. Presented evaluations demonstrate that does not only provide accurate approximative solution LES. It also outperforms traditional with respect time certain LESs.