Comparison of a sensor fusion algorithm implementation on a C674X DSP and a Cortex A8 core

Inertial sensors used for real time motion capturing of human movements usually consist of triads of gyroscopes, accelerometers, and magnetometers. Using these sensors for motion capturing therefore requires computational intensive sensor fusion algorithms, as drift free orientation in 3D space is desired rather than sensor RAW data. Kalman filters, comprising numerous floating point matrix operations, are a common algorithmic approach for this sensor fusion. Filter algorithms in literature heavily vary in achieved accuracy and computational effort. Performing sensor fusion on the TI Integra processor allows filter partitioning, either to the RISC core or to the DSP core. Therefore, either the Cortex A8 processor with its SIMD unit NEON or the C6A816x DSP could be used for efficient calculation. The paper presents implementation results for utilizing the TI Integra in a challenging biomedical application for movement sonification.