Heterogeneous architecture for parallel real-time spectral estimation in Doppler blood flow instrumentation

The work described here is concerned with the implementation of a heterogeneous architecture, interfacing a transputer to a TMS320C30 digital signal processor. As an appropriate matching of hardware granularity to task size is essential to fully exploit the parallel computational power available, fusing the finer granularity of a DSP with the transputer´s coarser granularity and management of parallelism, can lead to a versatile and more effective system. The modified covariance algorithm has been implemented on the resulting heterogeneous architecture, in order to take advantage of the resolution benefits of this method to be applied to the diagnosis of cardiovascular diseases. A brief description of the algorithm and strategies for partitioning and allocating the resulting algorithm tasks on the heterogeneous architecture are given. Performance analysis and results are presented.