High performance parallel-DSP computing in model-based spectral estimation

摘要

Doppler blood flow spectral estimation is a technique for non-invasive caridovascular disease detection. Blood flow velocity and disturbance may be determined by measuring the spectral mean frequency and bandwidth, respectively. The work presented here, evaluates a high performance parallel-Doppler Signal Processing architecture (SHARC) for the computation of a parametric model-based spectral estimation method known as the modified covariance algorithm. The model-based method incorporates improvement in frequency resolution when compared with Fast Fourier Transform (FFT)-based methods. However, the computational complexity and the need for real-time response of the algorithm, makes necessary the use of high performance processing in order to fulfil such demands. Sequential and parallel implementations of the algorithm are introduced. A performance analysis of the implementations is also presented, demonstrating the effectiveness of the algorithm and the feasibility for real-time response of the system. The results open a greater scope for utilising this architecture in implementing new and more complex methods. The results are applied to the development of a real-time spectrum analyser for pulsed Doppler blood flow instrumentation.