ALTERNATIVE PARALLEL IMPLEMENTATIONS OF AN AR-MODIFIED COVARIANCE SPECTRAL ESTIMATOR FOR DIAGNOSTIC ULTRASONIC BLOOD-FLOW STUDIES

The work described here is part of a research program aiming to increa se the sensitivity to disease detection using Doppler ultrasound by re ducing the effects of the measurement procedure on the estimation of b lood velocity and detection of flow disturbance. The paper presents a summary of autoregressive spectral estimation, focusing the attention on a specific spectral estimator - the modified covariance method. Thi s method has been realized in parallel to achieve a fast computer proc essing. The new parallel version of this algorithm has been developed and implemented on a multiprocessing transputer-based system. Two diff erent approaches to the problem of parallel partitioning the algorithm into a number of tasks were considered - a fine and a medium grain ta sk scheme. The medium grain scheme is mapped onto a transputer-based s ystem, by means of a processor farm computational structure. Two appro aches to this farm model were adopted: a linear and a tree topology. F or different model parameters, performance measurements were obtained revealing that the tree topology offers a higher performance.