Efficient Implementations of Ultrasound Color Doppler Algorithms on VLIW Architectures

摘要

DSP chips are gaining importance in ultrasound applications as the need for portability and low power grows. One of the more computationally demanding applications for ultrasound involves estimating blood flow characteristics using Doppler techniques. This ultrasound mode, called color Doppler ultrasound, is used to diagnose many conditions like blood clots, valve defects and blocked arteries. This work looks at mapping some typical color Doppler algorithms onto Texas Instruments' (TI's) high performance C64x+™ core. The algorithms include RF demodulation, wall filtering and flow power, velocity and turbulence estimation. This paper starts with a general technique for analyzing algorithm complexity in terms of CPU instruction cycles on VLIW architectures like the C64x+™. It then applies this technique to Doppler processing algorithms, explains their mapping to the C64x+™ architecture and derives lower bounds for the computational complexity for these algorithm kernels. For each of these algorithms, these estimates are finally compared to actual implementations, and various implementation tradeoffs will be illustrated. Based on these implementations, it will be shown that these algorithms can run on TI's C64x+™ based DSPs using a fraction of the available processing power.