Summary: | A system comprising of a personal microcomputer and a Digital Signal Processor board has been assembled and programmed for real-time spectral analysis of Doppler ultrasound signals. Three spectrum analysis techniques were implemented to run in real-time on the system: The fast Fourier transform (FFT), the autoregressive (AR) model, and the moving average (MA) model. The FFT and the AR techniques were investigated in some depth. The advantages of using such a system are that it is entirely programmable, cheap, reliable, and that the processed information can be stored on diskettes. The inputs to the system are the forward and reverse components of the Doppler ultrasonic signal, and the outputs are the sonogram, the frequency envelope, and the intensity weighted mean frequency curve, which are presented on the screen and can be saved to diskette. Five frequency ranges can be selected by the operator, from 1.28 kHz to 20.48 kHz, corresponding to sampling frequencies from 2.56 kHz to 40.96 kHz. Most commercial systems for real-time spectral analysis of Doppler ultrasonic signals implement the modified FFT-periodogram technique for power spectral density estimation (PSDE), which is computationally very efficient but has some shortcomings, especially for the analysis of relatively short records. With the AR model approach the spectra can be estimated from short segments, no antileakage window is necessary and the spectral resolution is better than for the FFT. A study of some methods for order selection used with the autoregressive model for the spectrum analysis of Doppler signals is reported and the use of a fixed order of around 12 is suggested for the AR model. The implementation of the AR PSDE approach in real-time, in a reasonably priced system, is a step towards the practical use of the so called 'modern techniques' for spectral analysis of Doppler ultrasonic signals, but further work has to be done on the validation of the technique in clinical usage.
|