Search results
(1 - 1 of 1)
- Title
- EVALUATION OF TIME-FREQUENCY DISTRIBUTIONS FOR ULTRASONIC IMAGING APPLICATIONS
- Creator
- Lu, Juan
- Date
- 2013-05-01, 2013-05
- Description
-
This thesis presents the performance evaluation of generalized time-frequency distributions (GTFD) and model-based time-frequency (TF)...
Show moreThis thesis presents the performance evaluation of generalized time-frequency distributions (GTFD) and model-based time-frequency (TF) estimation of ultrasonic signals. Two new TF distributions which are related to generalized time-frequency distribution have been examined. These methods are singular value decomposition of Choi-Williams distribution (CWD-SVD), and 2D (time and frequency) Gaussian kernel applied to generalized time-frequency distribution. The application of Short-Time Fourier Transform (STFT) is studied for chirplets estimation. Then, the Wigner distribution (also called the Wigner-Ville distribution) of estimated Chirplets yield a precise TF representation. The performance of the STFT, the Morlet wavelet transform, the Wigner distribution (WD), the CWD and the CWD-SVD are compared. CWD-SVD is a very effective algorithm to keep the high clarity of the Wigner distribution and to suppress the undesirable cross-terms resulting from multi-component signals. The Gaussian echo model is used to obtain the analytical TF distribution. For CWD the proper range of exponential kernel parameter, , is attained. This range allows CWD to sustain a high concentrated auto-terms and significant suppression of cross-terms. For this range of the CWD-SVD extracts high clarity auto-terms and facilitate eliminating the residual cross-terms. To remove the cross-terms, singular value decomposition algorithm extracts basis functions corresponding to auto-terms. After discarding the basis functions and singular values of the cross-terms and noise, the basis functions and their singular values of auto-terms are used to reconstruct the TF distribution. The results of multi-component Gaussian echoes with significant time and frequency overlaps show that the CWD-SVD is able to eliminate residual cross-terms for xi which the CWD failed to eliminate. The numerical analysis of multi-component Gaussian echoes indicates that CWD-SVD has the ability to resist noise resulting in accurate estimates of center frequencies and arrival times. The generalized time-frequency distribution with 2D Gaussian kernel is able to separate two extremely close Gaussian echoes in the time-frequency domain. In this study, typical values of the 2D Gaussian kernel parameters for efficient cross-terms elimination are provided. The relationship between the kernel's parameters and Gaussian echoes' parameters is deduced. A practical method for TF analysis is to decompose the signal into sparse chirplets. Decomposition requires chirplet parameter estimation. In this study, the parameters of a signal which is composed of two overlapping chirplets are estimated using STFT. By this method the estimation results are found to be accurate confirming that the STFT is an effective method for decomposing and estimating chirplets in a multi-component signal.
M.S. in Electrical Engineering, May 2013
Show less