Comparative Study of Convergence of Optimization Algorithms with Chirp Signal Input

In structural analysis of the materials, for detecting the material defects, reflected echoes can be used. These backscattered echoes can also provide several information regarding material properties. Echo is modeled in this paper as a white-gaussian noise added chirp signal. The signal phase, centre frequency and sampling frequency are the parameters to be evaluated from the noisy chirp signal. Assessing optimized parameters from reflected reverberations is quite tough. Various algorithms such as, Trust Region Reflective, Levenberg - Marquardt, Quasi Newton and Sequential quadratic programming, Active Set are used to optimize necessary noise altered signal parameters. The research focuses on the study of goodness of fit of various estimated signals by using the aforementioned methods for a range of SNR values from 0-20 dB. Additionally, the noisy reverberations are denoised using wavelet before estimating the parameters. Sequential quadratic programming algorithm gave the optimized parameters for both noisy and de-noised echoes.