Estimating time delays for signal alignment is important for many applications. This paper extends a successful frequency domain cost function minimization algorithm capable of estimating time delays to within a fraction of sampling periods. Since a narrow basin of attraction around the global minimum of is typical for this function, this method often diverges when initial time delay estimates are not sufficiently close to the desired optimal time delays. We propose a second order successive minimization method with reduced sensitivity to initial guesses. Both an analytic expression for the cost function Hessian matrix and a condition guaranteeing positive-definiteness are presented. This condition facilitates the construction of sequentially modified cost functions whose nested minimization increases the basin of attraction around the global minimum. This successive minimization technique is more robust and yields higher accuracy when compared to the original method and the well-known method of Cross Correlator (CC).
Citation: Signal Processing
Pub Type: Journals
Signal Processing, Optimization, Minimization