Time delay estimation of GPR backscattered signals using matrix pencil based support vector regression

The need for estimating time delays arises in numerous applications where a multipath medium has to be identified from wavelets transmitted through the channel. This paper presents a method for estimating the time delays of ground penetrating radar backscattered signals. The proposed method consists of two steps: feature extraction through the generalized eigenvalue decomposition of a Hankel matrix pencil followed by training and prediction using support vector regression. The proposed feature extraction technique reduces the dimension of feature vectors and thus allows for more efficient testing and training. Simulation results show that the method succeeds in accurately estimating the time delays of the model.