In this paper, we employ a blind source separation technique to estimate the thickness of thin pavement layers. The linearity of the convolution operation along with a simple data reorganization permit to formulate a multidimensional signal model in accordance with the model adopted in blind techniques. We use this model to retrieve the reflectivity series and eventually the corresponding time delays and reflection amplitudes without any a priori information. Thus, this technique enables parameter estimation without incorporating a calibration stage which might be difficult in near-field conditions. Numerical simulations indicate the feasibility of the approach.