Depth resolution enhancement of terahertz deconvolution by autoregressive spectral extrapolation

This Letter presents a method for enhancing the depth resolution of terahertz deconvolution based on autoregressive (AR) spectral extrapolation. The terahertz frequency components with a high signal-to-noise ratio (SNR) are modeled with an AR process, and the missing frequency components in the regions with low SNRs are extrapolated based on the AR model. In this way, the entire terahertz frequency spectrum of the impulse response function, corresponding to the material structure, is recovered. This method, which is verified numerically and experimentally, is able to provide a "quasi-ideal" impulse response function and, therefore, greatly enhances the depth resolution for characterizing optically thin layers in the terahertz regime.