Abstract:In the traditional signal sampling based on Nyquist sampling theorem, the sampling rate must be at least two times of signal bandwidth in order to guarantee the non distortion in recovering the original signal; but for terahertz pulse signal, its pulse width is extremely narrow, up to picosecond level, therefore, its bandwidth is up to hundreds of GHz, that will bring great pressure to the hardware if we follow the traditional Nyquist sampling frequency, and sometimes cannot be achieved. A new sampling theory-Finite Rate of Innovation(FRI) theory is introduced. By choosing suitable sampling kernel function, sampling can be performed at the rate higher than the signal rate of innovation. The free parameters can be estimated through subspace algorithm, and the original signal can be reconstructed. The sampling accuracy of FRI sampling theory and the effectiveness of subspace algorithm for signal reconstruction are verified through the time delay estimation error of the THz pulse signal.