Abstract: To solve the problem of signal aliasing and structural noise in detecting small defects in the near-surface dead zone of austenitic stainless steel by time of flight diffraction (TOFD), a frequency domain sparse blind deconvolution method was proposed. The artificial defects with a depth of 3. 0 mm and a height of 3. 0 mm were processed in the dead zone of TOFD near surface detection of austenitic stainless steel test block with a thickness of 35 mm. The experimental signals of TOFD were sparsely decomposed based on Matching pursuit (MP) algorithm to decouple the noise parts. The objective function of inversion problem was established by homomorphic transformation combined with L₁ and L₂ norm constraints, and solved by interior point method. The results showed that without reference signal, the method can separate the direct wave and the diffraction wave on the upper and lower edge of the defect by TOFD. The calculated defect depth was 3. 04 mm, the height was 2. 65 mm, and the maximum error was less than 11.7%.