Time delay estimation for analog signals is more susceptible to noise than that for demodulated signals, which increases false positives and conceals the correct peak value. The singular spectral decomposition method combined with improved generalized cross-correlation function is proposed to reduce the influence of Gaussian noise on the time delay estimation results. Simulink simulation results show that the proposed method can obtain fault-type and distance information in the same order of magnitude accuracy for lower transmission frequency requirements than demodulation signal time delay estimation. Multiple experimental verifications in the -5 dB Gaussian noise environment can yield that, compared to the results of the secondary correlation method, the absolute value of the main peak side-lobe ratio increases by more than 0.6756 dB, and the ratio of the peak value of misjudgment to the peak value of the fault point decreases by more than 0.2710. This also improves in different degrees under other conditions.