Computation of linear prediction autocorrelation of each segment sampling signals are performed to suppress the influence of non-integer period sampling. Time domain averaging of autocorrelation signals is performed to improve the SNR. The reference signals are generated according to average autocorrelation signal. The error function is constructed through the least square method

and frequency estimation value is obtained by minimizing error function. Calculation verification indicate that the proposed method is not affected by the noninteger sampling and effectively improves anti-noise performance and the precision of frequency estimation.