The principle and algorithm of kernel independent component analysis（kernel-ICA） is introduced and it is applied to separate the mixed signal at the output of the current sensor.The phase difference of the sensor is ascertained by computing the one between the single-frequency testing signal and its original input.Moreover

the maximum likelihood estimation is applied to evaluate the separated results kernel-ICA.Experiment results show that the phase error generated by kernel-ICA is less than 0.002rad when the SNR of the input signal is 18.73dB.Therefore this algorithm is able to meet the practical demand of the engineering.