Abstract: Aiming at the problem that the detection method of rotational magnetic resonance was easy disturbed by the peak noise, we proposed  a collaborative filtering method to eliminate the peak noise in nuclear magnetic resonance (NMR) data. Firstly, the method of 3σ was used to determine whether the
re was peak noise in the measured data under a pulse moment. The measured data were divided into two groups: peak noise and non-peak noise. Secondly, the discrete cosine transform and the Hadamard transform were performed respectively to obtain two groups of transform domain coefficients. Wiener filtering coefficient was calculated by using the transform coefficient of non-peak noise data, and the coefficient was used to filter the data containing peak noise. Finally, the filtered coefficient containing the peak noise data was inversely converted by Hadamard and discrete cosine to eliminate the peak noise. The simulation results show that the method has high accuracy in eliminating the peak noise in the ground NMR data and improves the extraction precision of signal characteristic parameters.

Key words: collaborative filtering, nuclear magnetic resonance, peak noise