Despiking of magnetic resonance signals in time and wavelet domains
Stephan Costabel and Mike Müller-Petke
Journal name: Near Surface Geophysics
Issue: Vol 12, No 2, April 2014 pp. 185 - 197
Special topic: Magnetic Resonance in the Subsurface
Info: Article, PDF ( 2.84Mb )
Price: € 30
In this paper three different despiking methods for surface-NMR data are investigated and compared. Two of these are applied in the time domain: a threshold is determined that identifies and marks a spiky event. Afterward, the marked time sequence is substituted with zeros or with the mean value of the signal amplitude of the measurement repetitions for the same passage on the time axis. The third despiking approach takes advantage of the wavelet-like nature of spiky events. It isolates and eliminates spiky signals in the wavelet domain, i.e., after transforming a single record with the help of the discrete wavelet transform. The latter is able to reconstruct the original signal content in the (spike-caused) distorted time sequence to some extent. If the spiky noise in surface-NMR measurements consists mainly of single spiky events, the three despiking methods show very similar results and are able to remove spiky noise from data very effectively, as we can show with two real data examples. However, a synthetic study shows that, if a series of spikes within a relatively short period of time occurs, the wavelet-based despiking approach shows significant shortcomings. Because the NMR signal content cannot be restored completely in a single record, the fitting of the signal after stacking leads to underestimation of the initial amplitude up to approximately 10%. Nevertheless, we can show that, in principle, the processing of surface-NMR data in the wavelet domain works and can lead to the same results as straight-forward applications. Moreover, waveletbased strategies have some interesting properties and thus have some potential for further development regarding surface-NMR processing, which is discussed in detail.