Separation of the Superparamagnetic Response in Central-Loop Time-Domain Electromagnetic Surveys Based on the NG-MMI-ICA Method
Yanju Ji, Ming Yang, Yaoming Jia, Tiantian Wei, Huaishi Liu, Xuejiao Zhao · IEEE Transactions on Geoscience and Remote Sensing · 2025
The superparamagnetic (SPM) layer in time-domain electromagnetic (TDEM) surveys can induce characteristic –1 power-law decay in late-stage electromagnetic (EM) responses, leading to resistivity interpretation errors. To address this issue, this study used the natural-gradient minimizing-mutual-information independent component analysis (NG-MMI-ICA) algorithm to separate the SPM responses from TDEM data. To avoid the underdetermined blind source separation problem, a synthetic magnetization response was constructed based on the Levenberg–Marquardt (L-M) algorithm. The two input signals were the constructed signal and the measured total response. Then, the non-Gaussianity of the input signals was determined by mutual information and a minimization criterion, and the natural gradient algorithm was applied to maximize this non-Gaussianity, thereby obtaining the separation matrix and separating the SPM responses. To validate the effectiveness of the method, We conducted experimental verification using SPM layered models with different parameters and 3-D complex models. The EM responses obtained after separation showed a decibel error reduction of more than 20 dB compared to those before separation.Finally, based on the SPM equivalent circuit, we acquired field-measured TDEM induction-magnetic viscosity effect and performed separation on the measured data.The results showed that the applied method eliminated the interference of the SPM response,suggesting that the NG-MMI-ICA algorithm can be used to obtain accurate electromagnetic data in TDEM surveys.