Fast Fourier transform of electromagnetic data for computationally expensive kernels

SUMMARY 3-D controlled-source electromagnetic data are often computed directly in the domain of interest, either frequency or time domain. Computing it one and transforming via a Fourier transform to other is viable alternative. It requires evaluation many responses computational if standard transforms used. This can make prohibitively expensive kernel time-consuming as case modelling. The speed modelling obtained through such defined by three key points: solver, method implementation transform, gridding. faster will be. important that solver robust over wide range values (frequencies times). should require few evaluations possible while remaining robust. As ranges span orders magnitude, required ideally equally spaced on logarithmic scale. proposed fast uses digital linear filter together with careful selection points interpolation. In frequency-to-time tests this methodology typically 15–20 frequencies cover offsets. gridding be frequency- time-dependent, which accomplished making function skin depth. Optimizing for least number cells combined optimizing speed. Looking carefully at these resulted much smaller computation times speedup factors ten more previous methods. A followed transformation therefore an alternative corresponding grids chosen.