Recent results on seismic deconvolution

One of the important steps in seismic imaging is to provide suitable information about boundaries. Sharp variation of physical properties at a layer boundary cause reflection the wavefield. In previous work done by C. M. Dupuis, seismic signal characterization is divided into two steps: detection and estimation. In the detection phase, the goal is to find all singularities in a seismic section regardless of their order and then to categorize the data to different events by windowing each singularity. In the estimation step, we determine the order of singularity more precisely by using a rough estimate based on the detection phase. Traditionally, a redundant dictionary method is employed for the detection part. However, we attempt to instead use a new L1-solver developed by D.L. Donoho: the Stagewise Orthogonal Matching Pursuit (StOMP). It approximates the solution to inverse problems while promoting the sparsity in the solution vector. This algorithm will allow us to experimentally confirm the recent analysis by S. Mallat on spiky deconvolution limits, which imposes a required minimum distance between spikes. This required minimum distance between different spikes is dependent on the number of spikes as well as the width of the chosen source wavelet used in convolution with the train. These results allow for the design of more robust and accurate detection schemes for seismic signal characterization.