Memory-assisted compression of seismic data: Tackling a large alphabet-size problem by statistical methods
Ali Payani, Afshin Abdi, Faramarz Fekri · 2017
In this work, we propose a memory-assisted universal compression framework for seismic signal compression. We assume that there is a common memory between the encoder and the decoder, obtained using past transmissions of seismic signals. As such, we propose to build a model, using common memory, at both encoder and decoder to be used as an engine for compression and decompression respectively. Further, since the common memory is consisted of non-stationary seismic traces, we cluster the memory and build a single model for each cluster. The alphabet size in modern seismic data is 232 which makes the universal compression very challenging. We propose two methods to deal with the large alphabet problem for two cases of lossless and near lossless compression. In the first approach, we use a lossless statistical compression model. However, this method would require a very large training data to correctly estimate the model parameters in the large alphabet regime. As such, we modify the Context Tree Weighting algorithm to be more suitable for large alphabet settings. In the second approach, we use the concept behind single-bit oversampling A/D to decrease the alphabet size and hence making the signal more suitable for near lossless compression. Simulation results on real seismic data demonstrate the effectiveness of the proposed algorithms. Presentation Date: Monday, September 25, 2017 Start Time: 1:50 PM Location: Exhibit Hall C/D Presentation Type: POSTER