Optimization of sub‐band coding method for seismic data compression

Abstract

A B S T R A C TSeismic data volumes, which require huge transmission capacities and massive storage media, continue to increase rapidly due to acquisition of 3D and 4D multiple streamer surveys, multicomponent data sets, reprocessing of prestack seismic data, calculation of post-stack seismic data attributes, etc. We consider lossy compression as an important tool for efficient handling of large seismic data sets. We present a 2D lossy seismic data compression algorithm, based on sub-band coding, and we focus on adaptation and optimization of the method for common-offset gathers. The sub-band coding algorithm consists of five stages: first, a preprocessing phase using an automatic gain control to decrease the non-stationary behaviour of seismic data; second, a decorrelation stage using a uniform analysis filter bank to concentrate the energy of seismic data into a minimum number of sub-bands; third, an iterative classification algorithm, based on an estimation of variances of blocks of sub-band samples, to classify the sub-band samples into a fixed number of classes with approximately the same statistics; fourth, a quantization step using a uniform scalar quantizer, which gives an approximation of the sub-band samples to allow for high compression ratios; and fifth, an entropy coding stage using a fixed number of arithmetic encoders matched to the corresponding statistics of the classified and quantized sub-band samples to achieve compression. Decompression basically performs the opposite operations in reverse order. We compare the proposed algorithm with three other seismic data compression algorithms. The high performance of our optimized sub-band coding method is supported by objective and subjective results.

Compression and coding for cable-free land acquisition systems

“…Distortion introduced by coding will be also referred to as quantization error, because it plays a predominant role compared to typical instrument noise. In this paper, the levels of interest for SQNR range between 10 dB and 30 dB as typical values for lossy seismic data compression (Rosten, 2004). It is understood that these values might not comply with the requirements of cabled acquisition systems currently on the market.…”

“…Seismic signals also have an extensive oscillatory nature and suffer from a significant amount of coherent noise (Averbuch, 2001). Compression of seismic data acquired from land (target application for cable-free systems) is also more complex than marine data due to the predominant role of ground-roll (Rosten 2004). The statistics of seismic data are also highly nonstationary.…”

“…Data sorting aims at exploiting the coherence properties of the multidimensional seismic data set (2D-3D sets) to highlight any data redundancy that might improve the compression quality (Averbuch, 2001). Compared to data sorting by common-shot, a more convenient rearrangement of seismic traces to improve the compression factor is by common-offset/azimuth (Rosten, 2004). Common-offset/ azimuth sorting provides the largest compression quality because the reflection events are less dipping: this increases the spatial (lateral) coherence and provides larger data redundancy.…”

“…Moreover, encoding of such components can be optimized to assign the largest number of bits to represent the highest energy components of the seismic signal. A similar approach can be obtained by subband encoding where the input signal is decomposed by analysis filter bank (Rosten, 2004), while each filtered data stream is encoded independently. Time-domain methods are based on differential coding techniques such as differential pulse code modulation, abbreviated as DPCM (Davis, 1985;Jericevic, 1998).…”

Reducción de los tiempos de cómputo de la Migración Sísmica usando FPGAs y GPGPUs: Un artículo de revisión

“…En esta investigación, se utiliza el algoritmo de compresión propuesto por T. Røsten [66] para comprimir los datos sísmicos antes de ser enviados a la GPGPUs.…”