IEEE Signal Processing - March 2018 - 60

the refinement of the compression limit as well as the design of
the seismic traces could be segmented into informative (coneffective lossy and lossless compression algorithms.
taining reflected signals from shots) and silent parts (mostly
noise). By allocating more bit budget at the signal part and
fewer bits at the noisy segments, we can achieve higher comAcknowledgments
pression gains without degrading the interpretation results.
This work is supported by the Center for Energy and Geo ProAnother approach to further reduce the overall bit rate is
cessing at the Georgia Institute of Technology and by King Fahd
to exploit the redundancy among different sensors. As the
University of Petroleum and Minerals, Dhahran, Saudi Arabia.
scale of seismic exploration increases, so does the density
of the sensors. Thus, a strong correlation is formed among
Authors
traces recorded by the sensors that are in the proximity of
Ali Payani (payani@ece.gatech.edu) received his B.Sc. degree
each other in the grid. However, this article focused on algoin electrical engineering from the University of Tehran, Iran, in
rithms for compression of traces at each individual sensor,
2001 and his M.Sc. degree in electrical engineering from the
independent of other sensors in the netGeorgia Institute of Technology in 2011,
work. Hence, distributed source coding
where he is currently pursuing a Ph.D.
Further research in better
techniques such as distributed source coddegree in electrical and computer engineerstatistical modeling
ing using syndromes [40] and nested lattice
ing. His research interests are in the areas of
of the seismic signal
quantizers [41] can further reduce the overdigital signal processing and compression, in
will potentially lead to
all communication cost in a seismic data
particular, seismic compression; dictionary
acquisition. Incorporating distributed comlearning; and machine learning.
the refinement of the
pression techniques into the compression
Afshin Abdi (abdi@ece.gatech.edu)
compression limit as well
methods presented in this article is another
received his B.Sc. and M.Sc. degrees in elecas the design of effective
research direction we seek to investigate.
trical engineering from the Sharif University
lossy and lossless
Finally, lossless compression algorithms
of Technology, Tehran, Iran, in 2001 and
compression algorithms.
were not covered. The majority of the ex2003, respectively. He is currently pursuing
isting methods are either extensions of the
his Ph.D. degree in electrical and computer
lossy methods to the lossless case or ad-hoc modifications of
engineering at the Georgia Institute of Technology. His current
existing universal compression algorithms. For example, diresearch interests are in the areas of digital signal processing and
rectly encoding and transmitting the residue (from a linear
compression, in particular, seismic compression; dictionary
predictor output) would result in lossless reconstruction at the
learning; and model compression.
receiver. Another class of lossless algorithms are based on the
Xin Tian (xin.tian@whu.edu.cn) received his B.S. degree
existing universal compression methods such as CTW. Alfrom the Department of Electronic Science and Technology, and
though direct application of these algorithms on seismic traces
the Ph.D. degree from the Institute for Pattern Recognition and
is not efficient (if not impossible) due to the large alphabet size
Artificial Intelligence, Huazhong University of Science and
problem and nonstationarity of the signal, we can improve the
Technology, Wuhan, China, in 2004 and 2010, respectively. From
performance of existing universal schemes with careful modi2015 to 2016, he was a visiting faculty member at the School of
fications. We have developed such an approach [referred to as
Electrical and Computer Engineering, the Georgia Institute of
CTW for large alphabet size (CTWL)] based on CTW in [42]
Technology. He is currently an associate professor with the
by redefining the memory context for each bit separately and
School of Electronic Information and Collaborative Innovation
incorporating a clustering scheme based on the minimum deCenter of Geospatial Technology, Wuhan University, China. His
scription length principle. We have shown that the algorithm
current research interests include image/seismic compression,
can achieve compression gain of close to three, which is a sigdictionary learning, sparse coding, and computational imaging.
nificant improvement over compression gain of two achieved
Faramarz Fekri (fekri@ece.gatech.edu) received his B.Sc.
by current state-of-the-art universal lossless compression aldegree in electrical engineering in 1989 from Sharif University
gorithms. Table 3 compares the performance of the CTWL
of Technology, Tehran, Iran, and his Ph.D. degree from the
method to other universal compression algorithms for the lossGeorgia Institute of Technology (Georgia Tech) in 2000, where
less compression of seismic signal.
he has since been with the faculty of the School of Electrical and
In spite of the efforts for developing efficient lossless comComputer Engineering. He serves as the technical program chair
pression algorithms, not only is the compression limit of seismic
of several IEEE conferences. He is an associate editor of IEEE
signals unknown, but it is also unclear what the optimum lossTransactions on Molecular, Biological, and Multi-Scale
less compression algorithm for seismic signals is. In this article,
Communications. His research interests are in the areas of source
inspired by the mechanism of the seismic signal generation, we
and channel coding, information theory in biology, machine
developed a model based on noisy ICA and GMM to estimate the
learning, and statistical inference. He received the National
minimum average code length. However, as our estimate heavScience Foundation CAREER Award, Southern Center Young
ily depends on the model, there might exist a better model that
Faculty Development Award, and Electrical and Computer
could yield a lower average code length. Further research in better
Engineering Outstanding Young Faculty Award at Georgia Tech.
statistical modeling of the seismic signal will potentially lead to
He is an IEEE Fellow.
60

IEEE Signal Processing Magazine

March 2018

Table of Contents for the Digital Edition of IEEE Signal Processing - March 2018