Multi-window Real-valued Discrete Gabor Transform
for Long and Infinite Sequences
Liang Tao
School of Computer Science and Technology
Anhui University
Hefei, Anhui 230039, China
taoliang@ahu.edu.cn
H. K. Kwan
Dept. of Electrical and Computer Engineering
University of Windsor
401 Sunset Avenue, Windsor, Ontario, Canada N9B 3P4
kwan1@uwindsor.ca
Abstract—The length of windows used in the existing multi-
window complex-valued discrete Gabor transform (M-CDGT) is
restricted to be the length of analyzed sequences. Consequently, if
the lengths of analyzed sequences are long, to solve the windows
requires high computation burden and memory and even leads to
numerical instability. To overcome this problem, a multi-window
real-valued discrete Gabor transform (M-RDGT) for long and
infinite sequences is presented in this paper based on the
biorthogonal analysis approach, and its corresponding
biorthogonality constraint between analysis windows and
synthesis windows is derived. The lengths of the analysis and
synthesis windows are independent of the lengths of analyzed
sequences so that one can apply finite (or short) windows to
process any long (or even infinite) sequences. The completeness
condition of the M-RDGT is proved to be equivalent to its
biorthogonality constraint between analysis windows and
synthesis windows. In addition, the M-RDGT can utilize the fast
discrete Hartley transform algorithms for fast computation and
has a simple relationship with the M-CDGT such that the M-
CDGT coefficients can be directly computed from the M-RDGT
coefficients.
Keywords—discrete Gabor transform; multi-window; discrete
Hartley transform; biorthogonality; synthesis window; analysis
window.
I. INTRODUCTION
The Gabor transform [1] has been recognized as being
useful in diverse areas such as order tracking in mechanical
systems, system identification, speech and image processing,
radar data processing, sonar and seismic data processing and
interpretation; However, the traditional Gabor transform with a
single window [2-4] suffers a limitation of the constrained
time-frequency localization governed by the Heisenberg
uncertainty principle and thus it is not enough to analyze the
dynamic time-frequency contents of signals that contain a wide
range of time and frequency components, the resolution of
which is usually low. Therefore, the complex-valued discrete
Gabor transform with multi-windows (M-CDGT) based on the
frame theory were presented in several references such as [5-
9]. Some typical applications for evolutionary spectral analysis
and macromolecular sequence analysis can be found in [7] and
[9], respectively.
However, complex-valued Gabor transforms are complex
to be implemented in software or hardware compared with the
real-valued Gabor transforms and the computational load also
increases with the number of windows used in M-CDGT. For
real-valued signals, an alternate real-valued formulation of the
M-CDGT permits a computationally efficient implementation
[4]. In addition, the existing M-CDGT can only deal with the
finite (short) sequences. The analyzed sequences, analysis and
synthesis windows in the finite M-CDGT must have an equal
length. But, in some real applications, this is quite inconvenient
and even impractical because the lengths of the analyzed
sequences could be long, the number of either analysis
windows or synthesis windows is more than one, and to solve
the windows requires huge computation time and memory and
sometime leads to numerical instability. In such cases, the
existing finite M-CDGT is no longer adequate. To overcome
this problem, we will present a new multi-window real-valued
discrete Gabor transform (M-RDGT) for the long or infinite
sequences based on the biorthogonal analysis approach similar
to that proposed in [2-3]. In the M-RDGT, the lengths of the
analysis and synthesis windows are independent of the lengths
of the analyzed sequences so that one can apply finite (or short)
windows to process any long (even infinite) sequences. We
will prove that the completeness condition of the M-RDGT is
equivalent to its biorthogonality constraint between analysis
windows and synthesis windows. The M-RDGT is defined by
replacing the complex-valued Gabor basis functions (kernels of
the discrete Fourier transform (DFT)) of M-CDGT with real-
valued Gabor basis functions (kernels of the discrete Hartley
transform (DHT)), which allows the M-RDGT to utilize the
fast DHT algorithms [10-11] for fast computation. Moreover,
the M-RDGT has a simple relationship with the M-CDGT such
that the M-CDGT coefficients can be directly computed from
the M-RDGT coefficients. Therefore, the M-RDGT also offers
an efficient method to compute the M-CDGT.
II. M-RDGT
FOR LONG OR INFINITE SEQUENCES
A. Definition of M-RDGT
Suppose that P is the number of synthesis windows
][
)(
kh
p
and analysis windows ][
)(
k
p
γ
(
10 −≤≤ Pp
), the
length of a real finite sequence x[k] is L
s
, and the lengths of
This work was supported in part by the National Natural Science
Foundation of China under Grant No. 61372137.
978-1-4799-8391-9/15/$31.00 ©2015 IEEE 2604