766 IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, VOL. 28, NO. 3, MARCH 2017
Exponential Stability of Complex-Valued Memristive
Recurrent Neural Networks
Huamin Wang, Shukai Duan, Member, IEEE, Tingwen Huang, Senior Member, IEEE,
Lidan Wang,
Member, IEEE, and Chuandong Li, Senior Member, IEEE
Abstract—In this brief, we establish a novel complex-valued mem-
ristive recurrent neural network (CVMRNN) to study its stability. As a
generalization of real-valued memristive neural networks, CVMRNN can
be separated into real and imaginary parts. By means of M-matrix and
Lyapunov function, the existence, uniqueness, and exponential stability
of the equilibrium point for CVMRNNs are investigated, and sufficient
conditions are presented. Finally, the effectiveness of obtained results is
illustrated by two numerical examples.
Index Terms—Complex-valued memristive neural networks
(CVMNNs), exponential stability, Lyapunov function, M-matrix.
I. I
NTRODUCTION
In the past decade, complex-valued neural networks (CVNNs) with
complex-valued states, connective weights, and activation functions
have attracted increasing research interest in analyzing the dynamical
behaviors of CVNNs because of their wide applications, includ-
ing wireless communications, electromagnetic wave imaging, the
processing of ordinary images and signals, engineering optimization,
pattern recognition, and so on [1], [2]. Some related important
theoretical achievements have been published [3]–[8].
Since the invention of the first physical memristor device [9]–[11],
it has attracted much attention because of its similar features as the
neurons in the human brain [12], [13]. In recent years, memristive
recurrent neural networks (MRNNs) have attracted more and more
attention. It is very important to investigate the dynamical properties
of MRNNs, because the dynamical behaviors of MRNNs play a key
role in the design of this system. After the construction of a delayed
Manuscript received April 30, 2015; revised December 26, 2015; accepted
December 26, 2015. Date of publication January 6, 2016; date of current ver-
sion February 15, 2017. This work was supported in part by the Fundamental
Research Funds for the Central Universities under Grant XDJK2016A001 and
Grant XDJK2014A009, in part by the Program for New Century Excellent
Talents in University under Grant [2013]47, in part by the Qatar National
Research Fund through the National Priorities Research Program (a member
of Qatar Foundation) under Grant NPRP 4-1162-1-181, in part by the
Excellent Talents Program in Scientific and Technological Activities for
Overseas Scholars within the Ministry of Personnel, China, under Grant 2012-
186, in part by the National Natural Science Foundation of China under
Grant 61372139, Grant 61503175, Grant 61571372, Grant 61374078, Grant
61101233, and Grant 60972155, in part by the High School Key Scientific
Research Project of Henan Province under Grant 15A120013, in part by the
Spring Sunshine Plan Research Project within the Ministry of Education of
China under Grant z2011148, and in part by the University Excellent Talents
Supporting Foundations of Chongqing under Grant 2011-65.
H. Wang is with the College of Electronic and Information Engineering,
Southwest University, Chongqing 400715, China, and also with the Depart-
ment of Mathematics, Luoyang Normal University, Luoyang 471022, China
(e-mail: wwhmin2002@163.com).
S. Duan, L. Wang, and C. Li are with the College of Electronic and
Information Engineering, Southwest University, Chongqing 400715, China
(e-mail: duansk@swu.edu.cn; ldwang@swu.edu.cn; cdli@swu.edu.cn).
T. Huang is with the Department of Electrical and Computer Engi-
neering, Texas A&M University at Qatar, Doha 23874, Qatar (e-mail:
tingwen.huang@qatar.tamu.edu).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TNNLS.2015.2513001
MRNN model [14], numerous results of its dynamical analysis have
been published [15]–[18].
It is well known that a new MRNN model can be built by using
memristors instead of resistors in VLSI circuits of neural networks,
which implies that it is a state-dependent switching neural network.
As an extension of CVNNs, complex-valued MRNNs (CVMRNNs)
can also be obtained by replacing resistors with memristors in the
VLSI circuits of CVNNs. According to the properties of memris-
tor and the study of CVNNs [19]–[21] and MRNNs [13], [22],
CVMRNNs have a wide range of potential application in many
areas, including image processing, engineering optimization, pattern
recognition, and so on. Since the practical application of CVMRNNs
depends on their dynamical properties, it is important and neces-
sary to study the dynamical behaviors of CVMRNNs. Although
many results of MRNNs and CVNNs have been derived, very
few research results of CVMRNNs have been obtained. Recently,
there were several results on this novel neural networks [23]–[26].
Li et al. [23] analyzed the dissipativity of complex-valued memristor-
based neural networks (CVMNNs) by using the differential inclusion
theory, the Lyapunov–Krasovskii functional method, and the linear
matrix inequality technique. The problem of finite-time stability of
the fractional-order delayed CVMNNs was extensively investigated
in [24]. Some sufficient conditions for the passivity of CVMNNs are
obtained in [25] and [26]. However, in addition to these results, there
are no more results on the dynamical behaviors of CVMRNNs.
Based on the above analysis and discussions, the stability of
CVMRNNs is concerned in this brief. A novel CVMRNN model is
first established. Then, the existence, the uniqueness of equilibrium
point, and the exponential stability of this CVMRNN model are
investigated. By separating CVMRNN into real and imaginary parts,
a real-valued system is constructed. Using the differential inclusion
theory and the definition of homeomorphism, the sufficient conditions
of the existence and the uniqueness of the equilibrium point are
derived. According to the properties of M-matrix and Lyapunov
techniques, the sufficient criterion of exponential stability for the
CVMRNNs is obtained. Two numerical examples are given to illus-
trate the effectiveness of the presented result.
Notation: Throughout this brief, solutions of all the following
systems are intended in Filippov’s sense [27]. Let R and C denote
the set of real numbers and the set of complex numbers, respectively,
R
n
and C
n
denote the n-dimensional Euclidean and unitary space,
respectively, R
n×n
and C
n×n
are the set of n × n real matrix
and the set of n × n complex matrix, respectively, ·denotes
the Euclidean vector norm, and co{
1
,
2
} denotes the closure of
the convex hull of C
n
, which are generated by complex numbers
1
and
2
.WhenP ∈ R
n×n
, P > 0(< 0) denotes a positive
(negative) definite matrix, and P
T
denotes the transpose of matrix P.
If z = (z
1
, z
2
,...,z
n
)
T
∈ C
n
,then|z|=(|z
1
|, |z
2
|,...,|z
n
|)
T
.
II. M
ODELS AND PRELIMINARIES
Replacing resistors with memristors in the circuit realization of
neural networks, we can obtain an MRNN model, which is described
2162-237X © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.