Extended Dissipative Criteria for Delayed

Semi-discretized Competitive Neural Networks

B. Adhira1 and G. Nagamani1*
1 Department of Mathematics, The Gandhigram Rural Institute (Deemed
to be University), Gandhigram, Dindigul, 624302, Tamil Nadu, India.

*Corresponding author(s). E-mail(s): nagamanigru@[Link];

Contributing authors: adhirabalu@[Link];

Abstract
This brief investigates the extended dissipativity performance of semi-discretized
competitive neural networks (CNNs) with time-varying delays. Inspired by the
computational efficiency and feasibility of implementing the networks, we for-
mulate a discrete counterpart to the continuous-time CNNs. By employing an
appropriate Lyapunov-Krasovskii functional (LKF) and a relaxed summation
inequality, sufficient conditions ensuring the extended dissipative criteria of dis-
cretized CNNs are obtained in the linear matrix inequality (LMI) framework.
Finally, to refine our prediction, two numerical examples are provided which
demonstrate the sustainability and merits of the theoretical results.

Keywords: Competitive neural networks, Semi-discretization technique, Extended

dissipative analysis, Lyapunov-Krasovskii functional.

1 Introduction
Over the past few decades, the study on the dynamical behavior of neural networks
(NNs) has acquired widespread consideration in the domain of computer vision, object
detection, image recognition, fixed-point computations, pattern classification, quan-
tum communication, and so on [1–4]. As the structure of NNs is concerned, varying the
connectivity and learning algorithms produces distinct dynamic behaviors. Moreover,
in the context of NNs, the learning process is commonly perceived as the task of mod-
ifying the network architecture and connection weights to achieve efficient dynamic

1
behaviors. In general, there are three main learning paradigms for NNs namely super-
vised, unsupervised, and hybrid, from which the unsupervised learning algorithm
has efficient real-life applications since it does not require an output associated with
each input pattern to train the data set. Further, the unsupervised learning algo-
rithm explores the inherent structure or correlations among patterns within the data,
categorizing them based on these correlations. Bearing this fact in mind, in [5] the
authors incorporate competitive architecture, typically a type of unsupervised machine
learning algorithm to update the neuron states in the NNs and as a result develop com-
petitive NNs (CNNs). The major key to the design of competitive networks is the idea
of excitatory and inhibitive influences on an artificial neuron. In most of the existing
NNs, we might have noticed that neurons with activation functions receive inputs (xi )
and generate outputs (yi , i ∈ Nn ) accordingly as shown in Figure 1. Comparatively,
neurons within CNNs engage in inter-neuronal competition to attain activation. In
CNNs, the output layer is designed with lateral connection, where each neuron is fed
back to itself in a self-excitatory manner and to the other neurons in an excitatory or
inhibitory manner as shown in Figure 2. As a consequence, the output layer competes
with each other attributes to the prevalence of both feed-forward and feed-backward
connection layers. In such idealization, the synaptic weight is time-varying and hence
it can be modified by external stimulus. Consequently, the network operating within
this framework encompasses the dynamics of both neural activity levels, referred to
as short-term memory (STM), and the dynamics of synaptic modifications, denoted
as long-term memory (LTM). This type of unique structure formulation allows the
network to incorporate the dynamics of both neural activity levels and synaptic modi-
fication, making it different from conventional NNs. In consequence, numerous reports
based on the dynamics of CNNs have been published recently [6–11].

Fig. 1 Learning Paradigm of NNs Fig. 2 Learning Paradigm of CNNs

In the parallel perspective, NNs are typically modeled and analyzed using
continuous-time differential equations. It is common practice to approximate these
networks using discrete-time difference equations in practical implementation. These
approximations provide solutions at discrete time points that are expected to represent

2
samples of the solutions of the original differential equations. Such approximations are
commonly used in numerical integration techniques like Euler and Runge-Kutta meth-
ods for simulating continuous-time networks on computers. However, it is important
to note that the dynamics resulting from the numerical discretization of differen-
tial equations can lead to misleading steady-state solutions and asymptotic behavior.
These artifacts are not inherent to the original form of the differential equations.
Owing to these reasons, the discretization of continuous-time NNs is vital and essen-
tial due to its easier modeling and practical implementations [12, 13]. Additionally,
there are two major benefits to analyze NNs in the discrete-time context. First, the
digital controller can be implemented directly by using the proper technique rather
than using the analog controller. Secondly, the synthesized network is executed directly
within the digital processor. These two advantages make the discrete-time NNs easy
to implement in reality [14, 15]. Inspired by these compelling facts and recognizing
that all the previously mentioned studies rely on continuous-time CNNs and no out-
comes have been reported yet regarding discrete-time CNNs (DT-CNNs). Even though
there exist numerous techniques (such as Runge-Kutta and Euler schemes) in the
existing results, to discretize the continuous-time dynamical systems, these numeri-
cal schemes can reveal fictitious steady-state responses and asymptotic behavior that
are not inherent in the original continuous-time counterparts. To solve these limi-
tations, more recently in [14] and [15], the authors adopted the semi-discretization
techniques to determine the discrete counterpart of impulsive Cohen-Grosberg and
quaternion-valued NNs, respectively, and endorsed that the resultant discrete-time
network preserves the dynamics of their continuous-time version.
Meanwhile, it has been recognized that the qualitative behavior of NNs is usu-
ally characterized by an input-output relation and storage function consideration. The
evaluation of input-output correlation within the examined network is accomplished
using an energy-like function (referred to as the storage function) and input-power-like
functions (referred to as the supply rate). The concept of dissipativity is theoretically
delineated by these storage functions and supply rates, signifying that the growth of
stored energy remains bounded by the quantity of energy introduced by the external
environment. In essence, dissipative systems have the inherent capability to only dissi-
pate and not generate energy. Consequently, dissipativity has gained recognition as a
fundamental tool for analyzing and stabilizing large-scale systems, including NNs [16–
18]. However, when there is no sufficient information about the external disturbance,
then it leads to an unpredictable supply rate of the network, which no longer produces
satisfactory performance. To deal with these type of problems various robust perfor-
mances like H∞ and L2 − L∞ was introduced. In a generalization of all the above
robust performance, the authors introduced a novel performance index called extended
dissipativity, which contains passivity, dissipativity, H∞ , and L2 − L∞ performance
as its special case. In the existing literature, the study on robust performances like
passivity and dissipativity has been analyzed for CNNs due to their numerous appli-
cations [19–21]. It is noteworthy to mention that there are only a few results available
in the existing literature based on the robust performances of continuous-time CNNs

3
and there are no such results in DT-CNNs. In this paper, the generalized robust per-
formance called extended dissipativity performance is newly introduced to DT-CNNs,
which increases the novelty of this study.
Based on the aforementioned discussions and the fact that the numerical analog
is essential to perform the computational task, in this paper semi-discretization cri-
teria are used to study the dynamics of DT-CNNs. Moreover, as mentioned earlier,
obtaining the generalized performance index is an effective tool for achieving less con-
servative results. With this motivation in mind, this paper accomplishes the extended
dissipativity performance for the delayed DT-CNNs. In implementing the stability cri-
terion, an arduous problem is to construct an appropriate LKF of delayed DT-CNNs
and to reach the tighter upper bounds of the estimated summation terms of LKF. In
this paper, an appropriate system dependent LKF is considered such that it contains
the more system information. In particular, the LKF is constructed based on both the
STM and LTM state vectors, time-varying delay of the DT-CNNs, and the activation
function. In addition, the augmented LKF is constructed based on the state vectors
and delay bound components of the discretized CNNs which provide various cross-
terms of the augmented vectors. Consequently, to obtain the tighter bound, in recent
days the relaxed summation inequality which is obtained by the combination of tighter
summation inequality and the matrix bounding techniques is attracted high research
interest [22, 23]. For instance, in [22], the relaxed summation-based inequality was
derived by combining Wirtinger-based summation inequality and reciprocally convex
matrix inequality (RCMI). Similarly, this combination technique is used in [23], where
the auxiliary function-based summation inequality (AFSI) is combined with extended
RCMI (ERCMI) [24] to analyze the non-fragile extended dissipativity of DT-NNs. It is
noteworthy that the inequalities AFSI and ERCMI are the generalization of Wirtinger
and Jensen summation inequalities.
This paper aims to provide more insight into investigating the extended dissipativ-
ity criteria for DT-CNNs through relaxed AFSI. The key contributions of this article
are summarized as follows:
i) The discrete-time analog of the continuous-time CNNs is formulated using the
semi-discretization technique, where the corresponding system parameters of the DT-
CNNs are obtained accordingly from its continuous-time counterpart. It demonstrates
the appropriateness of the developed discrete-time counterparts of CNNs as mathe-
matical models.
ii) The concept of extended dissipativity has been introduced to the discretized
CNNs with time-varying delay. Accordingly, the established criteria not only ensure
the discretized network is stable but also satisfies the common input-output energy
functions like passivity, dissipativity, H∞ , and L2 − L∞ performances.
iii) Finally, the effectiveness of DT-CNNs is investigated in both theoretical and
simulation aspects.
Notations: The notations are same as those in our previous paper [23]. Nn denotes finite
set of natural numbers {1, 2, . . . , n}; the vecn {ai (·)} represents [a1 (·), a2 (·), . . . , an (·)]T .

4
2 System Formulation
In this section, the discrete-time analog of the continuous-time CNNs is formulated
by adapting the semi-discretization technique. Let us consider the following CNNs
formulated by continuous-time differential equations:

dxSi (t) n
P n
P n
P
 ϵ dt

 = −ai xSi (t) + bij fj (xSj (t)) + cij fj (xSj (t − ρ(t))) + Dij xLi (t),
j=1 j=1 j=1
dxLi (t) Pn
= −αi xLi (t) + βij fj (xSj (t)), i ∈ Nn ,


dt

j=1
(1)

where the first system is called STM and the latter is called LTM; n indicates the
number of neurons; ϵ > 0 is the time scale of STM state; xSi (t) is the neuron current
activity level in the system; fj (xSj (t)) and fj (xSj (t − ρ(t))) denote the activation and
delayed activation functions of the j-th neuron, respectively; xLi (t) is the synaptic
efficiency, ρ(t) is the time-varying delay satisfying 0 ≤ ρ(t) ≤ ρ; ai > 0 is the self-
feedback constant; bij and cij denote the synaptic weights of activation functions; Dij
denotes the strength of the external stimulus; αi > 0 and βij are scaling constants.
To obtain the DT analog of the considered continuous-time CNNs (1), based on
the semi-discretization method we initiate by reformulating
   the network based on the
t t
step size and obtain the following for t ∈ r, r+r
r r
n n
          
dxSi (t) P t P t t

 ϵ = −a i xSi (t) + b ij fj xSj r + cij f j xSj r − ρ r
dt r r r



 j=1 j=1
n
   
 P t
+ Dij xLi r ,

 j=1 r
n
 dxL (t)    
t

i
P
= −αi xLi (t) + , i ∈ Nn ,


 βij fj xSj r
 dt j=1 r

where
 r > 0 is the positive constant denoting uniform discretization step size  and
t t t
represents the integral part of r . For notation simplicity, let us assume = k,
r    r
t
k = 0, 1, 2, . . . and xi r = xi (kr) ≜ xi (k). Thus, the above system can be
r
reformulated as

dxSi (t) Pn
P n
P n
 ϵ dt = −ai xSi (t) + bij fj xSj (k) + cij fj xSj (k − ρ(k)) + Dij xLi (k),


j=1 j=1 j=1
n
 dxLi (t) P
 = −αi xLi (t) + βij fi (xSi (k)) , i ∈ Nn ,
dt

j=1

for t ∈ (kr, (k + 1)r), k = 0, 1, 2, . . . . The aforementioned system can be restated as

follows:

5
 ai " #
 d h ai i
t eϵt P n
P n
P n
 dt xSi (t)e = bij fj xSj (k) + cij fj xSj (k − ρ(k)) + Dij xLi (k) ,

 ϵ
ϵ

" j=1 # j=1 j=1
d  n
αi t 
= eαi t
 P
 dt xLi (t)e βij fi (xSi (k)) , i ∈ Nn , t ∈ (kr, (k + 1)r) , k = 0, 1, 2, . . . .



j=1

Now, by integrating the above over [kr, t), where t < (k + 1)r, we have
 ai ai "
ai ai e ϵ t − e ϵ kr n n
t kr
 P
P

xSi (t)e ϵ − xSi (k)e ϵ = bij fj xSj (k) + cij fj xSj (k − ρ(k))


ai




 j=1
# j=1

 Pn
+ Dij xLi (k) ,


 j=1 " #
eαi t − eαi kr P

 n
αi t αi kr

 xLi (t)e − xLi (k)e = βij fi (xSi (k)) , i ∈ Nn .


 αi j=1

By using the continuity of xSi (t) and xLi (t), letting t → (k + 1)r, and after some
simple mathematical calculations, the above equations can be written as follows:
 "
 −ai n
P n

xSi (k + 1) = xSi (k)e ϵ r + ϕi (r)
P
bij fj xSj (k) + cij fj xSj (k − ρ(k))





 j=1 j=1

 #
 Pn
+ Dij xLi (k) ,

 j=1

 " #

 n
 xLi (k + 1) = xLi (k)e−αi r + φi (r)
P

βij fj xSj (k) .



j=1

−ai
1−e ϵ r 1 − e−αi r
For convenience, we let ϕi (r) = and φi (r) = , since ai > 0, αi > 0,
ai αi
and r > 0, we have ϕi (r) > 0 and φi (r) > 0. Moreover, the above model converges to
the continuous time CNNs (1) as r → 0+ .
We assume that the external disturbance in both STM xS (k) and LTM xL (k) as
ωS (k) and ωL (k) ∈ L2 [0, ∞), respectively. Thus, the equivalent form of (1) together
with external disturbances can be written as follows:

 ST M : xS (k + 1) = ĀxS (k) + B̄f (xS (k)) + C̄f (xS (k − ρ(k))) + D̄xL (k) + ωS (k),

yS (k) = ExS (k),


 LT M : x L (k + 1) = ĀxL (k) + B̄f (xS (k)) + ωL (k),
yL (k) = F xL (k),

(2)

where yS (k) = vecn {ySi (k)} ∈ Rn and yL (k) = vecn {yLi (k)} ∈ Rn are the mea-
sured output vectors; E and F ∈ Rn×n are the real constant matrices; xS (k) =
vecn {xSi (k)} ∈ Rn ; and xL (k) = vecn {xLi (k)} ∈ Rn are the STM and LTM
state vectors of the DT-CNNs (2); f (xSm (k)) = vecn {fi (xSmi (k))} ∈ Rn denotes
the activation functions, which are not delay-dependent and f (xSm (k − ρ(k))) =
vecn {fi (xSmi (k − ρ(k)))} ∈ Rn denotes the delay-dependent activation function of the

6
ith neuron at k th instant; Ā = diag{ā1 , . . . , ān } ∈ Rn×n is the state feedback connec-
−ai
tion weight matrix with āi = e ϵ r ; B̄ ∈ Rn×n and C̄ ∈ Rn×n with b̄ij = ϕi (r)bij
and c̄ij = ϕi (r)cij are the connection and delayed connection weight matrices, respec-
tively; D̄ ∈ Rn×n represents the strength of the external stimulus with D̄ij = ϕi (r)Dij ;
−αi
Ā = diag{ᾱ1 , . . . , ᾱn } ∈ Rn×n with ᾱi = e ϵ r ; and B̄ ∈ Rn×n is the matrix with
β̄ij = φi (r)βij . The time-varying delay function ρ(k) is supposed to be bounded as
ρ1 ≤ ρ(k) ≤ ρ2 . Further, the initial conditions for the discretized CNNs (2) are defined
by xS (℘) = ξS (℘) and xL (℘) = ξL (℘), where ξS and ξL ∈ C(Z[−ρ2 , 0], Rn ).
Remark 2.1. Contemporary consensus holds that our world is characterized by a
“continuous” flow of time and a seamlessly interconnected spatial arrangement. Conse-
quently, the evolutionary mechanics within CNNs are often described using non-linear
differential equations within the context of Euclidean space. Although this portrayal
is conceptually clear, obtaining the exact dynamical state behavior of continuous-time
CNNs in a simulation context is very hard. To surmount this hurdle, one can employ
the time discretization technique, where the sequences of points (referred to as tra-
jectories) are generated through iterative mappings. Obtaining discrete-time systems
through the semi-discretization technique involves the following steps: (i) Divide the
time domain into discrete intervals by using a uniform step size, often denoted as
‘r’. This step size determines the time resolution at which the continuous system will
be approximated. (ii) From the original continuous-time CNNs, derive corresponding
difference equations that approximate the system’s behavior within each discrete inter-
val. This involves approximating derivatives with finite differences. It should be noted
that the choice of the time step ‘r’ is crucial, smaller steps may lead to more accurate
approximations.

3 Preliminaries
This section considers the following definitions and assumptions for proving the main
results.
Assumption 3.1. For l ∈ Nn , the activation function fl (·) in (2) is considered to be
bounded and satisfy

fl (e1 ) − fl (e2 )
Fl− ≤ ≤ Fl+ , ∀ e1 , e2 ∈ R, e1 ̸= e2 ,
e1 − e2

where Fl− and Fl− are the known constants. For presentation convenience,
in the
 −following, we denote F1 = diag{F1− F1+ , . . . , Fn− Fn+ } and F2 =
+
F1 + F1 F − + Fn+
diag ,..., n .
2 2
Lemma 3.2. [23] For given scalars a > 0 and b > 0, delay function d(k) with a ≤
d(k) ≤ b, matrix R > 0, any matrix S, and a vector function ψ(·) : Z[a, b] → Rn , such
that the following inequality holds:
k−a−1  T  
X Λ1 Λ
l ψ T (i)Rψ(i) ≥ εT (k) RS 1 ε(k),
Λ2 Λ2
i=k−b

7
 " #
e + (1 − α)S1
R S e−1 T T e −1
holds, where RS = e + αS2 , S1 = R − S R S , S2 = R − S R S,
e e
∗ R
e = diag{R, 3R, 5R}, α = α1 , 1 − α = α2 , α1 = b − d(k), α2 = d(k) −
R
 l l 
a, εT (k) = εbT1 (k) εbT2 (k) εbT3 (k) εbT4 (k) εeT1 (k) εeT2 (k) εeT3 (k) εeT4 (k) in which εb1 (k) =
k−d(k)−1
P −d(k)−1
P k−d(k)−1 P −d(k)−1
P k+j
P
ψ(i), εb2 (k) = ψ(i), εb3 (k) = ψ(i), εb4 (k) =
i=k−b j=−b i=k+j j=−b i=k−b
−d(k)−1
P m
P k+j
P k−a−1
P −a−1
P k−a−1
P
ψ(i), εe1 (k) = ψ(i), εe2 (k) = ψ(i), εe3 (k) =
m=−b j=−b i=k−b i=k−d(k) j=−d(k) i=k+j
−a−1 k+j −a−1 m k+j 2
h
ψ(i). ΛT1 = ı1 ı1 −
P P P P P
ψ(i), εe4 (k) = ı2 ı1 −
j=−d(k) i=k−d(k) m=−d(k) j=−d(k) i=k−d(k) l1 + 1
6 12 i h 2 6 12 i
ı3 + ı4 , ΛT2 = ı5 ı5 − ı6 ı5 − ı7 + ı8 ,
l1 + 1 (l1 + 1)(l1 + 2) l2 + 1 l2 + 1 (l2 + 1)(l2 + 2)
for ıi = [0n×(i−1)n In 0n×(8−i)n ]T .
Definition 3.3. [25] Let Γi ∈ S2n + (i ∈ N4 ) such that Γi = diag{Γi1 , Γi2 }, satisfies
the following assumption:

Γ1 ≤ 0, Γ3 > 0, Γ4 ≥ 0,
(3)
(||Γ1 || + ||Γ2 ||) ||Γ4 || = 0.

Denote J (k) = Y T (k)Γ1 Y (k) + 2Y T (k)Γ2 W (k) + W T (k)Γ3 W (k). For prescribed
symmetric matrices Γi , i ∈ N4 satisfying the assumption in (3), the discretized sys-
tem (2) is said to attain the extended dissipative performance under zero initial
d yS (·)
J (i) ≥ sup Y T (k)Γ4 Y (k) holds for any d ≥ 0, Y (·) =
P
state, if and
i=0 0≤k≤d yL (·)
 
ω (·)
W (·) = S .
ωL (·)

4 Extended dissipativity of Delayed DT-CNNs

In this section, based on the augmented LKF, the sufficient condition ensuring the
extended dissipativity performance for the discretized CNNs (2) is obtained in terms
of LMIs. The following notations will be used in the subsequent
h sections:
 T
ιm = 0n×(m−1)n In 0n×(26−m)n , m ∈ N26 , Υ (k) = ΥS (k) ΥL (k) f T (xS (k))
T T T
i h
f T (xS (k − ρ(k))) , ΥTS (k) = xTS (k) xTS (k − ρ1 ) xTS (k − ρ(k)) xTS (k − ρ2 ) ηST (k)
k−ρ(k)−1 P1 −1
k−ρ −ρ(k)−1
P k−ρ(k)−1 −ρP
1 −1 k−ρ
P1 −1 −ρ(k)−1 k+i
xTS (i) xTS (i) xTS (j) xTS (j)
P P P P
i=k−ρ2 i=k−ρ(k) i=−ρ2 j=k+i i=−ρ(k) j=k+i i=−ρ2 j=k−ρ2
−ρP
1 −1 k+i i h
xTS (j) xTS (j) ωST (k) , ΥTL (k) = xTL (k) xTL (k−ρ1 ) xTL (k−ρ(k)) xTL (k−
P
i=−ρ(k) j=k−ρ(k)
k−ρ(k)−1 P1 −1 T
k−ρ −ρ(k)−1
P k−ρ(k)−1 −ρP
1 −1 k−ρ
P1 −1 T
T
xTL (i) xTL (j)
P P
ρ2 ) η L (k) xL (i) xL (j)
i=k−ρ2 i=k−ρ(k) i=−ρ2 j=k+i i=−ρ(k) j=k+i

8
−ρ(k)−1 k+i −ρP
1 −1 k+i i
xTL (j) xTL (j) ωL
T
P P P
(k) . Let Ξ1 = [ξ1 ξ2 ] Ξ2 = [ξ3 ξ4 ], in
i=−ρ2 j=k−ρ2 i=−ρ(k) j=k−ρ(k)
which ξ1 = [ι1 +ι5 ι2 −ι4 +(ι6 +ι7 )], ξ2 = [ι1 ι6 +ι7 ], ξ3 = [ι17 +ι13 ι14 −ι16 +(ι18 +ι19 )],
ξ4 = [ι13 ι18 + ι19 ], Ξ3 = [ι1 ι13 ι2 ι4 ι14 ι16 ], Ξ4 = [ι1 ι13 ι3 ι15 ], Ξ5 = [ι25 ι26 ],
 
Ξ6 = [ι1 ι25 ], Ξ7 = [ι3 ι26 ], Ξ8 = ι1 + ι5 , Ξ9 = ι13 + ι17 , Ξ bS = Ξ b S2 ,
b S1 Ξ
     
Ξ
eS = Ξ e S2 , Ξ
e S1 Ξ bL = Ξ b L2 , Ξ
b L1 Ξ eL = Ξ e L1 Ξe L2 , α
e = (ρ12 + 1)(ρ12 + 2),
αi (k) = (ρbi (k) + 1)(ρbi (k) + 2), i ∈ N2 , P1 = diag{P1 , −P1 }, Pb2 = diag{P2 , −P2 },
b
e 2 , −Q
Q  1 , −Q2 , −Q1 , −Q2 }, R = diag{(ρ12 +  1)R, −R},
b = diag{Q b1 + Q b2 , Q
e1 + Q b b e e b
F 1 Λ1 −F 2 Λ 1 F 1 Λ 2 −F 2 Λ2
Sb = diag{(ρ12 + 1)S, −S}, Fb1 = , Fb2 = ,
−F2 Λ1 Λ1 −F2 Λ2 Λ2
ιT T
 
3 − ι4
 ιT − ιT − 2
ρb (k)ιT T

3 − ι6
" # " # 
Te V Te X  3 4 b2 (k)+1 2
ρ
TS = 1 e , TL = 2 e , e = [In 0n ], ΞT

S1 (k) =  T T 6 T T 


,
∗ T1 ∗ T2 ι3 − ι4 − ρb2 (k)+1 − ι6 − ρb2 (k)ι4


ρ
b2 (k)(b ρ2 (k)+1) T
+ α 12 − ιT 10 − 2 ι4
2 (k)

ιT T
 
2 − ι3 ιT T
 
3 − ι4
 ιT − ιT − 2 T
ρb (k)ι2 − ι7  T
 T T 2 T T

 2 3 b1 (k)+1 1  ι3 − ι4 − ρ12 +1 ρ12 ι3 − ι6


ρ
ΞT
S2 (k) = ιT − ιT − 6 T T
 , ΞbTS1 =  T T ,
 ι3 − ιT 6 T

2 3 ρb1 (k)+1
− ι 7 − ρ
b 1 (k)ι 3 

4 − ρ12 +1 − ι6 − ρ12 ι4 
ρ
b1 (k)(b ρ1 (k)+1) T ρ12 (ρ12 +1) T
+ α 12 − ιT + 12 T



1 (k) 11 − 2 ι3 e − ι10 −
α 2 ι4

ιT T ιT T



3 − ι4
   
2 − ι3
T T

T T
T
 − ιT


bT  ι2 − ι3
− 2 − ι7
 , Ξ

eT  3
 ι − ι 4
− 2 6
 ,

Ξ S2 =  T T T  S1 = T T T
ι2 − ι3 − 6
− ι7  ι 3 − ι 4 − 6 − ι 

6
+6 − ι11 +6 − ιT 10
ιT T
 
T T
15 − ι16
 
ι2 − ι3
 ιT − ι T − 2 T T
  ιT − ιT − 2
ρb (k)ιT T

ρ12 +1 ρ12 ι2 − ι7
 15 − ι18 

T  2 3 T  15 16 b2 (k)+1 2
ρ
ΞS2 =  T T  , ΞL1 (k) = ιT − ιT −
,
 ι2 − ιT T 6
− ιT b2 (k)ιT


6
e 
3 − ρ12 +1 − ι7 − ρ12 ι3  15 16 18 − ρ 16 

 ρ
b2 (k)+1
ρ (ρ +1) ρ (k)(b ρ (k)+1)
+ 12 T
ιT + α 12 − ιT ιT
12 12

2 2


e − ι11 − 22 −
b
α 2 3 2 (k) 2 16
T T
 
ι14 − ι15 ιT T
 
15 − ι16
 ιT − ιT − 2 T
ρb (k)ι14 − ι19  T
 T
 ι −ι − 2 T T T

 14 15 b1 (k)+1 1 ρ12 +1 ρ12 ι15 − ι18


ρ  15 16
ΞT
L2 (k) = ιT − ιT −
 6 T T
 , ΞbTL1 =  T T 6 T T ,
 14 15 ρ
b1 (k)+1
− ι 19 − ρb 1 (k)ι 15 
 ι15 − ι16 − ρ12 +1 − ι18 − ρ12 ι16 
T ρ
b1 (k)(b ρ1 (k)+1) T
T ρ12 (ρ12 +1) T

+ α 12 − ι 23 − 2 ι 15 + 12αe − ι22 − 2 ι16
1 (k)

ιT T
ιT T



14 − ι15 15 − ι16
   
T T
T

ι15 − ι16 − 2 − ιT
T T


bT
 ι14 − ι15 − 2 − ι19  e T  18
 ,

Ξ L2 = 
 ιT − ιT − 6 − ι T 


, Ξ L1 = 
 ιT − ιT − 6 − ιT 


14 15
19 15 16
18
+6 − ιT 23 +6 − ιT 22
ιT T
 
14 − ι15
 ιT − ιT − 2 T T

ρ12 +1 ρ12 ι14 − ι19


eT  14 15
Ξ L2 =  T T 6 T T .
ι14 − ι15 − ρ12 +1 − ι19 − ρ12 ι15 
T ρ12 (ρ12 +1) T

+ 12e − ι23 −
α 2 ι15
Theorem 4.1. Let the Assumption 3.1 holds. For given scalars ρ1 and ρ2 , the semi-
discretized CNNs (2) attains the extended dissipativity performance, if there exist

9
positive definite real matrices Pi ∈ R2n×2n , Q b i , Ri , S, and Ti ∈ Rn×n (i ∈ N2 ),
ei , Q
positive diagonal matrices Λ1 and Λ2 , non-singular matrices ZS and ZL , any matrices
V and X, with appropriate dimension, such that the following LMIs hold:
   
Θ b S2 V T Ξ
b1 Ξ b L2 X T Θ
b2 Ξ
b S1 V Ξ
b L1 X
 ∗ −Te1 0  < 0,  ∗ −Te1 0  < 0, (4)
   
∗ ∗ −Te2 ∗ ∗ −Te2

eP1 eT − E T Γ41 E > 0, eP2 eT − F T Γ42 F > 0, (5)

where Θ b1 = Θ b −Ξ b S TS ΞbT − Ξ bT − Ξ
b S2 Te1 Ξ bT − Ξ
b L TL Ξ bT , Θ
b L2 Te2 Ξ b −Ξ
b2 = Θ eT −
e S TS Ξ
S S2 L L2 S
e e e T
ΞS1 T1 ΞS1 − ΞL TL ΞL − ΞL1 T2 ΞL1 in which Θ = Ξ1 P1 Ξ1 +Ξ2 P2 Ξ2 +Ξ3 QΞ3 +Ξ4 RΞT4 +
e e T e e e T b b T b T b T b
b T5 +ι5 (ρ212 T1 )ιT5 +ι17 (ρ212 T2 )ιT17 −Ξ6 Fb1 ΞT6 −Ξ7 Fb2 ΞT7 +Ξ8 (ZS )ΠS +(Ξ8 (ZS )ΠS )T +
Ξ5 SΞ
Ξ9 (ZL )ΠL + (Ξ9 (ZL )ΠL )T − ι1 (E T Γ11 E)ιT1 − ι13 (F T Γ12 F )ιT13 − 2ι1 (E T Γ21 )ιT12 −
2ι13 (F T Γ22 )ιT24 − ι12 Γ31 ιT12 − ι24 Γ32 ιT24 .
Proof: To prove the main result based on Lyapunov stability theory, let us consider
P 5
the LKF V (k) = Vl (k) for the semi-discretized CNNs (2), where
l=1

V1 (k) =ℶTS (k)P1 ℶS (k) + ℶTL (k)P2 ℶL (k),

2
X k−1
X
V2 (k) = χT (i)Qχ(i),
l=1 i=k−ρl
k−1
X −ρ1
X k−1
X
V3 (k) = χT (i)Rχ(i) + χT (j)Rχ(j),
i=k−ρ(k) i=−ρ2 +1 j=k+i
k−1
X −ρ1
X k−1
X
V4 (k) = f T (xS (i))Sf (xS (i)) + f T (xS (j))Sf (xS (j)),
i=k−ρ(k) i=−ρ2 +1 j=k+i
−ρ1 −1 k−1  T   
X X ηS (j) T1 0 ηS (j)
V5 (k) =ρ12 ,
ηL (j) 0 T2 ηL (j)
i=−ρ2 j=k+i

" # " #
P1 −1
k−ρ P1 −1
k−ρ
in which ℶTS (k) = xTS (k) xTS (i) , ℶTL (k) = xTL (k) xTL (i) , χT (k) =
i=k−ρ2 i=k−ρ2
 T
xS (k) xTL (k) , R = diag{R1 , R2 }, Q = diag{Q

e l }, ηS (k) = xS (k + 1) − xS (k),
bl , Q
ηL (k) = xL (k + 1) − xL (k), and ρ12 = ρ2 − ρ1 .

Now, we calculate the finite difference of V (k) together with the solution of the
P5
desired estimation error states xS (k) and xL (k) as ∆V (k) = ∆Vl (k), where the
l=1

10
finite difference of Vl (k) (l ∈ N5 ), can be computed as follows:

∆V1 (k) =ℶTS (k + 1)P1 ℶS (k + 1) + ℶTL (k + 1)P2 ℶL (k + 1) − ℶTS (k)P1 ℶS (k) − ℶTL (k)P2 ℶL (k)
n o
=ΥT (k) Ξ1 Pb1 ΞT1 + Ξ2 Pb2 ΞT2 Υ(k). (6)

Next, we can determine the forward difference of V2 (k) as follows:

2 
X 
∆V2 (k) =xTS (k)(Q b 2 )xS (k)+ xTL (k)(Q
b1 + Q e1 + Q
e 2 )xL (k)− xTS (k − ρl )Q
b l xS (k − ρl )
l=1
2 
X 
− xTL (k − ρl )Q
e l xL (k − ρl )
l=1
n o
T e T3 Υ(k).
=Υ (k) Ξ3 QΞ (7)

The finite difference of V3 (k) can be computed as follows:

∆V3 (k) ≤χT (k)(ρ12 + 1)Rχ(k) − χT (k − ρ(k))Rχ(k − ρ(k))

=ΥT (k) Ξ4 RΞ
b T4 Υ(k).

(8)

Now, we can compute ∆V4 (k) as

∆V4 (k) ≤f T (xS (k))(ρ12 + 1)Sf (xS (k)) − f T (xS (k − ρ(k)))Sf (xS (k − ρ(k)))
=ΥT (k) Ξ5 SΞb T5 Υ(k).

(9)

Now, ∆V5 (k) can be computed as

k−ρ
X 1 −1

∆V5 (k) =ηST (k)(ρ212 T1 )ηS (k) + ηL

T
(k)(ρ212 T2 )ηL (k) − ρ12 ηST (i)T1 ηS (i)
i=k−ρ2
k−ρ
X 1 −1
T
− ρ12 ηL (i)T2 ηL (i). (10)
i=k−ρ2

By utilizing Lemma 3.2, the first summation term of ∆V5 (k) can be obtained for any
matrix V as follows:

k−ρ1 −1 T "
 # 
X ηbS1 (k) Te1 + ρ1 (k)U1 V ηbS1 (k)
ρ12 ηST (i)T1 ηS (i) ≥ , (11)
ηeS2 (k) ∗ Te1 + ρ2 (k)U2 ηeS2 (k)
i=k−ρ2

where Te1 = diag{T1 , 3T1 , 5T1 } U1 = Te1 − V Te1−1 V T , U2 = Te1 − V T Te1−1 V , ρ1 (k) =
ρb1 (k) ρb2 (k) T h
2 6
i
, ρ2 (k) = , ηbS1 (k) = εbTS1 εbTS1 − ρb2 (k)+1 εbTS2 εbTS1 − ρb2 (k)+1 εbTS3 + α212(k) εbTS4 ,
ρ12 ρ12

11
 h i
2 6 12
T
ηeS2 (k) = εeTS1 εeTS1 − ρ eTS2
b1 (k)+1 ε εeTS1 − ρ eTS3
b1 (k)+1 ε + eTS4
α1 (k) ε in which
k−ρ(k)−1
P
ρb2 (k) = ρ2 − ρ(k), ρb1 (k) = ρ(k) − ρ1 , εbS1 = ηS (i), εbS2 =
i=k−ρ2
−ρ(k)−1
P k−ρ(k)−1
P −ρ(k)−1
P k+i
P −ρ(k)−1
P Pi k+j
P
ηS (j), εbS3 = ηS (j), εbS4 = ηS (l),
i=−ρ2 j=k+i i=−ρ2 j=k−ρ2 i=−ρ2 j=−ρ2 l=k−ρ2
P1 −1
k−ρ −ρP1 −1 k−ρ
P1 −1 −ρP1 −1 k+i
P
εeS1 = ηS (i), εeS2 = ηS (j), εeS3 = ηS (j), and
i=k−ρ(k) i=−ρ(k) j=k+i i=−ρ(k) j=k−ρ(k)
−ρP
1 −1 Pi k+j
P
εeS4 = ηS (l).
i=−ρ(k) j=−ρ(k) l=k−ρ(k)
Now, using a simple mathematical calculation, (11) can be rewritten as follows:

k−ρ 1 −1
( 2
)
X X
ρ12 ηST (i)T1 ηS (i) ≥ Υ (k)T
ΞS (k)TS ΞTS (k) + ρi (k)ΞSi (k)Ui ΞTSi (k) Υ(k),
i=k−ρ2 i=1
(12)
 
where ΞS (k) = ΞS1 (k) ΞS2 (k) .
By following the similar steps as above and utilizing the Lemma 3.2, for any matrix
X, the lower bounds for the second summation term of ∆V5 (k) can be computed as
follows:

k−ρ1 −1
( 2
)
X X
T
ρ12 ηL (i)T2 ηL (i) ≥ ΥT (k) ΞL (k)TL ΞTL (k) + ρi (k)ΞLi (k)Yi ΞTLi (k) Υ(k),
i=k−ρ2 i=1
(13)
 
where Te2 = diag{T2 , 3T2 , 5T2 }, ΞL (k) = ΞL1 (k) ΞL2 (k) , in which Y1 = Te2 −
X Te2−1 X T , Y2 = Te2 − X T Te2−1 X.
By substituting (13) and (12) in (10), the following upper bound for ∆V5 (k) can be
obtained:
2
(
∆V5 (k) ≤Υ (k) ι5 (ρ212 T1 )ιT
T 2 T T
ρi (k)(ΞSi (k)Ui ΞT
X
5 + ι17 (ρ12 T2 )ι17 − ΞS (k)TS ΞS (k) − Si (k)
i=1
o
−ΞLi (k)Yi ΞT T
Li (k)) − ΞL (k)TL ΞL (k) Υ(k). (14)

By considering the Assumption 3.1, there exist diagonal matrices Λ1 > 0 and Λ2 > 0
such that the following inequality holds:
 T   
xS (k) F1 Λ1 −F2 Λ1 xS (k)
≤ 0, (15)
f (xS (k)) −F2 Λ1 Λ1 f (xS (k))
 T   
xS (k − ρ(k)) F1 Λ2 −F2 Λ2 xS (k − ρ(k))
≤ 0. (16)
f (xS (k − ρ(k))) −F2 Λ2 Λ2 f (xS (k − ρ(k)))

12
Adding the above zero inequalities (15) and (16), we get

ΥT (k) Ξ6 Fb1 ΞT6 + Ξ7 Fb2 ΞT7 Υ(k) ≤ 0.


(17)

In addition to this, for any non-singular matrices ZS and ZL , the subsequent zero
equalities hold from (2):

T
(xS (k + 1)) 2ZS ĀxS (k) + B̄f (xS (k)) + C̄f (xS (k − ρ(k))) + D̄xL (k)

+ ωS (k) − xS (k + 1) = 0,
 
T
(xL (k + 1)) 2ZL ĀxL (k) + B̄f (xS (k)) + ωL (k) − xL (k + 1) = 0.

The above equations can be equivalently represented as

n o
ΥT (k) Ξ8 (2ZS )ΠS + Ξ9 (2ZL )ΠL Υ(k) = 0, (18)

where ΠS = [Ā − In 0n×3n − In 0n×6n In D̄ 0n×11n B̄ C̄] and ΠL = [0n×12n Ā −

In 0n×3n − In 0n×6n In B̄ 0n×n ].
For any real symmetric matrices Γi for i ∈ N3 , the supply rate J (k) is considered as
follows:

J (k) =Y T (k)Γ1 Y (k) + 2Y T (k)Γ2 W (k) + W T (k)Γ3 W (k)

n
=ΥT (k) ι1 (E T Γ11 E)ιT1 + ι13 (F T Γ12 F )ιT13 + 2ι1 (E T Γ21 )ιT12 + 2ι13 (F T Γ22 )ιT24

o
+ ι12 Γ31 ιT12 + ι24 Γ32 ιT24 Υ(k). (19)

Adding up the finite difference of Vl (k) (l ∈ N5 ) derived in (6) - (9), and (14),
along with inequality (17), equations (18), (19), and by performing some elementary
algebraic calculations, we get

∆V (k) − J (k) ≤ ΥT (k){Θ(k)}Υ(k), (20)

2
b − ΞS (k)TS ΞT (k) −
where Θ(k) = Θ
P
ρi (k)ΞSi (k)Ui ΞTSi (k) − ΞL (k)TL ΞTL (k) −
S
i=1
2
ρi (k)ΞLi (k)Yi ΞTLi (k).
P
i=1
Assume the LMIs listed in (4) are true. Then, the following inequalities are obtained
by utilizing the Schur complement lemma:

Θ b S2 (V T Te−1 V )Ξ
b1 + Ξ b L2 (X T Te−1 X)Ξ
b TS2 + Ξ b TL2 < 0, for ρ(k) = ρ1 , (21)
1 2

Θ e S1 (V Te−1 V T )Ξ
b2 + Ξ e L1 (X Te−1 X T )Ξ
e TS1 + Ξ e TL1 < 0, for ρ(k) = ρ2 . (22)
1 2

13
In accordance with the convex combination method [26], the above two inequalities (21)
and (22) are valid for any ρ1 ≤ ρ(k) ≤ ρ2 if and only if the inequality Θ(k) < 0 is
satisfied. Hence, from the inequalities (20) and Θ(k) < 0, we conclude that

∆V (k) − J (k) < 0, k = 0, 1, 2 . . . . (23)

On the other hand, to establish the extended dissipativity criteria, for the discretized
CNNs (2), we considered (23) under zero initial condition, described by

k−1
X
V (k) < J (i). (24)
i=0

To complete the proof, it remains to verify the inequality in Definition 3.3. From the
LMI conditions (5), one can ensure that

Y T (k)Γ4 Y (k) =xTS (k)E T Γ41 ExS (k) + xTL (k)F T Γ42 F xL (k)
≤xTS (k)(eP1 eT )xS (k) + xTL (k)(eP2 eT )xL (k) ≤ V (k). (25)

Combining the inequalities (24) and (25), we have the following

k−1
X
Y T (k)Γ4 Y (k) < J (i). (26)
i=0

d
P
If Γ4 = 0, the inequality in Definition 3.3 holds since J (i) > 0, d ≥ 0. If Γ4 ̸= 0,
i=0
it is evident from assumption (3) that Γ1 = 0, Γ2 = 0 and Γ3 > 0, which leads to
J (k) = W T (k)Γ3 W (k) ≥ 0. As a result, for any integer d ≥ 0 with 0 ≤ k ≤ d, we
have the following inequality:

d
X k−1
X
J (i) ≥ J (i) ≥ V (k) ≥ Y T (k)Γ4 Y (k).
i=0 i=0

By taking the supremum over 0 ≤ k ≤ d, we obtain the extended disipativity condition.

This completes the proof.
Remark 4.2. Theorem 4.1 provides an intuitive criterion for determining the robust
performance of the considered DT-CNNs (2). Here, to reduce the conservatism of
the stability performance, the augmented LKF is constructed, which provides valuable
structural insights into the DT-CNNs. One of the key challenges in estimating the
forward difference of the constructed LKF is bounding the summation terms through
summation-based inequalities. On the other hand, the matrix-based inequality, like
the RCMI technique has been widely used to avoid the occurrence of the variable
delay terms in the finite difference of the LKF. The combination of summation-based
and matrix-based inequalities provides a powerful tool for the stability analysis of the
dynamical systems [22]. Due to this consideration, in this paper, the relaxed AFSI [23]

14
 is used which is the generalization of well-known Jensen summation inequality and
Wirtinger-based summation inequality. As a result, the LMIs (4) and (5) obtained
from the Theorem 4.1 essentially reduce the conservatism of the proposed result.
It is worth pointing out that the analyzed extended dissipative criterion is a com-
posite index consisting of other robust performances. By tuning the weighting matrices
in Definition 3.3, some well-known performance indices can be obtained as follows:
(i) When Γ1 = −I2n , Γ2 = 02n , Γ3 = γ 2 I2n , and Γ4 = 02n , the extended dissipativity
in Definition 3.3 reduces to H∞ performance.
(ii) When Γ1 = 02n , Γ2 = I2n , Γ3 = γI2n , and Γ4 = 02n , we obtain passivity
performance.
(iii) The strict (Γ1 , Γ2 , Γ3 ) − dissipativity can be obtained by setting Γ4 = 02n . More-
over, (Γ1 , Γ2 , R) − dissipativity is obtained by setting Γ3 = R − γI2n and Γ4 =
02n .
(iv) When Γ1 = 02n , Γ2 = 02n , Γ3 = γ 2 I2n , and Γ4 = I2n , the L2 − L∞ performance
is deduced from Definition 3.3.

5 Illustrative Example
In this section, two numerical examples are provided to illustrate the validity and mer-
its of the theoretical findings. In the existing literature, almost all results concerning
the stability of CNNs are dealt with continuous-time analog. Generally, in the dis-
cretization process, the derived discrete-time analogs (2) of CNNs should faithfully
preserve the convergence characteristics of the continuous-time CNNs (1). In order to
demonstrate this effectiveness, in this paper, for the first example, we adopt the same
system parameters as in continuous-time CNNs [10] and for the second example, we
adopt the parameters as in [9]. From the simulation results, it is evident that the semi-
discretized CNNs have the similar dynamical behaviors as in their continuous-time
counterparts.
Example 5.1. In this example, we consider the same system parameters of
continuous-time CNNs in [10]. The corresponding system parameters of DT-CNNs (2)
are obtained
 by semi-discretization
  techniques for the stepsize r = 0.7 as follows:
0.4317 0 0.5209 −0.0568 −0.7104 0.9471
Ā = , B̄ = , C̄ = ,
 0 0.4025
  −2.2980 1.0111   −1.3788 −1.1490

0.0947 0 0.6126 0 −0.6641 0
D̄ = , Ā = , B̄ = , E = G = 0.5I,
0 0.1149 0 0.7047 0 0.6497
F = H = 0.4I. Let the activation function be fi (k) = tanh(0.1 k) for i ∈ N2 , for
which the Assumption 3.1 yields the bounds F1 = diag{0, 0} and F2 = diag{0.1, 0.1}.
The time-varying delay parameter ρ(k) and the disturbance input ωS (k) and ωL (k) are
h iT h iT
assumed as ρ(k) = 3 + cos( kπ 0.01 0.01
2 ), ωS (k) = 1+k2 1+k2 , and ωL (k) = 1+k0.04 0.04
2 1+k 2 ,
respectively. Accordingly, the bounds of time-varying delay can be taken as ρ1 = 2 and
ρ2 = 4.
Based on the above-mentioned system parameters, the state trajectories of STM
xSi (k) and LTM xLi (k), i ∈ N2 of discritized CNNs (2) are obtained as given in Figure
3. From the simulation results, it is evident that the discretized CNNs converge together

15
 Fig. 3 Simulation results for the discretized CNNs (2) for Example 1.

Fig. 4 Simulation results for the CNNs (2) under different initial conditions.

Table 1 Minimum robust performance level γ for

ρ1 = 2 and various ρ2

Performances ρ2 = 4 ρ2 = 6 ρ2 = 8 ρ2 = 10
Passivity 2.7846 2.8738 3.1467 3.6456
H∞ 2.4875 2.5632 2.7479 2.9867
L2 − L∞ 1.8740 1.9741 2.0137 2.1541

with the external disturbances ωS (k) and ωL (k) ∈ L2 [0, ∞). Moreover, in order to
prove the sustainability of theoretical results, the simulation figures are obtained for
distinct initial conditions under the interval [−2, 2] in Figure 4.
Based on the above parameters and by choosing the appropriate values of the
weighting matrices Γi , i ∈ N4 , we obtain the following dynamic performance:
(i) H∞ Performance: By choosing the weighting matrices Γ1 = −I4 , Γ2 = −04 , Γ3 =
γ 2 I4 , and Γ4 = 04 , the extended dissipativity criteria becomes H∞ Performance.

16
 Table 2 Maximum dissipativity level γ for ρ1 = 2 and
various ρ2

Performance ρ2 = 4 ρ2 = 6 ρ2 = 8 ρ2 = 10
γ-Dissipativity 5.8142 5.6540 5.3255 5.0174

For different values of upper delay bound ρ2 , Table 1 lists the minimum H∞ level
of γ by fixing the ρ1 .
(ii) Passivity Performance: The extended dissipativity index reduces to passivity per-
formance by choosing Γ1 = 04 , Γ2 = I4 , Γ3 = γI4 , and Γ4 = 04 . The minimum
performance level for passivity performance is listed in Table 1 for different values
of ρ2 .
(iii) L2 − L∞ performance By letting the matrices Γ1 = 04 , Γ2 = 04 , Γ3 = γ 2 I4 ,
and Γ4 = I4 , we obtain the L2 − L∞ performance. Table 1 displays the minimum
L2 − L∞ performance level γ for various ρ2 and fixed ρ1 .
(iv) (Γ1 , Γ2 , R) − dissipativity We obtain the dissipativity performance by choosing the
matrices Γ1 = I4 , Γ2 = 0.1I4 , Γ3 = 0.5I4 −γI4 , and Γ4 = I4 . Further, the maximum
performance level γ of (Γ1 , Γ2 , R) − dissipativity is obtained in Table 2 for different
upper delay bound ρ2 and fixed lower delay bound ρ1 .
Hence, from Table 1, one can observe that the larger value of ρ2 , we obtained the larger
performance index γ. On the parallel perceptive, for the same ρ2 , one can see that the
obtained performance level γ using the Theorem 4.1 is smaller. Hence from Table 2,
it is clear that the obtained performance level γ is inversely propositional to the upper
delay bound ρ2 , which verifies the advantage of the presented method.
Example 5.2. Consider the continuous-time CNNs (1) with the same parameters as
in the numerical example section of [9]. By employing the semi-discretization tech-
nique, which is demonstrated in the system formulation, we obtain the DT-CNNs (2)
of the corresponding continuous-time CNNs in [9], where the discretized coefficient
matrices are obtained with discretization
 step size 0.9s as follows:
 
0.7634 0 1.9718 −0.0789 −1.5775 −0.3944
Ā = , B̄ = , C̄ = , D̄ =
0 0.8353 −0.1235 2.8828 −0.2471 −1.6473
     
1.8930 −0.2366 2.4596 0 −0.2335 0.0018
, Ā = , B̄ = , E = G = I,
0.3706 0.4942 0 2.4596 0.0073 −0.1095
F = H = I. Here, the activation function is chosen as fi (k) = tanh(k) for i ∈ Nn , and
based on the Assumption 3.1 the corresponding lower and upper bounds are taken as
F1 = diag{0, 0} and F2 = diag{0.5, 0.5}, respectively. The time-varying delay param-
eter ρ(k) and the disturbance input ωS (k) and ωL (k) are assumed as same in Example
5.1. Further, by solving the LMI conditions (4) and (5) in Theorem 4.1, using Matlab
Yalmip toolbox, we can obtain the required feasibility matrices.
Based on the above-mentioned parameters the simulation results for the state
responses of the discretized CNNs (2) are obtained as shown in Figure 5 for several
initial conditions between the interval [−1, 1]. This clearly demonstrates the superi-
ority of the DT-CNNs in (2) over the existing continuous-time CNNs. Further, from
the simulation results, it is evident that the constructed discretized CNNs (2) exhibit
similar dynamical behaviors as in its original continuous-time CNNs and hence this
illustrates the validity of our proposed results.

17
Fig. 5 Simulation results for the discretized CNNs (2) for Example 2.

Remark 5.1. When discretizing continuous-time NNs, two critical concerns come
to the forefront: (i) formulating discrete-time analogs of continuous-time NNs along
the spatial coordinates only, and (ii) ensuring that these discrete-time analogs accu-
rately capture the dynamics of their continuous-time counterparts. Addressing the first
concern, according to numerical analysis theory, there exist numerous methods to
derive discrete-time analogs from continuous-time dynamical CNNs (1). However, as
emphasized in literature [15], discretization might fail to uphold the continuous-time
dynamics, even with a small sampling period. To overcome this limitation in this paper,
the technique of semi-discretization is utilized to achieve an accurate representation of
the dynamic performance exhibited by the continuous-time CNNs. Further, to address
the second concern (ii), in the aforementioned examples, we employed the identical
system parameters of the continuous-time CNNs. As a result, from the figures, we can
guarantee that the dynamics for both the continuous and DT-CNNs are uniform.

6 Conclusion
In this paper, the discrete-time analog of continuous-time CNNs has been formulated
by using the semi-discretization technique. The extended dissipativity performance has
been examined for the semi-discretized CNNs. In addition, the novel LKFs and relaxed
AFSI have been utilized to obtain a tighter upper bound of the summation terms in
the forward difference of LKF. Further, in order to ensure that the constructed DT-
CNNs retain the dynamics of the corresponding continuous-time analog, in numerical
simulation, we adopted the same parameters in continuous-time CNNs [9, 10] and
verified the effectiveness of theoretical results.

References
[1] T. Wu, J. Cao, L. Xiong, J. H. Park, and X. Tan, “Adaptive Event-Triggered
Mechanism to Synchronization of Reaction Diffusion CVNNs and Its Application

18
 in Image Secure Communication”, IEEE Transactions on Systems, Man, and
Cybernetics: Systems, pp. 1-14, 2023. doi: 10.1109/TSMC.2023.3258222

[2] W. Deng, H. Zhou, J. Zhou, and J. Yao, “Neural network-based adaptive asymp-
totic prescribed performance tracking control of hydraulic manipulators”, IEEE
Transactions on Systems, Man, and Cybernetics: Systems, vol. 53, no.1, pp.
285-295, 2022.

[3] S. Duan, X. Hu, Z. Dong, L. Wang, and P. Mazumder, “Memristor-based cellular

nonlinear/neural network: design, analysis, and applications”, IEEE transac-
tions on neural networks and learning systems, vol. 26, no. 6, pp. 1202-1213,
2014.

[4] S. Pandiselvi, R. Raja, J. Cao, and G. Rajchakit, ”Stabilization of switched

stochastic genetic regulatory networks with leakage and impulsive effects”,
Neural Processing Letters, vol. 49, pp. 593-610, 2019.

[5] A. Meyer-Base, F. Ohl, and H. Scheich, “Singular perturbation analysis of com-

petitive neural networks with different time scales”, Neural Computation, vol.
8, no. 8, pp. 1731-1742, 1996.

[6] K. Balasundaram, R. Raja, A. Pratap, and S. Chandrasekaran, “Impulsive

effects on competitive neural networks with mixed delays: existence and expo-
nential stability analysis”, Mathematics and Computers in Simulation, vol. 155,
pp. 290-302, 2019.

[7] X. Nie, J. Liang, and J. Cao, “Multistability analysis of competitive neural

networks with Gaussian-wavelet-type activation functions and unbounded time-
varying delays”, Applied Mathematics and Computation, vol. 356, pp. 449-468,
2019.

[8] A. Pratap, R. Raja, J. Cao, G. Rajchakit, and H. M. Fardoun. “Stability and

synchronization criteria for fractional order competitive neural networks with
time delays: An asymptotic expansion of Mittag Leffler function”, Journal of
the Franklin Institute, vol. 356, no. 4, pp. 2212-2239, 2019.

[9] Z. Shi, Y. Yang, Q. Chang, and X. Xu, “The optimal state estimation for com-
petitive neural network with time-varying delay using Local Search Algorithm”,
Physica A: Statistical Mechanics and its Applications, vol. 540, p. 123102, 2020.

[10] Yin Sheng, Zhigang Zeng, and Tingwen Huang, “Finite-Time Stabilization of
Competitive Neural Networks With Time-Varying Delays”, IEEE Transaction
on cybernetics, vol. 52, no. 11, pp. 11325-11334, 2021.

[11] S. Gong, S. Yang, Z. Guo, and T. Huang, “Global exponential synchronization

of memristive competitive neural networks with time-varying delay via nonlinear
control”, Neural Processing Letters, vol. 49, pp. 103-119, 2019.

19
[12] Y. Liu, Z. Fang, J. H. Park, and F. Fang “Quantized Event-Triggered Synchro-
nization of Discrete-Time Chaotic Neural Networks With Stochastic Deception
Attack”, IEEE Transactions on Systems, Man, and Cybernetics: Systems, pp.
1-11, 2023. doi: 10.1109/TSMC.2023.3251355

[13] H. Liu, Z. Wang, W. Fei, and H. Dong, “On state estimation for discrete time-
delayed memristive neural networks under the WTOD protocol: A resilient set-
membership approach”, IEEE Transactions on Systems, Man, and Cybernetics:
Systems, vol. 52, no.4, pp. 2145-2155, 2021.

[14] L. Li and C. Li, “Discrete analogue for a class of impulsive Cohen?Grossberg neu-
ral networks with asynchronous time-varying delays”, Neural Processing Letters,
vol. 49, no.1, pp. 331-345, 2019.

[15] X. Chen, Q. Song, Z. Li, Z. Zhao, and Y. Liu, “Stability analysis of continuous-
time and discrete-time quaternion-valued neural networks with linear threshold
neurons”, IEEE Transactions on Neural Networks and Learning Systems, vol.
29, no. 7, pp. 2769-2781, 2017.

[16] W. Tai, X. Li, J. Zhou, and S. Arik, “Asynchronous dissipative stabilization for
stochastic Markov-switching neural networks with completely-and incompletely-
known transition rates”, Neural Networks, vol. 161, pp. 55-64, 2023.

[17] Y. Liu, C. Zhang, and M. Li, “Quasi-Synchronization and Dissipativity Analysis

for Fractional-Order Neural Networks with Time Delay”, Fractal and Fractional,
vol. 7, no. 5, pp. 364, 2023.

[18] G. Rajchakit and R. Sriraman, “Robust passivity and stability analysis of

uncertain complex-valued impulsive neural networks with time-varying delays”,
Neural Processing Letters, vol. 53, pp. 581-606, 2021.

[19] C Aouiti, F Cherif, and F. Touati, “Dissipativity Analysis of a Class of Compet-

itive Neural Networks with Proportional Delays”, In International Conference
on Artificial Neural Networks, pp. 32-42, 2019.

[20] G. Rajchakit, P. Chanthorn, M. Niezabitowski, R. Raja, D. Baleanu, and A.

Pratap, “Impulsive effects on stability and passivity analysis of memristor-based
fractional-order competitive neural networks”, Neurocomputing, vol. 417, pp.
290-301, 2020.

[21] P. Anbalagan, R. Ramachandran, J. Alzabut, E. Hincal, and M. Niezabitowski,

“Improved Results on Finite-Time Passivity and Synchronization Problem
for Fractional-Order Memristor-Based Competitive Neural Networks: Interval
Matrix Approach”, Fractal and Fractional, vol. 6, no. 1, pp. 36, 2022.

20
[22] C. K. Zhang, Y. He, L. Jiang, and M. Wu, “An improved summation inequality
to discrete-time systems with time-varying delay”, Automatica, vol. 74, pp. 10-
15, 2016.

[23] B. Adhira, G. Nagamani, and D. Dafik, “Non-fragile extended dissipative

synchronization control of delayed uncertain discrete-time neural networks”,
Communications in Nonlinear Science and Numerical Simulation, vol. 116, p.
106820, 2023.

[24] C. K. Zhang, Y. He, L. Jiang, M. Wu, and Q. G. Wang, “An extended recipro-
cally convex matrix inequality for stability analysis of systems with time-varying
delay”, Automatica, vol. 85, pp. 481-485, 2017.

[25] Jinrong Yang, Guici Chen, and Shiping Wen, “Finite-time dissipative control for
bidirectional associative memory neural networks with state-dependent switch-
ing and time-varying delays”, Knowledge-Based Systems, vol. 252, pp. 109338,
2022.

[26] P. Park and J. W. Ko, “Stability and robust stability for systems with a time-
varying delay”, Automatica, vol. 43, no.10, pp. 1855-1858, 2007.

21