Abstract
The stabilization problem of the networked control systems (NCSs) affected by data quantization, packet losses, and event-driven communication is studied in this paper. By proposing two event-driven schemes and the extended forms of them relying on quantized states, zoom strategy is adopted here to study the system stability with time-varying logarithmic quantization and independent identically distributed (IID) packet losses process. On the basis of that, some sufficient conditions ensuring the mean square stability of the system are obtained here. Although zoom strategy has been utilized by many literatures to study the quantized stabilization problem of NCSs, it has not been adopted to analyze the stability of NCSs with data quantization, IID packet losses, and event-driven communication. Furthermore, the existing literatures relating to zoom strategy employ the quantizer with quantization regions holding arbitrary shapes, but here we use the logarithmic quantizer which holds better performance near the origin. In addition, the detailed comparisons of the system performance under different event-driven schemes are given here, which can guide the strategy selection according to the different design goals. The above three points are the main innovations of this paper. At last, the effectiveness of the proposed methods is illustrated by a benchmark example.
1. Introduction
The introduction of the network in the control system brings us a lot of convenience, such as increased flexibility and maintainability and reduced wiring cost. However, network inserted in the control system also brings some new challenges for studying the stability and stabilization problems of the closed-loop system. Under the network-induced imperfections including data quantization, transmission delay, packet losses, and communication constraints, the sampled signals can not be transmitted timely and accurately through the network, which results in the fact that the system performance may be degraded, or even instability. How to design suitable controller based on noncomplete information to stabilize the NCSs has gotten a lot of attention in recent years.
As two important issues in NCSs, quantization and packet losses have been researched by many literatures including [1–11]. Ji et al. study the mean square quadratic stabilization problem for NCSs with arbitrary packet losses and Markovian jump packet losses, respectively, in [1] and [2]. An optimal quantizer/controller design method is represented in [3]. For the systems with multiple networks, literature in [4] proposes a less conservative sufficient stabilizing bit rate condition. The problem of optimal tracking of multi-input multioutput systems with quantization and packet-dropouts is studied in [5]. Literature in [6] discusses the quantized control problem for the neural network affected by distributed time delay and packet losses. For the linear systems and fuzzy systems affected by delays, quantization, and packet dropout, observer-based output feedback controllers are designed in [7] and [8], respectively. Moreover, the design criteria for the static output feedback controller, optimal dynamic quantizer, and controller are researched in [9], [10], and [11], respectively.
Some special methods have been adopted to deal with quantization and packet losses issues, such as [12–19]. A sliding mode controller is designed in literature [12] to stabilize the NCSs. A switched dynamic controller is proposed in [13] by converting the system into a time-delay switched system. By adopting model predictive control method, literatures in [14] and [15] discuss the issues of quantization and packet losses for nonlinear systems and constrained feedback control systems, respectively. Literature in [16] also discusses the effect of time-varying delay on system performance by quantized predictive control method. By utilizing zoom strategy, we discuss the quantized control for linear systems and fuzzy systems with packet losses in [17] and [18], respectively, and consider the unknown disturbances simultaneously in [19].
Event-driven scheme is a useful method to reduce the communication resources in contrast with the periodic protocol, which means that just important signals are sent through the network; that is, if the signals being sampled have not altered sufficiently, there is no superiority to transfer them. Some papers have discussed the quantized control for the systems with event-driven scheme, such as [20–22]. By adopting an event-driven zero order holder to deal with the effects of stochastic packet losses, the stochastic stability of the system under quantized control is discussed in [20]. Under the event-driven communication, literature in [21] studies the synchronization for networked passive systems with data quantization. For the hybrid NCSs, [22] derives a sufficient condition of the stochastic stability by proposing a distributed event-triggered scheme and multiple quantization method.
The purpose of this paper is studying the issues of quantization and packet losses for NCSs under event-driven communication. In fact, the problem studied here has been researched in [23, 24]. By restraining the upper bound of the number of successive packet losses and adopting the logarithmic quantization strategy proposed in [25, 26], the sufficient conditions ensuring the asymptotical stability of the systems are shown in [23], in which an event-driven scheme is designed by the errors between the current sampling states and the last transmitted states. The differences of this paper with [23] include three aspects. (i) In this paper, the packet losses are modeled as an IID process, under which the mean square stability rather than the asymptotical stability can be guaranteed. (ii) Although logarithmic quantizer is also used in our paper, the parameters of which are time-varying based on zoom strategy. However, the parameters in [23] are all fixed. (iii) The event-driven scheme proposed here is according to the error between the current (quantized) sampling states and the last transmitted quantized states, which is more reasonable than [23]. Moreover, the stabilization problem discussed here is also studied in [24], where the quantized controller was given for NCSs with event-driven communication by modeling the Markov packet losses process. The different points between this paper and [24] include the following: (i) the models of packet losses are different from each other. (ii) Zoom strategy is used in both our paper and [24]. However, quantizer used in [24] is proposed in [27], but the one we adopted is a logarithmic quantizer proposed in [25, 26]. (iii) Only one event-driven scheme is adopted in [24]. But in this paper, we propose two event-driven schemes and compare them with each other. Furthermore, we also extend the event-driven schemes to the ones relating to an additional variable , the effect of which on the system performance is also studied.
Summarized above, our contributions with respect to earlier literatures are three aspects.
() Depending on the quantized states, two event-driven schemes and the extended forms of them are proposed in this paper, which are more applicable due to the quantized data being indeed transmitted through the network under the limited bandwidth.
() The detailed comparison of the event-driven schemes is given here, based on which users can choose different strategies according to their actual needs including minimizing transmission times, increasing convergence rate, or extending the applicable scope of the algorithm.
() Although there are many literatures using zoom strategy to deal with data quantization, the quantizer adopted there is the one proposed in [27]. Considering the advantages of the logarithmic quantizer, such as which holds better performance near the origin, we adopt zoom strategy to handle the stabilization problem for NCSs with logarithmic quantization, packet losses, and event-driven communication. It is worth mentioning that the logarithmic quantization combined with zoom strategy has already been studied in our paper [28], but in which the packet losses and event-driven scheme are all uncovered.
The organization of our paper is arranged as follows. The problem formulation including the detailed system, network, and closed-loop system formulations is described in Section 2. Mean square stabilities of the closed-loop system under two event-driven schemes are studied in Sections 3 and 4. Some analysis and comparison results are illustrated in Section 5. Simulation and conclusion are given, respectively, in Sections 6 and 7.
Notation. -dimensional Euclidean space is denoted as . and represent the positive real numbers set and positive integers set, respectively. The Euclidean norm in and corresponding induced matrix norm in are both illustrated by . For the positive-definite matrix , and are, respectively, used to denote the maximum and minimum eigenvalues of it. The transposed matrix is indicated by . The least integer not less than is signified by the signal . Zero vector with dimension is denoted as .
2. Problem Formulation
2.1. System Formulation
We considers the following continuous-time system:where is the system state, denotes the control input, and and are constant matrices with appropriate dimensions. Given sampling period , system (1) can be rewritten as discrete-time one: with and , where represents continuous-time instant and is the time step satisfying .
2.2. Network Formulation
2.2.1. Packet Losses
The packet losses process is modeled as an IID process, in which denotes the successful transmission of data packets at time instant , and represents the occurrence of packet losses. Moreover, we assume that the random variable satisfies and .
2.2.2. Data Quantization
For the th component of the state , , the time-varying logarithmic quantizer adopted in this paper is represented aswith , , , for any , . It is obvious that , , , and are the critical parameters of the logarithmic quantizer, which are all assumed to be undetermined. On the basis of , , the quantization value of the vector is defined as
Remark 1. Existing articles involving zoom strategy mostly use the quantizer defined in literature in [27]. Using logarithmic quantizer to deal with quantization and packet losses based on zoom strategy is one of the innovations of this paper.
2.2.3. Event-Driven Scheme
To reduce the network burden, we set the sampling data to be transmitted just at some specific times. Denote the transmission times as with satisfying . Two event-driven schemes are designed in this paper, in which is selected, respectively, asin which is a piecewise constant function denoted by where and are defined below.
2.3. Closed-Loop System
Under the above network constraints and event-driven schemes, the controller is defined as
The following closed-loop system can be obtained by combining (2) and (7):
The purpose of this paper is proposing suitable design methods for logarithmic quantizer such that the system (8) with the above-mentioned network-induced effects and event-driven schemes (5a) and (5b) to achieve the mean square stability.
3. Stability Analysis for the System (8) under Event-Driven Scheme (5a)
The aim of this section is to design suitable quantizer parameters including , , , and , under which the mean square stability of the system (8) under event-driven scheme (5a) is guaranteed. To obtain this purpose, we rewrite (8) as with and .
Theorem 2. Assume that the packet loss rate is given, if there are positive-definite matrix and matrix satisfyingthen the feedback matrix defined by with can ensure that the system (8) is mean square stability under event-driven scheme (5a) if quantizer parameters , , and , are selected satisfying andfor any given , where and are, respectively, given as and with . Moreover, the variables and are determined by (15).
Proof. Let . Taking the fact into consideration, we get by the inequalities and for any matrices , , and positive-definite matrix . Combine with (12) and the facts and , which directly derives that If the linear matrix inequality (10) holds, then it must give a positive-definite matrix satisfying . Let ; then the inequality (13) gives us Zooming-Out Stage. In this stage, it set the system as uncontrolled form. The aim is designing a suitable quantizer parameter such that it gives a time instant , at which the system state is unsaturated, that is, . To achieve this purpose, we let , , where is an arbitrary given positive real number and the selection of satisfies . Hence there must exist a time instant satisfying indicating that and thus hold, which accomplishes the goals of this stage.
Zooming-In Stage. At time instant , zoom strategy is transferred to zooming-in stage, and the mode of the system is changed from open-loop to closed-loop. The aim of this stage is to design appropriate quantizer parameters and time instants such that . In order to obtain this aim, we first prove that is an invariant set. In fact, if , then , and thus and hold for any by using . Hence we get Moreover, event-driven scheme (5a) tells us Combined with (14), (17), and (18), we see Thus the set defined by with is obviously an invariant set. If we define and select satisfying , where , then it must be , and hence is an invariant set.
By defining and , we will next prove with if is selected as . In order to achieve this purpose, we will first prove If we assume that (24) is unsatisfied, this means Defining the set as then it must be . This gives the fact that is an invariant set, which combined with (25) tells us and thus for any . Similar to the analysis of (14)–(19), we know that for any . Hence we get But (16) and (25) give us which contradicts formula (29). So the inequality (24) must be established, which means indicating that (23) holds. Moreover, we define as Similar to the analysis of , it is easy to see that is an invariant set and , . Similarly, for any , by definingwe get that , holds, and thus , Taking the limit on both sides of above inequality, it gives . This completes the proof.
4. Stability Analysis for System (8) under Event-Driven Scheme (5b)
In order to design quantizer parameters , , , and such that the system (8) under event-driven scheme (5b) is mean square stability, we first rewrite the system (8) as with , by which the following theorem can be derived.
Theorem 3. We assume that the packet loss rate is fixed, if it gives positive-definite matrix and matrix such thatthen the mean square stability of the system (8) with event-driven scheme (5b) under the feedback matrix defined by with can be ensured if quantizer parameters , , and , , satisfy andfor any given , where and are, respectively, given as and with . Moreover, the variables and are also described by (15).
Proof. Set ; direct calculations give If (35) is satisfied, there must exist a positive-definite matrix such that Let ; from (37) and (38) we can easily get As can be seen from the above formula, there is no need to discuss the impact of quantization errors on system stability in the following analysis. However, since the quantizer has a saturation boundary, the distance between and may always be greater than . In order to ensure that events under the event-driven scheme (5b) are not triggered frequently, it is necessary to discuss and restrain the range of values for in the analysis process. Based on this, the selection method of in the zooming-out stage of Theorem 2 is adopted here to ensure the boundedness of ; that is, Combining with (39) and (40) gives us With the analytical ideas of Theorem 2, we obtain that if variables , , , and , are defined as and , , then we get and thus . The proof is completed.
5. Analysis and Comparison
The aim of this section is analyzing and comparing the above two theorems, which includes the following four items.
() It can be seen from the proof process of above two theorems that the two event-driven schemes (5a) and (5b) proposed here can effectively simplify the system stability analysis, which means that (5a) and (5b) are with specific high practicability. In addition, from the theorem derivation process, we see that the analysis process under the event-driven scheme (5b) is simpler than the one under (5a).
() Both the parameters given in Theorems 2 and 3 can ensure the system stability. But, by comparing conditions (10) and (35), we know that the result of Theorem 2 must be more conservative than the one of Theorem 3 from the point of view of the maximum allowable packet loss rate.
() Contrasting (11a), (11b), (36a), and (36b) gives the fact that, in general, defined in Theorem 3 is less (greater) than the one in Theorem 2. This means that the decreasing frequency of the mean squared value of the system state in Theorem 3 is slower than that of Theorem 2, but the decreasing rate in Theorem 3 is higher than the one in Theorem 2. Therefore, since the convergence time is affected by the combined effect of the decreasing frequency and the decreasing rate, we can not directly compare the magnitude of the convergence time between Theorems 2 and 3. The convergence time will be compared by simulation in Section 6.
(4) Obviously, for any given , if defined in event-driven schemes (5a) and (5b) is changed as then the stability analysis process can still be simplified. Taking Theorem 3 as an example, if is defined by (44), then the variables and defined by (36a) and (36b) should be changed as which means that if , comparing with the case of , we have
(i) the range of values of becoming smaller;
(ii) the decreasing frequency of the mean squared value of the state becoming faster; that is, decreases;
(iii) The decreasing rate of the mean squared value of the state decreasing; that is, increases.
Summarized above, comparing with Theorem 3, if is defined by (44) with , then item (i) above tells us that the range of values of quantizer parameters is more conservative. Moreover, items (ii) and (iii) show that the increasing of is a double-edged sword, and thus the value of and the actual convergence time can not be established a direct proportional relationship. To illustrate the effect of the variable on the system stability, we will compare the state mean square curve and analyze the convergence speed under different values of in the simulation.
6. Simulation
A well-known benchmark example shown below is adopted here to represent the usefulness of the main results: with By selecting the discretization interval as , the continuous-time model above can be converted to the discrete-time one given as with If the initial state is selected as , and other variables are selected as , , and given in Table 1, we can obtain the variables including , , , and the applicable scope of and under Theorems 2 and 3, respectively. The total transmission times between the 9000 iterations can also be obtained. Moreover, the mean square values of the states can be shown in Figure 1, from which the mean square stability is guaranteed, and the convergence time can also be estimated.
Comparison of Theorems 2 and 3. From Table 1, we see that
(i) the applicable scope of Theorem 3 is larger than the one of Theorem 2 due to the fact that it can tolerate the smaller coding length and the larger packet loss rate ;
(ii) the convergence time under Theorem 3 is less than the one under Theorem 2, which means that Theorem 3 is better than Theorem 2 from the system performance point of view. However, Theorem 3 needs more transmission times than Theorem 2 to ensure the stability, which represents the fact that Theorem 2 is superior to Theorem 3 in terms of saving the network bandwidth.
Comparison of System Performance under Different Values of . To illustrate the effect of different values of defined in (44) on system performance, we select and show the relating variables and state trajectories in Table 2 and Figure 2, respectively.
From Table 2, we know the following:
(i) With the increase of , the applicable scope of the theorem is reduced; that is, the allowable lower bound of is increased.
(ii) When increases, the decreasing interval decreases but the decreasing rate increases, under the coupling actions of which the convergence time is prolonged. However, the transmission times are reduced as long as increases. Therefore, in the premise of ensuring the convergence of the system, if it takes the minimum transmission times as the goal, then the value of should be increased. Otherwise, if the goal is increasing convergence rate or the applicable scope of the algorithm, then should be reduced appropriately.
7. Conclusion
The mean square stability of NCSs with data quantization, packet losses, and event-driven communication is studied in this paper. Some sufficient conditions are given to ensure the mean square stability of the closed-loop system. From Figures 1 and 2, it is seen that the convergence speed of the state trajectory is low due to the large value of ; how to improve it is one of our future research directions.
Conflicts of Interest
The author declares that they have no conflicts of interest.
Acknowledgments
The work was supported by the National Natural Science Foundation of China (61403125, 61773154, 41501562, 61304258, 61503123, and U1504616), the Natural Science Foundation of Henan Province Education Department (15A413012), and the Fundamental Research Funds for the Henan Provincial Colleges and Universities in Henan University of Technology (2015RCJH15).