The problem of iterative learning control (ILC) is considered for a class of time-varying systems with random packet dropouts. It is assumed that an ILC scheme is implemented via a remote control system and that packet dropout occurs during the packet transmission between the ILC controller and the actuator of remote plant. The packet dropout is viewed as a binary switching sequence which is subject to the Bernoulli distribution. In order to eliminate the effect of packet dropouts on the convergence property of output error, the hold-input scheme is adopted to compensate the packet dropout at the actuator. It is shown that the hold-input scheme with average ILC can achieve asymptotical convergence along the iteration axis for discrete time-varying linear system. Numerical examples are provided to show the effectiveness of the proposed method.

1. Introduction

Iterative learning control (ILC) is an attractive technique when dealing with systems that execute the same task repeatedly over a finite time interval [1]. This technique has been the center of interest of many researchers over the two decades [25] and covered a wide scope of research issues such as model uncertainty [68], disturbance uncertainty and stochastic noise [9], the initial condition and desired trajectory uncertainty [1012], continuous-time nonlinear system control [13], and parameter interval uncertainty [14].

On the other hand, the remote control systems have been the focus of several research studies over the last few years [1521]. In the remote control systems, one feature is that the control loops are closed through a real-time communication channel which transmits signals from the sensors to the controller and from the controller to the actuators [17]. The remote control systems eliminate unnecessary wiring reducing the complexity and overall cost in designing and implementing the control systems. However, the introduction of communication networks makes the analysis and control design more complicated than classical feedback loops. Data packet dropout can randomly occur due to node failure or network congestion and impose one of the most important issues in remote control systems. In [18, 19], the authors are concerned with the stability problem for remote control systems with the packet dropout. In the work [20, 21], decentralized stabilization of remote control systems with nonlinear perturbations is studied.

Besides the stability issue, trajectory tracking is a challenging issue for remote control systems. Fortunately, for periodic systems, iterative learning control offers a systematic design that can improve the tracking performance by iterations in a fixed time interval. ILC is in principle a feedforward technique; thus it can send the controller signals obtained from previous trials. It is still an open research area in ILC which is implemented via a remote systems setting, except for certain pioneer works [2229]. In [22, 23], the authors designed an optimal ILC controller for a class of linear systems with random packet dropouts. Bu et al. [26] studied the stability of first and high order ILC with data dropout when the plant is subject to measurement signal dropout. In [24, 25], the authors investigated the implementation of ILC in a remote control systems environment and specifically focused on compensation when both random data dropouts and delays occur at the communication network between the sensors and the controller. In [27], a sampled-data ILC approach was proposed for a class of nonlinear remote control systems to analyze the effect of packet loss. In [28], the author considered the problem of ILC for a class of nonlinear systems with control signal dropouts and measurement signal dropouts, but the convergence analysis needs controller and actuator to know the received signal whether lost or not. Huang and Fang [29] discussed the wireless remote iterative learning control system with random data dropouts.

In this paper, we proposed an ILC for a time-varying system with random packet dropouts. As depicted in previous studies [2229], there are two different kinds of packet dropouts in remote ILC systems: control input signal dropouts and output measurement signal dropouts. For the sake of convenience, we only consider the control signal dropouts in this paper, but the results can be extended to the measurement signal dropouts. The packet dropouts would be described as a binary sequence which is subject to a Bernoulli distribution taking the value of one or zero with certain probability. The ILC law adopts an iteration-average operator and a revised learning gain that takes into consideration the probabilities of data-dropout factors. As a result, the ensemble average of the output tracking errors can be made to converge along the iteration axis. In this paper, we consider a class of discrete time linear plants with output matrix and input matrix ; our results refer only to of full-column rank.

The paper is organized as follows. Section 2 formulates the system problem. Section 3 formulates the hold-input scheme with average ILC algorithm and proves the convergence property of ILC for linear varying discrete-time plants. Section 4 presents numerical examples, and Section 5 draws the conclusions.

2. Problem Formulation

Consider the ILC system with network communication depicted in Figure 1. The discrete time linear plant with actuators and sensors is described by where denotes the iteration index; is a given finite time; , , and are state, control, and output, respectively; is unknown matrix, while and are known; and are random noises with and ; for all , the initial state is a random variable of with a fixed point . Assume that has full-column rank. The discrete time controller consists of a ILC algorithm and a memory. The controller and the actuators are connected via a communication network through which the controller transmits data to the actuators, while the controller is directly connected to the sensors. The plant and the controller are assumed to be time driven and synchronized.

At each of the th iteration stage, the controller output is computed, the controller transmits to the actuators through the network. The transmission may succeed or fail. For a successful transmission, it is assumed that the transmission delay through the network is negligible. With the negligible delay, the actuators can employ , when is transmitted successfully. Of course, when the transmission fails, the actuators receive no and have to employ (this paper prescribes ). Overall, the scheme of actuators is where Specially, this paper assumes that, for all , for all , is a random variable of with a constant as well as that and are independent either when or when . In addition, TCP-like protocol is assumed, in which there is an acknowledgment for a successful transmission, and hence the controller has indicators of whether the current controller output is received or not by the actuators.

Assumption 1. Given an output reference trajectory , which is realizable; that is, there exists a unique desired control input such that

The purpose of this paper is to design an iterative learning control law for the above plant with network communication such that tracks as closely as possible when is large enough.

3. ILC Algorithms and Convergence Analysis

Denote . The control law is a D-type ILC with average operator that employs updating mechanism: where the gain matrix . From (2) and (5), the hold-input scheme with average ILC is expressed as Define the input and state errors And subtracting from both sides of (6) yields where (this paper prescribes and hence ) . Noticing that is independent of and and taking expectation on both sides of (8), we have

Expanding expression (9) from to , we have The above expression can be arranged later below (this paper prescribes when )

From (1) and (4), we have

Taking expectation on both sides of (12) and expanding expression from to , we obtain The above expression can be arranged later (this paper prescribes when )

For any and any , denote

Lemma 2. For all , for all , and for all ,

Proof. From (17), we have

Theorem 3. For the system with network communication described in Section 2 and the iterative learning controller (5), suppose Then for all , for all , there exist and such that

Proof. From definition of average operator, note the relation
Applying the ensemble operator to both sides of (21) and substituting the relationship (8), we can obtain Substituting (11) into (22) leads to the following relationship:
Now let us handle the third term on the right hand side of (23); we will express with . Substituting the state error dynamics (14) into (23) leads to the following relationship: where .
Next, combining analogous terms on the right hand of (22), we obtain The relationship (25) can be rewritten as follows: To simplify expression of , , and , we choose and such that where
Taking -norm on both sides of (26), we obtain
Using Lemma 2, it can be proved that, for all, , for all , and for all ,
Combining Lemma 2, (31) and (32) yields There exists such that when . Now for , (28) and (33) imply that Consequently, we obtain
According to the relationship (14) between the input error and output error, we have
Similar to the proof of Lemma 2, one can prove that
Finally, from (29), (35), and (37), we can obtain because lambda can be chosen arbitrarily large in (38).
This completes the proof.

Remark 4. In this paper, we consider D-type iterative learning control with average operator, and the result obtained can be extended to P-type iterative learning control with average operator.

4. Numerical Examples

In this simulation test, let us consider system (1) and matrices given by The random noises and have uniform distribution on the intervals and , respectively. In this control problem, the desired output trajectory for is given to be , and the initial states and have uniform distribution on the intervals and , respectively. The fixed time interval is . The control profile of the first iteration is . Random packet dropout in controller-actuator channel is subject to Bernoulli distribution of expected value ( means transmission success while means transmission failure).

For expected value , we compare our algorithm with the other algorithms.

Algorithm 1 (classic ). The control signal is constructed as with satisfying .

Algorithm 2 (zero-input scheme with average ILC). The control signal is constructed as with .

Algorithm 3. Now, we consider the proposed algorithm. From (2) and (5), the control signal is constructed as where the learning gain and expected value .

As shown in Figure 2, the tracking error profiles for the proposed algorithm are much lower than the other two algorithms with 5% packet dropout. In Figure 3, the mathematical expectation of the tracking error versus iterations is shown, and the proposed hold-input scheme with average ILC achieves the convergent performance.

5. Conclusion

In this work we address a remote control system problem with random packet dropout in controller-actuator channel. The hold-input scheme with average ILC is applied to handle this remote control problem with repeated tracking tasks. Through analysis we illustrate the desired convergence property of the hold-input scheme with average ILC. In our future work, we will also explore the extension to more generic stochastic process such as Markov chain.


: The set of all real numbers
: The set of all positive integers
: The average operator
: The expected value of a random variable
: The probability of an event
: The maximal singular value of a matrix
: The Euclidean norm of a vector
: Identity matrix of appropriate dimensions
: Zero matrix of appropriate dimensions.


This work is supported by the 973 program of China (Grant no. 2009CB320603).