Wireless Sensor/Actuator Network Design for Mobile Control Applications

Published in: Sensors , vol.7, no.10, pp.2157-2173, 2007. Open Access at http://www.mdpi.org/sensors/papers/s7102157.pdf Wireless Sensor/Actuator Network Design for Mobile Control Applications Feng Xia 1,2 , Yu-Chu Tian 2 , Yanjun Li 1 and Youxian Sun 1 1 State Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou 310027, China E-mail: f.xia@ieee.org. 2 Faculty of Information Technology, Queensland University of Technology, GPO Box 2434, Brisbane QLD 4001, Australia E-mail: y.tian@qut.edu.au. Abstract: Wireless sensor/actuator networks (WSANs) are em erging as a new generation of sensor networks. Serving as the backbone of control applications, WSANs will enable an unprecedented degree of distributed and mob ile control. However, the unreliability of wireless communications and the real-tim e require ments of control applications raise great challenges for WSAN design. W ith emphasis on the reliability issue, this paper presents an application-level design methodology for WSANs in m obile control applications. The solution is generic in that it is independent of the underlying platforms, environment, control system models, and controller design. To capture the link quality characteristics in terms of packet loss rate, experiments ar e conducted on a real W SAN system. From the experimental observations, a simple yet e fficient method is proposed to deal with unpredictable packet loss on actuator nodes. Trace-based simulations give prom ising results, which demonstrate the effectiveness of the proposed approach. Keywords: wireless sensor/actuator network, sens or network, control application, link quality, packet loss. 1. Introduction Recent advances in pervasive computing, comm unication and sensing technologies are leading to the emergence of wireless sensor/actuator networks (W SANs) [1,2]. A WSAN is a distributed system of sensor nodes and actuator nodes that are inte rconnected over wireless links. Sensors gather information about the physical world, e.g., the envi ronment or physical system s, and transmit the collected data to controllers/actuators thr ough single-hop or multi-hop com munications. From the received information, the controllers/actuators pe rform actions to change the behaviour of the environment or physical systems. In this way, rem ote , distributed interactions with the physical world are facilitated. Depending on the type of the targ et application, nodes in a WSAN can be either stationary or mobile. In many situations, however, sensor nodes are stationary whereas actuator nodes, e.g., mobile robots and unmanned aerial vehicles, are m obile. Sensor nodes are usually low-cost, low- power, small devices equipped with limited sens ing, data processing and wireless com munication capabilities, while actuator nodes typically have stronger computation and comm unication powers and more energy budget that allows longer battery life [3 ]. Regardless, resource constraints apply to both sensors and actuators. WSANs are not just an enhancement or com plement to the intensively-investigated wireless sensor networks (WSNs) [4-8], but go beyond. They are a new generation of sensor networks [2,3]. While WSANs and WS Ns share many common considera tions concerning network design, such as reliability, connectivity, scalability and energy efficien cy, the coexistence of sensors and actuators in WSANs causes substantial difference between these two types of networks. Applications in which some actions are introduced for the purpose of enhancing the m onitoring capability of the sensor networks do not embody the essential characteristics of WSANs. On the contrary, actuators in a WSAN should be an integral part of the network and perform actions interacting with the physical world. As a consequence, WSANs have the ability to change the physical world, but W SNs do not. In WSNs, power consumption is generally the prim ary concern; however, this may not be the case in some WSANs where m eeting the real-time, re liable com munication requirements m ay be more important [9]. Although there are many situations in which only WSNs are required, for exam ple, environment monitoring, product quality monitoring, and the like, th ere are an increasing number of applications that necessitate the use of actuators along with sens ors [10-12]. That is, the network system needs to interact with the physical system or environment. Examples of application areas of WSANs include disaster relief operations, intelligent building, hom e automation, sm art spaces, pervasive computing systems, and cyber-physical systems. Because of the use of both sensors and actuato rs, WSANs, by definition, exploit the m ethodology of feedback, which has been recognized as the central element of control systems [13]. The advent of WSANs has the potential to revolutionarily promote existing control applications. It can be envisioned that WSANs will become the backbone of m any control applications enabling an unprecedented degree of distributed control. The use of WSANs in control applications has m any advantages compared to wired solutions, which are dominant at the m oment [14,15]. For instance, WSANs allow more flexible installation and maintenance, fully m obile operation, and monitoring and control of equipments in hazardous and previously difficult- to-access environments. Another im portant factor that instigates the deployment of W SANs is their relatively low costs [11]. Despite many advantages, WSANs also raise challe nges for control applications. Wireless channels have adverse properties, such as path loss, multi- path fading, adjacent channel interference, Doppler shifts, and half-duplex operations [16]. WSANs are known to be notoriously unpredictable and inherently unreliable. This is especially true in the case of low-power comm unications and in the presence of node mobility. With these characteristic s, the quality of servi ce (QoS) of the network cannot be always guaranteed. A natural result is that control applications will suffer from time-varying delay and packet loss, both of which could signifi cantly degrade the control performance, or even cause system instability. Therefore, W SANs must be well designed when deployed to control applications. The design of WSANs featuring node mobility is inves tigated in this paper for control applications. The overall goal is to enhance the reliability of WS ANs so that the required perform ance of control applications is guaranteed in dynamic, lossy environm ents. In particular, our focus is on dealing with unpredictable packet loss caused by unreliable link quality in mobile WSANs, without considering the effects of time-varying delay. The reasons behind th is choice of focus can be briefly explained as follows. Firstly, since a packet loss can be equivale ntly regarded as a delay with a magnitude of infinity, from the viewpoint of cont rol, packet loss is a factor that often has more significant im pact on the resulting control performance than delay. Secondl y, several methods have been presented in the literature to cope with time-varying delay in cont rol loops closed over wireless sensor networks [17- 21], while the packet loss problem that arises in WSANs is yet to be investigated. An application-level design methodology will be de veloped for WS ANs based on an experimental study of the link quality properties and a compensation method for packet loss. The m ain contributions of this paper include: • The link quality of WSANs is characterized in te rm s of packet loss rate through experiments on a real deployment. The experimental results provide im portant insight into how the WSAN should be designed from the application point of vi ew. Moreover, they are al so of great value to the design and evaluation of sensor network protocols and algorithms. • A simple yet efficient method is developed to deal with unpredictable packet loss at actuator nodes. It can significantly improve the QoS of WSANs under unreliable channel conditions, and facilitates the implem entation of (mobile) control applications over WSANs. • The proposed approach is evaluated and verified using trace-based simulations that extract data from the real experiments. In this way, real ch aracteristics of the wireless links are taken into account in performance evaluation. Promising results are presented and analyzed. The proposed design methodology is a generic solution in the sense that: 1) It does not require any modification to low layers such as physical layer, MAC layer, and transport layer within the network protocol stack. Only the application layer is invol ved. Therefore, it is independent of the underlying communication protocols, fo r example, MAC and rou ting protocols, utilized in the WSAN. 2) It does not require any knowledge about the models of the phys ical systems to be controlled or the design of the control algorithms. It is suitable for a wide range of control applications. 3) The proposed algorithm is computationally cheap yielding only a small runtime overhead. This m eets well the general WSAN design requirements stem ming from th e constraints on data processing capacity and energy consumption in actuator nodes, thus making the proposed approach applicable to various WSAN platforms subject to resource constraints. This paper is organized as follows. Section 2 brie fly reviews related work with respect to WSAN, control over WSNs, link quality analysis, and packet loss handling. Section 3 discusses the architecture and design challenges of WSANs from an application perspective. In Section 4, the properties of link quality are captured in terms of packet loss rate through experim ents on a real WSAN. Aiming to improve the reliability and QoS of WSANs, Secti on 5 proposes a m ethod to handle packet loss. Section 6 evaluates the performance of the proposed approach using trace-based simulations. Finally, Section 7 concludes the paper. 2. Related Work While significant effort has been made in resear ch and development of WSNs in recent years and tremendous advancements have been achieved w ith respect to deploym ent, localization, MAC protocols, power control, topology control, routi ng, distributed signal processing, and security [22], WSANs are a relatively new research area with limited progress. Akyildiz and Kasim oglu [1] described research challenges for coordination and communication problem s in WSANs. Rezgui and Eltoweissy [2] discussed the opportunities and ch allenges for service-oriented sensor/actuator networks. Ngai et al. [23] studied the route design problem for m obile actuators and developed a practical algorithm to reduce the waiting tim e of sensors. Melodia et al. [3] presented a sensor-actuator coordination model based on an event-driven partitioning paradigm. Sikka et al. [24] deployed a large heterogeneous WSAN on a working farm to explore sens or network applications that can help manage large-scale farming systems. A power-aware m any- to-many routing protocol can be found in [9]. Despite their contributions in WSAN, none of the ne tworks designed in these works are particularly for real-time control applications. Sensor networks have started to attract the attention of control engineers. Kumar et al. [10] developed distributed ad-hoc network algorithms to facilitate executing control procedures in a distributed manner. Li [11] prot otyped a light monitoring and control application as a case study of WSANs. Oh et al. [25] illustrate the m ain challenges in developing real-time control systems f or pursuit-evasion games using a large-scale sensor network. A mixed m odel for design, analysis, and synthesis of control algorithms within sensor networks has been presented in [26]. Korber et al. [27] dealt with some of the design i ssues of a highly modular and scalable im plementation of a WS AN for factory automation applications. Considering networked control systems (NCSs) over W SNs, Nikolakopoulos et al. [17] developed a gain scheduler to cope with tim e-varying delay induced by dynamic changes in the number of hops in m ulti-hop communications. W itrant et al. [18] also considered the effect of time-varying dela y caused by m ulti-hop communication, and proposed a predictive control scheme with a delay estimator. Various design challenges associated with control over wireless networks have been addressed in these pa pers, but the im pact of p acket loss as a result of unreliable communications in W SANs, particularly those with mobile nodes, on the performance of the control applications remains an open issue, and needs to be investigated systematically. Experiments have been conducted for analysis of the link quality in sensor networks, e.g., [28-31]. The authors reported, respectively, their measurements of the packet delivery performance of sensor networks of different sizes in di fferent environments. Some im portant characteristics with respect to, e.g., packet loss rate in WSANs, have been capture d in these papers; but none of the analysis has intended for real-time control applications. For exam ple, the relationship between the resulting control performance and the link quality has not yet been characterised. Also, no methods have been developed in these papers to address the observed unreliable packet delivery. In the control community, effort has been m ade for packet loss compensation. A recent survey on this topic can be found in [32]. Despite their diffe rences, most of existing packet loss comp ensation methods have the comm on features that: 1) they depend heavily on the knowledge about the accurate models of the physical systems to be controlle d, and, possibly, the controller design; and 2) the relevant algorithms are computationally intensive. Du e to these reasons, they are impractical for real systems lacking well-established mathem atical models. In particular, they are not the desirable solutions for resource-constrained WSANs because of too large computational overheads. In summary, the field of W SANs is emerging; but the full potential of W SANs for control applications is yet to be explored. For this purpose, the characteristics of the link quality of real-world WSANs in terms of packet loss rate, which may significantly affect the performance of control applications, should be analysed. Resource-effici ent paradigms addressing the packet loss problem need to be developed when designing WSANs for (mobile) control applications. 3. WSAN for Control Applications This section describes the architecture of WSANs as a backbone for constructing control applications. The main challenges in design of W SANs will also be discussed briefly. In general, there are three essential components in a W SAN: sensors, actuators, and base stations. The roles of sensors and actuators have been descri bed previously, while the base stations are often responsible for monitoring and managing the overall network through com munications with sensors and actuators. Depending on whether or not there are explicit controller entities within the network, two types of architectures of WSANs for control app lications can be distinguished, as shown in Fig. 1 and Fig. 2, respectively. These two architectures are called automated architecture and semi-autom ated architecture, respectively, in [1]. In the first type of architecture as shown in Fi g. 1(a), there is no explicit controller entity in the WSAN. In this case, controllers are embedded into the actuators and control algorithm s for making decisions on what actions should be performed upon the physical systems will be executed on the actuator nodes. The data gathered by sensors will be transmitted directly to the corresponding actuators via single-hop or multi-hop comm unications. The act uators then process all incoming data by executing pre-designed control algorithms and pe rform appropriate actions. From the control perspective, the actuator nodes serve as not only the act uators but also the controllers in control loops. From a high-level view, wireless comm unications over WSANs are involved only in transm itting the sensed data from sensors to actuators; contro l comm ands do not need to experience any wireless transmission because the controllers and the actuators are logically integrated, as shown in Fig. 1(b). Base Station Sensor Actuator Physical System Actuator Sensor Controller WSAN (a) Network topology (b) Abstraction of control application Figure 1. WSAN Architecture without explicit controllers. Fig. 2(a) shows the second type of architecture, in which one or more controller entities explicitly exist in the WSAN. The controller entities could be functional modules embedded in the base stations or separated nodes equipped with sufficient com putation and com munication capacities. With this architecture, sensors send the collected data to the controller entities. The controller entities then execute certain control algorithms to produce cont rol comm ands and send them to actuators. Finally, the actuators perform the actions. In this context, both the sensor data and control com mands need to be transmitted wirelessly in a single-hop or multi-hop fa shion. A high-level view of the applications of this architecture is depicted in Fig. 2(b). Base Station Sensor Actuator Physical System Actuator Sensor Controller WSAN (a) Network topology (b) Abstraction of control application Controller Figure 2. WSAN Architecture with explicit controllers. In combination with the unique characteristics of WS ANs, control applications pose the following main challenges associated with the design of WSANs: • Reliability . From the control perspective, packet lo ss degrades control perform ance and even causes system instability. Because practical control applications can only tolerate occasional packet losses with a certain upper bound of a llowable packet loss rate, WSAN design should minimize the occurrence of packet losses as m uch as possible. Ideally, every packet should be transmitted successfully from the source to the de stination without loss. However, due to m any factors such as low-power radio comm unication, variable transm it power, multi-hop transmission, noise, radio interference, and node mobility, packet loss cannot be com pletely avoided in WSANs. The challenge then becomes how to im prove the reliability of the network system in the presence of packet loss. • Real-time constraint . Control systems are inherently real-time system s in the sense that control actions must be performed on the physical syst em s by their deadlines. It is worth mentioning that real-time does not necessarily mean ‘fast’ . For real-time control applications, both delay and its jitter should be limited and predictable in favor of control perform ance improvement. However, the use of dynamic routing protoc ols and random MAC protocols (e.g., CSMA/CA), as well as the mobility of nodes, makes the WSAN-induced delay time-varying and unpredictable. The challenge here is how to guarantee the delay is sufficiently small and deterministic with small jitter so that it will not significantly degrade the control performance. As mentioned in Section 2, WSAN-induced tim e-va rying delay has been addressed in [17,18]. In the recent work by Tian and colleagues [19-21], a r eal-time queuing protocol has been developed that can be used to tackle the second design challenge of WSANs. Therefore, in this paper, we concentrate our attention on the first design challenge concer ning the reliability of WSANs. The existence of mobile nodes in the system undoubtedly m akes this task even more difficult. In the following, we will restrict our description to WSANs with the first type of architecture as shown in Fig. 1, since it is more resource-efficient a nd is m ore representative of the next generation of sensor networks. It is, however, noteworthy that our design method is applicable to a wide range of WSANs with arbitrary architectures. Typical exampl es of (ongoing) real-world application scenarios of the considered WSANs include the pollution sour ce location problem and the fire in a road tunnel scenario where mobile robots must be controlled in a W SAN to accomplish certain jobs [33]. W hen illustrating our approach, we will exploit a high- level abstraction of the WSANs fo r various application setups in order to maintain the app lication independency and wide applicability of our solution. 4. Experimental Analysis of Link Quality In order to address the challenge of unreliable communication in W SANs, it is necessary to understand first how unreliable practical WSANs really are. That is, the packet loss behavior of the network should be studied. This is done in this work through conducting experiments on a real WSAN system and collecting quantitative data to capture th e channel characteristics. This section reports our experimental measurem ents that characterize the link quality in terms of packet loss rate of a practical WSAN, since in this context the packet loss rate is a critical factor affecting the perform ance of the control applications. Instead of an exhaustive study of link quality with respect to a lot of factors, e.g., platform, environment, deploym ent, time, etc., which has been done e.g. in [27-30], we focus on a simple yet sufficiently illustrative characterizati on of the link quality that help make decisions concerning WSAN design for mobile control applications. 4.1. Experimental Setup The sensor nodes used in the experiments is th e MICA2 m otes from Crossbow [34]. The MICA2 mote, designed specifically for deeply embedded se nsor networks, is based on the Atm el ATmega128L microcontroller. Each sensor node contains 128K B program flash memory, 512KB m easurement flash, and 4KB configuration EEPROM. MICA2 uses the Chipcon CC1000 wireless transceiver, and supports multiple channels (868/916 MHz), hardware encoding (Manchester), frequency shit keying (FSK) modulation, and up to 38.4 kbps data rate. The RF power of MICA2 is programm able from -20 to +5 dBm, and the receive sensitivty is -98 dB m. The MICA2 51-pin expansion connector supports analog inputs, digital I/O, I2C, SPI and UART interf aces. These interfaces make it easy to connect to a wide variety of external peripherals. Any MICA2 Mote can function as a base station when it is connected to a standard PC (personal computer) in terface or gateway board. A base station allows the aggregation of sensor network data ont o a PC or other computer platform. MICA2 runs TinyOS, an open-source embedded opera ting system developed at UC Berkeley. It provides basic system services, such as comm uni cation and simple process scheduling, and access to hardware components such as sensors and actuato rs. The MAC layer im plements a simple CSMA/CA protocol. A link-level acknowledgement can be sent by the receiver for each successful packet. Base Station MICA2 Motes PC Figure 3. Experimental deployment. The experiments are conducted on an open gr ound, as shown in Fig. 3. Nodes are placed equidistantly along a line with a spacing of 0.5m. One PC is connected with the base station using a MIB510 mote interface board. The PC is used to configure network param eters and collect experimental data via the base station. In the e xperim ent, the first node, i.e. the mote on the position marked as ‘1’ in Fig. 3, transmits continuously a 13 byt e data packet at a rate of 8 packets per second. The packet loss rates associated with the remaini ng nodes with different distances from the first node are measured. 4.2. Observations In a WSAN with mobile (actuator) nodes, the distan ce between the mobile actuator and the sensor that are involved in transmitting the sensed data w ill change over tim e, thus aggravating the variability of link quality. This is why we pay a special atte ntion to examining how the link quality varies over distance. Fig. 4 shows the packet loss rates on different nodes when the transmit power is set to 0dBm. Within the distance of 7m, the packet loss rate s rem ain less than 10%. Beyond 30m, the packet loss rates are fairly close to 100%, implying that almost a ll packets sent to the nodes are lost. This indicates that the radio range is approximately 30m. For links with distances ranging from 7 to 30m, the packet loss rates could vary drastically. For example, the p acket loss rates vary nearly from 0% to 100% in the area between 9 and 13m. It can be observed that nodes far away from the transm itter may possibly undergo smaller packet loss rates than those near the transmitter. 0 5 10 15 20 25 30 35 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Distance (m) Packet Loss Rate Figure 4. Packet loss rate versus distance (0dBm). Fig. 5 plots the recorded data when the transmit power is set to be -5dBm. Sim ilarly, the packet loss rates show great variability and irregularity both ove r different distances and at a given distance. Compared to the higher power case shown in Fig. 4, alm ost 100% packet loss rates are observed at a smaller distance, and the radio range decreases to 26m. 0 5 10 15 20 25 30 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Distance (m) Packet Loss Rate Figure 5. Packet loss rate vs. distance (-5dBm). We have observed that the packet loss rates ove r the wireless channel in real W SANs are highly variable and irregular. The assumptions that link quality is exclusively based on distance, which is often made when modelling the link quality of sensor networks [35,36], sim plify system analysis but can be practically questionable. This could be fu rther justified by the fact that the link quality of WSANs depends also on many factors other than distance and transmit power, such as the environment, noise, radio frequenc y, modulation schem e, and (hardware) platform in use, just to mention a few. Due to the unpredictability of the link quality in term s of packet loss rate and the impracticability to model it accurately, it is now imperative to develop a platform-independent paradigm to enhance the reliability of WSANs unde r lossy conditions. A desirable solution should be widely applicable to diverse application scenario s with different system and environment setups. 5. Dealing with Packet Loss on Actuators To meet the above requirement, we attem pt to develop an application-level design methodology for WSANs in mobile control applications. The principles in our development are: 1) to modify only the application layer of the networks without explo iting any application-specific (lower layer) network protocols, 2) not to use any statistic information about the distribution of packet loss rate in any specific WSAN, and 3) not to use the knowledge about the m odels of the controlled physical systems and the controller design of the target control applications. We propose to use a simple yet efficient m ethod on the actuator nodes to cope with packet loss occurring in WSANs. The basic idea is: whenever a se nsor data packet is lost, the actuator will still produce a control command (usually called control input in control term s) by means of prediction from previous control command values. For a control loop within a (possibly large-s cale) system shown in Fig.1, suppose that the k- th sensor data is lost. In this case the actuator (to wh ich the last data should be sent) will calculate an estimate of the control comm and using the PID (propor tional-integral-derivative) algorithm, the most popular control algorithm in the control comm unity, as follows: 1 () ˆ () ( 1 ) (( 1 ) ( 2 ) ) k PI D ik m ui uk K uk K K uk uk m − =− =− + + − − − ∑   ( 1 ) where û ( k ) is the estimate of the k -th control comm and u ( k ), P K  , I K  , D K  and m are user-specified parameters. Using (1), the actuator predicts u ( k ) based on the previous m consecutive control commands (which are also possibly predicted values ) in the case of packet loss and perform s the actions corresponding to the value of û ( k ). Given that the accuracy of the prediction of control commands is sufficiently high, proper actions will be performed on the controlled physical system in every sampling period, regardless of the loss of the sens or data. In this way, the effect of packet loss on the performance of the control applications can be substantially reduced. In other words, the reliability of the WSAN is improved, from the application point of view. The PID algorithm is here used to predict the una vailable control comm ands that result from sensor packet loss. In the control community, in contrast, it is typically used in control system design, as will be shown later in Section 6. Varma et al. [37] have used a sim ilar method called nqPID to predict CPU workload in dynamic voltage scaling systems. It proved quite effective and insensitive to param eter changes. The work flow of the actuator can be illustrated as follows: Input : Sensor data Output : Control command Begin If the sensor data is lost then Compute û ( k ) using (1) Else Compute u ( k ) using pre-designed control algorithm(s) End if Store u ( k ) or û ( k ) in memory Discard u ( k-m ) in the mem ory Perform actions corresponding to u ( k ) or û ( k ) End It can be seen that this design method is quite simple. The m ajor overhead is a small fraction of memory to tem porarily store the previous m control comm ands. Despite this, it does not depend on any knowledge about the underlying platform, environmen t, link quality characteristics, models of the controlled systems, or controller design. Furtherm ore, only a very limited amount of computations have been introduced, which fulfills well the general requirem ents of WSANs concerning the constraints on computational capacity and energy expe nditure. In addition, it is worth mentioning that although zero delay is assumed in this paper, the proposed design method can be easily com bined with the real-time queuing protocol developed in [19-21] to deal with sim ultaneously time-varying delay and packet loss. 6. Performance Evaluation In this section, we conduct trace-based simulations using Matlab to evaluate the performance of the above-proposed design methodology for WSANs. 6.1. Control Application Overview In the simulations, a comm onplace control system desi gn is used to keep the results as general as possible. The model of the controlled physical system is given below, which may represent an inverted pendulum system, a com mon benchmark problem in the control field [38]: 32 4.546s () 0.182 31.182 4.454 Gs ss s = +− − The controllers use the PID control law, the most popular control law in practical control applications, with the following parameters: K P = 120, K I = 1000, and K D = 5. The PID control algorithm is implem ented in the actuator as follows [14]: () ( ) () () () ( 1 ) ( () ( 1 ) )2 ( ) ( ( )(1 ) ) / () () () () () P I D rk y k Pk K e k Ik Ik K h e k e k Dk ek ek h uk P k I k Dk ek K − = =− + + − =− − =+ + = / where r(k) is the desired system output (i.e ., reference input or set-point), y(k) is the sensed value of the system output (i.e., measurem ent), h is the sampling period of the sensor. In the simulations, h is set to 20ms. As the major perturbations on the controlled system, the reference input changes over tim e as a square wave with a period of 2s. To measure the perf orm ance of the control application, the integral of absolute error (IAE), one of the widely us ed control performance metric defined as 0 () | ( ) ( ) | t Jt r y d τ ττ =− ∫ , is recorded. The bigger the IAE value the worse the control performance. 6.2. Simulation Results and Analysis Because of the mobility of the actuator node, the di stance between the sensor and the actuator varies during runtime according to Fig. 6. Th e RF power of the sensor is assumed to be 0dBm . Accordingly, the packet loss rates with respect to different distances will be randomly extracted from the data set reported in Fig. 4. For instance, when the distance is 10m from t = 16 to 20s, the packet loss rates will take random values out of {0.7160, 0.2716, 0.6790, 0.9136, 0.9259, 0.6543, 0.3827, 0.6543, 0.4691, 0.3333, 0.2963, 0.1358, 0.5062, 0.6790, 0.5802}. 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 0 5 10 15 20 25 30 Time (s) Distance (m) Figure 6. Variable distance in simulations. The performance of the control application is co m pared with respect to two different WSAN design methods: 1) traditional design method without packet loss handling m echanism on the event-triggered actuator; and 2) the application-level design me thodology proposed in this paper. Som e relevant parameters are set as follows: 0.3 P K =  , 0.2 I K =  , 0.5 D K =  , and m = 3. 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 0 10 20 30 40 50 60 70 80 IAE Time (s) Traditional Design Method Our Design Method Figure 7. Control performance in terms of IAE. Fig. 7 shows the control performa nce in term s of the IAE associated with different design methods. The cumulated IAE value of the system designed us ing our m ethod is only 19.4% that of the system using traditional design method. The rapid increase in IAE from tim e t = 16s to 20s implies that the system becomes unstable during this period of tim e. This can also be seen from Fig. 8 (the upper part), where the measured/sensed system output is depict ed. The instability is m ainly caused by the large packet loss rates at the distance of 10m, which has been shown in Fig. 4. 16 16.5 17 17.5 18 18.5 19 19.5 20 −100 −50 0 50 100 System Output Traditional Design Method 16 16.5 17 17.5 18 18.5 19 19.5 20 −2 −1 0 1 2 3 System Output Our Design Method Time (s) Figure 8. System output. In contrast, the system remains stable all the tim e, when the design method proposed in this paper is employed. This is justified by the quite slow increas e in IAE throughout the simulation, see Fig. 7. As also shown in Fig. 8 (the lower pa rt), the performance of the control application is satisfactory even when the system may encounter considerably severe packet loss. Since no nodes are allowed to work beyond its radio range (i.e., 30m in this case) in prac tice, the above results demonstrate that the method proposed in this work is effective in mobile c ontrol applications where the com munication distance may change over time. 7. Conclusion This paper deals with the design of WSANs for c ontrol applications. The related design challenges have been discussed with respect to reliability a nd real-time constraints. W ith focus on improving the reliability of WSANs to provide control applications with network QoS guarantees, a generic application-level design methodology has been pr esented. The link quality of WS ANs has been examined in terms of packet loss rate through experim enting on a real WSAN system . From the experimental observations, a simple yet effec tive method has been developed to deal with unpredictable packet loss on the actuator nodes. The proposed design methodology has also been verified through trace-based simulations. It enables mobile control applications over W SANs since it can guarantee satisfactory control performance even in the presence of significant packet loss. The design methodology proposed in this paper is independent of the com putation and communication platform s upon which the WSAN is built, and the environm ent in which the WSAN is deployed. Also, it does not rely on the system mode ls and controller design of the target control applications. Furthermore, it is computationally cheap since only a sm all overhead is introduced. Therefore, the proposed design methodology can be a pplied in a wide range of WS AN-based control applications. Acknowledgements Authors Xia and Tian would like to thank Aust ralian Research Council (ARC) for its support under the Discovery Projects Grant Scheme (grant ID: DP0559111). The authors are grateful to Jing Yu at Zhejiang University for her assistance in collecting experimental data. References and Notes 1. Akyildiz, I. F.; Kasimoglu, I. H. Wireless sensor and actor networks: research challenges. Ad Hoc Netw. 2004 , 2 (4), 351-367. 2. Rezgui, A.; Eltoweissy, M. Service-oriented se nsor-actuator networks: Prom ises, challenges, and the road ahead. Computer Communications 2007 , 30 , 2627-2648. 3. Melodia, T.; Pompili, D.; Gungor, V. C.; Akyildiz, I. F. Comm unication and Coordination in Wireless Sensor and Actor Networks. IEEE Transactions on Mobile Computing 2007 , 6 (10), 1116-1129. 4. W ang, X.; Ding, L.; Bi, D.; Wang, S. Energy- efficient Optim ization of Reorganization-Enabled Wireless Sensor Networks. Sensors 2007 , 7 , 1793-1816. 5. Wang, X.; Ma, J.; Wang, S.; Bi, D. Time Se ries Forecasting Energy-efficient Organization of Wireless Sensor Networks. Sensors 2007 , 7 , 1766-1792. 6. Rosa, N.; Cunha, P. Using LOTOS for Forma lising W ireless Sensor Network Applications. Sensors 2007 , 7 , 1447-1461. 7. Wang, X.; Wang, S.; Bi, D.; Ma, J. Distributed Peer-to-Peer Target Tracking in Wireless Sensor Networks. Sensors 2007 , 7 , 1001-1027. 8. Hande, A.; Polk, T.; W alker W.; Bhatia, D. Se lf-Powered W ireless Sensor Networks for Remote Patient Monitoring in Hospitals. Sensors 2006 , 6 , 1102-1117. 9. Cayirci, E.; Coplu, T.; Emiroglu, O. Power aware many to many routing in wireless sensor and actuator networks. Proc. of the Second European Workshop on Wireless Sensor Networks 2005 , 236-245. 10. Kumar, R.; Kansal, A.; Srivastava, M. Dist ributed Control over Ad-hoc Sensor Actuator Networks. NESL Technical Report NESL-TR-Jan- 2004, Department of Electrical Engineering, UCLA, 2004 . 11. Li, S.-F. Wireless sensor actuator network for light monitoring and control application, Proc. 3rd IEEE Consumer Communications and Networking Conf. 2006 , 2 , 974- 978. 12. Pagilla, P. Real-time scheduling of sensor a nd actuator networks, Master thesis, Oklahom a State University, 2005 . 13. Murray, R. M.; Astrom , K. J.; Boyd, S. P.; Broc kett, R. W.; Stein, G. Control in an information rich world. IEEE Control Systems Magazine 2003 , 23 (2), 20-33. 14. Xia, F. Feedback scheduling of real-tim e cont rol systems with resource constraints, PhD thesis, Zhejiang University, 2006 . 15. Ploplys, N.; Kawka, P.; Alleyne, A. Closed-Loop Control over Wireless Networks. IEEE Control Systems Magazine 2004 , 24 (3), 58-71. 16. Willig, A.; Matheus, K.; W olisz, A. Wireless technology in industrial networks. Proceedings of the IEEE 2005 , 93 (6), 1130-1151. 17. Nikolakopoulos, G.; Panousopoulou, A.; Tzes A.; Lygeros, J. Multi-hopping induced gain scheduling for wireless networked control systems. Proc. of IEEE Conf. on Decision and Control (CDC05) 2005 . 18. W itrant, E.; Park, P. G.; Johansson, M.; Fisc hione C.; Johansson, K. H. Predictive control over wireless multi-hop networks. Proc. of IEEE CCA 2007 . 19. Tian, Y.-C.; Tade, M. O.; Levy, D.; Gu, T.; Fidge, C. Queuing packets in communication networks networked control systems. Proc. of the 6th World Congress on Intelligent Control and Automation (WCICA'06) 2006 , 210-214. 20. Tian, Y.-C.; Levy, D. Configuring the Worst-Case Com munication Delay in Real-Time Networked Control Systems. Proc. of the 2007 Int. Conf. on Embedded Systems and Applications 2007 , 114-120. 21. Fidge, C.; Tian, Y.-C. Functional Analysis of a Real-Time Protocol for Networked Control Systems. Lecture Notes in Computer Science 2006 , 4218 , 446-460. 22. Heidemann, J.; Govindan, R. Embedded Sensor Networks. In Handbook of Networked and Embedded Control Systems , (B. Levine, D. Hristu Ed.), Birkauser, 2005 . 23. Ngai, E. C.-H.; Liu, J.; Lyu, M. R. Delay- Minimized Route Design for W ireless Sensor-Actuator Networks. IEEE WCNC'07 2007 . 24. Sikka, P.; Corke, P.; Valencia, P.; Crossman, C.; Swain, D.; Bishop-Hurley, G. W ireless Adhoc Sensor and Actuator Networks on the Farm. Proc. of IPSN'06 2006 . 25. Oh, S.; Schenato, L.; Chen, P.; Sastry, S. Tr acking and coordination of multiple agents using sensor networks: system design, algorithms and experim ents. Proceedings of the IEEE 2007 , 95 (1), 234-254. 26. Sinopoli, B.; Sharp, C.; Schenato, L.; Schaffert, S.; Sastry, S. Distributed control applications within sensor networks. Proceedings of the IEEE 2003 , 91 (8), 1235-1246. 27. Koerber, H-J.; Wattar, H.; Scholl, G. Modular W ireless Real-Time Sensor/Actuator Network for Factory Automation Applications. IEEE Trans. on Industrial Informatics 2007 , 3 (2), 111-119. 28. Zhao, J.; Govindan, R. Understanding packet delivery performance in dense wireless sensor networks. Proc. 1st ACM Conf. Embedded Network Sensor Systems 2003 , 1-13. 29. Cerpa, A.; Busek, N.; Estrin, D. SCALE: A Tool for Simple Connectivity Assessm ent in Lossy Environments, Tech. Rep. 0021, Center for Embedded Networked Sensing, UCLA, 2003 , 1-16. 30. Woo, A.; Tong, T.; Culler, D. Taming the U nderlying Challenges of Reliable Multihop Routing in Sensor Networks, Proc. of SenSys 2003 . 31. Tang, L.; Wang, K-C.; Huang, Y.; Gu, F. Ch annel Characterization and Link Quality Assessment of IEEE 802.15.4 Compliant Radio for Factory Environments. IEEE Trans. on Industrial Informatics 2007 , 3 (2), 99-110. 32. Hespanha, J.P.; Naghshtabrizi, P.; Xu, Y. A Survey of Recent Results in Networked Control Systems. Proceedings of the IEEE 2007 , 95 (1) 138-162. 33. http://www.ist-runes.org, accessed on September 30, 2007 . 34. http://www.xbow.com, accessed on September 30, 2007 . 35. Lal, D.; Manjeshwar, A.; Herrmann, F.; Uysal- Biyikoglu, E.; Keshavarzian, A. Measurem ent and characterization of link quality metrics in en ergy constrained wireless sensor networks. Proc. Globecom 2003 , 446-452. 36. Leskovec, J.; Sarkar, P.; Guestrin, C. Modeling Link Qualities in a Sensor Network. Informatica (Slovenia) 2005 , 29 (4), 445-452. 37. Varma, A.; Ganesh, B.; Sen, M.; Choudhury, S.R.; Srinivasan, L.; Bruce, J. A control-theoretic approach to dynamic voltage scheduling. Proc. ACM CASES 2003 , 255-266. 38. http://www.library.cmu.edu/ctms/, accessed on Septem ber 30, 2007 .