Chinese Journal of Electronics Vol.20, No.3, July 2011

Smart Building Control Based on Wireless Sensor-Actuator Networks∗ SUN Yan, ZHAO Guotao and LUO Hong (School of Computer Science and Technology, Beijing University of Posts and Telecommunications, Beijing 100876, China) Abstract — Wireless sensor-actuator networks can bring flexibility to building control. We develop a building control prototype system to exhibit how to apply the wireless sensor-actuator networks technologies to the building control. The general purpose is energy saving and emission reduction through the collaboration among a mass of lowcost sensor nodes and actuator nodes. In this paper, we detail the hardware and software in our building control prototype system, and then we address the following key technical challenges: (1) a lightweight trust management framework for discrete data; (2) self-adaptive audio sampling and combining reception algorithm; (3) hierarchical synchronization strategy. Finally, a set of experiments are presented to evaluate the performance of our key technologies. Key words — Building control, Sensor, Actuator, Energy.

I. Introduction Wireless sensor-actuator networks refer to a group of sensors and actuators linked by wireless medium to perform distributed sensing and actuation tasks[1,2] . In this network, sensors gather information about the physical world, where actuators are usually longer battery life and perform actions on the environment based on the information gathered. Energy efficiency is the important factor for evaluating the performance of a modern work environment. The intelligent building control system able to make informed, autonomous decisions about energy consumption based upon real-world variables such as temperature, humility and light, required to deploy sufficient sensors and actuators to provide efficient control and sensing. Building is a harsh environment for reliable wireless communication. Traditional building control systems have suffered from robustness and reliability issues, as well as the inability to scale well in cost and network complexity. Wireless sensor-actuator networks can bring flexibility to building automation. Sensors and actuators are usually stationary in building environment, and cooperate to satisfy the requirements for saving energy without human intervention. A number of projects and research have developed building

control applications. Qiao Bing et al.[3] introduce some ongoing research on developing a multi-agent system that combines an EDA agent model, personalized space, policy management, building performance quotient, wireless sensor network, and building automation/management system to provide an intelligent work environment. A multi-agent framework[4] was described for an intelligent building control that is deployed in a commercial building equipped with sensors and effectors. J.Y. Adams[5] focus on the commercial building control environment and its needs for substantial energy consumption reduction. A practical example of green building construction on a small scale, with some level of linking between the disparate systems, is given. F.L. Zucatto et al.[6] describe some important aspects of Zigbee standards and the main directives of the stack profile. The network topology that have been selected for building control system, are also presented. Wireless sensor-actuator networks are introduced into Ref.[7], the authors present a wireless building control platform that provides a unified method to link up different kinds of building control applications. The platform is intended for low-power and low data rate autonomous communication between sensors and actuators. In this paper, we describe the building control prototype system deployed in our school building based on the wireless sensor-actuator networks in detail. In the building control prototype system, we first introduce the multi-layer architecture of the wireless sensor-actuator networks, in which multimedia information collected is transferred from sensor nodes to the sink node by multi-hops method and the control demands are transferred from sink node to the actuator nodes hop by hop. And then we present the model structure of the node hardware which we developed, thus describe the software part including sensing, acting, configuration and display. Finally, we detail the key technologies in the building control prototype system: a lightweight trust management framework for discrete data, self-adaptive audio sampling and combining reception algorithm, and hierarchical synchronization strategy. The remainder of this paper is organized as follows. Section II describes the building control prototype system deployed in our school building based on wireless sensor-actuator

∗ Manuscript Received May 2010; Accepted June 2010. This work is supported by the National Natural Science Foundation of China (No.61070205, No.61070206, No.60833009), the National 973 Project of China (No.2011CB302701), the Program of New Century Excellent Talents in University of China (No.NCET-08-0737), and the Cosponsored Project of Beijing Committee of Education.

438

Chinese Journal of Electronics

networks, including hardware part and software part. In Section III, we detail the key technologies in our prototype system. Experimental results are illustrated in Section IV. And we finally conclude this paper in Section V.

2011

of queue model. When the command is finished, actuator returns a confirmation signal to the sink node.

II. The Building Control Prototype System Aiming to save energy and promote a healthier, more comfortable workplace, we deploy a building control prototype system in our school building based on wireless sensor-actuator networks, as shown in Fig.1.

Fig. 2. The multi-layer architecture actuator networks

of wireless sensor-

2. Hardware part

Fig. 1. The building control prototype system deployed in our school building

In this building control prototype system, we deploy many sensor nodes (such as: temperature, humility, light, audio and image sensor nodes) and actuators (such as lighting and air conditioner controllers). There exist two service modes, namely proactive mode and reactive mode. In the proactive mode, wireless multimedia sensor nodes report information to the sink node automatically. Once the sink node judge it is the decision point according to the reports, then it sends the control demands to the corresponding actuators. In the reactive mode, the nodes are controlled via the user interface. 1. Multi-layer architecture of wireless sensoractuator networks We design a multi-layer architecture of wireless sensoractuator networks in order to design the building control prototype system, as shown in Fig.2. The bottom layer is made up of several aggregation areas and each area contains several wireless multimedia sensor nodes and actuator nodes. In the path from sensor nodes to the sink node, the routing node, named as aggregator, is responsible for fusing the multisource data and transferring the aggregated results to their aggregator in the higher level. If the nth level aggregator receives the compressed data from the (n−1)th level aggregators or wireless multimedia sensors, the nth level aggregator shall decompress the data first and perform the designated aggregation algorithm. The aggregated data is then compressed and routed to the (n + 1)th level aggregator, eventually to the sink node. In the path from the sink node to sensor-actuator nodes, the configuration data from the sink node to sensor nodes and the control commands from the sink node to actuator nodes are very important, we assign them with high priority. Our aim is to find an optimal traffic scheduling mechanism in terms

As shown in Fig.3, our developed wireless multimedia sensor nodes adopt the modular structure which can be combined, including the main processing module, the image capturing module, the wireless communication module, and the power supply module. The main processing module communicates with the image capture module and the wireless communication module through its serial ports 0 and 1, respectively. The power supply module of sensor node can select lithium battery of different capacity to satisfy different application requirements.

Fig. 3. The wireless multimedia sensor node

The main processing module is based on the Atmel AT91SAM7X256 48 MHz embedded chip with 256 kbytes flash and 64 kbytes memory. In order to save the economic cost, the audio, temperature, and humidity sensors are integrated in the main processing module. The image capturing module is designed with Omnivision chips OV7640 and OV528. The low-powered CMOS sensor OV7640 supports 640*480 image resolutions, 30fps rate and the standard YUV (4:2:2) and YCbCr (4:2:2) format output. Because the chip OV528 can provide the standard JPEG compression function, it is selected to compress the images. The compressed data is transmitted to the main processing module through the standard serial port. When the image capture module works at capturing, compressing and delivering the images of 160 × 128 resolutions with the speed of 1 fps and the other sensing modules are idle, the sensor node consumes

Smart Building Control Based on Wireless Sensor-Actuator Networks about 800 mW of power. The wireless communication module is based on the TI CC2431 chip which is comprised of an industry level 8051 core, 128kbytes flash, and 8kbytes RAM. This module supports Zigbee protocol stack and has the ability of localization. IEEE 802.15.4 radios have the potential to be the cost-effective communications backbone for sensory mesh networks that can effectively harvest data with relatively low latency and the ability to survive for a very long time on small primary batteries. We use the 8051 core of the CC2431 chip to achieve automatic control in the wireless communication module. Then, the device switch with the wireless communication module becomes the actuator by adding the GPIO control signal to control the switch, as shown in the Fig.4.

Fig. 4. The wireless actuator node

3. Software part In the building control prototype system, the collaboration process of multi-layer sensing and acting is shown as Fig.5. The wireless multimedia sensors collect environment information and report them to their aggregator hop by hop, the aggregator fuse the multi-source data and transfer the aggregated results progressively, till to the sink node. The sink node shows the final aggregated results and can make some decisions according to the final aggregated results. If the sink node judge an aggregated result exceeds a pre-set threshold, then sends demands to the corresponding actuator to control electrical equipment via multi-hop. The software of the building control prototype system include four-part: sensing part, acting part, configuration part and display part.

439

lected, first, we design a lightweight trust management framework to satisfy the high reliability requirements of discrete data. Furthermore, we adopt a self-adaptive audio sampling and combining reception algorithm mentioned above which can significantly improve the bandwidth utilization of network while maintaining high accuracy of audio stream. Finally, we use a self-adaptive wake-up mechanism which simple nodes wake complex nodes so as to save energy and prevent network congestion. If the temperature reported by several temperature sensor nodes exceeds a certain threshold, the image sensor node can be waked up for detailed monitoring. (2) Acting part The building’s intelligence reflects in controlling all the actuators with commands automatically based on the information gathered by sensors, and it’s important to have flexible rules ensuring that the environment is well controlled but not out-of-control. The acting part includes automatic control, manual control, and timing control. For example in automatic control, the sink node calculates the average of temperature values reported by several sensors in a room. Once the average exceeds the maximum threshold pre-defined (too warm here), then the sink node sends control demand to the corresponding actuator in the room hop by hop. The actuator receives the control demand and analyzes it, thus turns on the heater; On the contrary, once the average falls below the minimum threshold pre-defined (too cold here), then the sink node control the actuator to turn on the heater. In the period out of working, we can use timing control to save energy. (3) Configuration part This part provides user interfaces through which users can communicate with each sensor node. As shown in Fig.6, the configuration part is used to help users to select appropriate configuration parameters (for example: node number, color type, sampling frequency), and sends these parameters to the corresponding sensor nodes. Once the sensor node receives these parameters, then modifies its corresponding registers.

Fig. 5. The collaboration process of wireless sensor-actuator networks

Fig. 6. The configuration interface

(1) Sensing part We deploy various types of sensor nodes (including light, temperature, humidity, smoke, vibration, infrared detection, audio and image sensors) to collect the corresponding environment signals according to a certain sampling period. Aiming at the diverse QoS requirements of different kind of data col-

(4) Display part We can display the distribution of sensor nodes and actuator nodes in the building control prototype system, as illustrated in Fig.7, the sensors displayed in the left list are geographically distributed. Besides that, we can query some specific wireless multimedia sensors by choosing the target node

Chinese Journal of Electronics

440

in the left list, temperature curve, humility curve, audio curve and image data from the distributed sensor can be displayed in the right list.

Fig. 7. The display interface

III. Key Technologies In designing a building control prototype system based on wireless sensor-actuator networks, we had a number of design goals that we were trying to accomplish including: Reliability requirement of discrete data The control commands and the simple data collected may be interfered by the environment noise. If that, the accuracy and reliability of the data will be affected. We should avoid this problem and design a lightweight trust management framework for discrete data. Bandwidth requirement of streaming media Realtime audio and image data is delay-constrained and has a certain bandwidth requirement. Packet losses can be tolerated to a certain extent. In the premise of guaranteed quality, we must adopt efficient methods to reduce the data traffic in the resource-constrained networks. Lifetime requirement of network Since the nodes have a small form-factor and therefore can carry only a small battery, as a result, they have a limited energy supply and low-power operation is necessary. We have to design some mechanisms with a focus on energy efficiency. A plethora of inexpensive sensors and actuators, with most tied back to the sink node via Zigbee networking technology, promise to provide the ability to rapidly deploy sensors and actuators for any environment. Zigbee and IEEE 802.15.4 are standard-based protocols that provide the network infrastructure required for wireless network applications. The stack profile is selected by the Zigbee coordinator and is chosen on the basis of application areas, such as Building Automation, Home Control or Plant Control. Next, we introduce several methods in order to achieve our design goals. 1. A lightweight trust management framework for discrete data In a multi-layer aggregation network, all sensor nodes are divided into aggregating sets. In each aggregating set, the aggregator receives the data from sensor nodes periodically, aggregates them and forwards the result to the higher level aggregator until to the sink. In our proposed trust management framework, the aggregator acts as the trust manager. It shall calculate its area trust by using the propagation trust

2011

and the area trust of its child nodes. When the area trust of an aggregating set drops below a threshold, it predicts that there are some exceptions in the data transmission or data gathering of this aggregating set. Then the aggregator will identify these exceptions, and repair them locally. 2. Self-adaptive audio sampling and combining reception algorithm The impact of packet congestion can be reduced by adjusting network parameters such as data sampling rate. Each audio node adjusts its sampling rate according to the special correlation with its neighbors and the distance to the information source. In this way, the nodes in an area with higher density can keep lower sampling rate and lower data transmitting rate compared with those nodes in the sparse area. From the viewpoint of transport layer, lost or corrupted data packets cannot be retransmitted so as to guarantee the packet transmission sequence and the low bandwidth utilization. However, audio data of the same source can be reliably reconstructed by fusing multiple audio streams from nearby nodes. The fusion of multiple audio streams means that the aggregators recover high resolution audio data from several low ones. The recovery process is achieved by combining audio data from several audio streams with low resolution. As a result, the missing audio information, caused by packet loss or node failure, can be compensated by audio streams from other nodes and the demand of reliability of the upstream nodes is thereby reduced. The sink node adjusts the data sampling rates of sensing nodes by the feedback from the aggregator so as to reduce the data traffic and guarantee the fidelity of audio data. 3. Hierarchical synchronization strategy In order to prolong the lifetime of the wireless sensoractuator networks, we design an effective time synchronization mechanism to realize the synchronized sleeping of routing node and sensor node. In our building control system, routing node and sensor node work in different ways. Sensor node wakes up periodically, collects data, and reports data, thus sleeping immediately. Routing node requires long active periods to fuse the reports from sensor nodes and transfer the aggregated results to the sink node. Considering the different work mode of the two types of nodes, we design hierarchical synchronization strategy including route synchronization and node synchronization to guarantee the effectiveness of data transmission and achieve the maximization of network lifetime. In route synchronization phrase, the sink node broadcasts the synchronization frames periodically. Routing node wakes up and sleeps periodically according to the parameters (including ActiveDuration, SleepDuration, and SyncCycle) of the synchronization frame. The broadcast cycle of the synchronization frame is determined based on the clock drift model proposed in Ref.[8] and statistical data in our actual experiments. In node synchronization phrase, sensor node must send collected data in active state of the routing node.

IV. Experiment Results In this section, a set of simulations are presented to evaluate the performance of our key technologies: a lightweight trust management framework for discrete data, self-adaptive

Smart Building Control Based on Wireless Sensor-Actuator Networks audio sampling and combining reception algorithm and selfadaptive wake-up mechanism. 1. Trust guarantee for discrete data We assume a temperature monitoring network is already set up with several aggregating sets and evaluate the performance of this framework for discrete data in one aggregating set. Without loss of generality, an aggregating set includes one aggregator and 30 sensor nodes, and each sensor reports one data every sampling round. There are 60 sampling rounds during the experiment, and they are divided into three phases. Initially, the temperature is 26◦ C, and all sensor nodes work normally; Then, eight sensor nodes fail and fix their reports at 40◦ C from the 21st round; From the beginning of the 40th round, the temperature of environment raises to 100◦ C suddenly and maintains until the end of simulation.

Fig. 8. Self data reputation of aggregated results

Fig. 9. Aggregated results of aggregator

In the first phase (1–20 rounds), the data reported by all sensor nodes follows Gaussian distribution N (26, 2). The aggregated results are around 26◦ C as shown in Fig.9, and the area trust of the aggregating set is high relatively (as shown in Fig.8, which closes to 1). From the beginning of the 21st round, eight sensor nodes fail and falsely report 40◦ C suddenly. In the second phase (21–40 rounds), the aggregated results are around 28◦ C which is slightly larger than 26◦ C, and the area trust of the aggregating set descends to 0.98. From the beginning of the 40th round, the reports from the other 22 sensor nodes increase to 100◦ C suddenly and keep that value since the environmental temperature changes, but the eight abnormal nodes keep reporting 40◦ C. Even the system figures out the aggregated result is about 100◦ C, from the 41st to the 44th round, the area trust of the aggregating set is still low. As time

441

goes by, the area trust of the aggregating set approaches to 1. From the analysis above, the framework weakens the impact of the eight failure nodes effectively and can reflect the real change of the environmental temperature. 2. Improvement on audio signal quality In order to evaluate the performance of the proposed adaptive sampling strategy, we did the experiments on our simple platform shown in Fig.3, where there are several audio sensor nodes in one aggregating set. In this set of experiments, we use normal human voice as the audio source. When the volume of voice exceeds a given threshold, sensor nodes will collect the audio signals and report the samples to the aggregator. We verify the performance of the system under bad wireless environment with the low data reaching probability. In order to make the result more clear and readable, we use the reconstructed waveforms of the audio signal to illustrate the signal quality. In fact, each node in the network only need process the sampling data and need not reconstruct the waveform. Fig.10 shows the process of signal combination in noisy circumstances with different original SNR and data reaching probability. Fig.10(a) and (b) are the original audio waveforms gathered by the two sensors. Since the SNR of both channels are low, the two sensors do sampling at full rate. Because node 1 is farther away from the audio source than node 2, its signal strength is a little smaller and many samples are lost in transmission. Fig.10(c) illustrates the aggregated result at the fusion node. By using MRC diversity reception, the lost samples are recuperated, the signal strength increases, and the SNR of aggregated signal is desirable comparing with the two original audio streams. 3. Energy conserving In this experiment, we evaluate the performance of the proposed synchronization strategy on the energy consumption. We implement it on our test-bed which consists of 50 temperature and humidity sensors, 10 routing nodes and 1 sink node. The sensor nodes report a packet every 10 seconds. The sink node sets the ActiveDuration and SleepDuration to 30s, 120s respectively. We take the routing node which is nearest to the sink node as the target node and record the battery volt of this routing node. Then, we compare the energy cost with the conventional strategy in which the routing node works at full duty cycle.

Fig. 10. The result of diversity reception under bad channel environment, in which the SNR of original signal is high but the data reaching probability is low. (a) Original audio waveform from sensor 1; (b) Original audio waveform from sensor 2; (c) Combined audio waveform

Fig. 11. The comparison result of lifetime and voltage trends on routing node at different duty cycle

Chinese Journal of Electronics

442

As shown in the Fig.11, the lifetime of the routing node working at 20% duty cycle is longer than the node with a full duty cycle. The result illustrates that our proposed synchronization strategy can save energy effectively by reducing the duty cycle. Thus, it is desirable to extend the lifetime of network in many application.

[4]

[5]

V. Conclusion This paper presented a building control prototype system deployed in our school building based on wireless sensoractuator networks. Based on the multi-layer architecture of wireless sensor-actuator networks, we designed a lightweight trust management framework for discrete data, a self-adaptive audio sampling and combining reception algorithm and hierarchical synchronization strategy so as to make the building control prototype system providing flexible and intelligent service to end-users. Moreover, the building control prototype system which we deployed in our school building demonstrated that the wireless sensor-actuator network is efficient on building automation, and flexible to a variety of application-specific demands. Future work will be carried out to (1) design energyefficient multi-hop routing protocol; (2) research the flexible network management system, and (3) extend the prototype to implement intelligent building control.

[6]

[7]

[8]

2011

on Intelligent Agent Technology, IAT ’06. pp.653–659, Hong Kong, 18–22 December 2006. U. Rutishauser, J. Joller, R. Douglas, “Control and learning of ambience by an intelligent building”, IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, Vol.35, No.1, pp.121–132, 2005. J.Y. Adams, “Wireless sensors and controls make the organic building”, Proceedings of the 2006 IEEE International Symposium on Electronics and the Environment, pp.109–113, 8–11 May 2006. F.L. Zucatto, C.A. Biscassi, F. Monsignore et al., “Zigbee for building control wireless sensor networks”, SBMO/IEEE MTT-S International Microwave and Optoelectronics Conference, IMOC 2007, pp.511–515, Salvador, Brazil, 29 Oct.–1 Nov. 2007. M.N.K. Soini, L.T. Sydanheimo, M.A. Kivikoski, “Reliability and scalability of the Kilavi building control platform”, IEEE International Symposium on Consumer Electronics, ISCE 2007, pp.1–6, Dallas, Texas, USA, 20–23 June 2007. Jeremy Elson, Lewis Girod, Deborah Estrin, “Fine-grained network time synchronization using reference broadcasts”, Proceedings of the 5th Symposium on Operating Systems Design and Implementation (OSDI 2002), pp.147–163, Dec. 2002. SUN Yan Ph.D., associate professor in School of Computer Science and Technology, Beijing University of Posts and Telecommunications. Research Interests: IoT device and service, wireless multimedia sensor network and embedded communication devices. (Email: [email protected])

References [1] T. Wark, D. Swain, C. Crossman, P. Valencia, G. BishopHurley, R. Handcock, “Sensor and actuator networks: Protecting environmentally sensitive areas”, IEEE Pervasive Computing, Vol.8, No.1, pp.30–36, 2009. [2] E.C.H. Ngai, Jiangchuan Liu, M.R. Lyu, “An adaptive delayminimized route design for wireless sensor–Actuator networks”, IEEE Transactions on Vehicular Technology, Vol.58, No.9, pp.5083–5094, 2009. [3] Bing Qiao, Kecheng Liu, C. Guy, “A multi-agent system for building control”, IEEE/WIC/ACM International Conference

LUO Hong Ph.D., professor in School of Computer Science and Technology, Beijing University of Posts and Telecommunications. Research interests: IoT application and service, wireless multimedia sensor networks and wireless communication. (Email: [email protected])