Techniques for Minimizing Power Consumption in Low Data-Rate Wireless Sensor Networks

Sokwoo Rhee, Deva Seetharam and Sheng Liu

Millennial Net 201 Broadway, Cambridge, MA - 02139. {sokwoo, dseetharam, sliu}@millennial.net benet (freedom from wiring costraints and costs) of wireless networks be lost, but also many remote sensing applications may become impractical. Therefore, long battery life (several years) is essential in wireless sensor networks. The energy efciency of sensor networks is a critical issue and discussed in length by Goldsmith et al [1] and Min et al [2]. 2) Small Form Factor - It is obvious that devices must be small enough to be embedded in their operating environment. This requirement affects the choice of batteries - even AA batteries are too bulky to power the sensor node, so using coin cell batteries is the only option in many situations. 3) Low-data Rate - Since the sampling rate (rate at which variables such as temperature are sensed) is usually small, the number of bits transmitted per second by individual nodes is low. 4) Low Cost - Since the number of sensor nodes in a system can be large (e.g., a large scale industrial automation application can use several hundred sensors), the cost of individual nodes must be minimal to be of practical use. 5) Centralized Archictecture - Most of these applications consist of a sophisticated central node and several simple end nodes. For instance, several different types of sensors in a building are controlled by a single building automation controller. We have developed an ultra-low power sensor networking platform called the i-Bean Network to support such low datarate sensing applications. We have devised several techniques for minimizing power consumption in the i-Bean Network. This paper discusses these techniques in detail. The rest of the paper is structured as follows: Section II reviews works related to the research reported in this paper. Section III introduces the different components of the i-Bean Network. Section IV presents the techniques for minimizing power consumption. Section V concludes the paper with preliminary results. II. P REVIOUS W ORK As predicted by D. Estrin [3], energy concerns of wireless sensor networks have inspired several energy efcient protocols, processors, designs and algorithms. In this section, we present some of the key developments.

Abstract Wireless sensor networks can enable numerous embedded computing applications. One of the primary benets of wireless sensor networks is their independence from the wiring costs and constraints. However, if the batteries must be replaced often (every week or every month), not only will the initial cost savings be lost, but also many remote sensing applications may become impractical. Therefore, long battery life (in the order of years) is essential; power efciency is a critical requirement for wireless sensor networks. We have devised several novel techniques for minimizing power consumption in wireless sensor networks. Based on these techniques, we have developed a highly power-efcient sensor networking platform called the i-Bean Network. In this paper, we describe these techniques in detail and the associated tradeoffs.

I. I NTRODUCTION Wireless sensor networks have immediate utility in a variety of industrial, medical, consumer and military applications. We studied several such applications and classied them based on their modes of acquiring and propagating sensor data. We nd the following three classes to be the most common: 1) Periodic Sampling - For applications where a certain condition or process needs to be monitored constantly, such as temperature in a conditioned space or pressure in a process pipeline, sensor data is acquired from a number of remote points and forwarded to a data collection center on a periodical basis. 2) Event Driven - There are many cases that require monitoring one or more crucial variables and transmit only when a certain threshold is reached. Common examples include re alarms, door and window sensors, or instruments that are used intermittently. 3) Store-and-Forward - In many applications sensor data can be captured and stored or even processed by a remote node before it is transmitted to the central base station. For example, a temperature sensor that periodically senses and records the cargo temperature during transit for several days, and when the cargo gets to its loading dock for unloading, the device can detect the presence of a network and transmit all the accumulated temperature data to the network base station. The critical requirements of these applications are: 1) Long Battery Life - In many applications, sensors are placed in locations that are not conveniently accessible. Moreover, if the batteries must be replaced often (every week or every month), not only will the primary

The WINS [4] and PicoRadio [5] projects are attempting to develop integrated circuits that contain sensing, signal processing and radio elements. Shih et al present Amps a sensor node that exposes the underlying parameters of the physical harware to the system designer. All layers of the node, including the algorithms, the operating system, and network protocols, can adapt to minimize energy usage. Guo et al [6] present a transmission scheme that balance energy consumption amongst nodes by considering transmission distance (between the nodes and the base station) and data density. TRAMA [7], T-MAC [8], and S-MAC [9] are energy efcient MAC protocols. TRAMA reduces energy consumption by ensuring that communications are collision free and by placing nodes in sleep mode when they are not communicating. T-MAC uses an adaptive duty cycle, where the active part of it is dynamically ended. This reduces the amount of energy wasted on idle listening. S-MAC uses three techniques to reduce energy consumption: 1. Nodes periodically sleep 2. Neighboring nodes form virtual clusters synchronize their sleep schedules, so that the entire neighborhood is not asleep. 3. Nodes sleep when they are not actively communicating. SPIN [10], LEACH [11] and directed diffusion [12] are energy efcient routing algorithms for wireless sensor networks. Nodes running SPIN protocol name tag their data with descriptors, called meta-data. They use meta-data negotations to eliminate the transmission of redundant data. Moreover, SPIN nodes base their communication decisions upon both application-specic knowledge of the data and upon knowledge of the resources available to them. This allows sensors to efciently distribute data. LEACH uses hierarchy to reduce the data sent from sensors to the base station. This reduces power consumption. In directed diffusion, data transmission is minimized by caching and processing data in network. While similar to some previous work, the work presented in this paper is different in several respects. Our techniques minimize energy consumption at all levels of this system - node design, physical layer communications, MAC and network protocols and system design. We have validated our techniques by utilizing them to develop an ultra low-power sensor network composed of tiny and inexpensive nodes. Most importantly, the network and devices have been proven to work reliably and effectively in real life applications. III. I -B EAN N ETWORK The i-Bean network is composed of three types of devices: 1) Endpoints (or i-Beans) - The Endpoints are the devices that are directly connected to sensors and embedded in the environment. These Endpoints are tiny (25 x 15 x 5 mm - slightly larger than a dime). They are powered by coin batteries and are extremely power efcient. Each i-Bean provides four analog input channels, four digital 8-bit I/O channels, and an UART port for interfacing with sensors and actuators. Multiple sensors and actuators can be connected to an i-Bean.

Fig. 1.

An Endpoint embedded in a nger ring.

2) Routers (or Repeaters) - Routers extend the transmission range of i-Beans. Routers are responsible for forwarding and routing data between i-Beans and base stations. These devices have a small form factor - 56 x 33 x 5 mm. They are also power efcient, but they require more power than i-Beans. 3) Gateway (or Base station) - This Gateway is also compact 64 x 51 x 5 mm and designed to be unobtrusive. It serves as the Gateway between the i-Bean network and host computers such as desktops and SBCs1 . A base station can be connected directly to a RS-232 serial port of a host computer and gets power from the serial port. We designed our nodes with limited resources using lowcost off-the-shelf components. The nodes have limited computing and capacity. This means the nodes can perform neither sophisticated data processing nor store large volumes of data nor run complex network protocols. Simplicity of sensor nodes does not cause any difculties, since most sensing applications dont require sophisticated terminal nodes. Further hardware and software details can be obtained from our website [13]. The following sections present the techniques employed to minimize power consumption in these devices. IV. P OWER E FFICIENCY IN I -B EAN N ETWORKS Power efciency is achieved through ultra-low power nodes and power-efcient communications between those nodes. The following sections describe the node design and networking schemes in detail. A. Power-Efcient Nodes All nodes must consume as little power as possible, however power efciency of Endpoints is critical. The reasons are: Endpoints are powered by coin batteries, whereas Routers and Gateway are powered using larger power sources such as AC mains or AA/A batteries. Endpoints must be tiny, since they must be unobtrusive to be embedded in unconventional environment. Figure 1 illustrates the importance of Endpoints being tiny - an Endpoint is embedded in a nger ring. Typically the number of Endpoints in a network exceeds the number of Routers. The proportion of Endpoints and Routers depends on the deployment site, trafc patern,
1 Single

Board Computers

etc. As a result, collective power consumption could be higher and could also require several batteries to be replaced. The following sections describe how power consumption by Endpoints is minimized. 1) Dual Processors: The tasks of a processor in a sensor node can be classied into two main categories: 1. conventional computing tasks (including I/O with sensors) 2. communication tasks that employ RF circuitry. To execute these two types of tasks, the processor must operate at a frequency greater than the highest frequency task. Since RF circuitry operates at a very high frequency, a very high speed processor would be necessary. Unfortunately, a high speed processor wastes a lot of power, since its power consumption is substantial even in idle mode, thus reducing battery life. We overcome this problem by employing two microcontrollers to execute these two different types of tasks and putting the high-speed processor in sleep mode as much as possible. Each Endpoint has two interconnected processors p1 and p2 . The processor p1 operates at clock frequency f1 and the processor p2 operates at clock frequency f2 and the frequency f1 is much smaller than f2 (f1 << f2 ). A process called coordinator running on one of these processors allocates tasks in such a way that most of the tasks are run on the slower processor and the faster processor is in sleep mode for most of the time. This usually means that computing tasks execute on p1 and communication tasks execute on p2 . Although the coordinator usually runs on p1 , it can execute on either of the two processors. Both p1 and p2 have a copy of the coordinator. Only one of the coordinator instances is executed at any time, except when the coordinator wants to transfer from one processor to the other. To transfer before going to sleep the coordinator awakens the copy that is currently sleeping. The coordinators mobility is critical, since it can execute on p2 , when the system load is high and minimize power consumption by executing on p1 , when the load is less. If further power saving is required, both processors can be in sleep mode, but one processor should wake up and run the coordinator from time to time. A possible timing diagram of two processors is presented in Figure 2. 2) Narrowband Radio: Power consumption of RF circuitry is highly dependent on the modulation scheme. Wideband RF chips, such as that of Bluetooth [14], consume much more power than typical narrowband radios because of the complexities of base band processing. Although wideband radios offer better immunity to interference, for many sensor network applications, narrowband radios remain a practical choice. We use RFMs TR-1000 chipset that operates in 916 MHz band. This chipset draws no more than 10 mA current at 3 VDC.

ON Processor P1 OFF

ON Processor P2 OFF
T0 T1 T2 T3 T4

Fig. 2. A sample timing diagram of two processors. At time T0 , the node and p0 are turned on; at T1 , p2 is activated by p1 ; between T1 and T2 , both the processors are active; at T3 , coordinator is transferred to p2 ; at T4 , coordinator is transferred back to p1 .

Fig. 3.

i-Bean Network. (E - Endpoint, R - Repeater, G - Gateway)

B. Power-Efcient Network As previously mentioned, most sensing applications require sensor nodes to transmit data to a powerful base station. To support such applications natively, i-Bean Network is designed such that data communications are strictly between the Gateway and the Endpoints. That is, the data packets originate and terminate either in the Endpoints or in the Gateway. Although a Router can communicate with any other Router to relay packets, they never exchange data packets. The i-Bean network design exploits this communication pattern to minimize power consumption. The following sections describe the power-efcient network design. 1) Star-Mesh Hybrid Topology: Among wireless sensor networking topologies, the star system is the lowest in overall power consumption, but is limited by the transmission distance of the radio (typically 10 to 30 meters in the ISM band) in each node back to the base station. On the other hand, mesh topology consumes more power than star topology due to its higher duty ratio - nodes must always listen for messages for relaying packets for other nodes in the network. However, mesh networks are highly fault tolerant since there are multiple paths between any two nodes. As shown in Figure 3, the topology of the i-Bean network is a star-mesh hybrid. That is, the Endpoints are connected to

Sampling Interval Listening Window Sleep Mode Listening Window

t1

t1 + dt

t2

t2 + dt

Fig. 4.

i-Beans duty cycle.

the nearby Routers using a star topology (Router is the central hub and the Endpoints are the edge nodes), whereas Routers are inter-connected using a mesh topology. Endpoints never relay packets for other nodes. Routers can communicate with any other node in its vicinity and they do relay and route packets for other Routers and Endpoints. Using this hybrid topology, power consumed by Endpoints is minimized, while the network is fault-tolerant, since multiple paths can exist between the Endpoints and the Gateway. 2) Multihop Routing: The free space loss models of RF signal show that the radiated power is proportional to the square of required transmission distance. Hence, it is more power efcient to emit low strength RF signals to travel a short distance and be relayed a number of times than transmitting high strength signals for a longer range. i-Beans transmit packets using low-power signals and the packets are forwarded by Routers. As pointed out by Min and Chandrakasan [15], power consumption in wireless networks can increase with number of hops due to various overheads. In this multihop structure, Routers end up consuming more power. As explained above, higher power consumption by Routers is not a critical issue and our primary objective is to minimize battery power consumed by i-Beans. 3) Heterogeneous Nodes: As described in Section III, an iBean network is heterogeneous. The Endpoints are connected to sensors and actuators and they can be either source or destination of network data, but cannot forward data for any other nodes. The Routers are solely responsible for routing data in the network. Since Endpoints are not relaying packets, they can conserve power by turning off their communication circuitry except when they send or receive packets. Using dedicated Routers, power consumption of Endpoints is minimized. 4) Low Duty Ratio: It is well known that typical radio devices consume a fair amount of power even when they are just in the listening mode. We avoid such active listening and consequent power wastage by employing a small duty ratio - an Endpoint does not waste precious power listening to periodic beacon signals; instead it stays in power-saving idle mode most of the time and wakes up occasionally according to its own schedule. As shown in Figure 4, Endpoints wakeup periodically (period is determined by the sampling rate) to read the attached sensors and transmit their data. After transmitting the sensor data, they remain in receive mode for a short duration and go to sleep until the next sampling time. 5) Progressive Search: Power consumption is further reduced by using the Progressive Search technique. When an Endpoint attempts to join a network, or when it wakes up

after staying in the sleep mode for a long period of time, the Endpoint will rst broadcast a short hello to search its neighborhood for Routers. These hello packets contain only the identier of the network that the device would like to join or already belongs to. If the device receives a reply, which is also relatively short, then it will broadcast another hello packet that holds a different inquiry such as the ID of the destination node. After receiving a response that conrms a known route to the Gateway, the Endpoint will transmit a complete packet that contains all the information associated with the device; this information includes status of input/output channels, sampling rate, battery voltage, etc. The main advantage of progressive search is that an Endpoint will not waste power on sending multiple copies of lengthy hello packets when there is no device in the neighborhood to respond. An Endpoint will consume power to transmit the full packet load only when it conrms that there is a route to the destination and one of its neighbors knows the route. An additional advantage of progressive search is the efcient bandwidth usage. V. P RELIMINARY R ESULTS AND D ISCUSSIONS In this paper, we have presented several techiques to minimize power consumption at different levels of the system hierarchy of low data-rate wireless sensor networks. In some instances, we have shown how the individual nodes can be designed to be efcient. In other instances, we presented techniques for developing energy efcient network algorithms and protocols. Based on these techniques, we have developed the i-Bean Network, an ultra low-power wireless sensor network. From our preliminary studies, we nd that power consumption in this network is extremely low. For instance, when powered by a small coin battery (CR2032) with a capacity of 220mAh, the average current consumption is approximately 100 A, when the sampling interval is 1 second and therefore battery will last for about 80 days. If the sampling interval is increased to 120 seconds, average current consumption drops to 1.92 A and increasing the battery life to about 13.1 years 2 Despite these encouraging results, we need to perform more experiments to understand the impact of our design decisions and tradeoffs when the network is large (> 1000 nodes), since even simple protocols and algorithms can exhibit surprising complexity at scale [16]. VI. ACKNOWLEDGMENTS We thank Christy Fujio for reviewing the manuscript and providing us with valuable comments. We thank Ningya Wang and Jason Xiao for valuable discussions and interesting insights.
2 Any number more than 10 years may be meaningless, since the battery shelf life itself may be less than the computed time.

R EFERENCES
[1] A. Goldsmith and S. Wicker, Design challenges for energy-constrained ad hoc wireless networks, IEEE wireless communications, vol. 1, pp. 1010, August 2002. [2] R. Min and et al, Energy-centric enabling technologies for wireless sensor networks, IEEE wireless communications, vol. 1, pp. 1010, August 2002. [3] D. Estrin, Wireless sensor networks: application driver for low power distributed systems, in Proceedings of the 2001 international symposium on Low power electronics and design, 2001, pp. 194194. [4] G. Asada, T. Dong, F. Lin, G. Pottie, W. Kaiser, and H. Marcy, Wireless integrated network sensors: Low power systems on a chip, 1998. [5] J. M. Rabaey, M. J. Ammer, J. L. da Silva Jr., D. Patel, and S. Roundy, Picoradio supports adhoc ultra-low power wireless networking, Computer, pp. 4248, August 2000. [6] W. Guo, Z. Liu, and G. Wu, Poster abstract: an energy-balanced transmission scheme for sensor networks, in Proceedings of the rst international conference on Embedded networked sensor systems, 2003, pp. 300301. [7] V. Rajendran, K. Obraczka, and J. J. Garcia-Luna-Aceves, Energyefcient collision-free medium access control for wireless sensor networks, in Proceedings of the rst international conference on Embedded networked sensor systems, 2003, pp. 181192. [8] T. van Dam and K. Langendoen, An adaptive energy-efcient mac protocol for wireless sensor networks, in Proceedings of the rst international conference on Embedded networked sensor systems, 2003, pp. 171180. [9] W. Ye, J. Heidemann, and D. Estrin, An energy-efcient mac protocol for wireless sensor networks, 2002. [10] J. Kulik, W. Heinzelman, and H. Balakrishnan, Negotiation-based protocols for disseminating information in wireless sensor networks, Wireless Networks, vol. 8, no. 2/3, pp. 169185, 2002. [11] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, Energyefcient communication protocol for wireless microsensor networks, in Proceedings of the Hawaii International Conference on System Sciences, 2000. [12] C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann, and F. Silva, Directed diffusion for wireless sensor networking, IEEE/ACM Trans. Netw., vol. 11, no. 1, pp. 216, 2003. [13] Millennial, Millennial net, http://www.millennial.net. [14] J. Haartsen and et al, Bluetooth: Vision, goals, and architecture, Mobile Computing and Communications Review, vol. 2, no. 4, pp. 3845, Oct 1998. [15] R. Min and A. Chandrakasan, Top ve myths about the energy consumption of wireless communication, in ACM Sigmobile Mobile Communication and Communications Review (MC2R). ACM, 2002. [16] D. Ganesan, B. Krishnamachari, A. Woo, D. Culler, D. Estrin, and S. Wicker, Complex behavior at scale: An experimental study of lowpower wireless sensor networks, UCLA, Tech. Rep., 2002. [17] E. Shih, S.-H. Cho, N. Ickes, R. Min, A. Sinha, A. Wang, and A. Chandrakasan, Physical layer driven protocol and algorithm design for energy-efcient wireless sensor networks, in Proceedings of the 7th annual international conference on Mobile computing and networking, 2001, pp. 272287. [18] J. Rabaey and et al, Picoradio supports ad hoc ultra-low power wireless networking, Computer, July 2000.