Making Sense of Sensor Networks
What are Sensor Networks?
A sensor, in its simplest definition, is a device that is capable of observing and recording a phenomenon. This is termed as sensing. Sensors are used in various applications such as industry, military, healthcare, disaster relief, meteorology, etc. For example, sensors are used to study the formation of tornadoes by measuring the pressure, humidity, temperature, and wind direction when a tornado occurs. In rescue operations, seismic sensors are used to detect survivors caught in landslides and earthquakes.
With advances in electronics, sensors now have the capability to sense, process, and communicate data. These small-sized sensor nodes have low cost, low power needs, and the ability to communicate over short distances. This has led to the development of sensor networks, which capitalize on the sensor node's ability to communicate. A sensor network consists of possibly several hundred sensor nodes, deployed in close proximity to the phenomenon that they are designed to observe. The position of sensor nodes within a sensor network need not be pre-determined. Sensor networks must have the robustness to work in extreme environmental conditions with scarce or zero interference from humans. This also means that they should be able to overcome frequent node failures. Network topological changes thus become common. Sensor networks must conserve power since they are limited in power with usually the battery as the sole supply of energy. Sensor nodes may also have limited mobility, which allow them to adjust to topology changes.
Traditionally, a large sensing device is used instead a sensor network. But, the propagation laws of sensing [3] observe that a signal is faced with decay effects, dispersion, and scattering. To overcome some of these effects, a distant sensor has to take counteractive measures to reverse these effects. These steps are often complex and expensive. The laws of sensing thus dictate that if the system is to detect objects reliably, it has to be distributed. A sensor network achieves this distribution. A typical sensor network consisting of sensor nodes scattered in a sensing field in the vicinity of the phenomenon to be observed is shown in Figure 1. The nodes are connected to a larger network like the Internet via a gateway so that users or applications can access the information that is sent from the sensor nodes. An introduction to the merging field of sensor networks is provided in this paper with the aim to make the reader understand the challenges involved in designing these networks.

Figure 1: Sensor nodes connected in a network.
Challenges
The unique features of sensor networks pose challenging requirements to the design of the underlying algorithms and protocols. Several ongoing research projects in academia as well as in industry aim at designing protocols that satisfy these requirements for sensor networks. Some of the important challenges are presented below:
Ad hoc deployment
Sensor nodes are either placed one by one in the vicinity of the phenomenon or deployed in an ad hoc fashion by air or by some other means. However the mode of deployment, sensor nodes must be able to form connections to set up the network.
Self-Configuration
The position of sensor nodes is not determined apriori. This means that the sensor nodes must be able to configure themselves after deployment. They must possess some means to identify their location either globally or with respect to some locally determined position. Once the network is set up, sensor networks must be able to reconfigure themselves in case of connectivity changes.
Extreme Environmental Conditions
Environmental stress on sensor nodes cause frequent node failures leading to connectivity changes. These require frequent reconfiguration of the network and re-computation of routing paths.
Low Cost
The high probability of node failures in sensor networks requires that the cost of sensor nodes is minimal. This will enable redundancy of sensor nodes to account for node failures.
Low Mobility
In some cases, sensor nodes have the ability to move, although their mobility is restricted in range to a few meters at the most. Mobility of sensor nodes raises the possibility that nodes might go out of range and new nodes might come within the range of communication. The routing protocols for sensor networks must take these changes into account when determining routing paths.
Power Constraints
Limited battery life in sensor nodes requires sensor networks to manage power efficiently. Hence, algorithms and protocols developed for sensor networks must be energy-efficient.
Sensor Network Architecture
Given the above challenges for sensor networks, several architectures have been proposed for sensor networks. The Wired Integrated Network Sensors (WINS) [3] and Sensor Information Networking Architecture (SINA) [5] are examples of proposed architectures for sensor networks. SINA provides a framework wherein it enables easy querying, monitoring, and tasking of sensors. SINA is composed of three components:
Hierarchical Clustering
Sensor nodes are aggregated to form clusters based on their power levels and proximity. This process is applied recursively to form a hierarchy of clusters. A cluster head is then chosen to perform filtering, fusion, and aggregation.
Attribute-based Naming
Sensor nodes acquire the value of the attribute they are sensing. This type of naming scheme is applied to enable easy querying so that the sensors, which match the query response, will reply to the query.
Location Awareness
Sensor nodes also need to be aware of their location within their physical environment. A global positioning scheme with the use of GPS receivers can be employed, although the cost constraints of such a solution might result in only a subset of the sensor nodes having the GPS receivers. These nodes periodically transmit a beacon signal broadcasting their position as determined by their GPS receiver so that the remaining nodes can find their relative location. An alternate solution would be to use an optical tracking mechanism.
How are Sensor Networks Different?
Let us look at how sensor networks are different from traditional networks and why existing algorithms and protocols are not suited for sensor networks. Sensor networks make use of wireless communication and hence are different from traditional wired networks. They are also different from traditional wireless networks like cellular, Bluetooth, and mobile ad hoc networks (MANETS). In these networks, the objective is to optimize throughput and delay. Although MANETS share the characteristics of ad hoc deployment and self-configuration of the nodes, power conservation is not prioritized. Bluetooth also addresses similar power limitation problems, but the extent of power conservation that is required in sensor networks is much greater. Furthermore, sensor nodes are frequently exposed to extreme environmental conditions, making them prone to frequent node failures. This places a stringent cost constraint on sensor networks unlike the other wireless networks. The differences of sensor networks with existing networking technologies are discussed in detail in a previous study [1].
Physical Layer
Although the transmission media in sensor networks can be infrared, radio, or optical media, the 915MHz industrial, scientific, and medical band (ISM) has been popular for sensor networks. The deficiency of using infrared or optical media for transmission is that the laws of physics require that the sender and the receiving nodes maintain a line of sight for communication. However, various wireless specifications like Bluetooth, HomeRF, and Wireless LAN specified by IEEE 802.11b all operate in the license-free 2.4GHz channel. WINS architecture [3] makes use of radio frequency as the communication media.
Data Link Layer
The data link layer's responsibility is to establish a reliable link or a networking infrastructure above which data can be routed. Existing specifications like Bluetooth uses time-division multiplexing (TDMA) with frequency hopping whereas in wireless LANs specified by 802.11b, carrier sense medium access with collision avoidance (CSMA/CA) is used. A need for a new MAC layer protocol for sensor networks arises since the challenges of sensor networks are much different from the problems that the existing specifications aim to solve. For example, the range of a piconet, which is collection of eight nodes, in a Bluetooth network is 32 feet, whereas the required range is much smaller for sensor networks. In cellular networks, the base stations form a wired backbone providing a partial structure to the network. In sensor networks, there are no base stations. Previous research [6] describes a MAC layer protocol for sensor networks called Self-Organizing Medium-Access Control for Sensor Networks (SMACS) for link-layer setup. The Eavesdrop and Register (EAR) algorithm [6] then allows for mobile sensor nodes to interconnect with stationary nodes. SMACS proceeds on network startup with neighbor node discovery using transfer of messages. In SMACS, a channel is defined as a pair of time intervals. Neighbor discovery and channel assignment is combined into one phase, so that by the time all nodes hear all their neighbors, they would have formed a connected network. There is no hierarchy assumed in SMACS and hence it forms a flat topology. The EAR algorithm works with the problem of mobility management when mobile nodes are introduced in the sensor network.
Network Layer
Routing in sensor networks in quite similar to ad hoc routing protocols in MANETS. The difference is that in ad hoc routing algorithms, power conservation is of secondary importance. In Bluetooth, the specification calls for seven slaves with one master defining a piconet. Even when piconets are inter-connected to form scatternets, different topologies limit the nodes in the scatternet. For a sensor network with possibly hundreds of nodes, this will not suffice. Furthermore, the projected cost of a Bluetooth device is less than $4 whereas the estimated price of a sensor node is less than $1. In addition, the power requirements of sensor nodes are much less than for Bluetooth.
Various algorithms have been proposed for routing in sensor networks. The prime objective of the algorithms is to conserve power. Algorithms can be broadly categorized [1] as those that determine:
- Routes with the maximum available power in the nodes along the route
- Routes that consume minimum energy
- Routes that make minimum hops
- Routes in which the minimum available power is the maximum among all the other paths
Transport Layer
The need for a transport layer arises if the sensor network is to be connected to larger network like the Internet. In Figure 1, the sensor network is connected to the Internet via a gateway. The transport layer protocol connecting the user to the gateway could be an existing UDP or TCP protocol. However, the transport layer protocol connecting the gateway to the sensor node would have to be different since there is no global addressing scheme in a sensor network. Limited memory in sensor nodes becomes an important issue as protocols requiring less storage of state information are preferred over TCP-like protocols.
Application Layer
Sensor nodes have many diverse applications. Designing an application layer has the merit that the sensor network can be connected to a larger network like the Internet. Node addressing is an important issue here since, unlike other networks, sensor nodes do not have global identification. Application layer protocols like Sensor Management Protocol and Sensor Query and Data Dissemination Protocol [5] are currently being researched. The Sensor Query and Data Dissemination Protocol introduces a query interface to issue queries, reply to queries, and collect the query responses.
Although the applications for sensor networks are diverse, ongoing research suggests that some more work needs to be done to address the various protocols in all the layers. In addition, a common platform wherein proposed algorithms can be tested is required. Simulation of such sensor networks is difficult and is often tied to a particular application. Test beds are required to evaluate the various proposed solutions to sensor networks. The field of sensor networks is a growing field and we will definitely see several research efforts in this arena in the coming years.
Current Research Projects
Some of the most interesting research projects in sensor networks involve their use in various applications. Among these, the most inspiring project that I have come across has the been the study of natural habitats with sensor networks done by University of California, Berkeley [2] in collaboration with the Intel Research Laboratory at Berkeley and the College of the Atlantic in Bar Harbor. Wireless sensor networks have been deployed in Great Duck Island, Maine to collect climatic information without disturbing the natural habitat. The project website reports that nearly 1 million readings have been collected from 32 deployed "motes" where each "mote" has a microcontroller, low power radio, memory, and batteries [2]. The sensor node that was used is described as using a single channel, 915MHz radio. A pair of AA batteries was used for power. The sensor board that was used consists of temperature, photoresistor, barometric pressure, humidity and passive infrared sensors. Communication with the sensor nodes to a gateway with 802.11b (with embedded Linux) was compared with a mote-to-mote network. The results note that the sensor nodes have sufficient power for subsequent six months, even without human intervention [2].
Another application of sensor networks is in biomedical applications. The authors of a recent study [4] introduce embedded smart sensors, which operate within the human body to compensate for various diseases. As an interesting example, their ongoing project to create an artificial retina is discussed. The small and light smart sensors, each with 100 microsensors, are placed upon the retina. These sensors then produce electrical signals, which are converted by the underlying tissue into a chemical response. This chemical response is digital, forming a chemical serial communication [4]. Other challenging biomedical applications such as glucose level monitors, cancer detectors, general health monitors, and organ monitors are presented. But, the challenges here are much more than networking problems since human embedding is considered.
Lastly, the Smart Kindergarten [7] aims at applying sensor networks to develop a system which enhances the education process in early childhood by providing a learning environment that is customized to each child, adapts to the context, coordinates activities, and allows unobtrusive evaluation of the learning process. A wireless network of toys with sensory, processing, and communications capabilities is proposed as the application platform for the project. The challenges involved are in wireless communication as well as data management. The long-term goal of the project is presented as a system that triggers educational tasks based on temporal and spatial contexts and provides recorded data for evaluation.
Vision for the Future
In the Hollywood movie Twister, meteorologists chase surging tornadoes to place a barrel of sensors in their path to be picked up into the heart of the tornado. The scientists are then able to measure the tornado from inside with the information sent back by the sensors. National Severe Storms Laboratory (NSSL) reports that the movie was based on the work done by Dr. Al Bedard who developed an instrument named TOtable Tornado Observatory (TOTO), which was the size of a 55-gallon oil drum weighing 250-300 pounds. TOTO was built with robust weather sensors fixed on the outside to measure wind, pressure, and temperature. Batteries and recording equipment were fitted inside the barrel depicting a traditional large sensor device. The idea was for a tornado chaser to place TOTO down in the path of a tornado so that data from the passage of the tornado would be recorded. TOTO was withdrawn because of safety issues and the difficulty of getting such a large and heavy object in the path of a tornado. Although the movie Twister stretched reality at that time, it is not impossible to think of using sensor networks in such an application in the future. Light-weight sensor nodes could be suspended within the tornado with the capability to measure data and transmit back to the connected network where users can assimilate and analyze the recorded data. The robustness and cost of the sensor network could overcome the high probability of node failures to ensure the success of the project. However, whether sensors within the heart of the tornado can gather more information, is probably best left to the meteorologists.
To Know More
To read more about the habitat monitoring project at the Great Duck Island, Maine, refer to http://www.greatduckisland.net/.
To read about the implementation of a sensor network, check out the University of California, Berkeley's report on a live demonstration of a sensor network at http://today.cs.berkeley.edu/800demo/. The report also has some cool pictures of sensor nodes that they've developed.
Commercial sensor nodes are also available and the Rockwell Scientific Center has information about its successful deployment of sensor networks in various applications. Their website is at http://wins.rockwellscientific.com/.
Crossbow technologies in collaboration with University of California, Berkeley provide various solutions to challenges in sensor networks. Go to http://www.xbow.com/crossnet/ for more information.
To learn more about the TOTO project at NSSL, visit their site at http://www.nssl.noaa.gov/faq/vortex.shtml.
References
- 1
- I. Akyildiz, Su, W., Sankarasubramaniam, Y., and Cayirci, E. "A Survey on Sensor Networks," IEEE Commun. Mag., August 2000.
- 2
- Mainwaring, A., Polastre, J., Szewczyk, R., Culler, D., and Anderson, J. "Wireless Sensor Networks for Habitat Monitoring." ACM International Workshop on Wireless Sensor Networks and Applications, September 2002.
- 3
- Pottie, J. and Kaiser, W. J. "Wireless Integrated Network Sensors," Communications of the ACM, Vol. 43, no. 5, May 2000.
- 4
- Schwiebert, L., Gupta, S. K. S., and Weinmann, J. "Research Challenges in Wireless Networks of Biomedical Sensors," International Conference on Mobile Computing and Networking, 2001.
- 5
- Shen, C., Srisathapornphat, C., and Jaikaeo, C. "Sensor Information Networking Architecture and Applications," IEEE Pers. Commun., August 2001.
- 6
- Sorabi, K., Gao, J., Ailawadhu, V., and Pottie, J. "Protocols for Self-Organization of a Wireless Sensor Network," IEEE Pers. Commun., October 2000.
- 7
- Srivastava, M., Muntz, R., and Potkonjak, M. "Smart Kindergarten: Sensor-based Wireless Networks for Smart Developmental Problem-solving Environments," International Conference on Mobile Computing and Networking, 2001.
Biography
Rohini Krishnapura (rohini@cse.unl.edu) is a graduate student at University of Nebraska Lincoln in Lincoln, NE. Her research insterests are Ad hoc wireless networks, wireless communications in embedded systems, and clustered web servers. She is currently working with a research team at UNL on an embedded systems project involving ad hoc networks with 802.11b.
