Connecting Wireless Sensor Networks with the Internet
by Adam Dunkels, Thiemo Voigt and Juan Alonso
Wireless sensor networks enable numerous advanced monitoring and control applications. In this project scientists at SICS are connecting sensor networks with the Internet.
Networks of tiny sensors make it possible to monitor, unobtrusively and for long periods of time, natural phenomena such as sensitive wildlife habitats or deep oceans. Home automation, volcano exploration, and monitoring of patients in hospitals are some of the other applications that can be enhanced using wireless sensor networks.
A wireless sensor network consists of a large number of tiny sensor nodes, each of which is equipped with a radio transceiver, a small microprocessor and a number of sensors. These nodes are able to autonomously form a network through which sensor readings can be propagated. Since the sensor nodes have some intelligence, data can be processed as it flows through the network.
For many applications, the sensor networks cannot operate in complete isolation; there must be a way for a monitoring entity to gain access to the data produced by the sensor network. By connecting the sensor network to an existing network infrastructure such as the global Internet, a local-area network, or a private intranet, gaining remote access to the sensor network is straightforward.
At SICS we are looking at using TCP/IP, the Internet protocol suite, for communication within the sensor network. This enables the sensors to be easily integrated into TCP/IP networks. There are, however, a number of problems that currently prevent TCP/IP from being directly applicable to sensor networks. We are developing mechanisms in order to solve these problems. These include:
- Tiny TCP/IP implementation: it has often been said that the TCP/IP protocol stack is too heavy to be squeezed into such a tiny system as a wireless sensor. Our µIP TCP/IP implementation, however, is small enough to be useful in such systems.
- Spatial IP address assignment: In IP networks, each host is required to have an IP address. In large-scale sensor networks, it is not possible to manually configure the addresses and we cannot rely on a central server. Instead, we have designed a spatial IP address assignment scheme, whereby each sensor constructs its IP address from its physical location. Since most sensor applications already require the sensors to keep track of their location, this mechanism does not increase the complexity of the system.
- Shared context header compression: For TCP/IP, the overhead created by headers can be quite large, particularly for small messages. For example, a four-byte data message would have a header overhead of nearly 90%. We are developing a header compression mechanism that utilises the special conditions in sensor networks in order to reduce the header overhead to only a few bytes for messages carrying sensor data.
- Application overlay networking: The address-centric routing in IP does not match the data-centric applications of sensor networks very well. We are developing an application overlay network structure that lets distributed applications run on top of the network and decide how to process the packets.
- Distributed TCP caching: The TCP/IP protocol suite was developed for networks with very low error rates and does not work well in error-prone wireless networks. To remedy this, we are experimenting with a mechanism that lets the sensor nodes help each other in caching data segments. If segments are lost because of errors on the radio channel, neighbouring sensor nodes are able to re-transmit the lost segments.
|The Contiki network simulator.
Furthermore, we have developed Contiki, an operating system for small sensor nodes that includes the _IP TCP/IP stack and implements the above mechanisms. We have ported Contiki to a number of different hardware platforms and have developed a simulation environment that lets us run multiple instances of Contiki as processes on a PC. This simulates wireless communication and lets the simulated nodes communicate with the outside network using TCP/IP. Additionally, the simulator provides a graphic representation of the sensor network as shown in the Figure.
We have taken the first step towards bringing TCP/IP communication into the realm of sensor networking. Subsequently, this will enable us to integrate sensor networks into the Internet.
The DTN/SN project at SICS: http://www.sics.se/cna/dtnsn/
Juan Alonso, SICS
Tel: +46 8 633 1544