The Emerging World Of Wireless Sensor Networks

After trying to connect every computer, PDA, and telephone to the Internet, researchers are starting in on the world around us. Wireless sensors that can detect subtle changes in air temperature and soil quality, vibrations thrown off by machinery, or abnormal noise on a road are starting to appear in lab tests and industrial pilots. These sensors, lasting months or years in the field on a pair of AA batteries, can organize themselves into a network of hundreds of "motes" the size of a deck of cards, then transmit their findings back to a central computer.

Proponents of the technology say wireless sensor networks could lead to a more granular understanding of our surroundings. They could even be used in conjunction with radio-frequency identification to cost-efficiently identify and track items. InformationWeek senior writer Aaron Ricadela spoke with two prominent researchers in the field, Teresa Lunt, manager of the Palo Alto Research Center's computer science lab, and Hans Mulder, an associate director at Intel Research.

A full report on sensor nets will appear in our Jan. 24 issue.

A talk with Teresa Lunt, manager of PARC's computer-science labInformationWeek: Some of the sensor network pilots so far seem almost commonplace: monitoring vibrations from industrial equipment or moisture in the soil. How soon are we likely to see a killer app for sensor networks?

Lunt: We'll be surprised by the killer app just like we were with the Internet. We have the technology enabler, but people are just replacing things that exist, like temperature sensors in buildings, with things that are easier to deploy. We have the wireless technology, the sensors, and the computation. But so far there's been no single killer app to catapult the technology, like the Web did for the Internet.InformationWeek: What are the challenges to making sensor networks larger and more pervasive?

Lunt: You've got the sensors themselves--microphones, cameras, or other types. Then it takes some processing to extract some information from that. But the technology doesn't scale. If you're going to scale to tens of thousands of sensors in a network--and they have to be wireless to make it work in high numbers--then it doesn't make sense to have all the processing done centrally. With wireless, the more nodes you have, the less bandwidth you have. The throughput quickly goes down to zero. You can do "in-network" processing to reduce the data. That's been the problem with "ad hoc" networking, where every device broadcasts out to every other one. The local sensors flood the bandwidth and nothing will get though. And you don't want to post your information to the Web, because it may not be not time-sensitive enough. Say you're building a chemical cloud accident-warning system. You need to sense the direction and toxicity of that cloud, and get the information to the people who need it first.

InformationWeek: What else could we use these sensor networks for?

Lunt: The closest to a real world application today is in the military. Darpa [the Defense Advanced Research Projects Agency] has a lot of work in sensor networks. And sensors have been a huge part of the military forever. They've been part of intelligence gathering, like putting sensors in U2 spy planes. But these applications haven't had the flavor of millions of little sensors. You've had a few of them on these high-value things like satellites or planes. Now, they're using more technically interesting things overseas like unmanned aircraft. Darpa is also funding some small, autonomous airborne and ground vehicles.

Another place people are talking about using sensor nets is in vehicle networks, so cars can communicate with one another on the highway. But it's not deployed yet. The auto industry's been talking about that, and the Transportation Department has been interested in reducing the number of accidents by putting sensors on vehicles that could pre-deploy your airbag if another car's getting too close. Another example is a left-turn or merge assistant that could tell you if someone's in your blind spot or around the corner, or if you're driving off the road. All these things could be enabled through sensors.They could also be used in hospitals. Nursing stations have different networks and multiple displays, because the systems are made by different companies. It would seem to be a big opportunity if you had one network--the same camera used for a security application could also be used for some new application. It seems like low-hanging fruit. A talk with Hans Mulder, associate director, Intel ResearchInformationWeek: How inexpensive do wireless, networked sensors need to be before they're used in larger numbers and constitute a sizable market?

Mulder: Everybody is looking for a number of killer applications and getting their platforms to a state where they're economically feasible. One of the challenges is still cost, and not just the cost of the hardware. There are a number of apps where the hardware cost is acceptably low--the nodes with networking, storage, and CPU costs around $50 in volume. In the next four to six years, assuming higher volumes, you can get that to $10. But the cost of installation and maintenance are still too high. Our sensor nets are all done by electrical engineers and computer-science guys--and they're not cheap. You want to get to a state where those things can be installed by technicians.

InformationWeek: How could users of sensor networks reduce the huge volumes of data coming off them? Does software play a role?

Mulder: One of the interests for Intel is getting an end-to-end solution using Intel silicon. We're working with SAP for integration with their NetWeaver software. We're also working with Rockwell [Automation's] manufacturing software and Honeywell's building software. You need a software layer to get much better-fidelity data--both spatial and temporal--and at a better price. The volumes of data will require that you do a certain amount of processing on the sensor nodes. SAP has a vision of end-to-end control, and we're researching how to get a solution based on SAP software that has a view of sensor nets, so your business rules can receive triggers from the sensor nodes that warn of impending defects, or change the operation of your sensor nets, or let you query them. A lot of the added value is going to be in the software.

It's a volume issue on the back end, and also a scalability and energy issue on your network. An amount of local processing will be essential. Our networks don't do that, and this is one of the changes you'll see. We made a program that downloads all the data [from a sensor network] to a PC. But I don't see more balancing of the application [between the server and the nodes] happening right now.InformationWeek: What about hardware--what technology can Intel bring to this market?

Mulder: The most obvious thing we can do in silicon is low power. The radios may use a standard communication protocol like [IEEE standard] 802.15.4. The ZigBee standard defines the next two communications layers. But there are still interesting things to do in making extremely low-powered silicon.

InformationWeek: One concern some people have is that the sensor net market could be characterized by small, inexpensive chips--not the kind of market Intel typically plays in. Where are the profits in sensor nets going to be?

Mulder: It's hard to predict that. If it remains labor-intensive, there's money to be made in services. Clearly, there's money to be made in the intellectual property for data-processing algorithms--transforming the data into something useful. If you get 100 sensor-net companies whose products are all compatible, it's going to be a cutthroat business. If there are just a few, it will be a different story.

Is such a [profitable] market going to exist, or will it be a commodity market? I can't tell you that. That's all going to happen in the next 10 years. I'm a research director; I'm not the business guy. But if we say we don't care about these motes, that's a guaranteed way not to play in the sensor net silicon. Even if we don't get in, then at least we have the expertise to make sure our enterprise platforms play well with sensor nets. But in the meantime, I increase the probability that we increase the business in gateways [computers that help transmit data from the networks]. Many are based on [Intel] XScale silicon. And maybe we get to the point of making sensor nodes. Those will be sold by the billions. I have to keep the options open.0