Berkeley Electrical Engineering and Computer Sciences
One day soon, researchers hope to be able to sprinkle hundreds or thousands of sensor nodes liberally across buildings or landscapes and have them organize themselves into powerful wireless networks. But before this vision can be realized, sensor nodes must be pared down to a smaller, simpler, and less costly form. "Cheap components are not as reliable, but if one dies you have others, and you get a better total picture," notes Jan Rabaey, a professor of electrical engineering whose specialty is building tiny, low-power radios. "Lots of cheap components can do a better job than a few, very expensive ones."

UC Berkeley EECS Professor Jan Rabaey (Photo by Bart Nagel)

EECS Professor Jan Rabaey: A recent version of Rabaey's PicoRadio operates on a scant 60 microwatts of power, making it the lowest-power radio in existence.(Photo by Bart Nagel)
Having large numbers of cheap nodes also poses barriers to networking and wireless communication, however. Cheap nodes lack precise controls and, as the number of nodes increases, so does interference.

At Berkeley, several faculty members are devising smart wireless networking technology to overcome these hurdles and others. For example, Rabaey is devising algorithms to enable cheap radios to communicate, while David Tse, a professor of electrical engineering, has figured out a way that nodes can work cooperatively to increase the wireless communication capacity of large-scale, ad hoc networks.

The most expensive component of a wireless sensor node tends to be the radio. One way of lowering a radio's cost and power use, notes Rabaey, is to leave out the crystal—a very precise frequency element. Without a crystal, however, a radio cannot broadcast and receive at a predetermined broadcast frequency. In order to communicate, crystal-less nodes must somehow find a common channel.

Similar communication problems arise in the natural world. For example, for reasons that aren't completely understood, crickets have been observed to adjust their chirps to the surrounding ones until they sing in unison. Also, tropical fireflies synchronize the timing of their luminescent flashes by continually shifting to the average of their nearest neighbors.

Rabaey and colleagues are implementing a similar algorithm for clusters of wireless radios. Each radio pulses every second at some frequency. When a radio is not transmitting, it listens and centers its pulse relative to those of the other nodes. There is a tradeoff between how quickly the radios achieve synchrony and how accurately. To balance both, the researchers use a modal approach: first they achieve a rough synchrony quickly, and then they adjust to the desired level of accuracy.

Meanwhile, Tse is working on the information-theoretic challenges of building large-scale wireless networks. A famous paper by Piyush Gupta and P. R. Kumar identified a fundamental limit to the communication capacity of such networks: As the number of nodes increases, so does interference, pushing the information throughput per node down to zero.

Tse's recent work provides a possible workaround. Together with Ayfer Özgür and Olivier Lévêque at the Ecole Polytechnique Fédérale de Lausanne in Switzerland, Tse showed that if groups of nodes work together, the throughput scales linearly, which means that having many point-to-point conversations is just as easy as having only a few.

The Gupta and Kumar paper focused on the task of enabling multiple, simultaneous, point-to-point conversations in a bounded area, a problem that comes up in cell phone networks designed for emergencies and remote locales in which each cell phone relays calls to other phones. Tse's approach to the problem made use of MIMO, which stands for "Multiple Input, Multiple Output." MIMO provides a way to significantly increase data throughput, without additional bandwidth or transmit power, by using multiple antennas to send and receive the data.

Tse and his collaborators assumed that each node of the network has one antenna but can cooperate with nearby nodes. They subdivided the network into hierarchical clusters of nodes with each cluster at a given level sub-divided into smaller clusters at the next one. At the lowest level, nodes within each cluster take turns transmitting a message to all the others. The messages are then combined into a larger group message, which is exchanged among multiple clusters via MIMO. This enables cooperation within clusters at the next higher level of the hierarchy. Continuing this way, the researchers achieve cooperation across the entire network with minimal overhead. "Cooperation mitigates interference," Tse says.