"With advanced wireless technologies and MEMS sensors we can do that, but to integrate the network we had to take a lesson from neural networks and put memory alongside processing capabilities in each node," said Lupo.
MEMS provides the enabling technology for drone aircraft to drop thousands of wireless sensors onto an active battlefield, creating an instant network. But the amount of data streaming in from them would be impossible to digest in real-time by a central computer.
In a nutshell, Lupo discovered how to sidestep Metcalfe's Law for battlefield sensor networks by wiring them the way the spinal cord sends information up to the human brain from the eyes, ears, nose, tongue and skin-that is, by placing next to each sensor a neural-network memory element that "learns" what is normal.
Since the network manager (or brain, in the case of a human) also has a copy of what is normal, it can refer to its copy, rather than congest the network with redundant reports. That's possible because the smart sensors only transmit updates to the manager's copy when something abnormal happens.
In this manner, neural learning exponentially decreases the amount of network traffic and management overhead, thereby canceling out Metcalfe's exponential expansion law-and resulting in only linear overhead increases for each added node.
For instance, consider temperature. "Normal" for temperature would follow both the time of day and the time of year-cooler in the evenings, warmer during the day, cooler yet in the winter, warmer yet during the summer. With a neural network learning what the normal temperature is, the sensor need only transmit reports of out-of-bounds temperatures-the "delta"-thereby greatly decreasing the network traffic that is produced compared with a 24/7 continuous stream of reports.
According to neural researchers, if we wired our spinal cord the way that we do our computer networks, then no one would be able to hit a 100-mph fastball; we wouldn't have enough bandwidth. But by adding neural learning to make smart sensors, the network can get by with drastically reduced bandwidth, in much the same way as a hitter can slug a fastball.
The current program ($9 million during 2001) ends in October, but its eventual aim is to overlay five smart webs on the battlefield for images, weather, weapons, simulations and integration.
Civilian uses, too
However, just as the Internet was born of the military's ArpaNet, so has Lupo already envisioned a future for the smart-sensor web in the commercial sector. By melding single-chip MEMS sensors with the smart-sensor web for domestic neighborhood monitoring, almost any chemical or biological substance could be tracked in real-time-from pollen counts to natural-gas leaks and even terrorist-released contaminants.
The Office of Naval Research proposes that the government deepen the U.S. information infrastructure by installing a "chemical-lab-on-a-chip" in every traffic light in the United States and interconnecting them into a digital subscriber line network riding atop the ac lines. DSLs use high-frequency digital signal processors to multiplex data over normal twisted-pair telephone lines.
That approach works equally well for dual use of electric power lines in city traffic lights for always-on communication with MEMS sensors, said Lupo. With a national traffic-light network installed, civil and military authorities nationwide could monitor the atmospheric conditions on any street corner in America as easily as they monitor its visual appearance from satellites today.
"It's not just traffic lights either," noted Lupo. "Eventually we will probably update our building codes to include, say, a smart lightning rod on the roof."
It's already possible, said Lupo, to "build a smart MEMS weather station in the space of a cubic millimeter." In his view, "it will become increasingly easy to put smart sensors into all kinds of new construction."