BELMONT, Mass. — What if people could go out and buy their own piece of the Internet in the form of a $500 consumer-electronics black box instead of relying on a giant communications service provider to dole out pieces of the network to them for a monthly bill? Strange as it seems, that is essentially the idea proposed by Timothy J. Shepard, a self-employed communications engineer.

Just as people bought PCs and broke the hold that mainframe service centers had on delivering computing services, Shepard thinks people will someday buy networks in the form of personal radio transceivers that become their Internet-in-a-box. "Maybe we can all own our own telecom infrastructure when we own our own radios," Shepard said. "If everyone had this you could cancel your phone line."

Shepard describes a packet-based radio network made up of self-organizing radios that cooperate to form a metropolitan area network — and maybe someday even a global Internet. Signals pass from one neighbor to the next across thousands or potentially millions of links. The vision, the basis of his 1995 thesis at MIT, led him to leave a comfortable research job at GTE Internetworking last April, and is now the seed for a startup company Shepard hopes to build.

Since April, he said, "I've been having a great time and burning my savings account."

The 30-something engineer may not have to spend his own money much longer. Following a presentation of his ideas at a recent telecommunications conference, Shepard was courted by the likes of Hewlett-Packard, Mayfield Fund and other would-be investors interested in his vision that energized an elite crowd at the Telecosm Conference in Squaw Valley, Calif., last month.

"Imagine people buying boxes and putting them on top of their houses and those boxes organizing themselves into networks that can blanket a metropolitan network," Shepard said in an interview at the conference. "Everybody would own their own piece and as long as you were working across the metropolitan area it would be entirely end-user financed — some people say free."

More specifically, Shepard imagines a $500 consumer electronics box that includes 6-mm-wave downconverter/upconverter radios each pointing in a different direction working at a 2-GHz chipping rate and delivering a raw data rate of perhaps 200 Mbits/second with a 15-ms delay. "You can already get the components you need to make this, but they are very expensive," he said.

Cost reduction — a process that could take four to seven years — is just one of the hurdles Shepard sees ahead. "There's a lot of research that still needs to be done," he said. "I've solved the hardest 10 percent of the problems, but there are still plenty of issues about how to build the microwave radios and the network infrastructure.

"There's a routing problem I have glossed over," Shepard said. That involves inventing "an addressing architecture and a way to tackle the problem of how do you find what you want to get to and how do you get a route to that," he said. "It's a classical computer science problem in the tradition of scaling large information systems, and I have a lot of faith that problem is solvable."

Indeed, Shepard's background is steeped in just such problem-solving. As a graduate student at MIT, Shepard worked with the group that developed the Multinix operating systems and helped lay the groundwork for some of the Internet's architecture.

"They had no idea it would scale the way it has today," he said. "But they got a lot of things close enough to design so that it was able to grow."

Oddly enough, Shepard's thesis was born from this tradition of large-scale systems and the Internet in combination with his somewhat nerdy hobby: radio hunting.

"I was doing hidden-transmitter hunting for sport," Shepard said. "Someone would chain a box to a tree in the woods and it would transmit on the first minute of every 15 and I had some friends with whom I'd go hunting for it instead of writing our thesis as we should have done. We learned a lot about radio propagation in the process of doing that."

At one point, Shepard lost faith in his vision of a self-organizing packet network, believing it could not scale to a meaningful size. Indeed, theoretically, as the number of radios approaches infinity the signal-to-noise ratio on the network moves toward zero and no one can communicate.

"But if you look at what that ratio is at 10,000 stations or 10 million stations, you can see for practical purposes the signal-to-noise ratio stops declining and effectively bottoms out at around one part signal to 100 parts noise. You can plug that in to Shannon's capacity theorem and that tells you what are the possibilities in radio systems engineering. That calculation got me going again."

With that bit of math, and the probability of some venture-capital seed money not far away, Shepard is back on track in his latest radio hunt. He may not be alone. Word emerged at the Telecosm conference of at least two other radio startups in the works with ideas not far removed from those of Shepard.