The residents of Cambridge, MA, may soon be able to log onto the Internet from any bus stop or city park. The city is working with MIT to go wireless, with a special focus on giving low-income residents access to the Internet.
The project is based on an experimental system called Roofnet, an unplanned, multiroute mesh network developed at MIT’s Computer Science and Artificial Intelligence Laboratory. A mesh network is a series of radio transmitters and receivers randomly dispersed over an area. To get data from one part of the mesh to another, the network must figure out the best route between them, which can change depending on network traffic, data rates, and even the weather.
Roofnet has been operating for about three years across an area of roughly four square kilometers near MIT, using a few dozen transmitting/receiving nodes and one wired Internet connection through MIT. The nodes have been located in the homes and offices of volunteers, most of them MIT students and staff.
A node consists of a small box containing a hard drive, software written by the researchers, and the same kind of radio card used in laptops, operating on the Wi-Fi standard. There’s an Ethernet port, into which a user can plug his or her laptop, and a connection to an antenna. Generally, an antenna has been attached to a roof, with a cable running in a window. But that has required flat roofs and users who can get up on the roof to install the antenna. Now Roofnet is experimenting with antennas that can be placed in windows; they won’t get as much coverage, since the signal can’t pass through the building, but they’re easier to use.
The original idea behind Roofnet was to exploit the benefits of a random, unplanned network. “It’s not like making a cell-phone network, where you have to plan very carefully where the cell towers go,” says Robert Morris, associate professor of computer science at MIT, who heads the project. With simple-to-use equipment that requires minimal maintenance, the Cambridge-wide network could be inexpensive and grow organically. The downside is that in some areas, where a node is far away from its nearest neighbor, the service can be unacceptably poor.
Cambridge plans to remedy these coverage problems by attaching antennas to as many tall buildings as possible. Jerrold Grochow, MIT’s vice president of information services and technology, says the city views the project as a utility, like providing electricity for street lights. “It’s not meant to compete with someone in their home buying cable modem or DSL service,” he says.
A random mesh of radios works very differently from the wired Internet, and Morris’s group has had to figure out protocols to get the nodes to communicate. Like the Internet, the network relies on transmission control protocol, or TCP, where data is broken up into packets and both the sender and receiver work to make sure all the packets get through and are put back together in the right order. With radio, though, one has to choose the best among several possible routes between sender and receiver. “You’ve got a huge choice how you route your data through the network,” Morris says. “Unfortunately, many of those choices are much worse than others.”
Some nodes may get radio interference that drowns out part of the transmission, or may drop out entirely for all sorts of reasons. For instance, transmissions sent on the Wi-Fi frequency of 2.4 gigahertz are strongly absorbed by water, so the network operates differently in the summer, when trees can be covered in rain-laden leaves. In other cases, a passing truck can reflect radio signals and cause interference. Nodes may also be so far apart that a signal only gets through intermittently.
Roofnet tries to work around these problems. Its nodes constantly broadcast status reports that signal where they are and which nearby nodes they’re in communication with. By tracking these status reports, the network can select the best route between any two nodes at a particular moment. If a connection drops out in mid-transmission, the network sends the data along a different route.
Monitoring the connection between each pair of nodes also allows the network to decide how fast to send the data. If the rate of transmission is too low, users don’t get all the potential out of the network. But if it’s too high, the signal-to-noise ratio drops and the data can get lost. The radios use a brute-force approach, sending out test signals at all possible rates between one and 54 megabits per second, to figure out the highest rate at which data can be transmitted consistently.
The first antennas were erected last week in a test phase, and Grochow expects that widespread use by Cambridge residents will begin in late summer. “It won’t necessarily be the highest speed, but it’s going to be everywhere, just like air, and that’s pretty cool,” he says.