The problem with large, complex networks of sensors is that they're large and complex. A wired sensor network can be an expensive nightmare just to install, for example. Wireless networks save installation headaches, but they're notoriously unreliable.

But a new kind of network—the mesh network—is promising both ease of installation and reliability. A wireless mesh network builds itself—you put the nodes in place, and they form a network and start sending data. If some nodes fail for any reason, the network can even "heal" itself and keep working. Some real-world applications of wireless mesh networks have taken only hours to install and have operated with a reliability of essentially 100%.

Figure 1:  Small, wireless mesh-network nodes, such as this one from Dust Networks, can operate for several years on common batteries such as AA cells.

Using novel topologies and small, wireless modules that operate in the 900-MHz or the 2.4-GHz band (see Figure 1 and Figure 2), mesh networks are also power-miserly. The nodes of a mesh network typically run on common AA batteries that don't have to be replaced for three to five years—a big advantage in applications, such as building automation and industrial control, where nodes often go into places that are difficult to access. Mesh networks' combination of advantages just might, in fact, finally give wireless networks an industrial foothold. In the near term, however, the main markets will be less demanding applications like building automation and home security.

Figure 2:  The circuitry of mesh nodes fits on a very small board, such as this reference design from Ember Corporation. Batteries and an antenna occupy most of a node's volume.

Not coincidentally, mesh networks have many of the same target markets as the ZigBee technology, which became officially defined with the release last month of the first ZigBee specification. Until now, mesh networks have been available from a small handful of companies and have used proprietary technology, but one of the key players—Ember Corporation—has been deeply involved in ZigBee's development and offers both proprietary and ZigBee components. Now that ZigBee is set, says Ember's marketing vice president Venkat Bahl, many of the company's 100 or so OEMs will be moving to it.

Already, mesh networks have demonstrated some success. In Minneapolis, Supervalu supermarkets are using mesh technology from Dust Networks to gather and analyze energy-consumption data from refrigeration equipment. In Korea, mesh networks using technology from Ember are automatically reading utility meters. One network is in a large, high-rise apartment building; another covers half a city block.

So just what is a mesh network? It's a network with a topology that looks like a mesh (Figure 3). In a traditional star topology, many peripheral nodes connect to a central node. In a mesh network, however, nodes connect to each other. When the nodes of a mesh network send information, the data packets hop from one node to another until they finally reach their destination.

Figure 3:  In a mesh network, data packets hop from node to node until they reach their destination. This multihopping scheme is much more power-efficient than fewer, longer transmissions, and it increases reliability by providing many alternate paths.

Multihopping—the movement of data from node to node—is largely responsible for mesh networks' high reliability. Because data moves from node to node, and because a mesh network has many nodes, there is always more than one path from any one node to another. Consequently, if a node goes down for some reason, data packets from other nodes can go around it and take a different path to their destination.

Multihopping also contributes greatly to long battery life in mesh network nodes. The power required for transmitting is roughly proportional to the cube of the distance. So, three short transmission hops, for example, each of distance x, would use only 1/9 the power of a single hop of distance 3x. With more hops over greater distances, the power savings are even more impressive.

For additional power savings, mesh networks minimize radio use. In sensor networks, for example, a temperature sensor might need to acquire and send a temperature reading only once a minute and the node can be in sleep mode the rest of the time. Nodes do have to listen for other nodes' transmissions, though, and listening can use a lot of power. Consequently, timing algorithms control the schedule for radio operation to within milliseconds, so nodes don't have to listen any longer than necessary.

A mesh network can also be operational before all the nodes intended for it are installed. In the Supervalu supermarket energy-monitoring application, for example, the network was operating and sending information for analysis before the installer put the last node in place.

When nodes join a network, paths automatically form between those nodes and a destination node, which is often a gateway to a larger, traditional network. The nodes themselves establish routing simply by sending information that gets picked up and relayed by neighbor nodes. This relay process repeats, node to node, until the information reaches its destination. An acknowledgement from the destination node tells the originating node that a complete path was formed and that the information made it through.

Routing can also be a more complex process, with nodes using local intelligence to form routes rather than simply sending information to any other nodes that might be nearby. For example, nodes can measure the signal strength of neighbor nodes to see which one might offer a more reliable link.

Some mesh networks also include a "manager" node that provides additional routing control. For example, in a network based on Dust Network's SmartMesh technology (Figure 4), a manager node sees information from the entire network, whereas mesh nodes see only information from their nearby neighbor nodes. With this information, the manager can monitor reliability or latency over time, and it can dynamically establish different routing paths to optimize reliability, latency, or even overall network battery-power consumption.

Figure 4:  Dust Networks' SmartMesh technology uses a manager node to monitor overall network activity and provide deterministic routing.

The related concepts of redundant nodes and dynamic routing are enormously important for mesh networks because of the reliability they provide. According to Ember's Bahl, Ember networks consistently demonstrate "three nines" (99.9%) or "four nines" (99.99%) reliability. Rob Conant, vice president and cofounder of Dust Networks, goes even farther: "By figuring out which links are reliable and which are unreliable and modifying paths accordingly, we can establish 100% reliability in a mesh network."

That kind of reliability is a big deal for wireless networks, because point-to-point wireless networks historically have not been reliable. "You put a point-to-point network out there," Conant says, "and it's difficult to make it reliable unless the devices are very, very close together. In a typical deployment, point-to-point reliability between one node and another might be only 70% or less." But, Conant says, "With the mesh network, if you're within 30 meters of a few different devices, then even if you can't go down one path, you've got some other options." If it's highly likely that a node can talk to several other nodes, Conant adds, then it's extremely unlikely that it won't be able to talk to any.

Providers of mesh networks also increase reliability by building in some immunity to RF interference. Ember, for example, uses direct-sequence spread-spectrum (DSSS) transmissions, and Dust employs frequency hopping (FHSS). Each Dust node can communicate on any one of 25 different frequency channels in the 902- to 928-MHz range, and as new nodes join a network, each gets assigned a different channel. That way, if any path through the network gets blocked due to RF interference, it's still likely that another path will be open.

Fortunately, the low duty cycle of most sensor networks allows time to explore new network paths in case of interference. "You're sending maybe one packet per minute from a temperature sensor," says Conant. "If it doesn't go through, you try a different frequency."

To ensure that routing is deterministic in Dust mesh networks, the manager node gathers information about all the network's nodes and links and is aware of all possible routes. Then, by noting the percentage of packets that successfully reach their destinations, it can quantify network performance and assign paths accordingly.

Ember also quantifies network information, using a link-quality indicator (LQI) to predict a packet's likelihood of reaching its destination. Ember networks don't, however, use the LQI as a sole means of determining routing paths. "The problem," says Ember's Bahl, "is that link quality at time A is different from link quality at time B, because it's wireless. You could have a change in temperature, a change in moisture level, or a guy walking by at a particular time that you happen to be sending a packet." Changes such as those can change the link quality, Bahl says, "So we've chosen not to be deterministic to the point that this is the way a packet shall be routed every time."

In larger networks, with more nodes sending data, routing becomes more critical, of course. Although more nodes provide more alternate paths for data, they also produce more data that has to travel through the network. Most sensor networks don't sample data very often, however, so mesh networks don't—for now, at least—face a size limitation. The Korean application of an Ember network to read utility meters, for example, has over 1000 nodes.

Increasingly, says Dust's Conant, mesh networks will be used to bring real-world data into corporate IT networks, where it can be analyzed and put to use. Currently, a handful of companies make mesh network nodes, some with sensors included. Notable players in addition to Dust include Crossbow Technology, Millenial Net, and Zensys. Ember doesn't make nodes, but sells node component chips to companies that do.

As mesh networks proliferate, long-established semiconductor companies will be getting into the action, especially with ZigBee technology. Atmel and Microchip Technology both have ZigBee chips, for example. Freescale, a long-time ZigBee proponent, makes sensors in addition to its popular microcontrollers—a natural combination for mesh-network nodes. In addition, many other semiconductor companies have been active in the ZigBee Alliance and no doubt will soon join the ZigBee bandwagon.

Not everyone favors ZigBee, however, at least not for all applications. Dust, for example, expects to use the 802.15.4 radio technology, upon which the ZigBee network technology is built, but it doesn't plan on using ZigBee itself anytime soon. "We're trying to get ZigBee to go toward higher reliability enterprise-class applications," says Dust's Conant. "I think we'll get there at some point."