Connectionless Networks Enhance Sensor Efficiency

July 2006
By Henry S. Kenyon

 
The goal of the Defense Advanced Research Projects Agency’s (DARPA’s) connectionless networks program is to change the way remote devices such as this prototype sensor manage power and data transmissions.
Program seeks to develop smart, flexible remote monitoring equipment, data architectures.

A new network management technology soon may change the ways unattended ground sensors are designed and operated. By focusing on the radio systems that link individual devices, scientists hope to create an intelligent networking architecture that uses the radio’s full communications capability both to conserve energy in a passive mode and to provide brief high-bandwidth data streams. Such operational flexibility would allow the development of multisensor devices able to activate a variety of onboard applications from microphones to real-time streaming video to meet intelligence collection needs.

The goal of the Defense Advanced Research Projects Agency’s (DARPA’s) connectionless networks (CN) initiative is to change the way remote sensor systems manage power and data transmissions, explains Preston F. Marshall, the CN program manager in the agency’s Advanced Technology Office, Arlington, Virginia. He notes that with the availability of more sensitive and capable sensors, it is becoming more attractive to remove warfighters from some high-risk battlefield positions and replace them with networked robotic sensors.

But while smaller and more efficient sensors are available, the communications systems linking small, unattended devices have remained relatively unchanged. Marshall observes that this situation has led to 1-milliwatt sensors connected to radios requiring several watts of power. “Communications was driving persistence; it was driving the mission; it was driving the mass for the amount of batteries. We needed to rethink networking for these very small, low-energy systems,” he says.

DARPA’s goal is to build very low-energy systems that can hold a network together but that also can meet sudden spikes in use. Marshall explains that many sensors currently have limitations because they perform only a specific mission. Passive acoustic sensors can listen only, and dedicated video sensors use too much energy to constantly cover an area.

Persistent sensing presents several unique challenges. In the past, he notes, designers coupled their sensors with radios developed for other missions. “If we’re really going to be effective at creating persistent sensing over regions, we’re going to have to rethink the technology,” he says.

The goal of the CN program is to develop systems that can adapt to a variety of changing mission needs. CN also seeks to save power by reducing demand on the network and the individual sensors in it. Marshall notes that Internet protocol-based applications are not effective for unattended sensors because they were designed for continuous, asynchronous use. “In connectionless networking, we wanted to do the opposite—build protocols that allow the network to be highly synchronous. That is, it [the sensor] can predict when it should be on and when it could be off and not pay a performance penalty for that and adapt those periods based on the mission,” he says.

DARPA scientists have developed devices that can power up and shut off in sub-millisecond time frames, which will allow sensors to conserve energy until they detect something. When a target is identified, the sensors will activate the appropriate applications to monitor the activity and then shut down as soon as possible.

Another aspect that makes sensor networking unique is the assumption in traditional networking that events occur randomly. But sensor networks are correlated—if something happened in one part of the network it is probably taking place somewhere else. “If an event occurs at this millisecond, it’s probably occurring at the next millisecond. Unique among the networking community, we have the advantage of seeing the future because the future probably looks a lot like the very recent past,” Marshall contends.

Program researchers are working to further develop this predictive capability by having the network configure itself ahead of predicted demand. For example, the latency for the first new information packet in a connectionless sensor network may be two seconds, but all the following data packets relating to a specific event will transmit within milliseconds. The network finds the most efficient path to transmit the information, briefly becoming a high-speed network before returning to a power conservation mode.

At the heart of the research is the challenge of making every node on the network and its neighbors adapt individually. For example, in a sensor field, a cluster of devices in one area may detect movement. Instead of activating the entire network, only that small group of devices will increase their data transmissions, perhaps switching from passive audio to active video surveillance.

Software programming for connectionless networks also shares some similarities to DARPA research with disruption-tolerant networking. “In IP [Internet protocol] networks, the network is a dumb thing. It either delivers packets or throws them away. In a connectionless network, the network takes responsibility,” Marshall observes.

 
DARPA’s research will allow unattended sensors such as this video surveillance system to more effectively form and maintain networks. The program also seeks to create a new generation of multipurpose sensors capable of switching from passive audio to active video monitoring as the situation requires.
For example, an end node seeking to source information in an end-to-end network must manage the transfer of data. As a consequence, the node remains active to complete the transmission. If a packet is lost in the process, it must retransmit the entire message. If data must make multiple hops across a sensor network and information is lost, the nodes at the beginning and at the end of the transmission communicate about the packet loss. In connectionless networks, each hop between nodes is negotiated between adjacent devices. If there is a problem, the nodes resolve it locally. “Unless the network says it couldn’t deliver it, you assume it was delivered,’’ he says.

By transmitting data locally as opposed to end-to-end, connectionless networks can save large amounts of power. Internet protocol (IP) was designed for large file transfers, but sensor networks usually transmit only in small bursts of a few bits of information. Marshall explains that IP systems are roughly six orders of magnitude away from full efficiency. “That’s the equivalent of a car getting 12 inches per gallon,” he observes.

Connectionless networking is about four orders of magnitude away from true efficiency. He notes that, compared to wireless protocols such as 802.11 operating mobile ad hoc networks, the DARPA technology requires at least two or three orders of magnitude less energy to operate than current commercial wireless systems.

Phase one of the CN program worked out basic technical concepts. Phase two was a laboratory implementation that set the technology’s parameters by determining when the network would rest and when it would report data. Marshall notes that researchers were able to find sets of values for a radio’s different modes that could create desired performance ranges.

The goals for phase three are making the networks both self-managing and self-optimizing. At the end of this 15-month phase, the nodes will be able to automatically adapt themselves and change their settings to meet network requirements, Marshall predicts. Another part of phase three is to push the network management into the devices. At the end of phase three, the goal is to create a prototype form factor that can fit into a small sensor.

One challenge for the CN program is integrating how designers look at analog and network protocols. DARPA is working with the U.S. Army Research Laboratory (ARL), which is conducting its own investigation of sensor and low-energy radio technology. Marshall explains that DARPA is leveraging the ARL’s work on analog components. The partnership also is producing software and linking it to the hardware’s performance. “At the end [of the program], we want to demonstrate its military benefit. We’re going to put a sensor on it, both a high information rate and a low information rate and demonstrate not just the radio but also the military capability of the program. It will be a capability rather than a radio demonstration. The difference is having to go out and change the batteries every day when you could leave it out there for three or four months. The lifetime is in the batteries. It’s a shame to have that $500 sensor die because a $2 battery has run out,” he says.

Marshall explains that unless it is continuously monitoring video, CN technology should increase the lifespan of remote sensors’ communications systems by 100 percent. Although the concept of high-energy, high-data-rate/low-energy, low-data-rate networking has been accepted for some time, adaptive technology is an entirely new idea. “It [the network] can decide at any instant what’s the right mode, rather than a system engineer deciding years in advance,” he says.

Writing software that will allow sensors to manage power is a major challenge for the program’s next phase. Marshall says that one of the modes he hopes to develop will allow an entire network to use its energy evenly. This capability goes beyond individual devices sensing battery loss, it permits the entire network to prevent node failure and to retain battery power. The goal is for applications to select a variety of energy use options that will try to distribute the power burden as evenly as possible across the network.

The network itself is another important environment. For example, if the network’s sensors detect a large amount of vehicular traffic, it will try to optimize its data paths to provide the right amount of data for the mission. The difference between a connectionless network and other types of architectures is that the system identifies heavy use along certain data paths and attempts to optimize the efficiency of those transmissions. The network may in fact decide to use more energy to collect more data about a target. “It’s kind of like a little venture capitalist in each node thinking about investment strategies—what’s the best place to use energy and how fast will it pay back,” he shares.

Another program goal is for the network to be able to surge up to 1 megabit per second to provide real-time video and audio. This feature will allow engineers to develop multipurpose sensor systems and devices. Marshall believes that connectionless networks will change the nature of sensors. “You want the same sensor that heard the vibration to become the visual sensor. If I can get peak bandwidth—which I might use only five minutes in the sensor’s life—then I can open up the opportunity to think of the sensor as having multiple phenomenologies [sensors]. I can do more fusion, because I can pass data around locally, which I couldn’t before. So that dynamic range, which we can get with a radio, opens up the opportunity to build much more capable sensors,” he says.

 

Web Resource
DARPA Connectionless Networks: www.darpa.mil/ato/programs/CN/index.htm