San Diego, California - Most of us use many devices - perhaps too many - throughout the day: A smartphone at home and then at the cafe around the corner, a laptop computer at work and maybe a tablet or e-reader in the evening. That's not counting all of the other possible "smart" devices at our fingertips, such as health monitors, fitness trackers and smart watches.
Few people think about how all these devices could be efficiently networked via software and wireless technology; however, if you were trying to download a video and coworkers on the same network were attempting to communicate via Google Talk, you would want to be sure those limited wireless resources were allocated optimally.
Giorgio Quer, a postdoctoral researcher in electrical and computer engineering at the University of California, San Diego is on a quest to solve these and other networking problems. Quer has been working on "cognitive networking" and exploring related projects at the UC San Diego division of Qualcomm Institute for the past five years as a visiting scholar from the University of Padua in northern Italy. He works with Ramesh Rao, director of the Qualcomm Institute and principal investigator of the research, as well as Matteo Danieletto, another postdoc in the department.
The concept of cognitive networking, according to Quer, refers to "a way to apply cognition to wireless, where a network learns from past history." Such a process enables a network to "perceive" and learn about its current conditions and then plan, decide and act on those conditions, resembling (in a limited sense) cognition in the human brain.
Adds Rao: "The network elements gather and share vast amounts of data on the states they find themselves in, which opens up a rich set of possibilities using machine learning that can further enhance the functionality of the networks."
Within the past few months, Quer and Danieletto wrote two papers in IEEE electrical engineering publications on testing and implementing cognitive networking with wireless devices such as smartphones and tablets. They use devices that run the Android operating system, since these devices are relatively inexpensive, mobile and highly customizable.
They began with a theoretical framework in which a system collects relevant data about a network and uses "machine learning" tools to probabilistically identify patterns and assess structures within those data. These often turn out to be trends and relationships that develop between key variables or parameters, such as memory storage and data congestion. The system then determines how the parameters influence each other and the network's performance.
Quer's research, partly funded by the U.S. Army Research Office, began as he and his colleagues thought about utilizing wireless technologies in tactical networks, which in a military context often operated with extremely limited resources in hostile environments where the operators must avoid interference while maximizing the security of transmissions. In these conditions, one must make many decisions on-the-fly, such how to connect, which device to connect to and what to transmit.
Now the researchers are thinking bigger about many possible civilian and commercial applications of flexible and adaptable networks. They might have competition, however, as researchers at Google and elsewhere are also exploring the idea of incorporating some kind of intelligence or learning system in networks.
Devices with cognitive networking capabilities could also be designed to transmit more data while reducing energy consumption and efficiently using available bandwidth
Another example: In their recent papers, Quer and his collaborators describe a network comprised of a personal smartphone and physiological sensors, such as those detecting heart rate variability, involved in the "ambient assisted living" of elderly people. In the future, such networks would provide the elderly with digital environments that are sensitive, adaptive and responsive to their needs — say, a system that would alert them when their heart rate variability is less-than-ideal. Quer dreams of connecting wearable devices in this kind of network, including one comprised, in part, of wristbands and health monitors such as those popularized by Fitbit, Polar, and Zephyr.
The next challenge is to combine cognition with the "Internet of Things," networks that wirelessly connect electronics and software without requiring human interaction. Applications could include networking air and water quality monitors, tracking devices’ energy consumption, and optimizing the operation of next-generation “smart grid” electricity systems, in addition to the medical applications Quer and Danieletto examined. Some obstacles remain, such as the "admission control problem," in which the maximum number of nodes in a network is restricted in order to maintain another variable. An example is a network of Skype calls where sufficient call quality must be maintained at the expense of serving greater call volumes.
Quer anticipates that cognitive networking will pose possible privacy and security concerns. Once people's devices have cognitive networking enabled, "these networks start to make decisions, such as about which network to connect to," he predicts. “This might be "a bit scary," since personal information such as Facebook or physiological data could be compromised. Recent hacking scandals have demonstrated that even federal employees' Social Security numbers, health histories and other sensitive data are vulnerable.
In any case, assuming that the next generation of cognitive networks can maintain the security of apps and data, Quer and his colleagues at Calit2 look forward to extending and connecting their cutting-edge research to healthcare technologies and their myriad applications.
Notes Quer: "The work we’re doing is creating a network to readapt to new devices or new needs that pop up.”