With today's cell phone, the user's access is limited to the provider's network. Mitre computer scientist Joseph Mitola, currently working with both the National Security Agency and Defense Advanced Research Projects Agency (DARPA), thinks he has a system that will change that. He calls it "cognitive radio" and says his system will give wireless devices machine-learning capabilities. Mitola talked to CTO Media Senior Editor Loretta W. Prencipe about how his vision of "cognitive radio" would work, and how it could redefine cell phone technology.
IW: You coined the term cognitive radio. What is it?
JM: A cognitive radio (CR) has a computation model of itself. It knows that it is a smart radio, and it has a user who does certain things. If you're a journalist, you might be willing to pay a premium to get higher speech quality. If a CR detected an interview, and it would cost three cents a minute for a clearer signal, [a message] would pop up on the display asking if you want to pay three cents more a minute for clearer audio. Over time, it would learn and would build into the computation model that the user likes high-quality speech when doing interviews. This computation model of itself, of the user and the network, plus the machine learning means that the user doesn't have to reprogram it and keep telling it what to do. ... There isn't a true working model yet. It may be five to 10 years out.
IW: Define the architecture of a CR.
JM: It's a software radio. This would have enough flexibility in the hardware to be programmed to a band or mode. So instead of being stuck in the 800 to 900MHz band, it would be able to adapt over to an ISM band or up to an IEEE band or 5GHz. The CR would know what to do based on experience. It knows where home is. You get in the car to go to work. It's measuring the radio propagation, signal strength, the quality of the different bands as it drives around with you. It's building this nifty internal database of what it can do when and where.
IW: Could CR lead to convergence of wireless devices?
JM: Certainly. It would enhance convergence by simplifying the user's interaction with the increasingly complex devices. The vision of CR is that your PDA, laptop, and automobile would employ the mix of Bluetooth, IEEE 802.11, and cellular standards from 1G to 3G as needed by the user.
IW: What are the technical challenges?
JM: Affordability of the services. Right now there is a lot of pressure on service providers for profitability for [a] third generation of wireless. The big winners made a lot of money and captured the lead in [the] second generation. No one is making money in 3G. It costs a lot to deploy. The killer app has been slow to appear.
IW: What does a cognitive radio cell phone look like?
JM: In order for a CR to know things that are important, to deliver this level of service, it's got to have a lot more sensors on it than today's radio. For a CR to know that it just fell on the floor, to send a page to the user's Blackberry when it falls on the floor, it needs an accelerometer, a motion detector. A CR would wait a reasonable amount of time and then send a little page to my Blackberry saying, "Hey, you just dropped me. I'm back at the coffee table." [Because] GPS doesn't work well in buildings or urban settings, it would also need more sophisticated built-in sensors.
IW: CR would find the quickest and best spectrum to send video, to pick up audio?
JM: Not necessarily the quickest. If you are signed up to pay for the video, [your service provider] will be glad to give you the bandwidth. But if the user wants the most affordable way to get the video, then the CR could be an intelligent agent -- with an algorithm acting on the user's behalf -- and look around for the options and say, "Let's use this 802.11 over here because it's free." It's a matter of getting it done affordably and within the user's constraints. The free path might not be encrypted. But [if] the user wants what he's saying to be encrypted, he might be willing to pay a few more bucks on a 4G or 5G network to get the service encrypted. It's the user being able to define the parameters, and the CR will be able to pick that on his behalf once it has learned what his preferences are and be able to adapt.
IW: What are the potential enterprise uses of CR?
JM: The early practical value would be to allow the intelligent shaping of offered traffic. Say I'm on the subway and have 3G wireless at $1 a minute, and I have an e-mail with a 10MB attachment. Today's wireless would simply send and you'd be spending $1 a minute for megahertz on that network. With software-defined radio, you'd pick the network. A cognitive radio would pop up and say, "Hey, you're only 10 minutes from work, and the enterprise LAN is free. How about if I hold off on the attachment until I get to work?" Service providers are hoping that 3G and 4G will generate a lot of revenue. They are really competing with the office LAN for that traffic. And while software radios and software-defined radios will pretty much put the enterprise as a victim of whatever policies the network service provider wants to offer, the CR puts that power in the hands of the user.
IW: Can we clarify this example? Don't most users already have some level of control?
JM: Most users don't have the cognitive level of assistance in "shaping traffic," which is deciding what to pay for what, where, and when. Most users are not capable of understanding the complex tariff structures and trade-offs, which change on a frequent basis, as do the needs of the users. So users have to make a once-and-for-all choice, such as picking up the Blackberry to use that network or getting on the corporate LAN, which could in the future be a secure wireless network.
A CR that is managing your laptop could use the BlackBerry band and mode -- not just when you are away from work but also when you are at work and have a short but high-priority message when the network server is slow. In another situation, it could switch over to the LAN for a big message such as the 10MB attachment because it knows that the LAN is cheaper and that timeliness is not of the essence or that the server is sending messages quickly again. The dynamic knowledge needed to do this last scenario is not even readily knowable by a user, much less employed in an effective way on behalf of the user by today's networks. The CR of the future may be an "active agent" that has all the features of software agents, plus is integrated into the users' wireless device to act on behalf of the user.
IW: Some organizations support an open spectrum for radio communications as part of free speech rights. Others say the technology has not been developed for all communications to be handled on open spectrum -- in fact that open spectrum would impact some military and R&D operations. How do you view this debate?
JM: Both are right. This is like the Miller beer commercial -- less filling, tastes great. There was a city where a hospital was using heart monitors in an ISM band. The users were told that there would be a digital TV station in that band. Sure enough, when the digital station was put in that band, they had problems with heart monitors. That created a panic. A cognitive heart monitor would have been digitally told by the network to go over to another frequency. All that could have been done by algorithm and data, without people having to change the heart monitors. … A heart monitor with a CR would go looking around for a band without that much interference and would have a protocol between the heart monitor and base station. A cognitive [radio device] would generate a data network packet to the FCC and say, "Hey, I just got walked on by something that looks like a digital TV station, and I thought I had that frequency."
We would go from an era where it was all done manually, without any degree of automation, to an era that would be highly automated with measurements made by the radio itself. That would enable open spectrum.
IW: How would you protect legacy spectrum licensees?
JM: The policeman pushes the talk radio. When he lets go of the mike, the channels are clear. [One CR] would talk to another on [that policeman's] signal with what you call a listening mode. About 30 times a second, it would stop transmitting and listen a little bit. If it heard the carrier is up or the voice was starting up, it would simply stop transmitting and look for another clear channel. The legacy users just push to talk and start talking. The CR is polite. That's an etiquette.
IW: Would a CR create this spectrum knowledge?
JM: It would be preprogrammed with this basic spectrum stuff that would fit in a few 100MB. For a handset of the future, putting a 100MB of radio spectrum uses, like knowing that this spectrum is for satellite, would be okay. Let's say that you have a band allocated to air nav, and I'm in the middle of Arizona, and there are no pulses. A CR might pick that band and use it at a very low level, and would use a listening background function. As we get closer to a city, the CR would detect signals and switch to another band.
IW: Is that transparent to the user?
JM: Yes. For good etiquette, the CR should keep an audit trail of what it did so that spectrum managers of the future could diagnose what's right or wrong with CR, and how to make them learn better, adapt better, and know which have algorithms that are violating spectrum rights.
IW: What about security?
JM: These wireless networks are easy to hack into. All that is the tip of the iceberg of some real challenges in keeping data secure and protecting privacy rights of those who will be using future data networks. I envision CR as something that basically has to be trusted. If … DARPA wants CR to be a military radio, boy, what kind of policy changes [must] I have for my notion of information security to step up to the vision of CR in the future? That could be a strategic question for everyone.
IW: Privacy policies in effect?
JM: Yes, and for military uses, it's national security to protect and yet to provide a degree of access. Say you're fighting with a coalition partner, and you're shot, and your radio is lying next to you, and the coalition partner picks up the radio and says, "Hey, I need a medic." If it's a secure radio, and you have a coalition power trying to get on a secure network, that's a tricky circumstance. If the radio were a cognitive radio, it would look at this guy through its multimedia and say, "Hey, you don't look like Joe." The guy would point the face of the radio at Joe and say, "I need help." The radio might let him do a call for a medic and not let him do much else.