The sky isn't falling. Contrary to Chicken Little's apocalyptic hypothesis, there is no shortage of available spectrum upon which to deploy next-generation wireless services. In fact, there's an abundance. Study after study shows that only 2 to 6 percent of all available spectrum in the United States is being used at any one time.
Unfortunately, though, there's an equal abundance of regulations that bar access to that spectrum. So, what is in short supply are the policies, technologies and methodologies that would allow a suitably equipped radio to access spectrum intelligently worldwide without incurring the wrath of those now using it from TV broadcasters and cellular operators to radio astronomers and the various militaries. Enter cognitive radio.
A CR is "a really smart radio that would be self-aware, RF-aware, user-aware, and that would include language technology and machine vision along with a lot of high-fidelity knowledge of the radio environment," said Mitre Corp. researcher Joe Mitola, who coined the term in the late 1990s. And, for a Federal Communications Commission desperate to appease a spectrum-starved wireless industry, the concept was irresistible. In late 2003, the FCC's newly formed Spectrum Policy Task Force issued a notice of proposed rule making calling for input on how cognitive radio could be realized commercially. In the interim, the task force also issued regulations for the use of ultrawideband and opened up spectrum for unlicensed use in the 5-GHz and 60-GHz bands.
While the FCC may have adopted CR in principal, the wireless industry has unabashedly embraced it in the belief that cognitive is the next big thing after software-defined radio (SDR). "It's an extremely challenging and interesting application," said John Chapin, chief technology officer at Vanu Inc. (Cambridge, Mass.). "The FCC sees it as one of the few technologies that can have a major impact on their frequency crunch, and the military wants it because of the tremendous problems they face with frequency planning. Whenever they do an operation overseas or even domestically, for training they have tremendous frequency-allocation problems." With a dynamically adaptable radio, Chapin said, the military will no longer be confined to a static frequency plan and can adapt as needs change on the ground.
"Those are tremendous drivers and they've really brought cognitive radio along faster than I would have thought," he said. "Spectrum is a critical national resource that is now being underutilized."
One of the more immediate commercial applications of CR now being studied is the reuse of TV's UHF bands in the 80- to 850-MHz range, where the propagation characteristics allow for long-range, nonline-of-sight, last-mile Internet access.
While the concept of a smart radio that can dynamically adapt to its environment and reuse licensed spectrum on an opportunistic basis sounds attractive to many, incumbent spectrum licensees aren't among them. They perceive reuse as threatening to their business model. "If there hadn't been such strong government drivers, [CR] would have been dead on arrival," said Chapin, who credits the FCC for having pushed CR so far over the protests of existing spectrum users.
Pushback from incumbents, however, is only the start of CR's problems though the wireless-engineering community has, typically, recast those problems as innovation opportunities.
The opportunities start at the hardware level, where complex, frequency-agile RF front ends and software-defined radios, while not mandatory, are viewed as a foundation for CR in its most extreme form of a radio that can jump in and out of any band. The XG program of the Defense Advanced Research Projects Agency (Darpa) is the most ambitious dynamic spectrum-adaptation effort to date, calling for a radio that can cover the entire swath from 30 MHz to 30 GHz, although that's considered an endgame approach. In the interim, "hopscotching" whereby a radio can jump between specific bands of high interest, such as those between 800 MHz and 2.45 GHz is the more feasible approach.
Beyond the flexibility of the radio itself, CR faces interesting issues that are peculiar to its intended use. Aside from the business model, most questions revolve around how a radio senses its environment and provides feedback to the upper layers of the stack in real-time to avoid interfering with incumbents. Even more specific to CR is the debate over how, given the complexities of the decision making involved, such an intelligent radio can even be programmed. How that radio is to be certified in the face of multiple layers of system uncertainty is also fertile ground for innovation. The latter point makes the current certification requirement of out-of-band emission suppression seem tame by comparison.
All these issues are complicated by the fact that a clear definition of CR's characteristics and requirements remains elusive. How much intelligence is really needed? Exactly how flexible does the radio need to be? Aside from the actual air interface, how far into the wired network should cognition be implemented? All these questions are still to be resolved.
"We're still in the research and concept development stage," said Al Margulies, executive director of the SDR Forum. "The goal is to make more effective use of the spectrum, and for these ubiquitous networks there's a general feeling that it's not just about a cognitive radio but also a cognitive network."
Hardware problems receding
While the underlying SDR technology has been long in coming, Margulies is confident that many of the technical tough nuts are being cracked. Developments outlined at the SDR Forum's recent conference in Anaheim, Calif., support that belief (see Nov. 21, page 1). For example, front-end filtering technology from startups such as Discera Inc. could possibly be paired with the recently announced Softransceiver from BitWave Semiconductor Inc. to provide much-needed RF front-end flexibility.
In the meantime, the conference described advances in FPGA, DSP and structured-ASIC development tools and programming environments from the likes of Pentek, Spectrum Signal Processing and Mercury Computer that can accelerate waveform development and allow rapid baseband reconfigurability at ever-lower power and cost. In addition, the Software Communications Architecture (SCA) 3.0 promises to provide the security, application programming interfaces and code density required to make cross-platform waveform portability a reality.
"SCA is the operating system the Windows of radio which defines the interfaces, connections and plug-and-play," said Margulies. "It's a functional specification that can be implemented on Corba or Java."
SCA 3.0 is currently in a state of "strategic pause," according to Mark Turner, director of Harris Corp.'s RF communications software and security products engineering organization. But an SDR Forum spokesperson said the organization continues to plan what forms of support for SCA-related standards and extensions and associated compliance testing would be of greatest value.
Meanwhile, Margulies said, there is plenty to do beyond SCA in terms of developing the design processes and tools to simplify and streamline SDR system development. He pointed to PrismTech, Synplicity and Object Interface Systems as companies going in the right direction. "It's a market that's now being recognized and there's lots of emphasis on it," he said.
As the hardware underpinnings of the most extreme and flexible version of cognitive radio get ironed out, academia and industry are tackling a sense of dissonance left in the wake of the FCC's actions on cognitive radio. Where radio development once dealt with linear optimization of individual and well-defined bands, researchers must now think in terms of horizontal development across a wide, and possibly contiguous, swath of spectrum. For many particularly incumbents it's an uncomfortable transition. For others, though, it's a gaping opportunity.
Universities worldwide have started researching the theory, realization and practical application of CR. Most of the CR programs are based in the United States notably, at the UC Berkeley Wireless Research Center, Virginia Tech and Georgia Tech. Farther afield, the Universitaet Karlsruhe and RWTH Aachen University in Germany and Trinity College in Dublin, Ireland, have also mounted cognitive-radio initiatives.
On the industry side, the SDR Forum in January set up the Cognitive Radio Working Group and the Cognitive Applications Special Interest Group. "The CRWG is a natural extension of our work in SDR and is the next stage in developing intelligent radios," said Margulies. The working group will focus on the radio development, he said, and the SIG on applications.
Within the IEEE, the 802.18 Radio Regulatory Technical Advisory Group has spun out the 802.22 working group on wireless regional-area networks. The specific aim is to craft a standard for a CR-based physical layer and media-access control layer (PHY/MAC) air interface that license-exempt devices could use to access spectrum allocated to the TV broadcast service on a noninterfering basis.
"The singular advantage of this technology is the propagation characteristic in terms of range and power," said John Notor, wireless architect at Cadence Design Systems Inc. and an early participant in CR's development. According to Carl Stevenson, the IEEE group's chair, "the goal is do something equivalent to wireless DSL or cable but not 802.11a at 54 Mbits/second."
Covering the spectrum from 54 to 860 MHz, the PHY/MAC would allow for minimum download and upload rates of 1.5 Mbits/s and 384 kbits/s, respectively, at the fringe of what Stevenson predicts will be a coverage radius of 40 km. "The MAC will handle round-trip delays out to 100 km, and CPEs [customer premises equipment] down close could [get higher rates if they] use higher-order modulation schemes such as 64 or 256 QAM."
The group's work, according to Stevenson, takes a "divide and conquer" approach in a relatively narrow range of frequencies. More important, the incumbents are easy to detect and, therefore, to avoid, thanks to their well-documented bands of operation and in the case of digital TV their use of easily identifiable pilot tones. "I call it cognitive radio 'lite,' " he said.
Nevertheless, Stevenson is well-aware of broadcasters' ongoing concerns about potential interference issues, but he said that representatives from the National Association of Broadcasters, Fox and CBS were attending the group's meetings. "They believe something will happen [with spectrum reuse] and are helping to make sure we do it right."
The group will be reviewing proposals in January, and Stevenson is optimistic given the level of agreement he's seen so far that a baseline proposal will be ready by March. The many issues to be hashed out include TV signal detection and avoidance, with options including the use of spectral analysis, beacons and GPS.
While the FCC has stalled in its push of TV band reuse due to fears of interference, Cadence's Notor considers those concerns overblown and is adamant that reuse can be implemented with current technologies.
The IEEE working group's efforts may get a shot in the arm from work under way at Ireland's Centre for Telecommunications Value-Chain Research. Led by researchers from Trinity College, the CTVR is using orthogonal frequency-division multiplexing (OFDM) as the foundation for a spectrally sculpted CR scheme that it believes will aid signal detection and interference avoidance.
"You need good measurements to make a good estimate of what's going on as well as a means of controlling the spectrum [of the transmitted waveform]," said Keith Nolan, a professor at Trinity. "That's why we chose OFDM." OFDM, he said, can be modified at will, or "sculpted," to avoid interfering with incumbents.
The CTVR is setting up a test bed in the Dublin area in the bands between 1.6 and 2.5 GHz that it will open to other researchers worldwide for CR experimentation via a Web interface. More on the project can
be found at
At the SDR Forum's Anaheim conference, Nolan described how the Trinity group managed to perform frame synchronization, carrier offset estimation and subcarrier allocation in a single symbol, to reduce overhead.
Meanwhile, the IEEE Communications Society and IEEE Electromagnetic Compatibility Society early this year set up another CR-related body, the IEEE P1900 Standards Group. Its objective is to devise supporting standards related to new technologies and techniques being developed for next-generation radio and advanced spectrum management.
Programming and certification
As the intelligence level increases, the issue of programming a cognitive radio rears its ugly head. Language is problem No. 1.
According to Vanu's Chapin, for a sequence of imperative statements that occur in time (that is, A=B=C), it turns out for mathematical reasons to be much harder to prove the sequence's overall properties when using a procedural language such as Java, C and C++ than it is using a declarative language like Web Ontology Language or Prolog.
"A declarative language is a sequence of axioms and propositional logic, and we have very good methods for reasoning about whether you can conclude certain results from a sequence of axioms," Chapin said. "It is much harder to prove that an imperative program [developed with a procedural language] does or does not do a particular action."
However, at a panel on CR at the SDR Forum event, Chapin made the point that declarative languages, despite their flexibility, are not a silver bullet. "If you get some radio rules from the maker for the radio, then from the FCC, then the military and then the county in Nevada where you're performing the exercise, then it's entirely possible that you'll get conflicting rules," he said. "The process of proving there are no inconsistencies is in itself a problem." That's because "for a sufficiently large knowledge base, it becomes impossible to compute it in a reasonable amount of time," he said.
Of particular interest to Chapin is the issue of certification and assurance of CR radios, once developed. "Certifying a device means not only making sure the policy and policy reasoner are correct, but also that all of the things that provide input to that policy reasoner are correct," he said. This means ensuring that all the external sensors work correctly, there are no bugs in the software and the output conforms to what the reasoner dictates.
"It's much bigger than just getting the policy reasoner right," said Chapin, who is active in the IEEE P1900.3 group trying to resolve that issue.
While the CR hurdles are manifold, so are the paths to solutions. In typically optimistic fashion, Preston Marshall, director of Darpa's XG communications program, said at an SDR Forum panel that CR-based spectrum management will arrive in 18 months. However, according to Trinity's Nolan, adding only a few elements of intelligence to CR-like features already in use such as transmit power control and dynamic frequency selection may be enough.
"A small amount of this technology could provide 80 percent improvement," Nolan said. Beyond that, developers reach a point of diminishing returns.