Hi, John, this is your cell phone. You dropped me at the corner of 23rd and Fifth next to Starbucks. I've e-mailed you directions. I'll be waiting." Science fiction? Not if the FCC, the military and now the semiconductor and software industries have their way.

Driven by a need to more intelligently optimize spectrum use, the FCC issued a landmark request last December for input into how best to implement the concept of cognitive radio (CR). With the potential to endow radios with the ability to decide for themselves which bands to use based on availability, location and experience, CR has effectively surpassed software-defined radio as the Holy Grail of radio design-and sent the radio design and research community into overdrive.

Derived from seminal work undertaken by Joseph Mitolo, CR's ability to learn is what truly sets it apart conceptually. With sufficient artificial intelligence, it may have the ability to access experiences-such as dead zones, interference and usage patterns-and apply them to respond to real-time situations.

While the software and intelligence requirements have yet to be worked out, the implications for the radio itself are clear: If a CR is going to jump around the spectrum optimizing for power, range and required data rates, then an extremely flexible RF front end is required. This front end must operate over licensed and unlicensed bands and handle a plethora of interfaces. These interfaces range from UHF TV bands at 800 MHz; to Bluetooth, ultrawideband (UWB) and wireless LANs; and out to WiMax and the multiplicity of 2G, 2.5G and 3G cellular standards.

It's not that today's radios aren't already pretty smart. Lowly Bluetooth has enhanced its coexistence potential through adaptive frequency hopping that helps it avoid interference. Cellular radios adapt their power levels, and Wi-Fi wireless-LAN radios perform adaptive modulation to lower their data rates somewhat intelligently based on the environment.

Looking out further, researchers are exploring better ways of choosing the access points with which a Wi-Fi client associates. Instead of blindly choosing the AP with the highest signal strength, the client may soon interrogate the AP as to its data load and, if the load is too high, opt for an AP with a weaker signal but greater available capacity.

However, all these radios reside with a single band, so almost all the intelligence and flexibility stems from processing in the baseband and modem or media access control (MAC). For many years, particularly within the military, software-defined-radio has been a siren call to designers looking to complement this flexibility in back-end processing with highly flexible RF and mixed-signal front ends that could adapt to multiple bands. Success has come but has mainly been relegated to infrastructure or wall-powered systems; it has proved elusive in power-lite mobile devices.

Now, inspired by the potential of CR and its general recognition as the next big thing in wireless, designers, researchers and academics across the globe have reenergized their agile-RF efforts-though the hurdles remain massive. The power and performance issues associated with acquiring and processing a swath of spectrum ranging from hundreds of MHz (TV) to 6 GHz (Wi-Fi) in a low-power environment will require exploration of new concepts in antenna, switch and filter design, as well as down/upconverters and analog-to-digital conversion. These, in turn, must be combined with ever-evolving process technologies if the potential of CR is to be realized.

At an extreme, this will allow it to share licensed bands simultaneously with incumbents, though it is more likely that the cognitive radio will have to find another band once the incumbent restarts transmission.

However, determining the interference a radio may inflict upon other unknown radios is difficult. The research community is torn between a fully autonomous sensing technique or a database-driven alternative into which all radios would feed their own parameters and locations, then reference the database continuously for updates on which radios are within range and what their interference-mitigation capabilities are. Given the magnitude of the problem, a combination of intelligent sensing and a database may be a more appropriate design.

But not so fast. Sampling at RF frequencies automatically raises the power consumption. Also, the A/D will require much higher resolution to handle interferers, translating to more bits in the A/D. While the A/D's efficiency per bit has improved dramatically over the past 15 years-going from between 10 and 100 picojoules down to 1 to 10 -the fact that power scales exponentially with bit count means low-power 12-bit converters at 2 Gsamples/s are not on the horizon anytime soon.

The alternative is to use fewer bits and sample faster, as the power consumption scales linearly with clock rate (vs. the exponential relationship power has with the number of bits used). This has led researchers in the direction of 1-bit sigma-delta converters combined with oversampling techniques that sample at many times the frequency of the signal. While Nyquist calls for 2x sampling, oversampling reaches 8x or higher. The benefit is that much of the signal retrieval can be done through analog front-end feedback and digital signal processing, but the downside is the extremely high sampling rates. For example, for the equivalent of 2 Gsamples/s, the converter would have to oversample at 16x, translating to 32-Gsamples/s. That's not feasible with today's technologies. While flash converters can reach higher rates than the traditional sample-and-hold converter, the trade-off is higher power consumption. Consequently, research should focus on simpler sigma-delta converters with oversampling-with emphasis on raising the sampling rate.

Though RF sampling is in the very early stage of development, Texas Instruments did leverage its 90-nm high-speed CMOS capabilities to demonstrate an on-channel, RF-sampling direct-conversion Bluetooth radio this year. The radio was implemented using TI's Digital RF Processing (DRP) technology and sampled at the RF frequency of 2.45 GHz. Since Bluetooth has only a 1-MHz-wide channel, a relatively low data rate (1 Mbit/s) and relatively low performance requirements, it was a good place to start. That said, the company will use the same technique for a single-chip GSM radio that it plans to unveil by year's end.

The debate continues as to whether a cognitive radio should be the brains of a system that focuses on the vertical integration of cellular interfaces or migrate to agile RF front ends, in order to accommodate the myriad wireless standards that are used in the growing headset market.

That said, the drive toward seamless mobility between cellular and Internet Protocol (IP) networks such as Wi-Fi has introduced a nontraditional decision point. Designers of radios for cellular (GSM/GPRS/EDGE/W-CDMA and 2G/2.5G CDMA variants) and the multiplicity of wireless connectivity solutions (Bluetooth, UWB, Wi-Fi, WiMax) must decide whether to integrate vertically or horizontally. Startup Quorum Systems Inc. has chosen the latter, announcing that it has condensed a GSM and 2.45-GHz Wi-Fi/Bluetooth radio transceiver section into a single chip. Another startup, Berkana Wireless, has chosen the former, integrating a GSM/GPRS radio transceiver in CMOS. For Berkana-and almost all major semiconductor manufacturers-the attach rate on cell phones for Wi-Fi and Bluetooth will remain too low in the foreseeable future to justify their integration with the cellular radio.

However shrewd its strategy and inventive its design, Berkana is now going toe-to-toe with those mainstream semiconductor companies, while Quorum's horizontal-integration play may well be the differentiation a startup needs to succeed once the attach rate rises. Efforts such the recently announced Unlicensed Mobile Access (UMA) indicate operator interest in the rapid deployment of converged handsets, starting early next year. Such converged handsets play into Quorum's strategy.

This vertical integration strategy has one other major advantage: It is in line with the fact that until the cellular and wireless connectivity standards bodies get their efforts aligned-which won't happen anytime soon-there will always be a need for simultaneous operation of two radios for a handoff to take place from one to the other or for a data transfer over Wi-Fi while talking on the GSM connection. This prohibits the use of a single radio to do both.

As a result of this situation, when SDR front ends do emerge, we may have multiples of them-one each for the cellular, connectivity and even positioning (GPS and Galileo) connections.

While there is little to indicate that the RF front-end technology will advance sufficiently near-term to get rid of the multiple RF chains and conversion blocks forking out from the soon-to-be converged basebands, some promising technologies and techniques are being exercised to lessen their impact on size, power and cost.

In any case, the move to submicron CMOS has opened the door to fast, low-power signal processing that can be used to compensate for the limitations of CMOS when it comes to such parameters as noise, sensitivity and power output. It also has been a key enabler for the development of power-and space-conserving direct-conversion radios that now omit bulky SAW filters and a plethora of other passive devices.

High-speed, low-power CMOS is also being applied extensively in the analog portion of the radio. This has developed into a mass migration toward digitally assisted analog front-end blocks, where extensive digital calibration is applied to account for process and environmental variations and thereby improve the accuracy of such critical blocks as phase-locked loops and synthesizers.

Most interesting, however, is the ability of CMOS assistance to improve the performance of pipelined and bandpass delta-sigma A/Ds. This is an area of study that may enable cost-effective direct RF sampling.

While optimizing individual blocks in terms of power consumption and space is one route to mitigating the impact of multiple RF chains, two other paths exist to achieve that optimization. The first is the reuse of power-hungry portions of the chain such as the local oscillator. Intel has already achieved ranges of 1.8 to 5.8 GHz in CMOS.

The other route is to replace the front-end preselect filters with RF microelectromechanical systems. The MEMS use the electrically controlled movement of a cantilevered arm to modify the values of capacitive and inductive filter components . They can also be used to change the matching for antennas. The devices can change the characteristics of a filter within a millisecond and have the additional advantage of being almost a perfect wire, so there are no losses associated with them.

Reliability remains a question, as do bulk and cost. Nonetheless, many researchers are confident that MEMS are one of the keys to unlocking agile RF front ends.

See related chart While current interest is in software-defined radios that have a clear technology timeline, a 'cognitive' radio is still in its seminal stage. Such a radio would rely upon GPS and software to make it smarter and to endow it with location-aware capabilities that would predict the user's needs and wants.