Winner: Radio Revolutionaries

Software-defined radio is one of those promising but elusive ideas that’s been anticipated for quite some time. The concept is elegantly simple: get rid of the specialized electronics used to process radio signals and instead do everything with software. The result, we’ve been promised, will be a universal wireless device that can seamlessly handle a range of frequencies, modulation techniques, and encoding schemes. Just as the personal computer replaced the typewriter, the adding machine, and even the telephone, so too will a software radio one day replace your cellphone, Blackberry, and any other wireless device you may happen to use.

A cellphone based on software-defined radio would be lighter, smaller, cheaper, and more power efficient. What’s more, it would be better at making calls: instead of being stuck with one frequency or even one cellular carrier, it would automatically search out the best and least expensive way of connecting. And equipment ­makers wouldn’t need to overhaul their products to fit every new telecommunications standard. Wireless providers would be able to roll out new services easily and troubleshoot technical glitches with a simple download.

We aren’t there yet, but software-defined radio is definitely coming. Don’t expect an overnight transformation, though. After all, it took years for the PC to sweep aside the IBM Selectric typewriter. This revolution, too, is bound to happen in a series of incremental but significant steps.

Steps like this: Vanu, a small Cambridge, Mass., company, says that this year it will begin selling the first cellular base station that can simultaneously process two waveforms—CDMA (short for code division multiple access) and GSM (global system for mobile communications)—all in software running on off-the-shelf computer servers.

For cellular carriers, the advantage of such technology is clear. Base stations are an integral but largely invisible part of their infrastructure, routing voice and data traffic between cellphone users and the network of landlines. The market for base stations is huge, totaling some US $45 billion worldwide in 2004, according to ABI Research, in Oyster Bay, N.Y. It’s also hugely competitive, as carriers look for any edge that will allow them to operate more efficiently and at lower cost. Being able to add new services and adapt to new standards by merely tweaking software, instead of replacing or adding hardware, is just such an edge.

“In the software radio base-station race, the clear winner is Vanu,” says David Murotake, a telecom engineer whose company, SCA Technica, of Nashua, N.H., focuses on security issues in software-defined radio. Although telecom equipment makers have been incorporating elements of software-defined radio into their base stations for more than a decade, none has taken as radical an approach as Vanu. “They’ve already introduced award-winning technology, and in the process they’ve outdistanced the usual suspects, like Ericsson and Nokia”—two of the biggest ­manufacturers of base stations.

Or, as a top executive at one of the major cellphone carriers remarked after seeing the new Vanu base station in action, ”This could be game changing.”

To appreciate the significance of Vanu’s latest product, it helps to understand a few things about radio. Nearly every radio, be it a walkie-talkie or a pager or a portable AM/FM unit you take to the beach, works pretty much the same way. In a traditional cellphone, for example, there’s the radio front end, consisting of an antenna and a radio-frequency transceiver that picks up the analog radio waves, filters out the unwanted portions of the spectrum, and converts the remainder into a lower-­frequency signal, which is fed into an ­analog-to-digital converter. The resulting baseband signal is then processed—that is, it’s demodulated, decompressed, or otherwise decoded—by a special-purpose integrated circuit, a digital signal processor (DSP), a field-programmable gate array (FPGA), or some combination of the three. The resulting streams of bits—voice or data—could be the latest stock quotes, a “happy birthday” call from your mom, or a snapshot of your co-worker snoozing at his desk.

Software-defined radio aims to get rid of most of that hardware. A number of companies, including AsicAhead, BitWave Semiconductor, and TechnoConcepts, are working on reconfigurable RF chips that can directly convert any analog radio signal into a digital one, across frequencies from several hundred megahertz to several gigahertz.

But refashioning the radio front end in software is exceedingly difficult. For one thing, it’s hard to design a single antenna with good gain across a wide range of frequencies. “It’s just the laws of physics,” says John Chapin, Vanu’s chief technology officer. “An antenna tends to have a fairly narrow range at which it effectively resonates.” The U.S. military’s Joint Tactical Radio System, for example, has several antennas so that it can operate from 2 MHz to 2 GHz.

If you’re trying to cover all the commercially interesting wave bands, it’s also hard to filter the received signals, efficiently amplify the power of transmitted signals, and do the analog-to-digital conversion.

For those and other reasons, most efforts, including Vanu’s, focus on baseband signal processing. The typical approach is to build in reconfigurable components such as DSPs or FPGAs and then write code for whatever waveform you’re dealing with. Processing a different waveform just means writing new software. Once the hardware becomes obsolete, though, much of the software typically can’t be reused, because it was written for those specialized components. So you have to develop code from scratch.

Vanu takes a very different tack. Its engineers start with existing general-­purpose processors, such as the Intel Pentium or Xeon, and then design software to do all the intense computation required to demodulate and decode the waveform. Chapin likes to say that the company has been riding Moore’s Law, taking advantage of the billions of dollars that chip makers spend to make computers run faster. In November, Intel announced the first four-core processor, and by 2010 the company projects that it will be making 80-core versions. [For more on programming multicore processors, see "Cure for the Multicore Blues," in this issue.]

Speed isn’t the only advantage the new chips offer, Chapin says. The ­signal-­processing capabilities of general-purpose processors continue to improve with each generation, in large part because of all the video and multi­media applications that ordinary computers are expected to handle these days. Good thing, too, because wireless waveforms are only getting more complex. Handling signals from one of the old analog cellphones takes just 20 million instructions per ­second—a trifling amount for a modern Pentium. By comparison, the CDMA standard requires about a gigaflop, or a billion floating-point instructions per second. And the third-generation phones coming out now demand at least another factor of 5 beyond that, Chapin says.

Vanu’s new Anywave multistandard base station is a software update to a GSM-only version that the company introduced in 2004. While most traditional base-­station equipment is bulky, the Anywave is tiny. It consists of standard off-the-shelf servers (the number of servers depends on how many calls the carrier expects to handle), plus the radio-front-end equipment, which Vanu purchases from any of several vendors.

Being able to reuse its software is a top priority at Vanu. Whereas other software-defined radio systems require a lot of reworking whenever the hardware changes, any code that Vanu engineers write can easily migrate from one processor to another. That’s a lesson learned from one of the earliest forays into software-defined radio, a U.S. Army project launched in 1992 called SpeakEasy, which aimed to build something that could handle 10 different waveforms generated by more than a half-dozen military radios.

SpeakEasy’s developers designed the system around what at the time was the most advanced commercial ­signal-processing chip available, the Texas Instruments TMS320-C40. It took them three years, and the finished product filled the entire back end of a transport vehicle, but much to everyone’s amazement, the radio did everything it was supposed to do. The only problem was that SpeakEasy could not be upgraded—as ingenious as the software was, it was designed to run on the C40 and no other processor.

Vanu’s code, by contrast, is all written in a high-level language, C++, running under the Linux operating system. ”When we change platforms, we maybe rewrite 5 ­percent of the code, versus 50 or 60 percent for our competitors,” Chapin says. In real terms, using general-purpose processors can mean sacrificing some performance for portability. But the difference may not matter. Chapin mentions Clayton Christensen’s classic example of how the hydraulic shovel eclipsed the steam shovel back in the 1950s. “The steam shovel was clearly more capable, but the hydraulic shovel was cheaper, and it eventually became good enough,” Chapin says. “That’s what we’re seeing here.”

Nobody else has ever built a commercial device that uses a general-purpose processor to handle two waveforms simultaneously. But Vanu engineers have now done it twice. Five years ago, they developed a dual-standard prototype radio for the National Institute of Justice, the research arm of the U.S. Department of Justice. That device combined the functions of two radios commonly used by law enforcement: a Motorola two-way FM radio and a Datron unit that used a newer digital standard known as Project 25.

“When you run samples through the same pipe at two different sampling rates, managing the data flow so that you don’t drop data or get stuck turns out to be really, really challenging,” Chapin says. “It’s like listening to two tones that aren’t in harmony—the tones beat against each other and make a dissonant sound. In the same way, the sample streams move in and out of sync.”

After trying various approaches that all proved too labor-intensive, the engineers eventually hit on the idea of using Internet Protocol and sending the packets of data over gigabit Ethernet, which is designed to handle different rates of data, such as e-mail and streaming video. It’s also built into just about every server and computer these days, so there’s no need to construct a dedicated hardware link between the radio front end and the baseband processor.

Gigabit Ethernet has the advantage of being not just reliable and speedy but also a simple and well-understood networking standard, Chapin says, adding, “Wherever possible, our approach is: let’s use the commodity thing.” When 10-gigabit Ethernet rolls out later this year, the sampling capacity will jump by an order of magnitude.

Tackling CDMA was more complicated than the early GSM effort, because the waveform is more sophisticated. “We started by tearing apart the [CDMA] spec,” says Jeff Steinheider, who’s in charge of base-station engineering at Vanu. “Then we spent weeks or maybe months studying the document and identifying areas we knew how to do and figuring out where we needed new algorithms.”

Rather than going off and independently developing their chunks of code, the engineers repeatedly integrated the full system and ran it through an automated testing system. It’s an approach they cribbed from the creators of the Mozilla browser: because those developers weren’t always all in one place, they would periodically check their programming into the test system to make sure everything worked properly together.

“We added radio-specific modules, like one that pretends to be the radio [front end] and speaks gigabit Ethernet,” Steinheider explains. The company also uses its own base stations to route calls around its headquarters, which allows engineers to vet the product under real-world conditions. “That’s the ‘eat your own dog food’ approach,” Steinheider says.

Critics of Vanu’s strategy say that the general-purpose processors it favors won’t be able to cope with more complex waveforms, such as those that fourth-­generation cellphone standards undoubtedly will require. Chapin isn’t worried. “I’m prepared for a future in which we can’t just use general-purpose processors,” he says. At that point, he adds, it may make sense to supplement the basic software design by adding a digital signal processor, to off-load the overtaxed Pentiums or Xeons. “So far, though, we haven’t had to do that.”

Vanu the company grew out of work that its founder, Vanu Bose, did as a graduate student at MIT, just down the street from the company’s current headquarters. Bose formed the company in 1998 to commercialize his ideas, but without a clear idea of what type of product or which market to go after. He looked at public safety and automotive applications and landed some military contracts but found no takers in the commercial sector.

During the tech meltdown in 2001, Vanu avoided the fate of many a start-up because it hadn’t accepted any money from venture capitalists. (The company has gotten some angel funding over the years, totaling about $5 million.) Instead, Vanu grew slowly but steadily throughout the downturn and into the recovery, to its current staff of 55, most of them engineers.

In at least one way, the company even bene­fited from the collapse. Located in the heart of Cambridge’s tech zone, Vanu’s airy glass-and-blond-wood offices previously housed a dot-com that failed. It’s a big step up from Vanu’s previous digs, a former weight-loss center in a suburban strip mall.

To drum up work and show off their approach, Vanu engineers put together a number of clever demos, including a Compaq iPAQ PDA that they programmed to receive and play FM radio signals. ”Everyone told us to stay away from commercial [telecommunications] infra­structure,” Bose says. Plenty of start-ups had foundered trying to roll out new products for the telecom sector. But Bose ultimately decided to give it a try. ”We said it would take us five years to develop our first infrastructure product,” he says. ”I think that was accurate to within a few months.”

At this point the company breaks even, Bose says, with much of the revenue plowed back into R&D. That’s an important lesson he learned from his father, Amar G. Bose, founder of the audio-equipment maker Bose Corp., in Framingham, Mass. In 1981, that company suffered its first loss ever, and yet Amar Bose decided to invest $10 million in developing audio systems for cars, an area in which his firm had no experience. ”Everybody was freaking out and telling him he was crazy,” Vanu Bose says. ”Five years later, auto­mobiles were $100 million of his business.” He grins broadly. It’s clearly a story he never minds repeating.

The epicenter of the next phase in the ­software-defined radio revolution is likely to be De Leon, Texas, a town of about 2400 people that is roughly 220 ­kilometers southwest of Dallas. De Leon’s main attraction for the last 92 years has been its annual Peach and Melon Festival. The town also is the home of a rural mobile-phone provider called Midâ''Tex Cellular, which three years ago became Vanu’s first commercial customer.

“We were looking for new ­technology, something that was more flexible and maybe would allow us to avoid constant hardware changes,” says Toney Prather, Mid-Tex’s CEO. The company’s existing network was based on older, TDMA (time-division multiple access) technology, and to move to a newer standard, such as GSM or CDMA, the provider faced hefty upgrade costs. Standard base stations run about $50 000 apiece—which can add up fast. AT&T Wireless reportedly spent $10 billion to give its TDMA network a GSM makeover.

In early 2003, Prather agreed to let Vanu supply Mid-Tex with GSM software-defined base stations on a trial basis. At the time, Vanu didn’t actually have a working GSM base station, but its engineers got busy, and within seven months they had software that was good enough for a field trial [see “Mobile Phone System Passes Texas Test,” IEEE Spectrum, February 2004].

Since then, Mid-Tex has deployed 29 Vanu base stations, which all run on Hewlett-Packard ProLiant servers and handle about 6 million minutes of calls each month [see diagram, "Inside the Anywave”]. The carrier has since shifted most of its 11 000 customers from TDMA to GSM. In addition, more than 20 percent of its revenues come from the major ­carriers, through roaming agreements that give their users cell coverage when they travel through Mid-Tex’s 21 000-square-­kilometer territory.

In all, Prather figures he has spent about $2 million on his Anywave network. That’s about what he would’ve paid for traditional GSM equipment, but now he won’t need to buy an entirely new network when he adds CDMA. Nor will he need to lease additional T1 lines to get the signals from the cell site to the switch—which can run several hundred dollars a month per line. That’s because the base station architecture is IP-based, so the same T1 can handle any type of call from any carrier. A single T1 could even handle multiple sites, Prather says.

What’s more, the company can now perform diagnostics remotely over the Web, cutting maintenance costs. Eventually, the new network may even save Prather’s business. Rural cellphone providers always worry that one of the bigger carriers will extend its network or “overbuild” their territory. “You’re battling the giants,” Prather says. “But if we can provide a network efficiently and cheaply, then they won’t want to overbuild it.”

At press time, Mid-Tex was field-­testing the new dual GSM-CDMA software at two of its sites, and Prather was eager for Vanu to give him the go-ahead to upload the CDMA code throughout his network. “I hope it’s any day,” he told Spectrum. “We’re ready now.”

Anywave deployments so far have been limited to rural areas such as De Leon and certain military installations. But Bose expects to land some much bigger fish within the next year or two. Vanu executives have recently been getting a warm reception when they’ve demonstrated the new multistandard technology for some of the major carriers.

Another promising area is in smaller base stations, sometimes called “femtocells,” which are designed to offer wireless coverage within an office building or household. The worldwide market for femtocell products is expected to reach nearly 19 million units by 2011, according to ABI Research. “Our technology is a great fit for that,” Bose says.

But wait, there’s more: Bose says his ­company’s technology could just spark a restructuring of the  wireless-­infrastructure in­dustry. At present, it’s still mostly vertically integrated, with companies like Ericsson, Huawei, and Nokia producing nearly all the components, from the base stations to the antennas. Rather than having a single company do everything, Bose suggests, the ”horizontalization” of the industry would result in more agile companies—­including Vanu—doing whatever they’re best at.

It sounds like a big leap, but that’s precisely what happened in the computer industry 20 years ago, when Digital Equipment Corp., IBM, and other giants gave way to upstarts like Apple, Intel, and Microsoft.