Suppliers weigh software-defined radio solutions

Suppliers weigh software-defined radio solutions
MANHASSET, N.Y. — The road to one of the communications industry's most elusive goals, the software-defined radio, will probably not be a straight path. As the advent of second- and third-generation cellular networks turns up the heat in the quest for this crucial piece of technology for wireless handsets and basestations, chip makers are juggling a multitude of possibilities. Indeed, some say a mix-and-match approach seems more likely than a single, clear-cut solution.

If the answer is far from a slam-dunk, the need is not. "The [3G] deployment is going a lot faster than people expected," said Randall Fahey, vice president of marketing at startup Morphics Technology Inc. (Campbell, Calif.). "We're seeing the new networks taking off. It's going to happen next year."

Service providers don't want to install new basestations for each new air interface or wireless data scheme coming down the pike. And consumers won't buy multiple phones and PDAs to cover each wireless service they use, and each city or country to which they travel.

Software-defined radios let service providers reprogram basestations to reassign channels as standards change and the mix of analog vs. digital users shifts. And engineers envision handsets that someday will download from any network whatever code is needed to reprogram themselves to access a wireless service or run a mobile application.

Seven solutions

There's no shortage of silicon approaches to the problem. Mark Cummings, chief executive officer of enVia Inc. (San Jose, Calif.), has whittled them down to seven. From most to least conventional they include high-speed DSPs, multiple ASICs, parameterized hardware, switchable microcode, multiprocessor arrays (such as very long instruction word architectures), reconfigurable logic and a combination of any of the above, said Cummings, who was instrumental in the formation of the Software Defined Radio Forum.

"I think it's a little misguided to think about it as an ASIC vs. FPGA vs. DSP," said Jeff Bier, general manager of Berkeley Design Technology Inc., a Berkeley, Calif.-based DSP consulting and software development firm in . "I would rather think the solutions that will be successful will combine elements of some of those."

Bier said he expects to see solutions that might incorporate a programmable processor alongside hardwired circuits and reconfigurable logic of some sort. "The most successful designs will be pulling together pieces," he said.

To date, the development of software-defined radio has been confined to military applications, such as snooping or jamming communications, where cost and device size are not significant constraints.

But semiconductor advances combined with the imminent arrival of 3G wireless systems are driving software-defined radio toward commercial markets.

"I think the fundamental thing that's going to make software-defined radio happen is the current move to 0.13-micron technology," said Jim Gunn, an analyst with Forward Concepts (Tempe, Texas).

Processing punch

The key hurdle is providing the processing punch to convert analog RF functionality into the digital — and eventually the software — domain. The earlier in the chain the data is converted, the more processing power is required. Engineers see two approaches to that problem.

"The first way is to have a completely flexible RF front end that can handle any standard and give you the baseband no matter what format," said Srikathyayani Srikanteswara, a senior graduate research assistant at the Mobile and Portable Radio Research Group at Virginia Tech (Blacksburg, Va.). "The other is to sample the incoming waveform right at the antenna. Right now we can't do either.

"Theoretically, you could come up with an analog-to-digital converter that could sample up there, but it would be extremely expensive and bulky and unusable," she added. "The other problem is that even at baseband, the wide variety of standards, with different baseband processing, makes it a signal-processing nightmare in terms of flexibility, speed, power, cost, footprint and available processing horsepower."

Until recently, the choice of approach was dictated by the type of system. Basestations, generally more tolerant of power and heat, might tap banks of DSPs, while handsets would employ ASICs for gains in speed and size, with lower power. But now the lines are blurring.

"Thanks to microcells and picocells, at least a portion of the basestation market is moving toward what are known as 'shoebox' designs, meaning that now basestations are facing many of the same issues as handsets in terms of form factor, power consumption and cost," said Cummings of enVia.

Not surprisingly, chip makers tend to advocate approaches most in line with their product portfolios. Brad Taylor, director of applications at startup Chameleon Systems Inc. (San Jose), said reconfigurable CPUs like his company's CS2112, announced earlier this year, will outperform DSPs and FPGAs in soft radios.

Taylor admitted that the CS2112 may run a little more slowly than a DSP, a function of the fact that reconfigurable hardware techniques often entail more interconnect delay. "We tend to see a significant performance advantage, though, since we have more multipliers and we can keep them all busy," he said.

For a good fit with the classic software-defined radio application, the processing fabric must be able to switch functionality in real-time. "This is a shortcoming of FPGAs," said Taylor. "You can't change them in real-time, as it takes a half second to change the functionality, and the whole system goes down."

With the reconfigurable Chameleon chip, "we can load the new configuration while running the old one," he said. "We swap it over, and we jump right into the new protocol."

Cost goals

But cost is still an issue. "Our goal," Taylor said, "is to get flexibility and performance at least in the range of ASIC costs." Pricing isn't available yet on the 0.25-micron device.

Another startup eying software-defined radio is Morphics. Unlike Chameleon, which targets basestations only, Morphics is targeting both basestations and handsets — the former with chip sets, the latter with licensable cores. Founded in 1998 and still in stealth mode, the company plans to announce product in the first quarter of 2001. Morphics will be competing in the handset market against companies like Quicksilver Technologies and Sirius, a Belgian firm.

The complexity of wideband-CDMA networks will drive the need for software-defined radio, said Morphics' Fahey.

"With GSM, gate counts jumped to 30,000 to 50,000 gates for the convolutional coding and Viterbi coding," Fahey said. "And with IS95 CDMA, that ASIC is now 100,000 gates. The problem is when you go to wideband CDMA, now that ASIC functionality will be the bulk of the signal processing — in the order of 700,000 to a million gates."

DSP vendors are tackling the performance problem "with different approaches to the cores, VLIW, superscalar, multiprocessing — essentially homogenous multiprocessing," he said. "[They are] throwing more MACs [multiply-accumulate operations] at the problem, but it's tough to keep the MACs fed and you need to add huge caches to do that. So you start to run out of efficiency anyway."

Fahey said Chameleon's approach has efficiency advantages over regular FPGAs, but will not make inroads into handsets. And that is where Morphics sees its advantage.

The company takes a heterogeneous-multiprocessing approach. "Instead of having a bunch of MACs all running in parallel, we have some very specialized processors that have been optimized with advanced knowledge of the algorithms we're running on them," Fahey said. "We're not mapping [the functionality] onto a MAC but onto specialized processors, or specialized processing kernels."

Handset or baseband?

Morphics' approach raises the question of just where the soft-radio work should be handled — handset or basestation?

"Having this in the basestation is a very inefficient use of spectrum," said enVia's Cummings. "You have to allocate blocks of channels to a new user entering the cell site — you can't allocate just a single voice channel. It may also need a guardband, depending on how sensitive the channel is to what's happening in adjacent spaces. What happens is spectrum fragmentation."

Pointing to the scarcity of spectrum, Cummings said this is not the way to go. "We'd rather have the basestation be a particular air interface and have the handsets conform to that air interface. At the same time, we'd like to be able to change the air interface in the basestations" — hence the soft-radio requirement — "but at a much slower pace, in a more managed fashion as technology and the business evolve," he said.

Such an approach favors products such as Xilinx Inc.'s Virtex-II FPGA, which will incorporate a 300-MHz, 32-bit processor capable of 350 Dhrystone Mips. The device, which is set to go into production in 2001, tackles many problems, such as configuration (software-defined radio download) and network issues, which are best suited to a RISC processor.

"With the inclusion of a RISC tightly coupled to a high-performance FPGA fabric, software-defined radio designers will be able to realize all aspects of an SDR system — from demanding DSP functions to more control- and decision-intensive operations that are best handled by a processor," said Chris Dick, DSP group manager at Xilinx (San Jose). Xilinx's goal is to let OEMs design systems with the flexibility of an all-software solution and the performance of an ASIC approach, he said.

Whatever approach OEMs adopt, the move to soft radio could come soon, given that 3G cellular nets are almost ready to roll. Japan is on the cusp of activating its wideband-CDMA network, and Europe has seen a noticeable uptick in 3G activity. "Companies paid a high premium for spectrum licenses during the last round of auctions," said a spokesman for Ericsson Microelectronics (Kista, Sweden), "so they're now in a real push to start getting a return on these investments."