The effort could smooth the path for business networks to evolve gently from today's Gigabit Ethernet without having to retrofit offices with new kinds of copper or optical-fiber cable. But the participants — SolarFlare (Irvine, Calif.), Plato Labs (Campbell, Calif.), Cicada Semiconductor Corp. (Austin, Texas), Accelerant Networks Inc. (Beaverton, Ore.), Mysticom Inc. (Mountain View, Calif.) and startup Telicos — represent widely diverging ideas on cracking what some developers call the holy grail of wired networking.

"I expect there will be a vigorous debate on the issues" in Hawaii, said Vivek Telang, a senior engineer at Cicada. "The last few weeks have been quite a scramble to get people to agree just on the principles for how to address the problem and evaluate solutions."

The IEEE's "call for interest" on 10-G Base T had been planned for next year, but sufficient interest emerged this fall to slate the November meeting. If all goes well, an official study group will meet in Vancouver in January to lay the groundwork for drafting a formal specification. It could take two to three years to reach final ratification.

Telang said the SolarFlare and Cicada approaches rely heavily on digital signal processing and leverage multiple-input, multiple-output [MIMO] techniques that Lucent Technologies is using in antenna arrays for 3G CDMA basestations. By contrast, he said, Mysticom and Accelerant take a more hybrid approach comprising DSP and analog techniques, with Accelerant tapping its serializer/deserializer technology. And Plato Labs relies more heavily on analog design.

The group is expected to debate whether Category 5 wiring is up to the task of carrying 10 Gbits/s, but that's just one aspect of the complex design effort. "This is not a single 'Aha!' problem. Five or six major innovations need to be done well and in concert," said George Zimmerman, founder and chief executive officer of SolarFlare.

The use of MIMO implies new ways to handle near, far and alien crosstalk as well as equalization. Designers will also need to craft great echo cancellation devices, multilevel pulse-amplitude modulation (PAM) capability and highly linear, high-bandwidth analog circuitry. "A fair amount of the signal processing is done in the analog domain," said Zimmerman.

He said the SolarFlare approach requires an effectively 800-Msample/s A/D converter.

Plato Labs is using analog signal processing and DSP to implement all the necessary equalizations. Parts such as Maxim's MAX3802, a quad 3.2-Gbit/s adaptive cable equalizer for copper, show the way forward, said Plato Labs founder Joseph Babanezhad.

As for the MIMO techniques, Lucent has shown how it can actually use crosstalk in a wireless setting to improve channel capabilities. For 10-G Base T, MIMO involves treating the four pairs of Category 5 unshielded twisted-pair wiring as one channel with four inputs and outputs.

"This buys you a substantial savings in computational requirements," said Zimmerman.

That's important, because "if we address this problem in a conventional manner it's a 12-teraoperation/s problem," probably requiring a 20 x 20-mm die, he said. "That's not economically feasible. But we came up with an unconventional, proprietary architecture and algorithms that reduce this to a 1.5 tera-ops problem, so now it's in the ballpark." He said today's quad and octal 1,000 Base T parts need silicon performance in the same range.

In SolarFlare's approach, no major rework is needed on line codes or modulation schemes beyond some extensions to the current PAM spec, said Zimmerman. But the startup's most controversial concept is extending 100-MHz Category 5 wiring to Nyquist signaling rates of 400 MHz — well beyond the range of even higher-grade, Cat. 6 cable.

This may become a source of discord in the standards work. That's because all the proposals submitted to the 10-G Base T call for interest were PAM-based, but many opted for different numbers of coding levels.

"We are going to present proposals that talk about using 400 MHz of bandwidth on Category 5 wire with relatively moderate launch power," Zimmerman said of his planned tutorial at the meeting on the Hawaiian island of Kauai. "We won't go into detailed disclosure about our intellectual property for reducing the complexity of such a design, but we will talk about functions these techniques perform."

ISDN, digital subscriber line, Ethernet in the first mile and even 10 Base T have used information on characteristics of Category 5 beyond its 100-MHz rating, he noted. "People don't assume everything stops at 100 MHz; they build models that are continuous in frequency.

"For instance, near- and far-end crosstalk gets worse going up in frequency, but it still behaves by roughly the same physics, and we can model that," Zimmerman said. "We have taken substantial measurements of cabling and configurations from a wide variety of manufacturers to validate the physics."

Nevertheless, Zimmerman said he expects "a lot of resistance" from cable makers that want to sell more expensive copper grades and from chip makers that say the SolarFlare proposal simply can't be implemented.

According to Cicada's Telang, SolarFlare will recommend a Nyquist frequency of 416 MHz, but the actual signaling frequency will be at 833 MHz using 3 bits per symbol to achieve 2.5 Gbits/s over each of four copper pairs, adding up to 10 Gbits/s.

"I think we will need to ask the cable industry to recharacterize Cat. 5 for 400 MHz. Maybe they could relabel and sell it as Cat. 5ex or something," Telang said. But he said he doubts cable makers would be willing, because the costs of the testing could not be recouped by the small margin users would pay for the more thoroughly tested cable.

"If we change the media," Zimmerman warned, "it substantially lessens the business proposition." The SolarFlare founder said his company's technology could be used to craft practical parts that would comfortably fall between the costs of today's $10 to $30 1,000 Base T physical-layer devices and nearly $1,000 10-Gbit optical transceivers. Power dissipation should not be a problem, since 1,000 Base T PHYs started at 7 watts but have fallen to less than 1 W as process technology has improved. Zimmerman also expressed confidence his techniques would lend themselves to designing 1- and 10-Gbit parts similar to today's 10/100/1,000-Mbit devices.