Metro challenges
The case for EDC is even more compelling in metro networks, where chromatic dispersion is becoming a problem over longer spans and at higher transmission speeds. "Early metro networks were deployed with spans of 50-60 km, but that is now creeping up to 100-200 km," comments Castagnozzi. "Dispersion compensation is needed to support longer transmission distances and higher data rates, but carriers don't want to pay for extra optical components."
In contrast, incorporating EDC into linecards and modules minimizes both cost and complexity. "EDC is the only option for meeting price and performance requirements," claims Keating. "Our EDC designs can correct for typical amounts of chromatic dispersion at transmission distances of up to 150 km." And combining EDC with new modulation techniques could extend system spans to 200 km and beyond.
While EDC does not quite live up to the standards set by optical methods, Keating maintains that it offers good-enough performance affordably. "From the customer's perspective the issue is getting enough compensation for the system to operate at a price that makes it economically viable," says Keating. "What's necessary is to cut signal loss due to dispersion to acceptable levels."
Keating says Santel's device can recover about half of the signal lost through chromatic dispersion. That "gain" is enough to double the tolerance of a metro transponder to chromatic dispersion and so double the span of the system. Keating also claims that prototype EDC devices have been shown to offer similar PMD compensation performance to some optical-layer solutions.
The circuitry needed for EDC is typically integrated into a receive-end device that also performs clock recovery and data demultiplexing. Keating explains that EDC essentially creates a filter in the electrical domain that reverses the signal distortion created by the transmission process. In this way the EDC circuit "gives back" part of the signal lost due to optical distortion.
That's the theory, but some complex electronic engineering is needed to create a filter structure that can handle 10 Gbit/s data streams, and to build a clock-recovery circuit that can maintain a lock on severely distorted signals. "EDC is essentially an analogue function, but it must be adjusted adaptively - in a similar way to tuning in the radio," explains Armstrong. "The design of the analogue front-end and digital control is pretty tricky. Most chips exploit dynamic feedback equalization, which is difficult to implement in an analogue sense."
Unsurprisingly, chip vendors are reluctant to reveal their design secrets. "The algorithms are kept under wraps, but they dictate so many elements of the design, including the complexity, power dissipation and how effective it is for the application," says Armstrong. "The design must deliver adequate compensation for dispersion, but still be reasonable to implement in an analogue chip."
AMCC's solution is to exploit a dual-chipset design that combines EDC with forward-error correction (FEC). FEC is already widely used in optical networks to reduce bit errors in the data stream, and is typically achieved in a separate digital chip inserted after the demultiplexing function.
Castagnozzi says that integrating FEC with the compensation function addresses the problem of digital control. "We use the FEC to help with the adaptation of the EDC circuit," he explains. "Normally you don't know whether the decision circuit has made an error, but the FEC can detect these bit errors. Using this information in the EDC makes it possible to optimize the system."
Combining FEC with EDC delivers performance that could not be achieved with each technique on its own. "In some applications we can get 4-6 dB of extra gain from the EDC, which adds to more than 8 dB from our Niagara FEC device," says Castagnozzi. The Niagara device can be configured to offer standard FEC, as defined in G.709 by the International Telecommunication Union, or enhanced FEC, a proprietary method that Castagnozzi claims corrects for 100 times more errors than the standard approach.
Santel and AMCC have both implemented their EDC designs in silicon:germanium (SiGe), which delivers the bandwidth needed for high-speed operation at a reasonable level of power consumption. But SiGe devices still need more power than existing physical-layer chips based on complementary metal-oxide silicon (CMOS), which Armstrong believes will remain a barrier to widespread use of the technology.
"A typical mux/demux chip today consumes about 1 W of power, while EDC chips based on SiGe require about 5 W," he says. "It's a lot easier to do the analogue signal processing in SiGe, but the power consumption is not reasonable."
Castagnozzi and Keating believe that customers will accept higher powers for better performance in long-haul systems, but are taking steps to reduce power levels for metro devices. "We have a roadmap to optimize the design for metro and long-haul solutions," says Castagnozzi. "For the long-haul we will probably stick with SiGe, but we will shift towards CMOS for metro applications."
Castagnozzi says that AMCC has already halved the power consumption of its EYEMAX technology. "In the first iteration we wanted to demonstrate the capabilities of EDC, but we have been modifying the supply voltages to squeeze down the power requirements by a factor of two. The next product release will be a power-reduced version of our SiGe chipset, and then we will introduce a CMOS solution." Santel also plans to introduce a single-chip CMOS device for metro transponders in 2003.
In the meantime, AMCC is sampling its SiGe devices with customers, with general availability expected for the first quarter of 2003. Santel aims to release sample devices before the end of the year.
According to Castagnozzi, customer trials are going well, but the disruptive nature of the technology needs more customer testing to gain acceptance. "Customers were initially sceptical when we showed them simulated results, but tests of EDC in real optical systems have come close to those simulations," he says. "Both our EDC and enhanced FEC devices are gaining market acceptance, but design-in usually occurs earlier for Niagara because it is an improvement on existing techniques. But it is the EDC that is really attracting attention."
• This article originally appeared in FibreSystems Europe December 2002 p15