FEC coding for dispersion compensation in transponders
Current MANs typically comprise OC-12 and OC-48 SONET/SDH rings. As the cost of OC-192 line cards drops below 2.5 times the cost of OC-48, more OC-192 rings will be deployed to relieve the traffic bottlenecks between long-haul (LH) and access networks. The transition from OC-48 to OC-192 will not be painless, however, since there are many problems that need to be solved.
One key issue is the poor quality of the existing fiber plant, in light of the reach requirements of metro transport equipment. Unlike LH networks, metro networks typically do not use optical amplification, so the distance at which 3R regeneration is required becomes critical. At the same time, the expansion of metropolitan areas is extending the desired reach of metro equipment to 120 km.
For poor-quality fiber deployed before 1995, 120 km is a long distance even at OC-48 rates. Poor fiber quality can create severe light dispersion problems resulting from polarization-mode dispersion (PMD) and chromatic dispersion (CD). At OC-192 rates, these problems significantly worsen due to the smaller signal period. Some form of dispersion compensation is therefore mandatory.
Dispersion compensation is typically accomplished via optical means. These techniques typically require devices that are bulky and expensive—and the additional cost is one of the factors preventing widespread adoption of OC-192 in the metro. However, a frequently overlooked solution to metro fiber dispersion problems is forward error correction (FEC). While FEC coding does not address the dispersion problems directly, it does help mitigate them significantly. Low-cost and very-low-power CMOS chips are now becoming available that can be easily deployed in add/drop multiplexers (ADMs), multiservice provisioning platforms (MSPPs), and metro DWDM equipment.
Singlemode fiber, which serves as the backbone of the long-distance optical communications network, was first deployed around 1984. Initially used with 1310-nm optics, these networks moved to 1550 nm to take advantage of the lower fiber attenuation in that band (0.24 dB/km at 1550 nm compared to 0.4 dB/km at 1310 nm). To combat CD effects, a dispersion-shifted fiber (DSF) was later developed with carefully tailored dispersion characteristics that reduced dispersion to practically zero at about 1550 nm. This fiber was deployed in the networks of NTT and MCI/WorldCom, for example.
Unfortunately, the problem with DSF is that some amount of CD turns out to be beneficial for DWDM transmission, particularly at 10 Gbits/sec. Without CD, some nonlinear fiber effects, like four-wave mixing, significantly degrade fiber transmission performance. As a result, a new type of fiber was developed—nonzero dispersion-shifted fiber (NZDSF), which has a small CD profile at 1550 nm—and deployed over the last few years.
Consequently, today's worldwide fiber installations consist of a mixture of fiber types (standard singlemode, DSF, and NZDSF fibers), making optical interworking without optical-electrical-optical (OEO) conversion extremely difficult. With current capital-expenditure constraints, it is very unlikely that these older generations of fiber will be replaced anytime soon, so any networking solution must operate well in this diverse environment.
For most fibers, PMD characteristics are becoming important at high transmission rates, especially in cases where CD has already been addressed. Most of the fiber deployed before 1995 has poor PMD characteristics. This vintage fiber currently prevents the deployment of 40-Gbit/sec networks. In some cases, carriers are even unable to introduce 10-Gbit/sec links. With no new fiber currently being deployed, these poor-quality fiber links must be rejuvenated to solve metro bottleneck problems.
While dispersion effects can be addressed optically, equalization, FEC, and OEO regeneration also are options. While OEO regeneration is the most obvious and foolproof method, it is also the most expensive. Therefore, we'll focus on comparing the effectiveness of the first three options.
Optical compensation. Optical dispersion compensation is the most straightforward approach, since it corrects the problem at the source. Total CD is addressed by using fiber with dispersion compensation characteristics that utilize the properties of material dispersion and waveguide dispersion to cancel out each other.
Material dispersion arises from the dependence of refractive index on wavelength. Waveguide dispersion results from changes in light distribution within the core/cladding structure of the fiber that is also wavelength-dependent. By altering the core/cladding structure, the waveguide dispersion characteristics can be controlled.
The compensating element typically has a large negative dispersion coefficient on the order of 100 psec/ nm/km. However, this compensation is fixed, while the amount of actual fiber dispersion varies with time. Temperature variations change both CD and PMD (e.g., a fiber's zero-dispersion wavelength shifts by 0.03 nm/∞C). Wind and other forms of vibration alter PMD, making it a statistically changing phenomenon. Add/drop operations change optical power levels, triggering nonlinear effects. Finally, when wavelength routing is implemented, optical signals will travel along different paths, making fixed dispersion compensation ineffective.
Dynamic dispersion compensation that accommodates slowly changing dispersion characteristics is needed. A dynamic compensation system requires a feedback loop, using complex control algorithms to track and adjust the desired value of dispersion compensation. An optical dispersion compensation element that can be controlled by an optical or electrical signal is also needed. Various technologies—like fiber Bragg gratings, liquid crystals, and micro-optics—are currently being explored to perform this function. However, the one critical piece is the development of a reliable algorithm to perform optical-signal analysis in real time and drive the dispersion-compensating device.
Electrical correction (equalization). An alternative method to counteract dispersion is to correct the signal in the electrical domain through equalization. Both CD and PMD tend to close the electrical data eye; in fact, by the time the signal is converted to the electrical domain the two effects cannot be distinguished. By performing equalization in the receiver before clock and data recovery (CDR), it's possible to correct for the dispersion effects. Again, a closed-loop system is needed, in which a data eye detector controls the equalizer. This electrical correction can be accomplished by using either adaptive CDR or adaptive equalization.
FEC. In comparing optical compensation with electrical correction, a few observations are apparent. First, optical compensation is proven to work in real networks, while electrical correction is still being developed. Second, optical compensation devices are bulky and costly. While the large size and increased cost might be marginally acceptable in the LH network, it is not acceptable in metro applications. As a result, neither electrical correction nor optical compensation is ideal for metro networks. However, FEC offers a readily available and inexpensive alternative.
Rather than directly correcting dispersion, FEC devices implement numerical algorithms to detect and correct intersymbol interference (ISI)-induced errors, making the optical link inherently more robust. Direct dispersion techniques to prevent errors and FEC techniques to correct them are complementary and can be used simultaneously.
FEC techniques are well known and have been implemented in other fields such as wireless communications for years, but their application to optical communications is quite recent. The "power" of FEC techniques depends on the algorithm used for data coding. Simpler in-band techniques offer about 4 dB of coding gain, while more complex out-of-band techniques provide a 6- to 8-dB improvement. While higher-complexity algorithms offer improved performance, they also require more digital gates, so careful tradeoff considerations are required.
The placement of FEC devices on a line card presents an interesting problem for the optical-module maker and line-card designer. Figure 1 illustrates a standard OC-192 SONET line-card implementation. The optical signal is terminated by a 10-Gbit/sec transponder through the physical media device circuitry. The device circuitry consists of an optical receiver (avalanche or PIN diode) and a laser complemented with a laser driver. From the resultant electrical signal, both clock and data are recovered by the CDR phase-locked-loop circuitry. In a traditional 300-pin multisource-agreement (MSA) transponder implementation, a 1:16 serializer/deserializer (SerDes) is required to convert the 10-Gbit/sec serial signal into the 16-bit-wide SFI-4.1/XSBI parallel I/O interface (and vice versa in the transmit direction).
The FEC device is positioned between the 10-Gbit/sec transponder and the SONET framer. But as shown in Figure 2, it is also possible to place the FEC device inside the optical module. The addition of FEC increases module functionality and fiber reach. The cost is a larger module size and greater power dissipation. Interestingly, some new module vendors are providing even more functionality by including the framer in the transponder (see Figure 3).
The market viability of easier line-card design at higher component costs (due to low deployment volume) remains to be seen. Note, however, that larger power dissipation makes thermal stability of the laser difficult to achieve.
The trend toward greater functionality in SONET transponders seems to run contrary to the trend toward lower-cost and simpler functionality in optical modules for the enterprise and storage spaces. This optical-module evolution from transponder to transceiver is illustrated in Figure 4. The standard 300-pin MSA transponder (Figure 4a) requires a 1:16 SerDes at its output, which can be viewed as two stages of 1:4 SerDes. This module provides the 16-bit-wide interface standardized by the OIF (SFI-4.1) and IEEE (XSBI).
By eliminating one of the 1:4 SerDes devices, we have the transponder shown in Figure 4b, with a 4-bit-wide I/O operating at 3.125 Gbits/sec for 8B/10B XAUI encoded data and 2.488 Gbits/sec for SONET scrambled traffic (with a possible rate overhead due to out-of-band FEC). Various—possibly too many—commercial implementations of that are available: XENPAK, XPAK, and X2. More important, the power dissipation of the transponder shown in Figure 4b is actually higher than that of Figure 4a, since the high-speed parallel implementation requires dedicated data recovery units on each lane (not shown in Figure 4). The resulting connector size reduction due to the smaller number of pins is achieved at the cost of higher power.
To further reduce pin and module size, we need to move to a transceiver solution (see Figures 4c and 4d). Both XFP modules use serial 10-Gbit/sec signaling to communicate through the XFI interface. To achieve 8-12 inches of on-PCB reach, the XFP module shown in Figure 4c uses two CDR devices to clean up the 10-Gbit/sec signals in the receive and transmit directions. For shorter reaches of 2-3 inches, these extra power-hungry CDRs are not required (see Figure 4d).
Which of these four modules becomes a dominant player at 10 Gbits/sec is anyone's guess. Certainly, however, the number of different optical modules is too large and needs consolidation for the 10-Gbit/sec market to become truly cost-effective.
FEC techniques can be very useful for addressing the problems inherent in certain characteristics of deployed fiber in metro networks. They particularly help solve fiber dispersion problems in an efficient and cost-effective manner.
Combining FEC functionality with OEO conversion in the optical module enables a variety of transponder options. The widespread adoption of FEC techniques will help immensely in the required metro bandwidth rationalization to solve the bottleneck problems between the overbuilt LH network and growing enterprise/access networks. But to fully accomplish this goal, one or two flavors of optical modules need to become dominant, providing very low cost through high-volume deployment.
Dr. Krzysztof (Kris) Iniewski is product marketing manager at PMC-Sierra (Burnaby, British Columbia). He wishes to acknowledge numerous people at PMC-Sierra who contributed to this article, especially Dr. Andrew Wright, Kumaran Siva, Tyrone Jung, Rob Crestani, and Phil Northcott.