The best is not always the fastest

Fiber optics lends itself to the rabbit and the hare analogy, and often the hare is the hands-down favorite. Physical limitations in the existing fiber plant create an environment where 10-Gbit/second serial is at a disadvantage against the "slower" alternative, LX4. The LX4 standard defines a physical (PHY) layer that exploits data in the XAUI form, four lanes at 3.125 Gbits/s, where each XAUI lane transmits on a slightly different wavelength at 1,310 nanometers down one fiber.

LX4 wide wavelength-division multiplexing (WWDM) has risen, fallen and risen again in industry interest and activity. A couple of years ago LX4 had the potential to be the one-size-fits-all PHY layer to serve all 10-Gigibit Ethernet requirements. An LX4 transponder is unique in that it is suitable for all the fibers defined under 10 Gigabit Ethernet, from the 300-m multimode fiber (MMF) used in legacy networks, to single-mode fiber (SMF) used in long-haul communications.

However, the market impetus for LX4 dropped by the time the IEEE 802.3ae spec was approved. The rising popularity, and availability, of 10-Gigabet serial solutions appeared to represent the best long-term technology answer. The rabbit took the lead, and was well on its way to winning the race.

Then, legacy issues moved to the fore. New-generation systems are being deployed to replace systems installed in existing facilities using existing fiber-optic cabling. These 10-Gigabit Ethernet ports are not for long-haul applications only; they serve between platforms as well, with more than 50 percent of the 10-Gigabit Ethernet ports being consumed in local, short reach links. This legacy, installed multimode fiber (MMF) drawn in the Fiber Distributed Data Interface FDDI days, works well enough for 1Gigabit Ethernet and at data rates through 3.125 Gbits/s, but is practically opaque to 10-Gigabit Ethernet serial communications. Modal dispersion is the prominent issue, with different modes in the MMF spreading the received pulse.

Modal dispersion occurs at all data rates, in about the same absolute amount. The problem is that at 10-Gbit/s serial, modal dispersion takes up most of the 100-ps eye, causing eye closure after 20 or 30 m of MMF. Unfortunately, MMF was used both in horizontal applications, within a floor, and in risers. Up to 300 m of MMF can, and does, exist in many installations. This makes 10-Gbit/s serial a nonstarter for these applications, which is a major problem as data centers consider upgrading to next-generation systems.

LX4 implements 10-Gigabit Ethernet with each of the four lanes operating at 3.125 Gbits/s. Modal dispersion will still exist, however; the larger 320-ps eye permits the accommodation of much longer fiber lengths. Spec-compliant LX4 modules will achieve the 802.3ae spec limit of 240- to 300-m lengths with great margin. Vendors, such as Molex, have demonstrated links up to 2 kilometers on legacy MMF.

What is LX4?

The best illustration of LX4 is through an MSA transponder, such as a Xenpak. The electrical input to the Xenpak is a XAUI interface. The four lanes at 3.125Gbits/s are retimed, each is used to modulate a laser around 1,310 nm, and the four lambdas are combined in a WWD for optical transmission over a single fiber (Fig.1; illustration is from IEEE 802.3ae).

The four optical transmitters have challenging specifications to meet. The four wavelengths are centered 25 nm of each other with a width of ~13 nm. The wavelength characteristics of the light source must stay within these limits. A basic requirement is for the laser to have sufficient stability over time and temperature to stay within its assigned band. When a laser is turned on and off quickly, it can chirp, which will result in increased line width, a budgeted channel deficiency. These requirements dictate the use of distributed feedback lasers in LX4 modules. It would be nice to instead use the lower-cost lasers, but vertical-cavity surface-emitting laser and Fabry-Perot lasers do not yet have the characteristics required for this WWDM application. LX4 transponder companies will be working on ways to lower the cost, power consumption and complexity of their products, and advances are imminent in moving alternatives into the LX4 market.

The marketplace has a need for these 10-Gigabit Ethernet solutions that are delivered within MSA form factors, with Xenpak, X2 and Xpak the modules to target. LX4 was initially a packaging challenge for Xenpak, the largest of the three. Advances in the packaging of the laser sources, the detectors and the WWDM optical components have gone a long way toward reducing the physical space required to implement LX4. Also important are the advances in the ICs that are required to comply with both LX4 as well as the MSA's requirements.

Figure 1 shows a retimer function, and the 802.3ae spec is careful to note that this function is optional. There are two important reasons that such a retimer device will be required in the data path in practical systems.

The first reason is that the XAUI is being spread too thin. Typical XAUI I/Os are designed to drive a minimum load of 20 inches of PCB. Many can do better, but consider the elements of a LX4 link. The three elements of this link are two sections of PCB, one on the transmission end and one on the receiving end, plus the optical channel in between. On top of that, typical optical drivers and receivers use no equalization, reducing further the ability of XAUI to be used directly.

The second reason a retimer device will be required has to do with control plane issues. The 802.3ae spec has an extensive set of registers reporting the performance of the LX4 link. Registers are typically required by the MSA transponder definitions. An additional management interface is critical for operation of MSA modules, the two-wire interface used by the digital optical module (DOM) ICs.

The LX4 spec, as a part of the IEEE 802.3ae structure, builds upon the register definitions for other 10-Gigabit Ethernet interfaces. The LX4 spec includes appropriate registers that apply to LX4 related to link status and performance. These are typically accessed through a serial management interface, MDIO, or management data input/output.

Module-based implementations must also comply with the Xenpak MSA register definitions. The link-status alarm-interrupt (LASI) control and status registers provide critical indications to the host system. Additional registers containing PHY-specific information stored in EEPROM in the DOM are accessed by a two-wire serial interface.

A module implementation for LX4 will need to consider all of these register issues. Retimers not specifically designed to accommodate both the Xenpak and LX4 requirements will have to include a separate controller to serve as the MDIO interface to all elements. This would result in a mess wherein the microcontroller must be configured to intercept the MDIO signals and provide access to the nonvolatile register (NVR) as well as polling the LASI registers or pins (see Fig.2).

Some retimer ICs on the market are up to 23 mm per side. The microcontroller, if implemented in a FPGA, may be 12 mm per side. This complexity marginally fits into a relatively large Xenpak module. Given the size limitations of a smaller X2 or Xpak module, implementing such a module without relief from the IC side would be more than difficult.

The BBT3821 is an example of a retimer IC that provides such relief (see Fig.3). The retimer IC anticipates all required MDIO registers for either LX4 applications, in addition to the vendor-specific MDIO registers for its own use. Further, the retimer IC will need to have a two-wire interface for the external EE-PROM, plus the DOM ICs that an LX4 module would require.

The two reasons defined above for inserting an IC in the data path are compelling. Retiming is a fundamental 10-Gigabit Ethernet concept whereby the transmitted data is retimed to the local clock source, resetting the jitter budget to the local timing domain. This exploits the interpacket gap in 10-Gigabit Ethernet, which permits the insertion and/or deletion of idle columns to accommodate the minor frequency differences between the incoming data and the local clock source.

Thermal considerations are a major fact of life in optical transponder designs. The retimer IC needs to play its part as well. Low power is always beneficial. Also of great importance is the thermal path. A retimer IC whose primary thermal dissipation path is through the top of the package will be an asset in a transponder module design. This way, the heat can be conducted to the module's exterior in the most direct manner possible.

LX4 is the only viable solution today for the installed base of MMF. This is not a short- term issue. Today 10-Gigabit Ethernet plays a role at the high end of the data pyramid, but its role will expand as time passes. Intrabuilding links that started as 10/100 and moved to 1-Gigabit Ethernet will move to 10-Gigabit Ethernet quickly enough. These links will need to run over the installed fiber. Drawing new fiber will be one of the last options exercised.

One alternative to LX4 that is being developed is 10-Gbit/s serial with electronic dispersion compensation. EDC holds the promise of enabling 10-Gbit/s serial to operate over 300 m of legacy MMF. EDC solutions at 850 nm and 1,310 nm are being discussed, although they may have differences in implementation. Standards activities on EDC are just now starting, and will take a couple of years to complete them. EDC has other challenges in MMF. Given that the specific compensation that will be performed is based on the measured modal dispersion, and that the modal dispersion in a fiber is a function of its physical positioning, EDC will have to solve one difficult problem while needing to also accommodate the changes in the modal dispersion that come from normal fiber flexing caused by temperature and vibrations.

Will LX4 meet its original promise of being the one-size-fits-all solution? Maybe not. It is, however, a solid contender to be the implementation of choice to serve the existing MMF links. It's also a contender to serve new-generation systems that extensively use 10-Gigabit Ethernet links based on MSA transponders such as Xenpak, X2 and Xpak. Incumbency is a powerful force in this market. Once LX4 becomes established as the MMF solution, it will earn a place as the 10-Gigabit Ethernet MMF optical standard.

Bill Woodruff is vice president of sales and marketing at BitBlitz Communications Inc. (San Jose, Calif.).

See related chart

See related chart

See related chart