Gigabit interface converters increase network options
Pluggable, standards-based transceiver modules offer flexibility and cost savings to end-users and manufacturers.
Bob Mayer
Cielo Communications Inc.
In fiber-based networks, the basic function of the transmitter in the optical transceiver is to convert serial-electrical signals into serial-optical signals, which are then focused and sent across fiber-optic cable. On the receiving end, the transceiver converts the optical signal back into the electrical domain and recovers the original signal from any distortions that may have occurred along the fiber transmission path (see Figure).
Essentially, the receiver side of the transceiver must be able to accurately recover data from a signal with a specified minimum "eye-opening" where the amount of "closure" of the eye represents the maximum allowable signal distortion. Distortions to the optical signal result from three major factors:
Attenuation: The natural loss of signal strength caused by transmission through a medium.
Chromatic dispersion: The blurring of signal edges that results from transmission of multiple wavelengths in the optical signal. Different wavelengths of light travel at various speeds through an optical fiber.
Modal dispersion: In multimode fibers, depending upon how it is launched, the light travels down multiple signal paths or modes. Part of the signal may travel down the center of the fiber while the rest may bounce between the edges of the fiber, resulting in parts of the signal traveling different distances.
Both chromatic dispersion and modal dispersion cause portions of the initial signal to reach the receiver at different times. Multimode fibers are designed to compensate for modal dispersion by accelerating the speed of the light as it nears the outside of the fiber, thereby adjusting for the longer path it is traveling. But in real-world deployments, this acceleration cannot completely compensate for modal dispersion. A recent development is the discovery that the performance of multimode fiber is also dependent on the optical source: light-emitting diodes (LEDs) or lasers.
In slower-speed networks, the standards for fiber-optic transmission were designed around the technical performance of LEDs, which met the cost, manufacturing, and performance requirements of fiber-optic links for transmission rates of less than 1 Gbit/sec. At speeds of 1 Gbit/sec and above, LEDs cease to provide adequate performance and lasers are required.
Vertical-cavity lasers, which emit light at 850 nm, offer the promise of meeting current LED costs and volume production capabilities because of their physical similarities (both emit light vertically from the plane of the wafer). Edge-emitting lasers, which are more expensive due to the wafer processing and optical alignment required, are still needed for the transmission of gigabit signals at the longer wavelengths (1300 nm) where chromatic dispersion is minimized.
The different dispersion characteristics of today`s fiber affect the guaranteed distances of the various laser technologies. To provide price/performance flexibility, the gigabit networking standards specify guaranteed link lengths for different wavelengths. This presents a very different deployment situation from that which existed with the Ethernet and Fast Ethernet standards, where a single fiber-optic transceiver could transmit up to 2 km of multimode fiber and cover the requirements of the majority of customer installations.
Evolution to modular GBICs
Because one optical device could accommodate most installed multimode fiber-optic links at the lower speeds, manufacturers used hard-soldered physical media interface (PHY) components to implement the interface. For Ethernet, the common interface featured discrete drive and receive circuits coupled with optical transmitters and receivers, all soldered to the user`s board. For Fast Ethernet networks, the discrete circuits were incorporated into the fiber-optic transceiver, and these integrated modules were then soldered to the user`s board.
At these network speeds, the hard-soldered component model held up reasonably well. Requirements to reduce the board space of the optical transceiver led to the first generation of 1ٻ (one row by nine pins) transceiver components. These 1ٻ transceivers are used in Fiber Distributed Data Interface (FDDI) or Fast Ethernet environments, where designers can design their systems around the component, which incorporates an LED transmitting at 1300 nm to consistently achieve 2-km distances over multimode fiber.
Trying to use hard-soldered discrete transceiver components in gigabit-speed fiber-optic applications, however, presents another set of challenges. With the increase in speed, the bandwidth constant of the fiber-optic cable (i.e., bandwidth ¥ distance) requires that transmission distances become shorter. Therefore, at gigabit speeds, the installed base of multimode fiber is not able to maintain previous transmission distances while using the lowest-cost short-wavelength lasers, such as vertical-cavity surface-emitting lasers (VCSELs). But the cost of universally deploying singlemode fiber and long-wavelength lasers, such as Fabry-Perot, is prohibitive when a large part of the installed base does not require long-haul distances.
Therefore, a dual set of wavelength standards was established for gigabit-level fiber-optic installations to provide flexibility in balancing transceiver and fiber costs against distance requirements. For example, under the Gigabit Ethernet standard, short-wavelength 1000Base-SX links are specified to 220 m over multimode fiber, while long-wavelength 1000Base-LX links can be 550 m over multimode or up to 5 km over singlemode fiber. The total cost of long-wavelength implementation over multimode fiber installations is further increased by the need to use a singlemode offset patch cable to optimally target the launch of the light into the multimode fiber.
While including different wavelengths in the standard allowed greater cost/distance flexibility, this situation complicates network designers` choices when specifying and maintaining the configuration of a particular system. It quickly became impractical to populate every port on every interface card with the same hard-soldered 1ٻ components because it eliminated the flexibility to configure the transceiver for both long- and short-wavelength connections. The hard-soldered nature of the 1ٻ components also meant any configuration decisions made at the outset would limit future flexibility for reconfiguring the ports as network communications requirements changed.
In 1995, the industry adopted an open specification for a gigabit interface converter (GBIC--see photo on page 85). This pluggable transceiver module enables systems builders and network administrators to flexibly configure, incrementally populate, and cost-effectively reconfigure their fiber links as required. Initially designed to support Fibre Channel data networks, the GBIC standard was also quickly adapted for use with Gigabit Ethernet installations. By providing hot-swappable interchangeability, GBIC modules give network administrators the ability to manage their transceiver costs and to link distances and overall network topologies to current requirements. Hot-swappable GBIC modules also leave the door open for changing the network without wholesale replacement of system-level boards.
Benefits of GBICs
Some of the major advantages of GBICs to the system designer include easier manufacturability and a shorter time to market because the printed circuit board (PCB) does not have to be changed to accommodate long- or short-wavelength interfaces. System manufacturers can also offer customers multiple options for distance-versus-cost trade-offs, while still achieving the economies of scale associated with building a single type of PCB with pluggable GBIC ports.
The PCB manufacturing process itself is streamlined because much of the difficulty of assembling with sensitive fiber-optic components is eliminated. Many components cannot be subjected to the high temperatures and aqueous wash environments common to PCB assembly; they require special handling and secondary production operations to complete final board assembly. With hard-soldered components, there is potential for solder mask to get inside the optical subassembly`s ports and contaminate the lens, thereby reducing or eliminating its light-launching capability. With GBICs, the transceiver modules, including the optical interfaces, are plugged in after the entire PCB assembly process has taken place.
GBIC technology also allows users to better align their system configuration costs with real-world requirements. Network operators don`t have to "over-buy" at initial installation. With hard-soldered interface components, network operators would have to estimate all of their future port requirements at the outset and invest in fully populated systems, with the hope that they would "grow into" the full configuration with the passage of time. This approach has two downsides: the loss of opportunity-capital through premature investment and the risk that requirements will change so that the originally projected configuration will never become appropriate.
With GBICs, the network operator has to buy only the number and type of transceiver modules needed to support minimum requirements. For example, a company could invest in a networking system capable of supporting 36 transceivers but only initially purchase 18 transceivers to meet existing user requirements. As network demands change, the administrator can add more modules as required. Because the transceiver components are one of the highest-cost items in any fiber-optic networking environment, the ability to incrementally match deployments to actual growth curves represents a significant opportunity for overall cost containment.
Automated-configuration functionality
The GBIC`s hot-swappable design opens the door to new opportunities for intelligent configuration management, which didn`t exist with hard-soldered components. It uses three "module definition pins" to support hot-swapping, which allows the receiving system to check the type and status of the device that was just plugged in. Administrators can then easily resolve incompatibility issues through software management.
The latest version of the GBIC specification goes even further to support the concept of embedded "serial ID" data for each transceiver device. Using an on-module EEPROM (electrically erasable programmable read-only memory), every GBIC can provide information regarding its vendor, part number, standards, transmission distances, etc. Extended data fields can even contain more details, such as serial numbers and manufacturing date codes. As soon as a hot-swappable GBIC is plugged in, this embedded ID information enables the system to immediately check its status and determine whether the module is qualified for use within the switch and/or appropriate for the media link for which it is to be connected.
The on-module ID information and fault-detection signals also support remote diagnosis and configuration analysis across the network. The network administrator can conserve staff resources by remotely checking the capabilities of each GBIC port on a switch that is experiencing problems, potentially identifying incompatible or malfunctioning modules without an on-site service call.
Another feature of GBIC technology is the ability to selectively disable the transmit function during installation and/or diagnosis activities. A GBIC`s transmit capacity can be automatically held in static mode until the overall system comes up, thereby avoiding undue signal chatter across the network. An individual GBIC can often overcome an intermittent fault condition simply by toggling the transmit-disable function through an off/on sequence that does not affect the rest of the system.
Looking forward
Comparable in reliability to hard-soldered 1ٻs, GBICs improve overall network stability, reliability, and maintainability. The widespread support for the GBIC specification gives systems manufacturers and users a range of options and vendors to choose from.
In the future, more gradations of transceiver options will be implemented in the GBIC form factor. While the modular GBIC concept was driven primarily to provide options for cost-effectively mixing and matching long-wavelength, short-wavelength, and/or copper interfaces within the same system it can be leveraged to support an even wider variety of capabilities now that a standard form factor is in place. Future developments might include specialized GBIC classes to overcome differential mode-delay problems, thereby extending guaranteed multimode distances, or high-performance products to increase link-length capabilities over singlemode fiber.
In essence, the GBIC concept has effectively de-coupled the transceiver function from both the limitations of the PCB assembly process and the dictates of any particular network-transmission media. The result is a smoother implementation process as networks continue their heterogeneous migration from copper to fiber--and in the future, multi-gigabit interface speeds. u
Bob Mayer is the vice president of sales and marketing for Cielo Communications Inc. (Broomfield, CO).