Compatibility, cost shape fiber transceivers for local data communications
Compatibility, cost shape fiber transceivers for local data communications
Interoperability standards and opto-electronic products strive to achieve higher speed, cost-effective data links over optical fiber
kevin brown
hewlett-Packard Co.
The growth of fiber optics in local data communications has been driven largely by the emergence of PC-based local area networks that require high-speed, low-noise backbones. The evolution of opto-electronic transceivers has been largely determined by two often competing forces--the requirements of standardization for interoperability in a LAN environment and the need for lower cost.
Opto-electronic suppliers have struggled to reduce costs to meet the vision of fiber to the desktop, while being constrained by compliance to LAN standards based on positioning fiber primarily as a backbone solution. Ironically, while LAN vendors have spent years developing 100-megabit-per-second wire solutions to circumvent the high cost of fiber, optical technology is finally approaching cost parity with copper. Between 1994 and 1996, fiber`s share of desktop connections will rise from 7% to 19% of all users, according to Business Research Group in Newton, MA. This increase is driven by the desire for a future-proofed, high-bandwidth cable plant, and the shrinking cost difference between fiber versus wire network interface cards and hubs. Cost and compatibility will continue to drive optical transceiver development for the remainder of the decade.
Long after the commotion concerning fiber distributed data interface has faded, and as asynchronous transfer mode is being touted as the connection of the future, there are still more fiber-optic nodes in Ethernet LANs than in FDDI and ATM combined. If there is fiber in your LAN, it is most likely connected to an Ethernet port. Fiber is primarily used in Ethernet as a backbone, although it has been used to the desktop in security-conscious environments such as defense, government and financial applications.
The Institute of Electrical and Electronics Engineers 802.3 standard specifies short wavelength (800- to 900-nanometer) optical links using 62.5-micron diameter core multimode fiber. These optical links typically use ST-style connectors, although SMA and SC connectors are also found. FC connectors for Ethernet fiber connections are more common in Japan.
To minimize cost, designers of Ethernet links typically choose discrete components that offer simple functionality. The transmitters typically consist of an ST-connectorized light-emitting diode driven by a con- stant current driver. Receivers are based on a positive-intrinsic-negative photo diode and transimpedance amplif ier combination. The single-ended trans impedance amplifier is connected to a digitizer integrated circuit. This multi-component type of transceiver design is generally favored because the total cost of the various components is well below the cost of comparable integrated transceiver modules. A complete TTL-to-light transceiver function for Ethernet typically costs less than $40 in large production volumes.
Although the cost of fiber-optic components for Ethernet is less than half the cost of the integrated transceivers used in 100-Mbit/sec LANs, the same low-cost products can be used for 100-Mbit/sec links. Several LAN equipment vendors, including Thomas-Conrad, Plaintree Systems and Alfa Inc., sell 100-Mbit/sec LAN equipment that use 820-nm fiber-optic ports for lower cost. Although there is a distance limitation for short-wavelength technology, the combination of 1300- and 820-nm products allows them to offer a choice between a fully compliant, interoperable LAN system or a lower cost network that is not directly interoperable with equipment from other vendors.
The future of Ethernet
The success and longevity of Ethernet have led to distinct efforts to upgrade Ethernet for higher bandwidth. The fastest emerging option today is switched Ethernet, which substitutes a switched topology for the shared bus of traditional Ethernet. Because the physical layer is unchanged, switched Ethernet can use the same fiber-optic links mentioned above.
The implementations of fiber for Fast Ethernet and 100Base-VG systems require higher performance, because the data rate and physical layers are substantially changed. Fast Ethernet has adopted the physical layer of copper-based FDDI: 100-Mbit/sec data that is 4B/5B-encoded to 125 megabaud, operating over 100 meters. Fiber can be used to extend the distance to 500 meters. For unswitched Fast Ethernet, extensions beyond 500 meters would violate the round-trip delay time, a limitation of the carrier sense multiple access/collision-detect protocol, not of the physical layer.
The 100Base-VG system is built around splitting a 100-Mbit/sec data stream into parallel paths for transmission over four pairs of Category 3 untwisted-pair cable. However, vendors who are developing fiber links for 100Base-VG hubs intend to use a 100-Mbit/sec serial data stream that is multiplexed from the four 25-Mbit/sec unshielded twisted-pair streams for their fiber links. The IEEE 802.12 draft standard includes 2-kilometer links based on 1300-nm optics, and 500 meter links operating in the short-wavelength region.
Another area of growth for fiber in Ethernet applications is the emergence of singlemode Ethernet connections. Just as multimode fiber is being installed to the desktop for future needs, singlemode fiber is increasingly being installed in building wiring system backbones as a hedge against future bandwidth requirements. But users naturally want to leverage that fiber installation for the networks they use today. Although there are no standards for Ethernet transmission over singlemode fiber, a number of hub vendors are offering or developing singlemode Ethernet connections. Installation of singlemode fiber in end-user premises is particularly popular in Japan and Europe, and as American hub manufacturers ship more products overseas, they are adapting their hardware to the needs of those markets.
Singlemode Ethernet connections are typically based on a 1300-nm edge-emitting LED packaged for coupling light into singlemode fiber. The receiver uses an indium gallium arsenide PIN photodetector and transimpedance amplifier that are functionally identical to the receiver used with 820-nm Ethernet links. The advantage of this approach is hub vendors can simply substitute 1300-nm singlemode components into their existing 820-nm fiber designs, without developing new transceiver circuits.
More token ring fiber products
Compared to Ethernet, token ring networks have used relatively little fiber. For years, IBM recommended shielded twisted-pair wire for all but the longest links. Consequently, it is more common in token ring environments for STP to serve as the backbone cable. The completion of the 802.5J trial-use standard for token ring fiber connections has resulted in a sharp increase in token ring fiber product offerings.
The physical layer of the 802.5J standard is virtually identical to that of 10Base-F. Consequently, the fiber components used in token ring are the same LED and p-type-intrinsic-n-type/preamplifier combination used in Ethernet. Although fiber-optic receivers designed specifically for token ring are available, most hub vendors lower the cost of their component purchasing and qualification by buying one set of opto-electronic parts for both 802.3 and 802.5 hubs. Attempts to increase the bandwidth of token ring beyond 16 Mbits/sec have been unsuccessful, as token ring users turn to FDDI or ATM for upgrades.
FDDI--high-speed backbone
FDDI has secured a position as the high-speed backbone available today. However, the original vision of FDDI as the driver of fiber-to-the-desk largely disappeared. In the formulation of this first 100-Mbit/sec LAN, a conservative approach to specifications resulted in a fiber solution that had relatively high costs. The standards committee had the dual goals of defining an interface for 100-meter desktop links and for 2-km backbones. The adoption of 1300-nm technology permitted 100-Mbit/sec transmission over 2 km and promoted the use of 1300-nm transceivers for desktop connections. With it came the expectation that increased volume would dramatically lower costs. In 1990, a 100-Mbit/sec transceiver using 1300-nm LEDs cost approximately $500, 10 times more than a shorter-distance solution that uses the higher volume 820-nm LED components. Five years later, the cost differential has been lowered to approximately two to three times--still a significant premium.
Meanwhile, confusion over whether desktop connections should use fiber, STP or UTP caused many users to wait and see, and FDDI has not been widely implemented as a desktop connection. Using 820-nm products, a few vendors have implemented lower-cost fiber variations of FDDI. These products have found a niche for users who have pulled fiber to the desk and wanted to minimize their network interface controller card cost. However, because 820-nm solutions are not optically compatible with the FDDI standard, their acceptance has been limited.
A second issue of compatibility involves the connector choice. The MIC defined in the FDDI standard is polarized and keyed, but costs more than the ST connectors used in Ethernet and token ring. The X3T9 committee took a step toward remedying this with the newer low-cost fiber-PMD that specifies the duplex SC connector now being universally adopted for building wiring standards. While the LCF-PMD reduces connector and opto-electronic component costs, desktop FDDI (whether fiber or copper) has had limited success. Although the SC connector is a step toward lower cost compared to the MIC, many cable installers still view the SC connector as more difficult and costly than the ST connector. Fortunately, there are now low-cost transceivers with ST connectors that are compliant to FDDI optical specifications. These transceivers are pin compatible with the LCF-PMD duplex SC transceivers, so hub vendors can easily load ST or SC transceivers on their boards.
The ATM Forum took FDDI trans ceivers as a starting point when specifying multimode fiber links for premises applications. Most of the premises switches shipping today use 1300-nm transceivers with duplex SC connectors operating at the optical carrier level 3 (155-Mbit/sec) data rate. A number of vendors (Hewlett-Packard Co., Sumitomo Electric Fiber Optics Corp., AMP Inc., Siemens Co.) sell transceivers with a common footprint and a single row of 9 pins in back of the package (see "Industry Update" page 3). These "1ٻ" transceivers have become the de facto standard for ATM, and are beginning to show up in FDDI applications, as well. These transceivers cost less than $100, but are still more than twice the cost of short-wavelength solutions. As is the case with Ethernet, switch vendors also require singlemode transceivers that fit the same footprint, for easy implementation of singlemode links using boards designed for multimode fiber.
OC-12 data rate
The next area for standardization of optics in ATM premises is at the OC-12 data rate. The ATM Forum has considered several competing proposals for 622-Mbit/sec links over multimode fiber, based on 1300-nm LEDs, 780-nm compact disk lasers, and 850- or 980-nm vertical cavity surface-emitting lasers. The Forum`s Physical Layer Committee recently approved a 1300-nm LED interface specification.
Some opto-electronic suppliers have argued 780-nm lasers will be the lowest cost solution based on the high volume of CD laser production for consumer compact disk players. Others have noted the cost of the CD lasers that are used in data communications transceivers is substantially more than the garden variety CD laser used in consumer applications, because of the unique requirements of data communications. Consumer audio CD lasers have a typical mean-time-to-failure of approximately 20,000 hours, while data communications equipment requires MTTFs greater than 100,000 hours.
Which cost is lower?
The sensitivity of the human eye to radiation in the near-visible region of the optical spectrum imposes stricter requirements on eye-safe power levels for 780-nm lasers than for 980-nm lasers, while 1300-nm LEDs are inherently safer than lasers. The cost of special testing and screening to meet the performance and reliability requirements of ATM equipment results in CD laser costs similar to that of other technologies. Supporters of 1300-nm transceivers maintain 1300-nm LEDs will remain the lower cost technology--at least through the rest of the millennium.
Vertical cavity surface-emitting lasers operating at 980 nm offer the possibility of longer distance at 622 Mbits/sec and even lower cost, because of fundamental advantages in the fabrication process. In addition, the InGaAs photodiodes proposed for use with 980-nm vertical cavity surface-emitting lasers are identical to those used with 1300-nm LEDs. This enables 980- and 1300-nm transceivers to be interoperable, and links could be standardized to work interchangeably with both technologies. More importantly, these lasers are inherently higher bandwidth devices that could also be used for OC-48 (2.488-gigabit-per-second) links.
The use of lasers--whether CD or vertical cavity surface-emitting--with multimode fiber raises unique issues that are not present with LED-based links. The small spectral width of lasers (high coherency) can result in mode selective losses at fiber interfaces. The spectral width can be broadened by the excitation of simultaneous transverse modes. The spectral width of a Fabry-Perot laser is affected by the run length of the data being transmitted. ATM links use scrambled synchronous optical network data, which can result in bit run-lengths to 72 bits, much longer than with such block codes as 8B/10B or 4B/5B used in Fibre Channel and FDDI. Increased coherence with long run-lengths can increase the modal noise floor resulting from mode selective losses.
At the February 1995 meeting of the ATM Forum, supporters of the 1300-nm LED and 980-nm vertical cavity surface-emitting laser approaches presented a case that mode selective loss issues could be adequately dealt with in the formulation of link budgets based on vertical cavity surface-emitting laser technologies; while mode selective loss is not an issue with LEDs. The Physical Layer Committee voted in favor of a proposal for links based on 1300-nm LEDs; other technology options for multimode fiber interfaces are still under discussion.
The 622-Mbit/sec multimode links are limited to 500 meters with 1300-nm LEDs and 1 km with 980-nm vertical cavity surface-emitting lasers; singlemode fiber will be specified for longer distances. This validates the decisions of users who, anticipating the need for 622 Mbits/sec, 2-km links tomorrow, have installed singlemode in their premises backbones today.
Fibre Channel optics vary
Optics for Fibre Channel (American National Standards Institute standard X3T11) are available in different integration levels. Most equipment vendors today choose to buy a daughter card (optical link card) that contains not only a transmitter and receiver, but also integrated circuits with the clock and data recovery and the serializer/ deserializer function used to complete the FC-0 level of the Fibre Channel standard. Although the Fibre Channel standard specifies links at data rates as low as 133 Mbaud and as high as 1062 Mbaud over a variety of media, the optical link cards shipping today primarily use those links specified at 266 Mbaud, using 780-nm lasers, duplex SC connectors and multimode fiber. With FC-0 Fibre Channel integrated circuits now available, designers can place these directly on their boards, buying lower cost transceivers. New products use a low cost 1300-nm LED and PIN/preamplifier in a duplex SC package for 266-Mbaud operation.
Daughter cards similar to the 266-Mbaud optical link cards are being developed for the 531- and 1063-Mbaud rates. These are also typically based on 780-nm lasers with SC connectors. Fibre Channel specification FCSI-301-Rev 1.0 specifies a gigabaud link module operating at 1063 Mbaud with a standard 20-bit TTL interface, as well has half-gigabaud (531 Mbaud), quarter-gigabaud (266 Mbaud) and eighth-gigabaud (133 Mbaud) modules. Gigabaud link modules will provide complete FC-0 functionality in a package 20% smaller than the 266-Mbaud optical link cards currently in production. The anticipated widespread availability of gigabaud link modules will enable easy deployment of 1063-Mbaud Fibre Channel connections.
531 vs. 1063 Mbaud
At the February 1995 Fibre Channel standards committee meeting, solutions for 531 and 1063 Mbaud based on a variety of technologies were proposed. The 1063-Mbaud link specifications support 10 km with singlemode fiber and 1300-nm lasers, and support 500 meters with multimode fiber and 980-nm vertical cavity surface-emitting lasers. At 531 Mbaud, 1300-nm LEDs can be used for 50-meter multimode fiber links, while vertical cavity surface-emitting lasers extend multimode fiber links to 1 km. These proposals were forwarded for letter ballot for public review. Thus, a technology choice between short-wavelength CD lasers and longer wavelength LEDs and vertical cavity surface-emitting lasers faces Fibre Channel as well as ATM users. Because of its experience as a supplier of products based on 780-nm lasers, 1300-nm LEDs and 1300-nm lasers, Hewlett-Packard`s Optical Communication Division claims 980-nm vertical cavity surface-emitting lasers will be the lowest cost technology for high-volume 1063-Mbaud Fibre Channel links.
The lowest cost fiber links are those based on plastic optical fiber, which are very low cost for a variety of reasons. Plastic optical fiber has lower material cost than glass fiber, and consequently has a lower cable cost per meter. These links typically use 650-nm (red) LEDs for operation at a point of minimum plastic optical fiber attenuation. Because there are millions of red LEDs manufactured per day, they are dramatically less expensive than 820- or 1300-nm LEDs, or CD lasers. Finally, because plastic optical fiber is typically 1 mm in diameter, the optical coupling requires relatively low precision compared to standard 62.5-micron multimode fiber. u
Kevin Brown is product marketing engineer at Hewlett-Packard Co.`s Optical Communication Division in San Jose, CA.