Media conversion and short wavelength for Fast Ethernet
Joseph Coffey Ortronics
For fiber to fulfill its promise at the desktop, a smooth transition to Fast Ethernet must be created.
Fiber-optic technology has been in commercial use since the late 1970s. The inherent capabilities of fiber optics, such as noise immunity and high bandwidth, make it the "ideal" transmission media for a variety of applications. However, at the workstation level (the horizontal component), various technological issues that have kept the cost of using fiber very high have delayed the technology`s use. While some end users have been installing fiber to the workstation outlet since the 1980s as the ultimate in "future-proofing" their networks, this future-proofing was based on the expectation that cost-effective "fiber-to-the-desktop" products would begin to emerge. Thus the majority of this installed horizontal fiber cabling remains in the "dark mode," as the lack of cost-effective fiber-optic products, generally low end-user awareness, and the advent of Category 5 unshielded twisted-pair (UTP) standards caused interest in fiber to wane.
Yet recent changes in cabling standards; developments in optoelectronics, fiber-optic cables, and connectors; and escalating bandwidth demands have caused end users to reconsider "fiber-to-the-desktop" as a viable alternative to copper in the horizontal local area network (LAN) segment. However, for fiber to establish a strong role in these horizontal applications, there must be a clear and cost-effective upgrade path from 10 to 100 to 1000 Mbits/sec.
The 10Base-FL specification for IEEE 802.3 Ethernet has been available for some time as a 10-Mbit/sec fiber solution. The IEEE 802.5 Token Ring standard also has a fiber solution for up to 16 Mbits/sec, while the new Gigabit Ethernet standard defines the requirements for 1000Base-SX, a short-range, low-cost solution for 1000 Mbits/sec. But Fast Ethernet at 100 Mbits/sec has never had the support of a low-cost fiber standard. Instead, fiber development for Fast Ethernet has been cost-constrained by the long-wavelength requirement defined in the 100Base-FX specification. A new standard, 100Base-SX, is under development and should be out for ballot in the fourth quarter of 1998. This standard will promote the use of low-cost, short-wavelength fiber optics as the "fiber-to-the-desktop" solution for Fast Ethernet.
Fiber and electronics technology
"Short wavelength" refers to 850 nm, and "long wavelength" refers to 1300 nm. Table 1 shows two wavelength-dependent windows of operation that apply to current multimode optical fiber. A third window (not shown) has a center wavelength in the range of 1500 to 1550 nm and is not used in multimode fiber applications.
These windows of operation are defined by the attenuation characteristics of optical fibers at a specific wavelength. Attenuation in optical fiber is the result of energy absorption and scattering caused by impurities in the glass. Unlike attenuation in copper cable, the attenuation in an optical fiber is constant and depends upon the length of the optical fiber and the propagation wavelength. Impurities in commercial optical fibers such as ferrous and ferric ions absorb light in the visible spectrum. Another impurity is water in the form of hydroxyl (OH) ions, which causes attenuation to peak at certain wavelengths. These attenuation peaks can have such high magnitudes that the optical fiber is of little use in these regions and is a significant issue in optical fibers produced before 1996.
Loss peaks due to OH absorption have been reduced over the years such that the only significant peak occurs at 1370 nm. However, there is still a considerable installed base of optical fiber produced prior to 1996. This is why most optoelectronic devices still operate at a wavelength defined within one of the three operational windows. The post-1996 improvement in the attenuation characteristics of optical fiber makes the wavelength range of 720 to 1370 nm available for use. This is significant for developers of wavelength-division multiplexing (WDM) systems.
Table 2 is a comparison of 62.5- and 50-micron optical fiber at specific wavelengths. Both core sizes can be used for LAN applications. In 50-micron optical fiber, the bandwidth is the same regardless of wavelength. However, using cable of this core/cladding size will result in a 3-dB reduction in launch power due to aperture differences. If the power budget for the worst-case link configuration can accommodate the reduction in launch power, the increased bandwidth will support applications such as Gigabit Ethernet with sufficient margin.
Considering that the attenuation of an optical fiber is highest in the 820- to 920-nm range for 62.5-micron optical fiber, then why operate at this wavelength? Quite simply, the optoelectronics (LEDs and PINs) for the short-wavelength region are very low cost in comparison with the equivalent long-wavelength devices. They can also support the data rates required by Ethernet, Token Ring, Fast Ethernet, and some 155-Mbit/sec applications. Even the vertical-cavity surface-emitting laser (VCSEL) is lower in cost that the long-wavelength LED/PIN (see Fig. 1).
The LED and the PIN are the most common optical sources and receivers used for short-wavelength, multimode optical-fiber applications. The most common form of packaging used for LEDs and PINs is the 2 ¥ 4 cube that includes the optical connector. This cube is a PCB DIP-mountable device with two columns of four pins. It is sometimes referred to as a "sugar cube."
The LED is essentially a p-n junction that spontaneously emits light when a current is passed through it. LEDs can support data rates up to 125 Mbits/sec. The PIN is a photo-sensitive diode that consists of a negative and a positive region separated by a depletion region. This region is created by reverse biasing so little or no current can pass though the device. When light impinges on the diode, the photons absorbed create electron-hole pairs in the depletion region. The charge carriers then separate and drift toward their respective terminals. For each electron-hole pair created, an electron flows from the device to the external circuitry.
The current flow caused by the electron-hole pair process is very small and difficult to detect from the ambient electrical noise. To reduce the effects of noise and to improve signal detection, a transimpedance amplifier is included in the PIN package. These receiver configurations are referred to as PIN-FETs. The FET is a differential amplifier that provides an output voltage proportial to the changes in the PIN impedance. Current flow caused by the electron-hole process causes the impedance of the PIN to change. The change in impedance can be used to create a proportional output voltage by altering the transfer function of the PIN-FET combination. A transfer function defines the ratio of output to input and is controlled by impedances and the gain of the amplifier.
Lasers and ava lanche photodiodes (APDs) are another type of optical sources and detectors that can be used in an optical-fiber system. These devices can support extremely high data rates. The limited spectral width of the laser facilitates coupling of the output optical energy into small core fibers (singlemode). The design of APDs provides very good quantum efficiency, making them ideal for use in "low-light" applications.
However, both devices are considerably more complex and require a greater overhead in terms of electrical and temperature control to maintain stable operation. This complexity also results in very high cost, which limits their use to applications where the "bandwidth-to-dollar ratio" is low. Typical examples would be high-bandwidth backbones and long-haul telecommunications links between central offices. Due to the current cost of these devices, they are not considered viable components for use in "fiber-to-the-desktop" equipment.
The exception to the "laser rule" is the VCSEL, which operates in the short-wavelength region. This device, unlike the LED, is a semiconductor laser and can support a data rate of up to 2 Gbits/sec. The VCSEL requires a lower drive current than an LED but produces up to 1 mW (0 dBm) of optical power with a spectral width less than 0.5 nm. At data rates less than 200 Mbits/sec, the VCSEL can be driven directly by a positive emitter-coupled logic (PECL) gate. This greatly simplifies design requirements and makes the VCSEL an ideal device for applications where auto-negotiation from low to high data rates is required (see Table 2).
The superior bandwidth of the VCSEL has made the multimode fiber option for Gigabit Ethernet (1000Base-SX) possible. Table 38-2 of the IEEE 802.3z specifies the operating range for 1000Base-SX over the different types of optical fibers.
The development of the VCSEL and its application in 1000Base-SX has renewed considerable interest in the 850-nm region. The VCSEL is also superior to the LED in terms of packaging. Since the VCSEL does not require a lens, multiple VCSELs can be formed in an array on the same substrate. This is ideal for high-density ribbon cables and even for WDM applications. The PINs used with VCSELs can also be fabricated in an array, complementing the VCSELs for high-density ribbon-cable applications. The narrow spectral width of the VCSEL allows each emitter in the array to operate at a different wavelength. The output of each emitter can then be coupled onto a single optical fiber. If each VCSEL can support a data rate up to 2 Gbits/sec, then an array of four VCSELs could possibly support up to 8 Gbits/sec over a single optical fiber!
Optical standards
End users and network designers alike are becoming more concerned over issues such as electromagnetic and RF interference, bandwidth, link distance, data security, and network downtime. The only medium that addresses all of these points is fiber optics. These concerns have been reflected by the introduction of the EIA/TIA TSB-72 in 1995 and the formation of the Short Wave Length Alliance in 1998 under the TIA Fiber Optic LAN Section.
The EIA/TIA TSB-72 is a guideline for centralized optical-fiber cabling and is an addition to the current TIA/EIA-568A Commercial Cabling Standard. TSB-72 allows cable runs up to 300 m using optical fiber. This enables network designers to take advantage of long transmission distances to centralize network electronics such as routers, bridges, hubs, and switches in a single equipment room area. This architecture provides a migration path for users to evolve from the shared current bandwidth environment to an efficient, switched environment.
Centralized network architecture increases network flexibility, simplifies adds/moves/changes, reduces network down time, and simplifies network management. Most important, however, is that it significantly reduces installed costs. With centralized networking architecture, designers can effectively reduce the number of ports and chassis. However, it does increase the link distance between the workstation and telecommunications closet and has pushed the use of optical fiber to the forefront in network design. This has caused renewed interest in the use of optical fiber as the transmission medium for LAN applications, with media conversion being the optimal short-term solution.
In this context, one of the fastest growing LAN applications is 100-Mbit/sec Fast Ethernet. The IEEE 802.3u standard defines the use of fiber for this application under the 100Base-FX embodiment of 100Base-X. As shown in Figure 2, the 100Base-FX family uses the physical medium dependent (PMD) and the medium dependent interfaces (MDI) as defined in the ISO/IEC 9314 and ANSI X3T12 (FDDI).
The PMD and MDI have the greatest effect on the landed cost of a fiber-optic product for Fast Ethernet. The PMD specifies an operational wavelength of 1270 to 1380 nm and a duplex SC to be used as the MDI. This requires the use of 1300-nm LEDs and PINs. The landed cost of the 1300-nm devices using the 2 ¥ 4 package style is considerably higher than the 850-nm devices as shown in Figure 1. This high cost is one of the reasons the 100Base-FX devices are priced in the several hundred dollar range--too expensive for workstation applications. The duplex SC or MDI usually includes other electronics (quantizer, LED driver) in the package to form a 1 ¥ 9 transceiver module. This arrangement creates a compact package but it also increases the cost that is already burdened by the 1300-nm LED and PIN.
As a result, the Short Wave Length Alliance (SWLA) was organized during the first quarter of 1998 by a group of concerned manufacturers to address the need for a standard to support the development of low-cost, short-wavelength fiber-optic equipment for Fast Ethernet. Two working groups were formed during the Dallas meeting to develop the requirements for the 100Base-SX standard. These were the PMD and the Auto-Negotiation (AN) groups. Each group meets via teleconference on a weekly basis.
The alliance agreed to investigate a 100Base-SX initiative based on the following goals:
The primary emphasis is low cost. This constraint means that the standard would have to be developed so "off-the-shelf" optical components can be used. By using the existing short-wavelength optical devices (LEDs and PINs), a low-cost solution can be achieved.
The 100Base-SX standard will define AN for 10/100 over the fiber link.
A minimum 300-m (full-duplex) link distance will be supported. The link budget will be developed using the parameters associated with 62.5-micron multimode optical fibers.
The standard would ensure interoperability with existing 10Base-FL products. The FDDI PMD, ISO/IEC 9314-3, would be copied and modified where required for short-wavelength requirements. This also fits into the current IEEE 802.3u.
Finally the standard would be connector-independent and provide an easy upgrade path to 100 Mbits/sec.
Additional goals for the AN group include minimizing implementation costs, auto-negotiation between current 10Base-FL and new short-wavelength-based 100-Mbit/sec systems, parallel-detect non-negotiating devices, support of full-duplex operating modes, and optional implementation. The PMD group will have an interoperability demonstration hosted by TIA at the Networld + Interop show this month.
Media conversion
One very popular design option for achieving fiber to the desktop is through the use of media converters. These devices allow a LAN to be upgraded to fiber with minimal service disruption and without losing the investment in the copper-based LAN equipment. This is extremely important considering the investment that has been made in smart hubs and switches. Unless these units have very modular architectures, upgrading to fiber usually means wholesale equipment replacement. On the workstation side, replacement of a network interface card (NIC) to support fiber requires a software driver change, which is not always a simple task. In the end, who gets to keep the obsolete UTP equipment?
Every form of communication currently in use, whether for voice or data, involves some form of media conversion. Media converters used for copper-to-fiber conversion are both visible and accessible to the end user simply because they are physically external to the LAN equipment.
As shown in Figure 3, a basic media-converter model has a UTP connector on one side and optical-fiber connections on the other. The UTP side is interfaced to IC1 via the magnetic coupling module T1. T1 provides the necessary impedance matching between IC1 and the UTP cable. It also provides common mode attenuation of potential interference, which is necessary for EMI/RFI control. The purpose of IC1 is to convert the incoming UTP analog signals, which may be coded as Manchester or Multi-Level Tier-3 (MLT-3), into non-return to zero (NRZI) format. The NRZI signals are input to the LED driver (IC2), which has the task of converting the differential PECL signals into an LED drive current. The maximum drive current is usually set by an external resistor but the modulation follows the NRZI data.
On the receiver side, the PIN detects variations in the incoming optical signal and outputs an amplified electrical voltage that corresponds to the optical variations. The quantizer PECL output signal is input to IC1, where the reverse of the LED or transmit process occurs. The received NRZI signal is converted back into an analog Manchester or MLT-3 code. IC1 in conjunction with T1 drives the signal onto the UTP. In addition, IC1 may be equipped with equalization and level controls that ensure the signal is transmitted onto the UTP with minimal distortion.
Some manufacturers have designed IC packages that include the LED driver (IC2) and the quantizer (IC3) in a single package, which reduces the chip count from three to one. However, switching noise that propagates through the substrate caused by the LED driver limits these devices to use in 10Base-T/FL and Token Ring applications. High-speed devices that have both the LED driver and the quantizer do exist, but because of the switching noise problem the receiver sensitivity is greatly reduced. This reduction in receiver sensitivity reduces the link budget value, which in turn reduces the maximum transmission distance.
Short- and long-wavelength transceivers are units that contain the LED, LED driver, PIN, and quantizer in a single 1ٻ package. The package also serves as the media interface for the optical fiber connector. The transceiver does not reduce the total chip count but it does reduce the footprint required for PCB assembly.
Figure 4 shows two applications of media converters. In the first application, the workstation is connected to the hub via two media converters and optical fiber. The cable distance can range from 300 to 2000 m (UTP is 100 m only) depending upon the required bandwidth, type of optical fiber, and the type of media converter. The media converters in both applications are transparent to the equipment on either end. The second configuration shows two hubs linked together with media converters. This application is typical of basic backbone design and is the most common use of media converters.
Fiber-optic media converters have been available for most LAN applications (10Base-T/FL, 100Base-FX, Token Ring, etc.) for some time. The expected growth rate for media converters in general is compounded at 23% per year. Over the past two years, the number of installed 10Base-T/FL media converters has risen dramatically due in part to the low cost of the short-wavelength LEDs and PINs being used as the optical source and detector. 100Base-FX media converters have also been available but are considerably more expensive, since long-wavelength LEDs and PINs are used as the optical source and detector.
Media converter development
One challenge in the design of media converters for Fast Ethernet is that the equipment is not specifically addressed by any standard. In addition, the use of short-wavelength technology for Fast Ethernet is not addressed in any current standard, either. Thus, most Fast Ethernet media converters are based on the building blocks of the IEEE 802.3 model shown in Figure 2. They also use 1300-nm devices as specified in the 100Base-FX specification. Since the layers above the media-independent interface (MII) level are not required for a media converter, an MII cross-over is made as shown in Figure 5 to form a media converter.
This approach doesn`t always produce the most efficient design, however. For example, the initial development of one short-wavelength media converter, the TransOptix Fast Ethernet system, followed the IEEE 802.3 model as shown in Figures 3 and 7, but it was soon realized that this approach would not meet the cost target of the project. The only way the goal could be achieved was to eliminate the overhead required at the PCS and MII levels as shown in Figure 6. In this design, the MLT-3 data, already scrambled by the UTP NIC and/or hub, is received and converted to NRZI by the PMA. Once converted, it can be applied directly to the fiber PMD section. Likewise, when optical signals are received, they are converted back to NRZI by the fiber PMD. Then the PMA converts NRZI back to MLT-3. This design is totally transparent to the host UTP equipment.
General applications
One major advantage of using optical fiber for LAN transmission media is distance. The current copper cabling standards limits the distance to a total of 100 m. Both the 10Base-FL and 100Base-FX standards support a maximum distance of 2000 m using 62.5-micron multimode optical fiber. Both the 100Base-SX and 1000Base-SX standards will support a minimum distance of 300 m using short-wavelength devices. These incredible distances provide the network designer with the ability to build in design flexibility while controlling equipment costs. u
Joseph Coffey is director of research and development for Ortronics (Pawcatuck, CT).