Ethernet interoperability brings fiber to the desk
Ethernet interoperability brings fiber to the desk
Auto-negotiation promises to solve the problem of accommodating the different wavelengths associated with 10/100-Mbit/sec Ethernet transmission.
Michael Laudon Cypress Semiconductor
As users migrate to high-speed networking, they must confront the fact that the vast majority of networked PCs and workstations are still linked to 10-Mbit/sec Ethernet connections. The move from Ethernet to Fast Ethernet is simplified on twisted-pair cabling through the use of physical layer silicon that automatically negotiates between 10- and 100-Mbit/sec data rates.
The same is not true for fiber-optic Ethernet. ieee 802.3 10-Mbit/sec fiber-optic Ethernet, referred to as 10Base-FL, operates exclusively on 850-nm optics. Conversely, 100-Mbit/sec Ethernet, or 100Base-FX, is based upon the Fiber Distributed Data Interface (fddi) physical layer media-dependent (pmd) standard and operates on 1300-nm optics. The use of two different wavelengths makes it impossible to negotiate automatically between 10- and 100-Mbit/sec fiber-optic Ethernet. Furthermore, the high cost of 1300-nm optics has relegated 100-Mbit/sec fiber-optic Ethernet to backbone applications.
Both of these problems can be rectified through the use of 850-nm optics in 100-Mbit/sec Ethernet, permitting auto-negotiation between 10Base-FL and 100Base-SX. (100Base-SX, where "S" stands for "short wavelength," is not an ieee standard. However, it is used throughout this paper to describe 100-Mbit/sec fiber-optic Ethernet using 850-nm optics, eliminating confusion with 100Base-FX.) 100Base-SX uses the same fddi signaling method (4B/5B, non-return to zero inverted) currently used by 100Base-FX.
Background
Auto-negotiation, or N-way, is a function that allows two devices to exchange information over a link segment. The function allows the devices at both ends of a link segment to advertise abilities, acknowledge receipt, and understand the common mode(s) of operation that both devices share. Auto-negotiation is currently used by twisted-pair Ethernet link segments to select between the various physical layer standards (10Base-T, 100Base-T4, and 100Base-TX) as well as pass other information.
Auto-negotiation over twisted-pair cabling is described in the ieee 802.3u standard, clause 28. Auto-negotiation is performed using a modified 10Base-T link integrity pulse (lip). Instead of a single 100-nsec lip, auto-negotiation encapsulates local device information within a burst of closely spaced lips, referred to as a fast link pulse (flp) burst. The flp burst consists of a series of lips that form an alternating clock/data sequence, in which 17 clock pulses and up to 16 data pulses constitute a single flp. Extracting data bits from the flp yields link code words containing information about the transmitting physical layer`s capabilities.
The proposed 100Base-SX fiber auto-negotiation process is executed in a manner similar to that of twisted-pair auto-negotiation. 10Base-FL Ethernet requires that a constant 1-MHz idle signal be transmitted on the fiber when data packets are not being sent. Similarly, 100Base-SX constantly transmits a 62.5-MHz idle signal between data packets. The 10Base-FL idle signal is modulated so that auto-negotiation clock and data pulses are represented by 3 msec of logic high and the absence of clock/data pulses are represented by a continuous 1-MHz idle signal. A clock/data pulse length of 3 msec was chosen to eliminate confusion with the 2.1-msec logic high allowed during start of idle in 10Base-FL. The fiber auto-negotiation scheme also performs parallel detection for physical layers that are only for 10 or 100 Mbits/sec through logic that identifies the speed of the received idle signal. Parallel detection forces the link to operate at only half-duplex.
Figure 1(a) is an example of a fiber-optic auto-negotiation waveform. The auto-negotiating physical layer initially starts with a 10Base-FL 1-MHz idle signal. The link code word clock and data bits are sent to the receiving physical layer by placing a 3-msec logic high on the fiber. Note that in 10Base-FL, logic high is represented by low light at the media interface, which translates to 3 msec of low light on the fiber. The clock and data pulses are shown in Fig. 1(b). The typical timing between a clock and data pulse is 62.5 msec, the same as twisted-pair auto-negotiation. Figure 1(c) shows the burst-to-burst timing of 16 msec, which is also the same as twisted-pair flp burst timing.
This fiber auto-negotiation scheme was chosen because it works in twisted-pair applications and allows simple twisted-pair-to-fiber conversion with a minimum of buffering and control logic. Thus, a hybrid network of both twisted-pair and fiber is simple and inexpensive to implement.
The base link code word transmitted by the fiber-optic physical layer needs to include the following information on technology abilities: 10Base-FL half-duplex, 10Base-FL full-duplex, 100Base-SX half-duplex, and 100Base-SX full-duplex. Figure 2 shows an example of a fiber-optic link code word. Bits A4 to A7 are reserved for additional information that may be required in the future.
The requirements of the pmd layer are constrained by the fiber-optic power link budget. All of the following calculations are made assuming 62.5/125-micron fiber, based on ieee 802.3, clause 15.
The 10Base-FL standard requires a minimum emitter launch power of -20 dBm and a maximum bit error rate of 10-10 at an optical receive threshold of -32.5 dBm, resulting in a 12.5-dB link budget. The 12.5-dB link budget includes 5 dB for connector loss, which remains the same for 100Base-SX. Current integrated circuit technology can achieve a sensitivity of -27.5 dBm average optical receive power using 850-nm optics at 62.5 MHz (the maximum frequency content of 100-Mbit/sec Ethernet). This results in 100Base-SX having a total link budget of 7.5 dB, 5 dB less than the 10Base-FL budget.
The only place that the 5 dB difference can be rectified is in the length of fiber allowed. Because 100Base-SX must be backward-compatible, this results in a 7.5 dB minus 5 dB (connector loss) = 2.5 dB budget for fiber attenuation at 100 Mbits/sec. ieee 802.3, clause 15, allows a fiber attenuation of 3.75 dB/km at 850 nm, resulting in a 100-Mbit/sec maximum link distance of 670 m:
When using fiber in desktop Ethernet applications, the maximum collision domain is 320 m with a Class II repeater, far less than the 670-m maximum transmission distance. A half-duplex link distance of 300 m is easily achievable with 100Base-SX and is more than sufficient for desktop applications.
When using standard off-the-shelf 850-nm fiber-optic components, the average optical receive power level of -27.5 dBm can be translated directly into a single-ended output voltage. Equation 2 shows how dB is translated to power.
To achieve reliable links at -27.5 dBm, a margin of at least 1.5 dB is desirable. This moves the sensitivity requirement for the quantizer to -29.0 dBm. Over time, half the non-return to zero inverted data stream will be ones, resulting in peak-to-peak received power that is 3 dB higher than the average received power (i.gif., the receiver sees light 50% of the time). Adding the 3 dB to -29.0 dB results in a quantizer sensitivity requirement of -26.0-dBm received peak power, as shown in Equation 3.
The resulting power seen by the receiver is 2.51 msec, shown in Equation 4.
The worst-case performance of an industry-standard 850-nm receiver over temperature and voltage is 4.5 mV/mW. Equation 5 shows that the minimum sensitivity of the quantizer needs to be 11.3 mVp-p for reliable operation of a 100-Mbit/sec Ethernet fiber-optic link at -27.5 dBm.
However, because 300 m is sufficient for desktop 100-Mbit/sec Ethernet applications, the sensitivity requirements can be relaxed. Equation 6 shows that for reliable 300-m communications the quantizer needs to be error-free down to 15.5 mVp-p at 62.5 MHz.
Fiber-optic driver design
Figure 3 shows a block diagram of a system that implements fiber auto-negotiation as well as the fiber driver and receiver. A poorly designed fiber-optic light-emitting diode (led) driver can adversely affect the receive sensitivity. Fiber-optic leds can be modeled as an ideal diode with series resistance and parallel capacitance. The capacitance in a fiber-optic led makes it more difficult to drive than simply switching current on and off. Using a simple series switch will result in a waveform with a slow rise time and very slow turn-off time, unsuitable for high-speed communications.
The undesirable characteristics of a series-switched led can be eliminated with a properly designed driver circuit. The three most important techniques to enhance led performance are pre-biasing, peaking, and a low-impedance discharge path. "Pre-bias" is a small forward current applied to the led when it is in the "low" light state. The pre-bias current prevents the fiber-optic led junction and parasitic capacitance from discharging completely when the led is off. "Peaking" is an increase or overdrive of the led for a short period during the rising and falling edges of the current pulses used to modulate the emitter. The momentary increase in led drive current will dramatically improve the rise and fall time of the light output without causing overshoot of the optical pulses. The third technique, providing a "low-impedance discharge path" for the led capacitance, can be provided by either a resistive pull-up connected from the cathode to Vcc or an active pull-up such as a metal-oxide semiconductor transistor. The active pull-up provides superior performance, but requires more care to implement.
To implement an effective 10/100-Mbit/sec Ethernet system, the peaking must be adjustable and should turn off during 10-Mbit/sec operation. Peaking causes high amounts of crosstalk that are coupled into the fiber-optic receiver. This crosstalk can make it difficult to meet the 10Base-FL 10-10 bit-error rate at -32.5-dBm average received power. The crosstalk from peaking lasting 2-nsec with a magnitude 25% greater than the steady-state led drive current can lower the receive sensitivity by up to 1.5 dB. The peaking should also be adjustable in order to meet Federal Communications Commission and cispr 22 emissions standards. "cispr" stands for Comite International Special des Perturbations Radio Electriques.
Other factors, such as modal and chromatic dispersion, need to be taken into consideration when designing a fiber-optic network, but they are beyond the control of silicon vendors and need to be addressed by fiber led and cable manufacturers.
Benefits of fiber
One benefit of fiber-optic Ethernet is the increased network distance available in a single collision domain. Using a single Class I repeater, the maximum distance for a twisted-pair network is 200 m, while it is 272 m for fiber. With a single Class II repeater, the maximum distance for twisted pair remains 200 m, but for fiber this distance increases to 320 m.
Fiber optics also provides the ability to upgrade to higher-speed networks in the future. Twisted-pair has a limited bandwidth, and it is questionable whether Gigabit Ethernet over twisted-pair will ever be available. Because the main cost of a cable installation is in the labor required to pull the cable, a single installation of fiber will be less expensive over time than successive installations of twisted-pair.
Advances in fiber-optic cabling and connectors have also made fiber an attractive solution for fiber-to-the-desk, but in order for this to succeed in the workplace, revolutionary designs such as the 3M brand VF-45 duplex optical interconnect with its ferrule-less design will be needed. This product, expected to be globally commercialized by the first quarter of 1998, promises to lower the price of fiber-optic connections to a level competitive with twisted-pair. Other advances such as vertical cavity surface-emitting lasers (see Lightwave, April 1997, page 1), increased silicon integration, and low-cost fiber cabling continue to lower the price barrier. Fiber optics not only provides superior performance for 100-Mbit/sec Ethernet and beyond, but also has become a price-competitive solution.u
Michael Laudon is senior product planning and development engineer at Cypress Semiconductor, San Jose, CA.
