The purpose of this chapter is to discuss the resolution of some long-standing
and fundamental issues regarding laser-based multimode fiber transmission
that were addressed by the Gigabit Ethernet PMD sub-task force, in collaboration
with the Telecommunications Industry Association, Fiber Optic 2.2 standards
committee (TIA FO 2.2).
Before the telecommunications industry decided to use single-mode fiber, it had seriously considered standardizing on multimode
fiber. Multimode fiber was attractive because mechanical tolerances of link
components (connectors, splices, laser-coupling optics) are greatly relaxed
compared to the single-mode case. However, in the end, two basic technical
problems made multimode fiber unsuitable for the telecommunications industry:
The telecommunications industry had identified these problems, but it
never developed solutions for them.
Chapter 9, “The Gigabit Ethernet Optical
Link Model,” discussed how the Gigabit Ethernet task force resolved
the first issue. They were able to make a worst-case power penalty allocation for modal noise by using known worst-case models and extensive experimental testing. The
ATM forum also uses this approach.
The basic reason this worst-case approach works for data communication
links is that the total connection loss is relatively small. This made it
straightforward for the Gigabit Ethernet committee to conduct experiments
with mode selective loss much greater than is likely to be encountered in
the field. A large proportion of the committee's work during the first 18
months of the Gigabit Ethernet standardization process was spent ensuring
that modal noise would not be a problem.
However, during that period, little attention was paid to the second problemunpredictable
bandwidth. This was probably because the industry believed that any form of
restricted-mode launch would produce higher bandwidth than the overfilled launch (OFL) bandwidth specified for multimode fiber
(see Chapter 6, section 6.3.5 for descriptions
of OFL and restricted launches). Therefore, the Gigabit Ethernet task force
made the seemingly reasonable assumption that the worst-case bandwidth was
the OFL bandwidth specification for each multimode fiber type of interest.
In fact, many members of the Gigabit PMD sub-task force were also involved
in an investigation led by TIA FO 2.2 regarding the launch dependence on the
modal bandwidth of multimode fiber. It was believed that because all laser-based
transceivers produce a restricted-mode launch, compared to OFL, all laser-based
transceivers would perform much better than predictions based on the OFL bandwidth.
It was also hoped that a particularly good (high bandwidth) set of restricted-mode
launch types could be identified. These high bandwidth restricted launches
could then be used by both transceiver and fiber manufacturers to specify
much better performance.
The TIA FO 2.2 committee was conducting a round robin bandwidth evaluation
during the initial stages of the Gigabit Ethernet specification. Everyone
was confident that the results of the round robin would enable Gigabit Ethernet
to specify longer link lengths.
Unfortunately, when the first results from the modal-bandwidth round robin
began to be reported, it was clear that there was a major problem. Counter
to all expectations, it was discovered that many of the laboratory-defined
restricted launches produced lower bandwidth than an OFL. Worse, the effect
also occurred with real laser-based transceivers, especially when operating
at long wavelengths. Figure
10.1 shows measured bandwidth obtained with an unconditioned 1000BASE-LX
transceiver launch and a selection of fibers.
The fibers have an OFL bandwidth spread representative of the range
expected in commercial systems. The solid line on the graph represents the
expected result for the case where the measured bandwidth equals the OFL bandwidth.
The dotted line represents the specified worst-case (minimum) OFL bandwidth
specification for operation in the long-wavelength region with the fiber type
Clearly, the majority of the bandwidths measured with the transceiver
are well below the OFL bandwidth. Shockingly, for the Gigabit Ethernet committee,
many results don't even achieve the worst-case (minimum) OFL bandwidth specifications
for the fiber type used.
10.1. Measured Bandwidth Using
a 1000BASE-LX Unconditioned Launch Transceiver
Two of the fibers used for these tests have OFL bandwidths less than
the 500 MHz.km specification. These are only included for completeness, and
it is not expected that fibers with such low OFL bandwidth would be encountered
by Gigabit Ethernet.
Although Figure 10.1
shows results for only long-wavelength operation, the effect was also observed
for operation at short wavelengths. However, the effect was more dramatic
with long-wavelength operation.
The initial TIA FO 2.2 results only became available during May 1997.
At that time, it was planned to complete draft D3.1 at the July plenary meeting
of IEEE 802.3 and to ballot draft D3.1 as the final Gigabit Ethernet specification.
The expectation was that Gigabit Ethernet would become an IEEE standard by
The rest of this chapter describes the process by which the Gigabit
Ethernet committee resolved the bandwidth issue. Part of the solution was
to introduce new receiver conformance tests and specifications for laser-based
multimode fiber link design into the Gigabit Ethernet standard. Chapter
11, ·1000BASE-X: Optical Fiber and Copper PMDs,· provides
the specifications and details of these new tests.
It is important to understand the committee process so that you can
then understand the working environment and pressures that the committee was
under. Developing solutions to the bandwidth issue was not conducted in a
relaxed academic environment or to academic time scales. Rather, it was done
in a standards environment, under public scrutiny, to commercial time scales.
It is a great tribute to all concerned that such a difficult technical problem
was understood and resolved in time to make Gigabit Ethernet the success itis.
The poor performance of
laser-based multimode fiber links called the November 1997 completion date
for Gigabit Ethernet into question. Obviously, between May and July 1997,
the Gigabit Ethernet community put considerable pressure on the PMD sub-taskforce
with a view to forcing a solution to the bandwidth issue. The response to
this pressure was the formation of an ad hoc committee chaired by Dave Smith
(Honeywell). It was known as the Effective Modal Bandwidth Investigation
(EMBI). Between May and July1997, the EMBI met weekly (sometimes
biweekly) by phone and also had several face-to-face meetings. The members
of the EMBI performed many theoretical and experimental studies and then pooled
their results so that a reasonable solution could be found by the July plenary
meeting. The final results of the EMBI investigation were as follows:
Less than 5 percent of fibers were observed to have bandwidth less than 160 MHz.km at short wavelengths
with restricted-mode launch.
As many as 30 percent of fibers were observed to have bandwidth
less than 500 MHz.km for operation at long wavelength with restricted-mode
Table 10.1 summarizes some key bandwidth
results reported by the EMBI.
The EMBI also concluded that the Gigabit Ethernet link
model (see Chapter 9), as it existed then, could
be used to predict worst-case operating ranges. However, instead of using
the OFL bandwidth specification for each fiber type, a new value based on
the performance of restricted-mode launch should be used. Based on the measurements
conducted by the EMBI, Gigabit Ethernet chose the worst-case modal bandwidth
(WCMB) values for each link type shown in Table
Multimode fiber is abbreviated to MMF in tables and graphs throughout
Based on these worst-case effective modal bandwidths, the Gigabit Ethernet link model (with jitter due
to duty cycle distortion [DCD] equal to zero) was used to calculate the operating
ranges for D3.1 shown in Table 10.3.
At the July plenary meeting of IEEE 802.3, some members of the committee thought that the standard should not
go to its final sponsor group (IEEE 802) ballot. They felt that Gigabit Ethernet
had not met the objective of supporting the installed base of 62.5 μm MMF
within the building backbone and proposed that more work should be done with
the goal of achieving a 550 m operating distance on 62.5 μm MMF. In the
end, it was decided that the ability to achieve 550 m on 50 μm MMF was
sufficient, and draft D3.1 went to sponsor group (IEEE 802) ballot. The members
of the Gigabit Ethernet PMD, TIA FO 2.2, and the EMBI were greatly relieved
that the panic was over and the standard was done. We all believed we had
done the best job possible and that by reducing the bandwidth used to calculate
operating distances, we had picked a “better safe than sorry”
Between July and the September interim meeting of Gigabit Ethernet,
lightning struck the PMD group again. Results of both laboratory and customer testing indicated that the D3.1 draft
had not sufficiently solved the modal bandwidth problem!
Digital Equipment Corporation (DEC) reported at the September interim meeting
that there was a jitter budget problem that D3.1 had not addressed. They
reported that the additional jitter was due to excessive differential mode
delay (DMD) (see Chapter 6, section 18.104.22.168) due
to the restricted-mode launch of
laser-based transceivers. Furthermore, it was pointed out that the Gigabit
Ethernet link model does not take jitter into account. To illustrate their
point, the contributors from DEC showed measured step responses similar
to that of Figure
Figure 10.2. Step Response for a 500 m Length of Fiber Exhibiting Severe DMD
This step response was obtained using an unconditioned 1000BASE-LX transceiver
and a 500 m length of fiber having severe DMD. Due
to the DMD characteristics of the fiber, the unconditioned laser launch
excites mainly two mode groups with a DMD of 1,800ps/km. This bimodal DMD
causes the step response to have two transition regions separated by a plateau.
Jitter is very clearly a problem for this transceiver-fiber combination,
as can be seen from the eye diagram in Figure
Figure 10.3. Eye Diagram for a 500 m Length of Fiber Exhibiting Severe DMD
During the London meeting, Hewlett-Packard presented a graph of lowest
measured modal bandwidth, or worst-case modal bandwidth (WCMB), as it was
known (see Figure
10.4). The WCMB estimated from the peak-to-peak DMD data clearly indicated
a strong correlation between peak-to-peak DMD and WCMB.
Figure 10.4. Correlation between WCMB and Peak-to-Peak DMD Data