Erbium-doped fiber amplifiers control gain but also generate transients that can impair performance. Fast transient control using intelligent EDFAs provides protection and enables reliable, long-reach, dynamically reconfigurable network architectures.
In WDM optical networks, the number of input channels to an erbium-doped fiber amplifier (EDFA) can change abruptly—it might increase or decrease as a result of a system reconfiguration, or decrease because of a component failure. Any sudden change in the number of input channels can cause an EDFA to generate optical transients. Because these transients can cause loss of data, it is important to control them. And it will become even more important to control them in next-generation DWDM systems.
How much gain must an EDFA provide in order to generate an output signal with the required power? It depends on the strength of the input signal. The strength of the input signal depends on the length of the section of fiber that precedes the EDFA—a signal traveling through a longer section of fiber will experience greater losses, present a smaller input power to the EDFA, and require more gain.
Fortunately, carriers don't have to tailor the design or the operating conditions of each EDFA individually to match the length of the preceding section of fiber. Instead, they enable each EDFA to control its own gain automatically by operating it in saturation mode; they pump each EDFA strongly enough so that its output power is independent of its input signal power. The EDFA then provides whatever gain is necessary to keep its output power constant.
For example, if the number of input channels to an EDFA suddenly drops from eight to four, then the input power to the EDFA will drop by roughly 50%—and the EDFA will respond by doubling the power to each of the four surviving channels.
HOW TRANSIENTS OCCUR
A careful observation of a chain of first-generation EDFAs over short time scales (tens of microseconds) reveals that their response is neither smooth nor instantaneous. When the input power drops precipitously, the output power also drops within a microsecond. Then it rises sharply, overshooting its original value and peaking after 10 to 30 µs. Next it "rings," or oscillates above and below its design value, for several cycles until it finally equilibrates after approximately 100 ms. These short-term peaks and valleys of power that propagate down the fiber are known as optical transients.
If the carrier adds channels abruptly, the same phenomenon occurs in the opposite direction: as the input power to the EDFA increases, the output power from the EDFA increases, sharply at first. Then the output power drops, undershooting its original value. Finally it rings several times before settling (see Fig. 1).1
FIGURE 2: Transient control makes it possible to increase or decrease the number of channels without impairing signal integrity. Measurement of network/channel performance without transient control when eight channels are dropped and added in a 16-channel WDM system after transmission over a 320-km fiber shows a significant surge in surviving channel power to which a standard EDFA cannot respond in a timely fashion, resulting in a BER of ~0 to 4 (top images). Results of measurement of the same scenario with transient control activated indicates the intelligent EDFA responds within microseconds, with less than 0.5 dB change, and a significant reduction in errors, resulting in a BER of ~10 to 12 (bottom images).
The shape of the curve depends on the EDFA's position in the chain. The speed of the transient, the rate of change of the output power, the height of the peak, and the intensity of the ringing all increase in each successive amplifier down the line.
Transients can impair network performance in a number of ways. They can drive the first-generation EDFA's output power outside the power margins of the network. If the power drops below the minimum, then the system's signal-to-noise ratio becomes unfavorable. If the power rises above the maximum, nonlinear optical effects such as four-wave mixing can corrupt the signals.
Impairment can also occur when the carrier adds channels and the optical power at the receiver drops during the transient period. Low optical power at the receiver can cause eye-closure and severely degrade the bit-error rate. Another possible impairment is that optical signal-to-noise ratio can degrade during the transient period because of changes in the inversion ratio and related changes in the gain spectrum of the EDFA.
An EDFA, like a laser, works by the principle of light amplification by stimulated emission of radiation. It requires a pump to create a population inversion (an excess of excited electrons) in the EDFA. When an incoming signal photon stimulates an excited electron, the electron relaxes back into a lower energy state and emits a second signal photon with the same phase as the incoming photon.
This process of stimulated emission amplifies the signal: one signal photon comes in and two identical signal photons go out. The beauty of the EDFA is that the band structure of the erbium-doped silicon dioxide fiber causes the amplifier to emit a photon with precisely the wavelength at which the networks operate.
The problem is that the excited electrons in the EDFA can also relax back into the lower energy state spontaneously, without interacting with an incoming signal photon. This process of spontaneous emission also causes the EDFA to emit a photon. But spontaneous emission is a random process, and photons generated by spontaneous emission constitute noise in the system. The key point is that abrupt changes in the number of channels can change the ratio of excited electrons to normal electrons, thereby changing the ratio of stimulated emission to spontaneous emission in the EDFA and the signal-to-noise ratio in the network.
A fourth possible impairment involves signal power at the receiver that varies rapidly during the transient period, making it difficult to distinguish ones from zeros in the message bit stream. It would be possible to compensate by optimizing the threshold of the receiver on rapid time scales, but it's not clear that any of today's receivers could adjust fast enough.
CONTROLLING OPTICAL TRANSIENTS
To prevent performance penalties in a large-scale DWDM network, it is necessary to limit channel power excursions to no more than 0.5 dB when channels are added and 2 dB when channels are dropped.
The required speed of response increases with each successive amplifier downstream in the chain. For the 10th amplifier, the system must limit the transients to 0.5 dB within 850 ns when channels are added, and it must limit the transients to 2 dB within 3.75 µs when channels are dropped.
In a large network, and in the worst case when seven channels are dropped, the system would have to limit transients to 2 dB within 200 ns at the 100th amplifier.
Various research groups have demonstrated three methods of limiting the amplitude of the transients: pump control, link control, and laser control. In brief, pump control involves adjusting the pump power to control the gain of the EDFA during the transient period.2 Link control involves protecting surviving channels on a link-by-link basis. It requires the use of a control channel whose power is adjusted to hold constant the power of the signal channels.3 Laser control involves generating a compensating signal in the first amplifier with an all-optical feedback laser.4, 5 Next-generation, intelligent EDFAs can limit transients.
BENEFITS OF TRANSIENT CONTROL
The control of transients offers numerous system benefits. First, transient control enables flexible, dynamically reconfigurable network architectures. Intelligent EDFAs that limit transients make it possible to reconfigure networks dynamically, taking advantage of available fiber bandwidth wherever it may be. By limiting transients, these devices make it possible to increase or decrease the number of channels flowing through a chain of EDFAs without impairing signal integrity (see Fig. 2).
In addition, transient control enables longer-reach systems. Current networks contain expensive regenerators spaced by about 500 km. These regenerators convert the signal from the optical domain to the electrical domain, then retransmit optical signals in order to remove noise and jitter. Carriers are looking to install ultralong-haul networks that contain no regenerators at all, or few regenerators spaced several thousand kilometers apart. These networks will have to continue to contain many EDFAs. Since transients become stronger as they propagate down a chain of EDFAs, they would limit the length of an ultralong-haul network. Transient control can remove this limitation. However, since transients also become faster as they propagate down the chain, transient control mechanisms will have to respond very rapidly in future large-scale networks.
Transient control also prevents lost data on surviving channels. Whenever the number of input channels to an EDFA changes abruptly, for whatever reason, transients can cause loss of data on the surviving channels. Transient control prevents lost data by keeping the power levels within limits. In this way, transient control improves system reliability and performance.
Another benefit of transient control is that it enables system upgrades without service interruptions. Generally speaking, when a carrier introduces a new DWDM system, it doesn't start with a fully loaded system. If the system can carry 40 channels, it may use as few as four channels while bringing it up. If channels are added without transient control, data losses may be caused when the system is first shut down. Transient control enables the carrier to increase its number of channels without causing data losses or interrupting service.
Finally, transient control supports higher-capacity networks by making it possible to reconfigure a network dynamically to take advantage of available fiber channels. It also supports higher-capacity networks because the additional safety provided by transient control enables carriers to operate the system with smaller margins—for example, with higher power loading.
- J. L. Zyskind et al., Proc. ECOC, Post Deadline Papers, 49 (1996).
- A. K. Srivastava et al., Proc. OAA, PDP4 (1996).
- J. L. Zyskind et al., Proc. OFC , San Jose, PD31 (1996).
- M. Zirngibl, Elect. Lett., 278(560) (1991).
- J. L. Jackel et al., Proc. OFC, 84 (1997).
Atul Srivastava is chief scientist at Onetta, 1195 Borregas Ave., Sunnyvale, CA 94089. He can be reached at firstname.lastname@example.org.