New component technology and savvy design must combine to overcome linear and nonlinear effects efficiently.
DR. KATHERINE HALL, PhotonEx Corp.
Today and in the future, carrier networks must accommodate both static voice circuits and increasing amounts of dynamic traffic, often referred to as packet, Internet Protocol (IP), or data traffic. This traffic mixture puts new requirements on the design of core public networks and stresses their ability to scale to meet capacity demand at backbone distances. Future networks must support high-capacity IP networking across backbone distances, so providing reliable, cost-effective, ultra-fast long-haul optical transmission is the first step toward realizing the promise of optical networking.
However, it is difficult to design a transmission system that can tolerate the various impairments that limit speed and distance for optical-network systems. For example, "ultra-long-haul" solutions have emerged to reduce the need for electronic regenerators and thereby reduce the overall cost of the transmission system. Yet, despite some advantages, these initial solutions compromise optical-channel data rate or system capacity to achieve their goals. In the end, these speed and capacity reductions will limit the usefulness of both this equipment and the resulting infrastructure, since neither will be able to scale sufficiently to handle the continuing challenge of increased demand and unpredictable traffic patterns.
Next-generation optical infrastructures that can provide both high-speed channel rates and high spectral efficiency (high capacity) will be uniquely equipped to enable these new networks. An exploration of the key factors that limit transmission distance and data rates, as well as a look at a number of current methods for suppressing or avoiding these limiting effects, will highlight the promise and the challenge of ultra-fast optical data transmission.
WDM systems have been widely deployed in long-distance networks to boost capacity by packing a single fiber with multiple wavelengths, each carrying a different data stream (see Figure 1). The number of wavelengths deployed in these systems depends intimately on the system design. It's clear that there are economic benefits for ultra-long-haul transmission systems that can maximize the distance traveled before electronic regeneration is needed and for high-capacity systems that can carry as much data as possible on a single fiber.
Unfortunately, high-capacity optical signals transmitted over long distances in singlemode fibers encounter a number of limiting effects. Until recently, fiber-optic transmission-system performance has been limited primarily by transmitter and receiver design and linear effects in the fiber.
Linear fiber effects do not depend on signal intensity, which means that they affect low-intensity signals as much as they do high-intensity signals (where signal intensity in fiber is defined as the signal power divided by the area of the fiber core). The primary linear effects that limit system performance are loss and dispersion. Fiber loss or attenuation can be partially compensated by inline amplifiers that provide gain and boost the signal back to its original level. Dispersion causes signal distortions that increase proportionally with fiber distance but can be compensated using dispersion-management techniques.
Optical signals are attenuated as they propagate along a fiber. The transmitted signal is attenuated due to loss in the fiber as well as loss in fiber splices, connectors, and couplers. While this loss can be overcome using optical amplifiers such as erbium-doped fiber amplifiers (EDFAs) to boost the signal power, EDFAs also add noise to the signal and thus degrade the optical signal-to-noise ratio (OSNR). The number of amplifiers that a signal may traverse before the OSNR is degraded to an unacceptable level depends on the noise figure (NF) of the amplifier, the gain of the amplifier, and the performance of the receiver used to detect the signal.Figure 2a shows the calculated OSNR versus the number of amplifiers traversed for three different amplifier cases. In each case, the gain is the same but the noise figure is different. It is clear that for a given OSNR, greater transmission distances can be achieved with lower-noise amplifiers.
Figure 2b shows the calculated OSNR versus the number of amplifiers traversed for three different amplifier cases, each with the same NF (6 dB) but with different gains. It is important to note that noise accumulates more slowly in lower-loss spans (lower-gain amplifiers). The span loss is determined by the length and type of transmission fiber between two amplifiers and is fixed in terrestrial fiber systems. The amplifier spacing is determined by the route and where equipment huts are situated and is typically 45-100 km, or 11-25 dB. Clearly, longer transmission distances between electronic regenerators would be possible with closer amplifier spacings or shorter fiber spans.
One method for overcoming the loss and increasing transmission distances is to take advantage of a particular nonlinearity called stimulated Raman scattering (SRS), a nonlinear effect that can turn singlemode fiber into a broadband Raman amplifier and reduce the loss of a fiber span. SRS converts a small portion of a pump signal at a certain frequency (or wavelength), typically near 1,480 nm, into energy at a frequency that is downshifted from that pump signal by an amount determined by the vibrational modes of the fiber. For singlemode fiber, the SRS-induced frequency shift is centered approximately 13 THz below, or 100 nm above, the pump-laser frequency or wavelength.Figure 2c shows the improved transmission distances that may be achieved using Raman amplification. Note that Raman amplification has the same effect as lowering the EDFA NF or lowering the span loss of the link. Also note that for a given pump power, the Raman gain is higher in fibers that have smaller cores.
Because Raman amplification requires very high optical powers to be launched into the transmission fiber to achieve reasonable gains, these high power levels may damage splices and connectors in the fiber path and degrade the link performance. Also, high-power Raman pumps are not cost-competitive with EDFAs, so while Raman amplifiers alone may be used to overcome the loss in fiber transmission spans, it is more usual to see a combination of EDFAs and Raman amplifiers used to compensate loss in a link.
An important factor to consider for systems that stress spectral efficiency is gain tilt or gain flatness in the amplifiers. The gain of an amplifier is a specified parameter, but typically the gain varies as a function of wavelength. For EDFAs, the conventional, or C-band, extends from approximately 1,530 nm to 1,562 nm, and commercial amplifiers maintain the gain of the amplifier within 1 dB of the specified value over this wavelength range.
When the amplifier gain is not perfectly flat across the gain bandwidth, different wavelength channels experience different gain. This effect is exacerbated when amplifiers are cascaded as they are in long-haul transmission systems. These systems are typically designed assuming an average channel power constant across all wavelength channels. Consequently, if the signal intensity varies across the wavelengths, then some channels will suffer from uncompensated nonlinear effects because their intensities are too high, and others will suffer from poor OSNR because their signal intensities are too low. If these effects are not compensated or reduced, the system capacity and/or the transmission distance will be reduced.
Long-distance transmission of wavelengths across the entire C-band can be ensured through proper amplifier design and a thorough understanding of the link characteristics. For example, tradeoffs in EDFA performance can be balanced by optimizing the design to reduce NF, tolerate loss in the mid-stage, and restrict gain variations to a certain level across the entire C-band. For EDFAs, NF and mid-stage loss tolerance can be improved by appropriate pump-laser selection. Gain flatness is achieved through the use of specially designed filters and careful loss mitigation within the amplifier.
Raman amplifiers also can be gain-flattened using specially designed filters or by using more Raman pump components at different wavelengths. Remaining imperfections in gain flatness or tilt may be compensated by pre-distorting the launched power levels of the different wavelength channels.
In addition to loss, all singlemode fibers have group velocity dispersion (GVD) or chromatic dispersion, meaning that each wavelength of light travels at a different group velocity or speed in the fiber. Therefore, two signals of different wavelengths launched into the fiber at exactly the same instant will arrive at the other end at different times. The arrival time difference is determined by the magnitude of the GVD, the length of the fiber, and the wavelength separation between the two signals. The sign of the dispersion, positive or negative, indicates whether longer wavelengths travel faster than shorter wavelengths or vice versa.
All optical signals modulated with data have a finite spectral width, which means they comprise more than one wavelength. Therefore, as optical signals travel down the fiber and the different wavelength components travel at different speeds, the signal spreads out in time and becomes distorted. When the signal has spread out in time so much that it begins to overlap the next bit interval, bit errors are induced and the fidelity of the link is compromised. In this instance, the link is said to be "dispersion-limited."
The spectral width of an optical signal is related to the temporal width by the Fourier transform; in transform-limited signals, the spectral width of the signal is proportional to the signal data rate. Therefore, higher-speed optical data signals have broader spectra and are distorted more quickly by GVD.
Luckily, distortions induced by GVD are completely linear and can be compensated or undone by adding a different fiber to the end of the link that has equal magnitude and opposite-sign GVD. The fiber that is added at the end of the link to compensate the dispersion is called dispersion-compensating fiber (DCF). In most cases, the additional link loss that would be associated with adding the DCF at the end of the link is avoided by placing it in the mid-stage of the line EDFA. Careful design of the EDFA allows losses as high as 10 dB to be hidden in the mid-stage of the amplifier with little impact on the amplifier NF.
One complication to dispersion-compensation schemes for high-capacity networks is that the magnitude of the GVD is also wavelength-dependent. This wavelength dependence is often referred to as dispersion slope. Typically, a length of DCF can be chosen to compensate the second-order dispersion or GVD of a given transmission fiber at one wavelength. Unfortunately, because the dispersion slope is different for DCF than it is for transmission fiber, the dispersion cannot be compensated equally over a large range of wavelengths. Depending on the mismatch of the dispersion slope and the length of the transmission fiber, there may still be a significant amount of residual dispersion for various channels in the optical transmission system.
In long-haul DWDM transmission systems, the residual dispersion that results from imperfect slope matching between transmission fibers and DCFs can be compensated on a per-wavelength basis, after the different wavelength signals have been demultiplexed. Exactly how much residual dispersion compensation is necessary depends on the dispersion tolerance of the receiver. An optical transport system that does not compensate for residual dispersion may be limited in the distance and/or the capacity it can support.
The dispersion tolerance of the receiver, or the transmission system in general, may be optimized by choosing the appropriate signal transmission format and pulse shape as well as by proper receiver design. In addition to well-designed transmitters and receivers, new dispersion-compensating elements are being designed and marketed that simultaneously compensate chromatic dispersion and the dispersion slope. These new elements are based on novel fiber designs and fiber Bragg grating technology, and they may eliminate the need for per-wavelength dispersion compensation altogether. In addition, tunable dispersion compensators are being developed that could be used for more exact compensation of the chromatic dispersion in certain links. Such compensators also could be used to track dispersion changes due to changes in fiber temperature and the fiber plant itself.
Many of the linear compensation schemes discussed so far are well known and employed in current systems. However, schemes to compensate nonlinear effects are still being developed, and new techniques are necessary to ensure that future bandwidth demands can be met.
Nonlinear effects in fiber arise mainly due to the intensity dependence of the refractive index, which means that signals with different intensity levels travel at different speeds in the fiber. In fact, the speed of a pulsed light signal in a fiber will be different for different portions of the pulse, varying according to the intensity profile of the pulse. Such signals acquire a time-dependent phase shift because of the nonlinearity, and this phase shift results in spectral broadening.
The amount of spectral broadening induced by these nonlinear effects is proportional to the signal intensity and the nonlinear interaction length, a measure of the length over which nonlinear effects are significant. In fiber transmission systems, the intensity of a signal is continuously attenuated by the finite loss in the fiber, so the nonlinear interaction length is typically much shorter than the fiber span itself.
When the intensity of a given signal stream is high enough, it induces refractive-index nonlinearities that modulate its own phase. This phenomenon is referred to as self-phase modulation (SPM). When a high-intensity signal stream induces nonlinear refractive-index changes that are experienced by a different signal stream, the phenomena is referred to as cross-phase modulation (XPM). The magnitude of XPM effects can vary depending on the relative polarizations of the two optical streams, with the effect being largest when the two streams are polarized in the same direction. Also, the nonlinear interaction length for XPM is reduced in the presence of dispersion, because the interacting signals do not remain aligned in time over great distances.
Note that the impairment that results from XPM and SPM is spectral broadening. The severity of this impairment depends on the GVD in the system and the filter strength of wavelength-sensitive components in the link.
One way to avoid the effects of SPM and XPM is to reduce the optical signal power launched into the fiber. However, signal power reduction lowers the OSNR and reduces the distance that a signal can travel before electronic regeneration is necessary. More elegant techniques include manipulation of the intensity and spectral profile of the launched signal pulses to provide maximum tolerance to nonlinear phase modulation.
The intensity-dependent refractive index is also responsible for another common nonlinear effect in fiber referred to as four-wave-mixing (FWM). In FWM, three signal fields at various wavelengths interact to produce a fourth signal field at a different wavelength. In WDM systems, where the wavelength channels are equally spaced, as is the case for the standardized wavelength grid specified by the International Telecommunications Union (ITU), the newly generated fields coincide with other wavelengths on the ITU grid. Therefore, FWM may cause coherent interference by scattering energy from one wavelength channel into another channel.
The magnitude of this effect is proportional to intensity and inversely proportional to dispersion and wavelength separation between the interacting channels. Therefore, as in the case of nonlinear phase modulation, FWM may be avoided by lowering the signal power. Again, this solution will limit the transmission distance that can be achieved. A better solution is to achieve increased system capacity by increasing the per-channel data rate and increasing the corresponding wavelength spacing. In addition to reducing interchannel interactions, high-data-rate coarse-wavelength systems can achieve higher spectral efficiencies. Solutions that attempt to increase the system capacity by placing lower-rate data channels on 25-GHz and 50-GHz channel spacings are particularly susceptible to interchannel effects.
The nonlinear effects described here are due to the intensity-dependent refractive index. There are also intensity-dependent scattering processes that can affect transmission systems. For example, stimulated Raman scattering, the effect we took advantage of to make a Raman amplifier, may cause light from one wavelength channel to be partially scattered and downshifted in frequency. In a multichannel WDM system, this effect may cause the shorter-wavelength channels in the band to act as pump lasers for the longer-wavelength channels.
Therefore SRS can be viewed as an intensity-dependent gain tilt that may cause impairments similar to non-ideal gain flatness in the amplifiers and may limit the number of wavelengths that can be transmitted over long distances. Here, system solutions such as channel pre-emphasis may be employed to combat these effects. In addition, gain-equalization techniques distributed along the link may be used to reduce the impact of SRS.
While early-generation transport solutions have provided optical systems that overcome the linear effects in fiber, these solutions provide either high capacity over short distances or low capacity over long distances. These tradeoffs continue to limit service providers in their ability to build optical networks that push the limits of speed, distance, and capacity to realize the promise of optical networking.
Systems that provide both high-speed channel rates and high spectral efficiency (high capacity) are required to enable the efficient, flexible optical networks of the future. Straightforward extensions of today's techniques will not enable an optical infrastructure that is positioned to handle the bandwidth requirements of the coming years, so equipment providers will be required to provide optical transmission systems that increase capacity by increasing the single-channel data rates, not by increasing the number of parallel channels a single fiber can carry.
Future transport systems will have high single-channel data rates, relatively coarse wavelength spacings, high signal intensities, and transmitters that generate customized pulses that will experience robust propagation over long distances and be detected by receivers designed to compensate for some transmission impairments and tolerate others. Through careful design and a deep understanding of the effects that limit and improve transmission, next-generation high-capacity long-haul systems can be developed that achieve amazing results even in the presence of linear and nonlinear effects.
Dr. Katherine Hall is chief technology officer and co-founder of PhotonEx Corp. (Bedford, MA), where she leads the architectural design and optical system development of the company's services-aware smart photonics backbone system.