Tunable compensators master chromatic-dispersion impairments

Alan Willner

Tunable compensators provide significant enhancements at 10 Gbit/s and will be essential at 40 Gbit/s. Nonlinearly chirped fiber-Bragg-grating technology may meet the demand for tunable, multichannel chromatic-dispersion and dispersion-slope-mismatch compensation.

Just three years ago, in 1998, a debate was raging as to whether 10-Gbit/s systems were needed, given the simplicity of deploying more 2.5-Gbit/s wavelength-division-multiplexed (WDM) channels. That debate ended in 1999 when Nortel sold 10-Gbit/s equipment for revenues of approximately $6 billion. Although 2.5-Gbit/s equipment still accounts for a sizable portion of the optical communications market, we are squarely in the 10-Gbit/s phase of deployment. The new debate now raging concerns the timing of 40-Gbit/s systems deployment.

When moving from 2.5- to 10-Gbit/s systems, most technical challenges are less than four times as complicated, and the cost of components is usually much less than four times as expensive. One critical exception to this rule is the effect of optical-fiber-induced chromatic dispersion on the data bit stream. When increasing the bit rate by four times, the effect of chromatic dispersion increases by a factor of 16.

WHY IT'S A BIG DEAL
Chromatic dispersion is one of the most basic characteristics of fiber. The velocity of a photon down an optical glass fiber depends on the index of refraction of glass. Because the index of refraction is slightly dependent on the frequency (or the wavelength) of light, photons of different frequencies propagate at different speeds. In a single-frequency laser beam used to generate communications signals, data modulation causes the laser wavelength to spread out in frequency roughly on the same order as the data modulation. For example, approximately 1-GHz information bandwidth is produced from a 1-Gbit/s signal. That means that each photon in a single "1" bit exists at a slightly different frequency and will travel down the fiber at a slightly different speed. At the end of the fiber span, the "1" bit is spread out in time and is difficult for an optical receiver to detect clearly as a "1" bit, thereby severely limiting system performance (see Fig. 1).

Units of chromatic dispersion are expressed as ps/(nm.km), meaning that shorter time pulses, wider frequency spread due to data modulation, and longer fiber lengths all contribute linearly to dispersion. Higher data rates inherently have both shorter pulses and wider frequency spreads—as network speed increases, the impact of chromatic dispersion rises precipitously as the square of the increase in data rate.

It is possible to manufacture fiber that induces zero chromatic dispersion, in which all frequencies travel at the same speed. Such fiber, however, is incompatible with deployment of WDM systems because WDM actually requires a certain amount of dispersion for minimizing the effects of nonlinearities. With zero-dispersion fiber, all WDM channels travel at the same speed down the fiber, producing extremely deleterious effects such as cross-phase modulation and four-wave mixing. Whenever WDM is used, chromatic dispersion MUST exist and therefore must have compensation.

FIXED DISPERSION COMPENSATIONChromatic dispersion presents little problem for 2.5-Gbit/s signals, even over distances exceeding 500 km. In 10-Gbit/s systems, however, even links of roughly 100 km require compensation. The overwhelming amount of deployed fiber has positive dispersion, meaning that longer-wavelength photons travel slower than those of shorter wavelengths. As a result, one simple solution is to periodically place a lumped optical element that produces negative dispersion (for example, in which shorter wavelengths travel slower than longer wavelengths; see Fig. 2).¹ Two common fixed-value (or nontunable) negative-dispersion elements are dispersion-compensating fiber (DCF) and chirped fiber Bragg gratings (FBG), with the great majority of systems using DCF.

Dispersion-compensating fiber is a length of fiber producing negative dispersion that is four to five times as large as conventional single-mode fiber (SMF). Typically, a 15-km spool of DCF is required to compensate for 80 km of SMF, and the insertion loss of the DCF is roughly twice that of regular fiber. DCF is broadband, so many WDM channels can be compensated. Unfortunately, the spectral dependence of SMF and DCF is not exactly balanced—only one wavelength can be compensated exactly, with shorter wavelengths having a residual positive dispersion and longer wavelengths having a residual negative dispersion. This effect is known as the "dispersion-slope mismatch" and it's a separate dispersion issue that must be dealt with (see Fig. 3).

A chirped FBG is a grating written near the fiber core in which the periodicity of the etched grooves becomes shorter along the length of the grating. Light of only a specific frequency is reflected from that part of the grating where periodicity resonates with that frequency; with a proper grating design, faster-propagating photons are reflected later in the grating and incur a longer delay, whereas slower-propagating photons are reflected earlier in the grating. Therefore, a compressed and compensated optical pulse is reflected from the grating and can be removed with an optical circulator.²

THE NEED FOR TUNABILITY
In a perfect world, all fiber links would have a known, unchanging value of chromatic dispersion. Network operators would then deploy fixed dispersion compensators periodically along every fiber link to exactly match the fiber dispersion. However, three vexing issues require that dispersion compensators be tunable.

First, there is the basic business issue of inventory management. Network operators typically do not know the exact length of a deployed fiber link nor its chromatic dispersion value per km. Moreover, fiber plants periodically undergo upgrades and maintenance, leaving new and nonexact lengths of fiber behind. Therefore, operators would need to keep on hand a large number of different compensator models, and even then the compensation would only be approximate.

Second, we are experiencing the dawn of reconfigurable optical networking. In such systems, the network path, and therefore the accumulated fiber dispersion, can change in the following scenarios: SONET rings for which recovery from a fiber break necessitates the rerouting of a signal around the ring protection path; reconfigurable optical add/drop multiplexing for which a given wavelength channel may be dropped today at a node but then be routed tomorrow on to a more distant node; and full wavelength-dependent optical crossconnects of the future. Obviously, reconfigurable networks demand tunable dispersion compensation.

Third, we must consider the sheer difficulty inherent in 40-Gbit/s signals. The tolerable threshold for accumulated dispersion for a 40-Gbit/s data channel is 16 times smaller than at 10 Gbit/s. If the compensation value does not exactly match the fiber to within a few percent of the required dispersion value, then the communications link will not work. Moreover, the accumulated dispersion changes slightly with temperature, which begins to be an issue for 40-Gbit/s systems. Tunability is a must at this bit rate.

Given the requirements of systems integrators, practical dispersion compensators should have the following characteristics:

Limited tunability for point-to-point links and much larger for networking
Low insertion loss
Multiple-wavelength operation for cost effectiveness
Ms-range tunability speeds for reconfigurable networks
Uniform insertion loss upon tuning of dispersion
High reliability, easy maintenance and operation, and fail-safe design
Small footprint to fit on a standard telco linecard.

TUNABLE COMPENSATIONThere have been a few reports of new technologies that can achieve tunable dispersion compensation, including: a free-space virtual phase array,³ an FBG that requires several mechanical stretching elements,⁴ an FBG that uses a differential thermal tuning mechanism,⁵ and an integrated structure.⁶ These methods tend to fulfill some of the checklist items in the previous section, yet none of these fulfill all of the requirements, or are too new to be fully examined.

One new approach that may match all the requirements involves the use of a nonlinearly chirped fiber Bragg grating, in which the etched periodicity of the grating ridges is designed to provide continuous dispersion tunability when the FBG is stretched less than 0.1%. All fiber gratings offer the inherent advantages of fiber compatibility, low loss, low cost, and polarization insensitivity. By tailoring the wavelength-dependent delay (as measured in ps/nm) in the reflected optical signal of a chirped reflective fiber, an FBG can become a negative dispersion compensator. Further, if an FBG has a grating with a periodicity that varies nonlinearly along the length of the fiber, it will produce a time delay that also varies nonlinearly with wavelength—this is the key to tunability.

When a linearly chirped FBG is uniformly stretched by an external mechanical stretcher, the induced time-delay curve is simply shifted uniformly toward longer wavelengths with no change in slope or dispersion compensation. However, when a nonlinearly chirped grating is stretched uniformly by a single mechanical element (such as a piezo-electric tuner), the time-delay curve is also shifted toward longer wavelengths, but the slope of the curve at a specific channel wavelength changes (see Fig. 4).^{7, 8}

TUNING MULTIPLE CHANNELS
It's all very well to change the slope of a curve at a specific channel using a nonlinearly chirped FBG. But can a single FBG handle multiple channels in a WDM system?

It is true that a 10-cm grating typically compensates for only a single WDM channel. But multiple WDM channels can be accommodated by a single chirped FBG in one of two ways: fabricating a much longer (such as several meters) nontunable grating, or using a sampling function when writing the grating, thereby creating many replicas in the wavelength domain transfer function of the FBG (see Fig. 5).⁹

To coax a single 10-cm grating to compensate the dispersion of multiple WDM channels, the key is the combination of nonlinear chirping and sampling. Sampled gratings produce a multiplicity of identical negative-dispersion replicas in the wavelength domain.¹⁰ Each negative has a nonlinear chirp within it. Because each negative dispersion replica imparts dispersion compensation to a different WDM channel, the dispersion compensation for all the WDM channels can be tuned in unison by uniformly stretching the single nonlinearly chirped grating.

Figure 6 shows the results of such a tunable dispersion-compensation module in action, for one of four channels being compensated simultaneously. In this example, the dispersion can be tuned for four 10-Gbit/s WDM channels from 500 to 2100 ps/nm, resulting in a tunable range of <1-dB power penalty for SMF lengths covering 40 to 120 km. Modules that provide compensation for higher WDM channel counts should be available soon.

COMPENSATING FOR SLOPE MISMATCH
Transmission fiber, especially fiber with dispersion compensation built in, suffers from a dispersion slope in which a slightly different dispersion value is produced for each WDM channel. Compensating one channel exactly but leaving the other channels to progressively accumulate increasing amounts of dispersion severely limits the ultimate length of the optical link, as well as the wavelength range that can be used for all the WDM channels. Therefore, using a dispersion compensator that uniquely compensates each WDM channel with its own exact negative dispersion value can be a prime enabler for high-performance systems.

The same multichannel FBG sampling design used for chromatic-dispersion compensation can also provide dispersion-slope compensation. The only modification is to make the wavelength separation of the dispersion replicas smaller or larger than the channel spacing of the WDM channels themselves.¹¹ In this way, each WDM channel will be located at a different portion of a dispersion domain of each replica, experiencing a different amount of dispersion compensation. By tailoring the fabrication of the sampled nonlinearly chirped FBG, a module can be assigned a linearly varying amount of dispersion value to channels 1 through N; these values can then be tuned up or down by stretching the grating (see Fig. 7).

REFERENCES

A. E. Willner, IEEE Spect. Mag. 34(4), 32 (April 1997).
M. J. Cole, et al., Elect.. Lett., 33, 70 (1997).
M. Shirasaki, A.N. Akhter, and C. Lin, IEEE Phot. Tech. Lett. 11, 1443 (1999).
M. M. Ohn et al., Elect. Lett., 2000, (1996).
B. .J. Eggleton et al., IEEE/OSA J. Lightwave Tech. 18, 1418 (2000).
C. K. Madsen et al., IEEE Phot. Tech. Lett., 11, 1623 (1999); see also Conf. on Opt. Com. Tech. Digest, Post-Deadline Papers (2001).
A. E. Willner et al., Invited Paper, IEEE J. Sel. Topics Quant. Elect., Spec. Issue on Passive Fiber Optic Components, 5(5) (1999).
K.-M. Feng et al., IEEE Phot. Tech. Lett. 11, 373 (1999).
M. Ibsen et al., IEEE Phot. Tech. Lett. 10, 842 (1998).
J.-X. Cai et al., IEEE Phot. Tech. Lett., 11, 1455 (1999).
Y. Xie et al., IEEE Phot. Tech. Lett., 12, 1417 (2000).

Alan Willner is professor of electrical engineering at the University of Southern California, and founder and CTO of Phaethon Communications, 5005 Brandin Court, Fremont, CA, 94538. He can be reached at 510-445-3950 x228 or [email protected].