1504

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 11, NOVEMBER 2000

Optimization of the Average-Dispersion Range for Long-Haul Dispersion-Managed Soliton Systems

T. I. Lakoba and G. P. Agrawal, Fellow, IEEE
AbstractWe consider limitations on unfiltered transmission of dispersion-managed solitons, arising from the GordonHaus jitter, adjacent pulse interaction, and signal-to-noise degradation. We maximize the range of allowed values of average dispersion, thereby providing the first step in optimization of dispersion maps for wavelength-division-multiplexed lightwave systems. As specific examples, we consider dispersion maps made of several different types of optical fiber and study their performance for transmission of 1040 Gb/s channels over distances in the range from 3000 km to 10 000 km. Index TermsBroad-band optical fiber communications, dispersion management, optical noise, optical solitons.

I. INTRODUCTION HE GOAL OF this study is to compare the performance of various types of dispersion maps for long-haul soliton data transmission. Our basic idea is to find the maximum allowed range of values for the average dispersion, so that the maximum number of wavelength-division multiplexed (WDM) channels can be transmitted at a given bit rate per channel. Here we take , as seen by into account the fact that the average dispersion, different channels, is different due to the third-order dispersion. Clearly, the scope of the problem we have outlined above is too broad, and too complex, to be satisfactorily treated within one study. Therefore, we had to make a number of simplifying assumptions. First and foremost, we considered only those impairments that occur for single-channel transmission. That is, no limitations due to pulse collisions or gain fluctuations in different wavelength channels, were considered. It is clear that , while accounting only for singlemaximizing the range of channel impairments is the necessary first step in the design of any WDM system. This study aims at providing the guidelines in taking that first step by examining what types of dispersion maps can potentially yield high transmission capacity. Our second important assumption is that the dispersion-managed (DM) system under consideration has no in-line control elements, such as narrow-band filters or synchronous modulators. Both filters and modulators are known to be able to improve the quality of transmission; however, there also exist practical issues which make the use of these elements in a real long-haul system difficult.

Manuscript received September 27, 1999; revised May 31, 2000. This work was supported in part by the National Science Foundation under Grants PHY9415583 and ECS-9903580. The authors are with the Rochester Theory Center of Optical Science and Engineering, Institute of Optics, University of Rochester, Rochester, NY 14627 USA (e-mail: [email protected]; [email protected]). Publisher Item Identifier S 0733-8724(00)09816-9.

The lower limit for the average dispersion is known to be set by degradation of the signal-to-noise ratio (SNR) resulting from accumulation of spontaneous-emission noise from amplifiers [1]. The two principal impairments setting the upper bound are the GordonHaus (GH) timing jitter and the interacfor tion of adjacent pulses. Strictly speaking, the latter two effects are to be considered simultaneously, as the GH jitter affects the pulse center separation, which is a critical parameter for soliton interaction [2], [29], [30]. However, at present, there is no theory that correctly describes the interaction of DM solitons even neglecting the GH jitter (at least in the range of parameters that is of interest for this study). Therefore we have to adopt the following procedure. First, we calculate the upper bound for as given by the GH jitter alone [3], [31]. Then we numerically simulate the full nonlinear Schrdinger (NLS) equation, without determined at the previous the noise source and with the stage, and find how much the pulse separation would decrease due to the interaction alone. Finally, we adjust the value of in a certain way (cf. Section II). , It is well known that in a DM system with a given the GH jitter is suppressed, in comparison with the jitter in a by a uniform-dispersion fiber with the same dispersion so-called energy enhancement factor (EEF) [4]. The latter is known to depend strongly on the location of the amplifier(s) inside the dispersion map [5][8]. Furthermore, the strength of pulse interactions and the EEF follow roughly the same dependence on details of the dispersion map [8], [9]. That is, a larger EEF results in more effective jitter suppression but also in stronger pulse interaction. Thus, the location of the amplifier(s) inside the map must be a critical parameter in any DM optimization. However, since we do not have a formula for determining the effect of pulse interaction and thus have to resort to full numerical simulations, it is very important to reduce the number of free parameters in the problem. Therefore, we vary the amplifier location only within the shorter (compensating) section of fiber and only for one of the map configurations (cf. Section III). It is likely that not (fully) optimizing the location of the amplifier could change our results by a factor of order two or so. However, such accuracy is adequate for the main goal of this study, which is to reveal trends that can lead to design optimization rather than to obtain quantitatively correct results. We do not consider contributions to the timing jitter coming from the acoustic effect and the polarization mode dispersion (PMD) [10]. The reason for not considering the acoustic effect is that it only contributes a slow-time component to the jitter, which can be eliminated by a proper adjustment of the receiver [11], [32]. The jitter coming from the combined effect of the PMD and spontaneous-emission noise of amplifiers [12] can

07338724/00$10.00 2000 IEEE

LAKOBA AND AGRAWAL: OPTIMIZATION OF THE AVERAGE-DISPERSION RANGE

1505

be estimated to be much less than the GH jitter for the types ps/nm/km, of fiber considered here and for which we always find to be the case in this study. Also, we do not include effects of third-order dispersion and stimulated Raman scattering on single channel transmission, since these effects have been shown [13], [14] to be small for the combinations of bit rates and distances considered here. The remainder of this paper is organized as follows. In Section II, we present the theory from which the upper and are found. Section III describes our lower bounds for optimization procedure. Section IV contains the main results of this study: it shows the estimates for the maximum number of WDM channels that various types of dispersion maps can transmit, assuming only the single-channel impairments. Main results are summarized in Section V. II. THEORY A. Unperturbed DM Soliton Our model is based on the standard NLS equation that governs propagation of optical pulses in fibers (1)

for weakening of the nonlinear effects due to the fiber loss, and the nondimensional parameter is given by (5) This parameter measures the size of the nonlinearity compared to that of the local dispersion. In (4), the nondimensional dispersion coefficient is explicitly written as a sum of the constant average part (6) and the periodic part (7) . whose average vanishes: In the regime of strong DM, local dispersion is much greater than both the average dispersion and nonlinearity; hence in (4). The high local dispersion determines the functional form of the DM soliton in terms of chirped HermiteGaussian functions [15], [33], of which the largest is a chirped Gaussian (8)

, is the dispersion coefficient, Here is the operating wavelength (assumed to be 1550 nm), is the speed of light, and is the nonlinearity coefficient. The effect of fiber loss and its periodic compensation are included through , respectively. Changing their form, the parameters and we can study various cases of lumped amplification, as well as is the case of distributed amplification. In a DM system, and a piecewise-constant, periodic function with values in the two sections of the dispersion map. The lengths of these and , respectively, and , two sections are is the period of the map. where It is common to introduce normalized variables and write (1) in a nondimensional form. We introduce new variables as

where (9) is not relevant to this study. The and the form of the phase parameter is proportional to the chirp. The pulse width reaches . In Ref. its minimum, , at the points in the map where [16] it was shown that higher-order HermiteGaussian functions may contribute to the evolution of a perturbed DM soliton by no more than 56% (this does not pertain to the interaction of adjacent DM solitons). Thus in what follows we do not consider those higher order terms. With the Gaussian approximation for the pulse shape, and for a sufficiently small , the balance between the average dispersion and nonlinearity sets the following two conditions for the stationary propagation of a DM soliton (10)

(2) where is a time-scaling parameter chosen such that (3) is a reference power used for normalization The parameter and equals the peak power in an idealized lossless fiber. Its relation to the average pulse power in a fiber with periodically compensated loss is specified after (11). In terms of the normalized variables , , and , we obtain the following nondimensional form of the NLS equation: (4) where the periodic coefficient accounts

where the coefficient

is obtained from

(11) The first condition in (10) determines the pulse initial chirp, and the second condition determines the relation among the pulse amplitude, width, and the average dispersion parameter . These equations generalize those derived in [6], [17], [15], and [33] via including the next-order terms in . The factor 1/Re is proportional to the energy enhancement factor and is specified mentioned in Section I. The form of in the Appendix. The details of the periodic amplification

1506

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 11, NOVEMBER 2000

are included via the function in the integrals . This provides a uniform framework for dealing with both lumped and distributed amplification cases. Furthermore, the soliton amplitude, , can always be normalized to unity by a proper choice of the reference power , and we use this normalization in what follows. In such a case, the average DM soliton power , where is defined by (11). Alternatively, the equals DM soliton energy immediately after an amplifier equals , where is the value of at the amplifiers location. When the nonlinearity and the average dispersion are small ), (9)(11) indicate that the single parameter that de(i.e., termines properties of an unperturbed DM soliton is the normal. Here is the minimum width ized pulse width of the Gaussian pulse, related to the full width at half maximum . For easy comparison with (FWHM) as previous work, we use a related nondimensional parameter , called the map strength and defined as follows:

Gaussian approximation for the probability density of the detected signal [21] (14) (15) and 10 of BER, which we use later in this The values 10 and , respectively. Asstudy, correspond to , (14) yields SNR 17.2 and SNR 9.0, resuming spectively. Using (5), (6), (13), and the main-order terms in (10), one obtains the following inequality for the average dispersion :

(16) In the case of distributed amplification, this equation is modi. The nonlinearity coefficient fied using corresponds to either section of the dispersion map; the specific , so that the product in the numerator choice affects via was found of (16) is not affected. Since this lower bound for , the terms in (10) could to yield rather small values of be safely neglected. , is set by the GH The most restrictive upper bound for timing jitter and pulse interaction. The variance of the GH jitter is [3], [31] (17)

(12) . To support stationary where propagation of a DM soliton at zero average dispersion, one rewhere the quantity Re vanquires a specific value ishes [cf. the second of (10)]. Accordingly, the average disper. For both lossless and pesion should be normal for was found [6] to be riodically amplified cases, the value of approximately equal 4.7. When one takes into account the terms in (10) (which should be done when the pulse power is , the DM soliton sufficiently high), one finds that for can exist for either sign of average dispersion [18], [34], [35]. B. Upper and Lower Bounds for Since the soliton power is approximately proportional to the average dispersion [cf. (5) and (6)], the lower bound is imposed by the requirement that the SNR after. for amplifiers be greater than a threshold necessary to maintain a given bit error rate (BER). Here is the dimensional total propagation distance. From [19], we find that (13) where spontaneous emission factor (we assume ); energy of one photon; total gain within one amplification stage. is identified with the number of independent The parameter degrees of freedom in one polarization of the signal (cf. [20], is included because the re[21]). The factor two in front of ceiver is assumed to be sensitive to both polarizations. Equation (13) is written for the case of lumped amplification. For is replaced distributed amplification, the factor , the average loss coefficient. The relation between the with SNR and the BER is established via the parameter , using the

(18) where normalized distance; integer part of ; value of evaluated at the amplifier. Equation (17) is written for the case of one amplifier per map period; generalization to other cases is straightforward [3], [31]. The same equation can also be derived using the variational method [22] with a Gaussian ansatz. The leading-order term on the right-hand side of (18) equals , i.e., its value for the conventional soliton in a uniform-dispersion fiber. In all cases considered, we found that corresponds to . With the upper bound for , corresponding to a trans-oceanic distance, the size term is estimated to be on the order of 10% of the of the term is even smaller. Thus, when leading term, and the writing down the analytical expression for the upper bound (see (19), shown at the bottom of the next page), we retain only the term in order to make that expression more transparent. However, in the numerical calculations, whose results are presented in Section IV, we use the full expression (18) for the variance of the GH jitter.

LAKOBA AND AGRAWAL: OPTIMIZATION OF THE AVERAGE-DISPERSION RANGE

1507

Further, we assume, as done in most studies, that the probaof the deviation of the pulse from the center bility density . Then the probof the bit slot, is Gaussian with the variance ability of the pulse center to be found outside of the bit slot equals . For . this probability is approximated quite well by twice the right-hand side . Comof (15), where is replaced with bining (17) and (18) with (5) and (10), we arrive at the following upper bound for the average dispersion, imposed by the GH is the bit rate. jitter alone, as shown in (19), where , Since , appearing in the denominator, is proportional to . Our numerical (19) provides an implicit upper bound for code implements an iterative procedure to solve for the value of that upper bound, while taking into account all terms in (10) and (18). To account for pulse interaction, we use the following simple found trick. We numerically solve (4) with the parameter from (6) and (19). For a given propagation distance we find , in the pulse separation, that is due only to the change, to indicate the interaction. We set the condition pulse collision, when the interaction prohibits system operation that otherwise would have been allowed by the GH jitter alone. , we simply use inequality For smaller relative values of . Thus, (19), but with an adjusted bit slot, imposed by both the GH jitter and as the upper bound for pulse interaction, we use an estimate (20) In the absence of a theory that would have simultaneously accounted for both the GH jitter and pulse interactions, we have to resort to such a crude estimate to avoid time-prohibiting numerical simulations. The last restriction that sets an independent upper bound for is the following. It follows from (6) and (10) that depends on two independent parameters, and (or, equivalently, the map strength ). The parameter is related to the character, where is the avistic nonlinear length, erage pulse power, by (21) As we have mentioned earlier, stationary propagation of a DM soliton is supported by a balance between the nonlinearity and the average dispersion, the stronger local dispersion providing only the functional form of the pulse. This means that the characteristic length associated with nonlinearity and average dispersion should be much larger than the local dispersion length, in order to ensure stable pulse propagation. This results in the , or . In our numerical simulations condition , and we observe that pulses remain stable when

sometimes even when , with only a few percent of their energy being shed into radiation. However, already for we never found a stable DM soliton. Thus, in most cases, the condition (22) plays the role of an additional upper bound on tions (21), (10), and (6), this can be rewritten as . Using rela-

(23) which is solved iteratively using (6). Since this upper bound on is not as rigorously set as the one imposed by the GH jitter and pulse interaction, we use it as a guidance only. That is, we always verify whether the DM soliton is stable when (23) is not satisfied. III. OPTIMIZATION PROCEDURE AND FIBER PARAMETERS One map configuration that we consider has . There, we have two adjustable parameters. One is the location of the amplifier inside the compensating section of fiber. The , which we vary between 0.12 and other is the ratio 0.28 with an increment 0.01. We observe that the GH jitter and when the amplifier is located the SNR yield the optimal at either end of the compensating section of fiber, whereas pulse interaction is the least restrictive when the amplifier is close to the middle of that section. However, due to the strong dependence of the map strength (12) on the pulse width, the latter is found to affect the system performance significantly more than the amplifier location does. , where is Another map configuration has amplifiers exactly coman integer. There, each of the first of pensates the loss in the preceding segment of length , and the ( )st amplifier comthe fiber with dispersion pensates the loss in the shorther fiber section. The location of the amplifiers in this case is fixed, and the only adjustable parameter is the pulse width. For the third map configuration, with , we fix the amplifier location to be at the beginning of every th map period. There, as well as for the map with distributed amplification, the pulse width is also the only adjustable parameter. , and the We use (16) to determine the lower bound for more restrictive of expressions (20) and (23) to determine the upper bound. For fixed values of the bit rate, propagation distance, and BER, our main result is the dependence of the opon the map period , where optimizatimal range of tion is done with respect to the adjustable parameter(s). By the , for which the quantity optimal, we mean such a range , proportional to the allowed number of WDM , is the maximum. Indeed, if we assume that channels,

(19)

1508

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 11, NOVEMBER 2000

TABLE I FIBER PARAMETERS

the wavelength separation between channels is a fixed mul. tiple of the single-channel spectral width, then The maximum allowed number of channels is (24) is the average dispersion slope. Two obvious where are to either increase (or ways to increase ) or decrease by carefully designing the compensating fiber. In this study we focus on maximizing the former parameter. Now, dispersion management is most effective when the average dispersion is much less than the local dispersion in the map. Consequently, we expect that to , we need the map to be composed of achieve large fiber sections with high dispersion coefficients. (Note that using high-dispersion fiber is also beneficial for suppression of WDM-induced impairments.) Table I lists relevant parameters of the existing types of fiber, that we consider in this study. The acronyms in the left-most column stand for the standard single-mode (SSMF), dispersion-shifted (DSF), and dispersion-compensating (DCF) fibers. As we said in Section II-B, we do not have an analytic formula for the effect of pulse interaction and thus have to evaluate allowed by the it numerically for the maximum value of , we declare that CH jitter. Whenever we find the interaction prohibits the transmission. Since the interaction strength is proportional to the average dispersion [23], transmis. Howsion would still be possible for some lower value of ever, without an analytic expression for the effect of pulse interaction, we are unable to efficiently determine the corresponding upper bound. IV. RESULTS A. Transpacific Distance Here we consider propagation over 10 000 km and require that the BER at the output be at most 10 . We consider the bit rate of 10 Gb/s per channel and the map configuration with . The GH jitter suppression is most efficient in strong maps, since then the EEF, which mainly determines the suppression factor in comparison with the uniform-dispersion case, is large. On the other hand, pulse interaction rapidly increases as the map strength exceeds a certain value, which in the lossless case was found to be about 1.6 [24]. Moreover, the condition (23) is violated for too large values of the EEF (recall that EEF ). Thus, one needs to use a map with in the range between 12. (Let us note that the theory of DM soliton interaction, found in [8], [23], is applicable for larger values of the map
(a)

(b)

Fig. 1.

DCF, Z 10 000 km, B 10 Gb/s/channel, . (a) Range of D (solid) and N (dashed); (b) Ratios of the optimal pulse width (solid, left axis) and the decrease in pulse separation (dashed, left axis) to the bit slot, and the optimal map strength (right axis).

= L

SSMF

strength ( ) than we consider here.) As follows from (12) and Table I, the only combination of fibers for which this range of is possible, when the soliton width is about 1/5th of the is about 3040 km, is the SSMF DCF. The bit slot and is plotted in Fig. corresponding range of allowed values of 1(a). The reasons limiting this range can be understood from the , the behavior of the optimal pulse width parameter, associated optimal map strength, , and the pulse interaction [Fig. 1(b)]. For all values of considparameter, ered here, the optimal map strength is noticeably larger than 1.6, around which we expect pulse interactions to be the weakest. This rather large appears to be necessary to suppress the GH due to the exponenjitter, which increases with ; cf. (17). tial increase of the amplifier gain, , On the other hand, the increase of the parameter seen in Fig. 1(b), clearly indicates that, to limit pulse interaction, has to be reduced. However, this increase of the pulse width can limit the interaction via the reduction of only up to a certain point, because for too large a width, the pulses begin

LAKOBA AND AGRAWAL: OPTIMIZATION OF THE AVERAGE-DISPERSION RANGE

1509

Fig. 2. Same as in Fig. 1(a), but B = 5 Gb/s/channel. Squares (diamonds): configuration with L = 2L (L = 3 L ).

to overlap too much, causing the interaction to increase. The increase of the interaction parameter is clear from Fig. 1(b). In fact, the interactions render the system unusable soon after exceeds 40 km. The dashed line in Fig. 1(a) shows an estimate for the maximum number of channels, calculated using (24) and 0.8 nm for the interchannel separation. This corresponds to roughly five spectral widths of a single channel. For a rather , e.g., ps/nm/km at 35 small range of ) km, the number of channels is still very large ( due to the very small average dispersion slope of the configuration SSMF DCF. It is, therefore, quite likely that the limits for the total transmission capacity in this case are set not by the GH jitter and pulse interactions, but rather by the WDM impairments and the availability of amplifiers with a flat gain over the nm) and a high output power. bandwidth of 96 nm ( The same remark also pertains to the results shown in Fig. 2 below. We performed similar calculations for the same combinaDCF, with distributed amplification, which we tion SSMF assumed to exactly compensate for the fiber loss. For map periods of 25 and 30 km, the average dispersion range is about 40% larger than in the system with lumped amplification. This allowed by the GH is due to the larger upper bound for jitter (because the amplifier noise is reduced by a factor . However, already for 35 km, that upper bound becomes so high that pulse interaction prohibits the transmission. From this example, we make the following observations as to how the performance of the SSMF DCF configuration could 17 be further improved. 1) If, instead of the SSMF with ps/nm/km, one could use a fiber with a lower value of dispersion, one would effectively decrease , and thus would be able to use pulses with smaller widths. Then pulse interaction would limit the transmission less severely. 2) One could use the idea of dense DM, proposed in [26], where the map period is an in. teger fraction of the amplification distance: , narrower pulses (for the same bit rate) With small enough could be used. This gives one the freedom to operate at values of

the map strength that are high enough to yield significant jitter suppression. At the same time, even though pulse spreading is larger in those stronger maps, pulse interaction can still be not too strong owing to an increased pulse separation relative to the pulse width. We defer detailed examination of this venue of performance improvement until the next section. In the remainder of this section, we examine another way of increasing the total transmission capacity, which is evident from (19) and (23). 3) For a fixed map strength, the upper bound for set by the GH jitter is inversely proportional to the cube ). If the alof the bit rate (since were set by the GH jitter alone, the total lowed range (single-channel bit rate)) transmission capacity ( would increase by eight times if one uses 5-Gb/s channels instead of 10-Gb/s ones. Similarly, the other upper bound given by (23) would yield a four-time increase in that case. The lower bound would also decrease by a factor of two [cf.(16)], but this as much as the increase of the two upper would not affect bounds [cf. Fig. 1(a)]. Thus, the total capacity can, at least in principle, increase by four to eight times, if one decreases the bit rate per channel by a factor of two. To verify this, we modified our map so as to keep the map 10 strength approximately the same as. in the case of Gb/s/channel, which meant we had to use the map period about three to four times larger than that in the latter case. The most , appropriate configuration is then the one with is the length of the first fiber section in the map. We where and , performed calculations for the cases when 0.4 nm. The results for and are and set presented in Fig. 2, from which we see that using 5-Gb channels increases the total capacity by a factor of about five compared with the case of 10-Gb channels. B. Transatlantic Distance For the single-channel bit rate of 20 Gb/s, one cannot use a dispersion map composed of the combination of the SSMF DCF fibers and having , simply because the corresponding map strength is too high ( 7) for all realistic values of the amplifier spacings. To reduce the map strength, one option is to use the combinations DSF SSMF, in which the longer section of fiber has significantly lower dispersion than the SSMF. However, even in that case, it is difficult to find values of adjustable parameters in our optimization scheme . This occurs because which would yield a positive range , set by the GH jitter, is proportional the upper bound for and thus is dramatically decreased for 20 Gb/s comto pared with the case of 10 Gb/s. In order to increase that upper bound to a level where it would again make sense to consider optimization of the system, we need to reduce the propagation distance by about the same factor by which we increased the 6000 km, bit rate [cf. (19)]. In this ection, we consider which is a typical transatlantic distance. In addition, we set the allowed BER to a lower value of 10 , assuming that forward error correction can bring it down to an acceptable level. Expectfor the DSF+SSMF combination edly, we then find that is as large as that for the map considered in Section IV-A [compare Fig. 1(a) with the first line in Table II]. Note, however, that the number of WDM channels for the DSF SSMF combina-

1510

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 11, NOVEMBER 2000

TABLE II AVERAGE DISPERSION RANGE FOR VARIOUS MAPS

tion must be very small (about 2), due mostly to the much larger average dispersion slope of this fiber combination. One might be tempted to consider the same trick as that used 10 Gb/s instead of 20 Gb/s per in Section IV-A, i.e., to use channel, to boost the total transmission capacity of the system. However, this does not work for the DSF+SSMF combination, cannot be increased much without violating simply because . To get around this problem, one the condition needs to use the combination SSMF DCF, where the local dispersion is much higher. In order to keep the map strength not too high, one has to choose the map period to be smaller than the amplification spacing. Thus we are naturally led to considering , the dense DM [26] configuration with 10 Gb/s/channel, should be about 10 km where, for in order to have the map strength not too high. for the combination The maximum allowed ranges of SSMF DCF and for the bit rates 20 and 10 Gb/s, are reported 20 Gb/s, we set 5 in Table II (lines 2 and 3). For accordingly. For 10 Gb/s, we set 3 km and vary 30 and 35 km, 4 for 40 and 45 km, for 5 for 50 and 55 km, so that is always and between 1012 km. For the 20 Gb/s case, the optimum pulse width is set mostly by a compromise between the GH jitter and being close to 20% of the bit pulse interaction, with slot. As in Section IV-A, we note a significant increase of the total transmission capacity when using the single-channel bit rate 10 Gb/s instead of 20 Gb/s. For the 10-Gb/s case, we find that the low-power condition (22) severely decreases the upper that is otherwise allowed by the GH jitter. The bound for is found to be between 0.150.16 optimum value of ; for smaller pulse widths (which imply for all values of ), the DM solitons are not stable. Thus, if smaller smaller were used in this case, the allowed range values of would have been larger. C. Terrestrial Distance If we want to increase the single-channel bit rate to 40 Gb/s, we need to consider shorter propagation distances, as explained in the preceding section. As a challenging terrestrial scale, we 3000 km, and the amplifier spacing is the standard take 80 km. To have values of the map strength in the range between 12, we are required to consider dense DM configuration even for the combination DSF SSMF. For this combination, we can to be anywhere between 1016 km to transmit two take or three 40-Gb channels spaced 1.6 nm apart (cf. Table II). Similarly to the results of the preceding subsections, using 20-Gb instead of 40-Gb channels significantly increases the total trans-

mission capacity. Note that the condition is not violated in this case. Finally, we report on an interesting observation, which we made when trying to increase the transmission capacity by using the combination SSMF DCF for the dispersion map. Transmission of 40-Gb channels in such a map would require extremely short map periods of less than 1.5 km. For this reason, we concentrate on 20-Gb channels. Recall that we have already demonstrated viability of 20-Gb/s transmission DCF maps in the previous section. Surprisingly, in SSMF now, for twice as short a distance, we could not find a range of parameters where transmission would be possible near or set by the GH jitter. The reason at the upper bound for behind this is the following. The amplifier spacing of 80 km, used for terrestrial systems, is about twice of that considered in Section IV-B. The pulse power at the amplifier, that is needed to guarantee a given average power, increases exponentially and leads to a corresponding increase in pulse with interaction, which appears to be the main limiting factor in this case. We expect that the use of distributed amplification would lead to a decrease of the pulse peak power and thus make transmission possible. In fact, we expect that, it would provide than we found in Section IV-B for the a greater range of same map configuration. V. CONCLUSION This study addresses optimization of parameters of long-haul DM soliton systems while considering the impairments set only by the amplifier noise and pulse interaction. From the results presented in Section IV we draw the following four main conclusions. sufficiently First, in order to have the upper bound for large, the smaller of the dispersion coefficients of the two fiber , so as to ensure the consections has to be much larger than . If the latter condition is not satisfied, the dition dispersion map cannot be expected to significantly improve the system performance over that in the uniform-dispersion case. Thus, the SSMF DCF combination is expected to provide a than a combination involving some type of a DSF. lager In addition, WDM impairments are suppressed by high local dispersion. Second, in order to avoid large pulse spreading (which leads to increased pulse interaction) when operating at more than 10 Gb/s in a SSMF, one needs to use dense DM configuration where the map period is chosen to be a fraction of the amplification spacing. This allows one the freedom of using sufficiently

LAKOBA AND AGRAWAL: OPTIMIZATION OF THE AVERAGE-DISPERSION RANGE

1511

narrow pulses. The latter leads to more efficient jitter suppression via the increased EEF, and at the same time, pulse interaction is reduced because of an increased pulse separation relative to their width. Let us note, however, that using dense DM does not always help to reduce pulse interaction. Even though the may be satisfied for very short maps, condition the nonlinear length can still be hundreds of times shorter than the transmission distance. Then pulse interaction, whose effect is accumulated over that many nonlinear lengths, may become strong enough to prohibit the transmission. Thus a compromise , based on the above considerations, should be value for chosen. Third, using distributed amplification can increase the , because the GH jitter in a disallowed range tributedly-amplified system is reduced by a factor of ) compared with that in a system ( . However, this increase of with amplification spacing is less than the maximum possible increase, given by the above exponential factor, because the upper bound for may be set by pulse interaction rather than by the GH jitter alone. On the other hand, as we noted in Section IV-C, maximum pulse power in a distributedly amplified system is less than that in a system with lumped amplification, which may help to reduce pulse interaction. Fourth, using a lower bit rate per channel can significantly increase the total transmission capacity. This occurs because increases dramatically with the maximum allowed range of the decrease of the bit rate. Moreover, WDM impairments are also ameliorated in systems using lower bit rates per channel. Specifically, tolerance to the timing jitter induced by collisions of DM solitons in different channels is inversely proportional to the bit rate [27]. For the nonreturn-to-zero (NRZ) transmission format, the impairments due to four-wave mixing and interchannel Raman cross-talk were shown to also be less severe for a WDM system based on a lower bit rate per channel [28]. Since these types of impairments are similar in character for the soliton and NRZ systems (compare, e.g., [13] and [28]), we believe that the conclusion stated in the beginning of this paragraph will hold true when all types of transmission impairments are taken into account. We note that this conclusion is not specific to DM systems; it certainly was made earlier in regards to both the soliton- and NRZ-based transmissions in uniform-dispersion systems. A new twist which is specific to DM systems is that the value of the bit rate per channel must be carefully (cf. chosen so as to ensure the main condition, Section IV-B). Finally, we note that the results presented in Section IV for the SSMF DCF map and indicating the possibility of having a total transmission capacity of several terabit per second should be interpreted with caution. Consider, for example, Fig. 2. It is not realistic that transmission of thousands of 5-Gb/s channels spaced at 0.4 nm would ever be possible in a single-mode fiber, because of the limitations imposed by the requirements on the amplifier gain uniformity and output power. Instead, Fig. 2 shows that single-channel impairments do not impose significant restrictions on transmission in that particular case. Thus, further optimization of the dispersion maps should focus on minimizing the WDM impairments.

APPENDIX Here we present the explicit form of the terms and in (10). They are calculated by extending the method of [16] through the second order in and taking into account only the zeroth- and second-order HermiteGaussian functions. The same results could also be obtained by extending the equations of the variational method for the DM soliton through the same order. and that is valid for map We only write the form of , , and the disconfigurations with tributedly-amplified case. For the dense DM configuration, , these expressions can be easily generalwith ized.

(A1)

(A2) are defined in (11). We note that for sufwhere functions ficiently high power of a DM soliton (equivalently, suffiently large ) it is more accurate, and probably even easier, to find the initial value for the soliton chirp and the relation between and by directly solving the variational equations, as in [18], corrections [34], [35]. In this study, however, finding the to (10) using (A1) and (A2) was quite adequate. REFERENCES
[1] F. M. Knox, W. Forysiak, and N. J. Doran, 10-Gbt/s soliton communication systems over standard fiber at 1.55 m and the use of dispersion compensation, J. Lightwave Technol., vol. 13, pp. 19551962, 1995. [2] C. R. Menyuk, Non-Gaussian corrections to the GordonHaus distribution resulting from the soliton interactions, Opt. Lett., vol. 20, pp. 285287, 1995. [3] R.-M. Mu, V. S. Grigoryan, C. R. Menyuk, E. A. Golovchenko, and A. N. Pilipetskii, Timing-jitter reduction in a dispersion-managed soliton system, Opt. Lett., vol. 23, pp. 930932, 1998. [4] N. J. Smith, W. Forysiak, and N. J. Doran, Reduced GordonHaus jitter due to enhanced-power solitons in strongly dispersion-managed systems, Electron. Lett., vol. 32, pp. 20852086, 1996.

1512

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 11, NOVEMBER 2000

[5] M. K. Chin and X. Y. Tang, Quasistable soliton transmission in dispersion-managed fiber links with lumped amplifiers, IEEE Photon. Technol. Lett., vol. 9, pp. 538540, 1997. [6] T. I. Lakoba, J. Yang, D. J. Kaup, and B. A. Malomed, Conditions for stationary pulse propagation in the strong dispersion management regime, Opt. Commune., vol. 149, pp. 366375, 1998. [7] R.-M. Mu, V. S. Grigoryan, E. A. Golovchenko, A. N. Pilipetskii, and C. R. Menyuk, Optimizing the arrangement of optical amplifiers in a dispersion-managed soliton system, in Proc. Opt. Fiber Commun. (OFC98), San Jose, CA, Feb. 2227, 1998, paper ThC5. [8] M. Matsumoto, Analysis of interaction between stretched pulses propagating in dispersion-managed fibers, IEEE Photon. Technol. Lett., vol. 10, pp. 373375, 1998. [9] A. M. Niculae, W. Forysiak, and N. J. Doran, Optimal amplifier location in strong dispersion managed soliton systems, in Proc. Conf. Lasers and Electrooptics (CLEO99), Baltimore, MD, May 2328, 1999, paper CWC6. [10] A. F. Evans and J. V. Wright, Constraints on the design of single-channel, high-capacity ( 10 Gb/s) soliton systems, IEEE Photon. Technol. Lett., vol. 7, pp. 117119, 1995. [11] L. F. Mollenauer, Method of nulling nonrandom timing jitter in soliton transmission, Opt. Lett., vol. 21, pp. 384386, 1996. [12] L. F. Mollenauer and J. P. Gordon, Birefringence-mediated timing jitter in soliton transmission, Opt. Lett., vol. 19, pp. 375377, 1994. [13] T. I. Lakoba and D. J. Kaup, Influence of Raman effect on dispersionmanaged solitons and their interchannel collisions, Opt. Lett., vol. 24, pp. 808810, 1999. [14] T. I. Lakoba and G. P. Agrawal, Effects of third-order dispersion on dispersion-managed solitons, J. Opt. Soc. Am. B, vol. 16, Sept. 1999. [15] T. I. Lakoba and D. J. Kaup, Shape of stationary pulse in the strong dispersion management regime, Electron. Lett., vol. 34, pp. 11241126, 1998. , HermiteGaussian expansion for pulse propagation in strongly [16] dispersion-managed fibers, Phys. Rev. E, vol. 58, pp. 67286741, 1998. [17] S. K. Turitsyn, A. B. Aceves, C. K. R. T. Jones, and V. Zharnitsky, Average dynamics of the optical soliton in communication lines with dispersion management: Analytical results, Phys. Rev. E., vol. 58, pp. R48R51, 1998. [18] A. Berntson, N. J. Doran, W. Forysiak, and J. H. B. Nijhof, Power dependence of dispersionmanaged solitons for anomalous, zero, and normal path-average dispersion, Opt. Lett., vol. 23, pp. 900902, 1998. [19] N. A. Olsson, Lightwave systems with optical amplifiers, J. Lightwave Technol., vol. 7, pp. 10711082, 1989. [20] J. W. Goodman, Statistical Optics. New York, NY: Wiley-Interscience, 1985, sec. 6.1, pp. 238256. [21] D. Marcuse, Derivation of analytical expressions for the bit-error probability in lightwave systems with optical amplifiers, J. Lightwave Technol., vol. 8, pp. 18161823, 1990. [22] T. Okamawari, A. Maruta, and Y. Kodama, Reduction of GordonHaus jitter in a dispersion-compensated optical transmission system: Analysis, Opt. Commun., vol. 149, pp. 261266, 1998.

>

[23] S. Kumar, M. Wald, F. Lederer, and A. Hasegawa, Soliton interaction in strongly dispersionmanaged optical fibers, Opt. Lett., vol. 23, pp. 10191021, 1998. [24] T. Yu, E. A. Golovchenko, A. N. Pilipetskii, and C. R. Menyuk, Dispersion-managed soliton interactions in optical fibers, Opt. Lett., vol. 22, pp. 793795, 1997. [25] L. F. Mollenauer, P. V. Mamyshev, and J. P. Gordon, Effect of guiding filters on the behavior of dispersion-managed solitons, Opt. Lett., vol. 24, pp. 220222, 1999. [26] A. H. Liang, H. Toda, and A. Hasegawa, High-speed soliton transmission in dense periodic fibers, Opt. Lett., vol. 24, pp. 799801, 1999. [27] P. V. Mamyshev and L. F. Mollenauer, Soliton collisions in wavelengthdivision-multiplexed dispersion-managed systems, Opt. Lett., vol. 27, pp. 448450, 1999. [28] F. Forghieri, R. W. Tkach, and A. R. Chraplyvy, Fiber nonlinearities and their impact on transmission systems, in Optical Fiber Telecommunications IIIA, I. P. Kaminow and T. L. Koch, Eds. San Diego, CA: Academic, 1997, ch. 8, pp. 248251. [29] T. Georges, Bit-error rate degradation of interacting solitons owing to non-Gaussian statistics, Electron. Lett., vol. 31, pp. 11741175, 1995. [30] A. N. Pinto, G. P. Agrawal, and J. F. da Rocha, Effect of soliton interaction on timing jitter in communication systems, J. Lightwave Technol., vol. 16, pp. 515519, 1998. [31] V. S. Grigoryan, C. R. Menyuk, and R.-M. Mu, Calculation of timing and amplitude jitter in dispersion-managed optical fiber communications using linearization, J. Lightwave Technol., vol. 17, pp. 13471356, 1999. [32] A. N. Pilipetskii and C. R. Menyuk, Suppression of the acoustic effect in soliton information transmission by line coding, Opt. Lett., vol. 22, pp. 2830, 1997. [33] S. K. Turitsyn and V. K. Mezentsev, Dynamics of self-similar dispersion-managed soliton presented in the basis of chirped GaussHermite functions, JETP Lett., vol. 67, pp. 640646, 1998. [34] Y. Kodama, Nonlinear chirped RZ and NRZ pulses in optical transmission lines, Osaka Univ., Preprint. [35] V. S. Grigoryan and C. R. Menyuk, Dispersionmanaged solitons at normal average dispersion, Opt. Lett., vol. 23, pp. 609611, 1998.

T. I. Lakoba, photograph and biography not available at the time of publication.

G. P. Agrawal (F96), photograph and biography not available at the time of publication.