System vendors may say that it's wrong to use modulation formats other than return-to-zero (RZ). Beware—the long-standing debate whether RZ or non-RZ (NRZ) modulation provides better performance for long-haul (LH) terrestrial DWDM transmission systems is not over.
The usual hype of the telecommunications industry has contributed a lot to this heated debate. Some startup companies wanted to realize an explosive growth of transmission capacity—and their own profit margins—by pushing new technologies into the market. Hence, RZ technology was identified as a marketing vehicle to attract the attention of carriers, service providers, and market analysts.
The first deployment of RZ for high-capacity transmission of live traffic took place in transoceanic submarine systems some years ago. However, requirements and constraints of terrestrial LH systems differ considerably from their submarine counterparts. Lengths of currently deployed terrestrial systems are usually much shorter. In addition, capacities per fiber are higher if the state-of-the-art terrestrial systems carry the maximum number of designed channels. As a consequence, it is necessary to take a closer look at which terrestrial systems might potentially benefit from adopting RZ.Figure 1 shows a comparison of RZ and NRZ coded signals for a given data pattern. For NRZ, optical power is switched on when transmitting a "1" bit and it is kept on until a "0" bit is transmitted. In contrast, optical power is switched on and back off for every "1" bit for RZ modulation, regardless of the following bit. In the case of NRZ, the impulse width of an isolated "1" is completely defined by bit rate and corresponds to the bit period. The width of RZ impulses is a design parameter and usually selected shorter than the bit period. It is specified by the duty cycle, defined as the ratio of impulse width and bit period.
The debate over RZ and NRZ cannot be settled easily—there is actually no simple answer. Looking through the vast literature describing successful system experiments does not really help. Usually, what's found are reports describing transmission of X channels over Y spans with an amplifier spacing of Z km using a given modulation format, data rate, and so on.
Very few authors even address the question of what would be possible with a different modulation format. In most cases, the configuration sought after probably differs in some important system aspect, such as span length or fiber characteristics. We tackled this problem by concentrating on guidelines instead of selected system configurations. We first identified relations between important system parameters: fiber type, data rate, and channel spacing as well as the limiting effects of self-phase modulation (SPM), cross-phase modulation (XPM), four-wave mixing (FWM), and pattern-dependent crosstalk induced by stimulated Raman scattering (SRS-XT). Then we sought which modulation format provided better performance when system reach is limited by one of these effects.
Our results are based on a vast number of numerical simulations, experimental verifications, and analytical derivations. There is definitely room for more investigation, for example, in the direction of more exotic dispersion compensation schemes that may change the picture for certain subsets of configurations. Nevertheless, we're convinced that these guidelines are general and can be applied to systems that are found in current backbone networks or that are good candidates for next-generation optical networks.
Figure 2 illustrates some major results. The interpretation requires some background information on the underlying system-design rules. The major goal was to find operating parameters that achieve a maximum system length on a given type of fiber span depending on channel spacing and channel data rate.
Figure 2a shows a plot for standard-singlemode-fiber (SSMF) spans with a span loss of 25 dB. The power launched into the fiber spans must be chosen as high as possible to reduce signal degradation by optical-amplifier noise. Upper limits are set by nonlinear effects. Which nonlinear effect actually limits particular launch power levels depends on channel spacing, data rate, the dispersion coefficient (naturally, a function of the fiber type), and the dispersion compensation scheme. Dispersion compensation was optimized individually for each configuration.
We found that lower-bit-rate systems such as <40 Gbits/sec are generally limited by multichannel nonlinear effects such as FWM, XPM, and SRS-XT. XPM and FWM are strong if the group or phase velocity difference between two channels is small. This situation occurs for very narrow channel spacing or in wavelength regions with small second-order chromatic dispersion coefficients (local dispersion).
The strength of SRS-XT, however, depends mainly on total launch power, the number of channels, and the total width of the wavelength region allocated by channels. Therefore, SRS-XT is the limiting effect if XPM and FWM are sufficiently suppressed by wider channel spacing.
On the other hand, SPM is a single-channel effect. Its impact on system performance does not depend on the number of channels or channel spacing. The impact of SPM increases with increasing data rate in a "step-like" fashion. The position of the step-edge above 10 Gbits/sec depends on the dispersion coefficient. For SSMF, there is a strong increase in the impact of SPM between 10 and 40 Gbits/sec, whereas on nonzero dispersion-shifted fiber (NZDSF), it takes place slightly above 40 Gbits/sec.
It is quite illustrative to evaluate these relations for system configurations of practical interest. As an example, we discuss a state-of-the-art system offered by several system vendors as a product. It features 160-channel transmission with a channel data rate of 10 Gbits/sec and channel spacing of 50 GHz. The channels are located in the C- and L-band wavelength regions and deliver a total capacity of 1.6 Tbits/sec.
According to Figure 2a, the system reach or maximum number of spans is limited by SRS-XT if SSMF is used as transmission fiber. SPM stays relatively weak due to the low channel data rate of 10 Gbits/sec if appropriate dispersion compensation is applied. The strong local dispersion of SSMF in the C- and L-band helps suppress XPM and FWM. In Figure 2b, the same system is limited by FWM when using NZDSF. This fiber provides weaker local dispersion, resulting in less phase velocity mismatch between the channels and hence stronger mixing.
The next step in our analysis was to compare the performance of RZ and NRZ modulation formats for systems limited by a particular nonlinear effect. The result was surprisingly clear and simple. Using RZ, considerable performance improvements could only be observed if the system was limited by SPM. In any system limited by multichannel effects, performance with RZ was comparable to or even slightly worse than the performance with NRZ.
This finding has a huge impact on the attractiveness of using RZ in combination with terrestrial 10-Gbit/sec LH transmission systems. Our analysis clearly demonstrates that 10-Gbit/sec systems of high practical importance are limited by multichannel effects. Therefore, no performance improvement can be expected by using RZ signal modulation for this class of system. The validity of this statement does not depend on transmission fiber type. All transmission fibers of practical interest feature dispersion characteristics in the range of the two examples depicted in Figure 2.
Performance improvements are possible when using RZ modulation in combination with channel data rates of 40 Gbits/sec. As shown in Figure 2a, systems using 40 Gbits/sec in combination with SSMF are limited by SPM. However, the advantage of using RZ is comparatively small when optimizing the dispersion compensation scheme individually for the respective modulation format. For a given configuration, the optimum system using NRZ was only a few spans shorter than the optimum system using RZ. Furthermore, no advantages in using RZ could by found for systems transmitting 40-Gbit/sec channels on NZDSF, because these systems were limited by multichannel effects (see Figure 2a).
In addition to transmission performance, the cost of terminal equipment will also influence the choice of a modulation format. Implementation of RZ increases the complexity of the transmitter. Figure 3 illustrates a typical optical-transmitter design. Several input data signals with lower bit rate are electrically multiplexed into a single signal with the output bit rate.
The multiplexer is driven by a clock generator synchronized to the lower clock frequency of the input signals. The output signal of the multiplexer is boosted by a driver amplifier and coupled into the electro-optical modulator. A continuous wave optical signal is generated by a laser diode. In an NRZ transmitter, the laser signal is directly coupled into the data modulator.
Generation of RZ signals requires either a pulsed laser source or an additional pulse modulator. Due to stability issues and commercial availability of components, most implementations will probably select the second option with the additional modulator. The drive signal must be derived from the clock signal with proper phase adjustment and boosted by an additional driver amplifier.
The additional modulator impacts the cost of the transmitter considerably. The cost of an NRZ transmitter is by far dominated by the contribution of the data modulator and the laser diode. Usually, the modulator is even more expensive than the laser. Pulse modulators are less expensive than data modulators, because they are narrowband devices. Further price reduction may be achieved by an integrated solution. Nevertheless, the cost of an RZ transmitter can exceed the cost of an NRZ transmitter by nearly 50%. As this increased price occurs per channel, the result is a considerable increase in the transmit terminal's cost. This increased cost terminal might be compensated by reduced amplifier cost, but only for very-long-haul systems. For terrestrial systems with a typical length of several hundred kilometers, it will be difficult to realize an RZ system that is attractive from a cost point of view.
The performance improvement achievable by adopting RZ modulation for systems limited by SPM depends on the duty cycle. Considerable improvements over NRZ can only be expected for relatively short duty cycles. A value in the range of 30% would seem to be a good choice. However, reducing the duty cycle increases the bandwidth of the data signal. As a consequence, using RZ comes with a tradeoff between improvement in system reach and reduction in spectral efficiency. Reducing the duty cycle increases the bandwidth required per channel, but it does not increase the number of bits transmitted per second.
Spectral efficiency of RZ signals is an important issue for systems using data rates of 40 Gbits/sec. A channel spacing of 100 GHz does not pose any problems for a 40-Gbit/sec NRZ system. The same spacing is already challenging for 40-Gbit/sec systems using RZ modulation due to the wider channel bandwidth. It is possible to transmit 40-Gbit/sec RZ signals on a 100-GHz channel grid, but the crosstalk between channels degrades system performance. This crosstalk can diminish the advantages of using RZ compared to NRZ.
It is also important to mention that most applications in LH terrestrial networks do not require maximum reach. The constraints for a link are usually dictated by the distance between the terminal equipment located in major cities and potential amplifier sites defining possible span lengths. In many cases, the required system length is much shorter than the maximum achievable system length with the available amplifier spacing.
In such cases, both modulation formats can be used to realize the link with the same number of optical amplifiers. From an economical perspective, these systems should be equipped with NRZ transmitters because they are the lower-cost solution. Even if the link is close to its reach limit and an NRZ system needs some additional optical amplifiers and shorter spans, NRZ may be the more attractive solution.
Our analysis has shown that blindly equipping every terrestrial LH DWDM system with RZ modulation in hopes of better performance is not advantageous. Most systems are limited by multichannel nonlinear effects such as XPM, FWM, and SRS-XT. If a system is limited by these multichannel effects, NRZ provides comparable or even better performance than RZ. Moreover, NRZ provides better bandwidth efficiency and enables realization of lower-cost transmitter designs.
No doubt RZ is beneficial for transoceanic WDM transmission systems, even at 10 Gbits/sec. Similar improvements might be achievable for ultra-long-haul terrestrial systems with lengths of several thousand kilometers. However, a large share of applications found in current and future terrestrial backbone networks requires much shorter system lengths, typically in the range of several hundred kilometers. For these systems, NRZ is definitely the better choice.
RZ is sometimes mentioned in combination with soliton transmission. Solitons are a beautiful theoretical concept balancing nonlinearity and dispersion. Unfortunately, ideal solitons are not very well suited for real-life applications, because they require lossless transmission media. In practical applications, lossless transmission has to be approximated by short amplifier spacing. The concept might be applicable to submarine transmission, because submarine systems typically employ relatively short amplifier spacing of about 40-50 km. Nevertheless, no submarine system installed today uses solitons.
Terrestrial systems, on the other hand, typically use amplifier spacing of 80-100 km or even longer. The resultant span losses are considerably too high for ideal solitons. Dispersion-managed solitons have been proposed as a more suitable format for such systems. The idea is to use soliton-like pulse propagation in the section following the fiber input where power levels are sufficiently high. The remaining part of the fiber with sub-nonlinear power levels provides linear transmission and requires dispersion compensation. Nonlinearity-assisted RZ transmission is just another name for it. Since the concept basically requires optimization of launch power levels and dispersion compensation, it has been covered by our analysis of nonlinear effects. Performance improvements can only be expected in case of system configurations limited by SPM. Again, this concept is not a helpful replacement for NRZ in most terrestrial systems.
Still think NRZ is too old-fashioned? Longing for something more advanced? Then go one step further than just looking at RZ. Other modulation formats combining amplitude and phase modulation, such as differential phase-shift keying (DPSK) or differential quadrature phase-shift keying (DQPSK), have recently attracted a lot of attention. These formats and others are successfully employed in the field of microwave or radio transmission systems. Application in optical systems might help increase bandwidth efficiency or transmission of 40-Gbit/sec signals using 10-Gbit/sec components in combination with polarization multiplexing. Potential performance improvements compared to NRZ are under investigation.
Peter Krummrich is director of ultra-high-capacity transmission and Berthold Lankl is vice president of advanced technology at Siemens AG (Munich, Germany). They can be reached via the company's Website, www.siemens.com.