When deciding to deploy high-speed 40 Gbit/s or 100 Gbit/s technologies, operators are faced with an important dilemma in terms of spectral planning: should the new wavelengths be deployed over an existing network, with existing traffic (brownfield deployments), or should they be rolled out over unused or new fibers (greenfield deployments)? While the answer to that question will depend on the context, type of network (metro, long-haul, etc.), capacity requirements, and other factors, operators should keep in mind that each approach has its pros and cons, and particular challenges. This article explores the trade-offs of greenfield and brownfield next generation architectures, the impairments these approaches give rise to, as well as ways to reduce the operational costs that these impairments generate for service providers, based on data from the ITU-T G. 697 standard®. The topic of proper spectral planning and managing capacity will underlay this entire article.
This article begins with a definition of coherent vs. non-coherent, as well as greenfield vs. brownfield. Next, it presents the trade-offs that each approach entails in terms of bandwidth, scalability and impact on quality of service (QoS). Impairments commonly found in these high-speed deployments, such as polarization mode dispersion (PMD), non-linear effects and inter-channel crosstalk, will also be discussed, as will their impact on bit error rates (BERs) and operational costs. Following this, a recommended way to analyze these impairments and reduce their financial impact will be explored in detail. In addition, common mitigation approaches are highlighted (use of guard bands, power adjustment wavelength planning, etc.).
A Few Definitions
The following definitions and concepts will be referenced throughout this article, and in particular, the notion of a coherent detector versus a non-coherent detector. First, a coherent detector, or coherent system, refers to a detector that uses a local oscillator, or a laser, to properly recover the signal. Coherent detectors are commonly used today for phase-modulated schemes, such as dual-polarization quadrature phase-shift keying (DP-QPSK) operating at 40 Gbit/s or 100 Gbit/s. Coherent detectors can also be used to recover amplitude modulated signals, and that capacity will be mostly leveraged in 200 Gbit/s or 400 Gbit/s systems. Figure 1 shows the content of a coherent detector.
Direct detection (also called non-coherent detection), on the other hand, relies solely on a measurement of signal power, which is proportional to the square of the signal amplitude. This is the detection method of choice for legacy systems such as 10 Gbit/s using on-off keying (amplitude modulation). Although non-coherent 100 Gbit/s systems are also available, their market penetration remains very low. The content of a non-coherent detector is a lot simpler than that of a coherent detector, because its main component consists of a simple photodiode. These two technologies, coherent and non-coherent transmission, have several benefits and drawbacks, which will be explored in the next paragraph.
40G/100G Deployment Key Decision: Greenfield or Brownfield?
When deciding to deploy high-speed 40 Gbit/s or 100 Gbit/s technologies, one of the key decisions that operators must make in terms of spectral planning is whether the new wavelengths should be deployed over an existing network and with existing traffic (brownfield deployments), or whether they should they be rolled out over unused or new fibers (greenfield deployments)? Although this decision might seem simple, it will have a significant impact on the maximum capacity of the fiber and QoS. Other significant issues may also arise.
A greenfield deployment will typically include only 100G coherent wavelengths, although it might have some 40G coherent wavelengths too. The most common example of a brownfield system is a fiber featuring 10G channels with 100G channels. Other examples of brownfield systems, though less common, include 10G mixed with coherent 40G, or non-coherent 40G mixed with 100G.
Infonetics published an interesting survey showing that in about two thirds of all cases, service providers (including operators and telecom companies) prefer to roll out brownfield technologies. The reason for the popularity of each approach resides, to a great extent, in their respective merits for spectral planning, as highlighted in Figure 3.
Each characteristic will now be examined more closely. Greenfield technology offers a higher maximum bandwidth, because, in theory, each channel can support 100G service, whereas brownfield technology will have some channels operating at a lower data rate, such as 10G or 40G. In addition, greenfield technology does not require any dispersion compensation modules to cancel out chromatic dispersion effects: this is due to coherent systems that have robust chromatic dispersion compensation capabilities. This is the opposite case with regards to brownfield systems. Greenfield systems also show higher tolerance to PMD issues than brownfield systems because of the PMD compensation capabilities offered by the digital signal processor (DSP), even though several instances of PMD compensation failure in coherent systems have been reported in cases of high PMD or fast-changing PMD, which is common in aerial cables. Finally, greenfield is more easily scalable because guard bands do not need to be planned (this concept will be explained later). Considering the pros of greenfield, it may seem like the holy grail of coherent deployments; however, greenfield comes at a higher price, mainly because it requires laying new fiber and installing all the components that come with it (amplifiers, reconfigurable optical add/drop multiplexers, mux, demux, etc.), whereas brownfield makes use of existing fiber, and with all of these components already in place.
Brownfield deployments do not entail such steep initial costs, since the main capital expenditure involved is just the cost of the transmitter and receiver. This means that brownfield systems are, in general, faster to implement than greenfield ones. Moreover, no service interruptions occur in conjunction with brownfield deployments, as long as a proper commissioning procedure is followed. Inter-channel crosstalk, a topic that will be covered later in this article, also occurs less often in brownfield systems.
However, brownfield deployments do have a few drawbacks that are worth taking into consideration. First, a typical brownfield system requires a guard band (this concept is illustrated in Figure 4), which involves leaving some channels empty to avoid bit errors. This requirement stems from the fact that:
• 10G will easily cohabit with non-coherent 40G
• 10G will NOT easily cohabit with coherent 40G
• 100G coherent will NOT cohabit easily with 10G
The reference to “easily cohabit” means that two channels (e.g., 10G and non-coherent 40G) can be transmitted side by side without creating unnecessary bit errors, whereas a 10G channel right beside a 100G channel will cause bit errors due to cross-phase modulation, which can usually be prevented by using a guard band.
Second, a brownfield deployment will create cross-phase modulation, an impairment that will be explained in the next section. Third, brownfield technology is not so easily scalable, because nonlinear effects, most notably cross-phase modulation, have to be taken into account whenever a new wavelength is added, something that is much less critical for greenfield deployments.
Impact of Common Impairments in Coherent Deployments
Now that the impact of choosing greenfield or brownfield deployment strategies has been explored, the next topic of discussion will be the most common impairments in coherent deployments and their associated costs. Some of these impairments are more common in brownfield deployments, while others are more prevalent in greenfield systems.
One of the best references on this topic comes from the International Telecommunications Union (ITU)®, with its recommendation G.697 v3.02®: Optical Monitoring for DWDM Systems, which lists the most common impairments in 10G systems, as seen above in Figure 5.
In Figure 5, high frequency means “10 events per year,” medium means “one event per year,” and low means “one event per ten years.” The standard only includes the first two columns; the third column has been added by EXFO to help readers identify the right test instrument for each impairment.
The relevance and importance of this data is that this list does not concern a single service provider or single geographical area: it comes from an independent industry standards body, comprised of members that are service providers and systems vendors (NEMs) from all regions of the world. Therefore, this data is quite representative of the telecommunications industry.
The ITU also states the following in the recommendation: “At present, there is not enough experience to prepare a similar table for optical channels with bit rates up to 40 Gbit/s.” Although it would be best to have such a table for 40G or 100G, this data can be extrapolated using well-known technical considerations.
Most of these impairments happen at the same frequency, whether the system is greenfield or brownfield. However, the choice of greenfield vs. brownfield has a notable impact on the prevalence of four of these impairments in particular: polarization mode dispersion, chromatic dispersion, inter-channel crosstalk, and cross-phase modulation. Chromatic dispersion has already been discussed in conjunction with figure 3, so it will not be covered herein.
For each of the three remaining impairments, the cost of a system failure due to that particular impairment will be evaluated. A brownfield system featuring ten 10G channels and four 100G channels with unprotected routes (meaning there is no alternative route that can be used as an immediate backup in the case of system failure) will be taken into consideration. Despite rigorous efforts, it was not possible to obtain industry-wide data showing the service-level agreement (SLA) penalties associated with network downtime, most likely because such information is highly sensitive to network operators for competitive reasons. Instead, cost estimates will be used that are based on the best information available: commercial proposals that EXFO has received from service providers for guaranteed business services at its offices in Canada and the United Kingdom. The SLA penalties guaranteed by this contract for our dedicated bandwidth have been multiplied to obtain the SLA penalty of a 10G channel. For a downtime of between one and six hours on a single 10G wavelength, SLA penalties range from $5,000 to $175,000, for which an average of $50,000 has been estimated. For downtimes exceeding 16 hours on a single 10G wavelength, SLA penalties range from $50,000 to $700,000; for which the estimated average is $250,000.
It should be noted that these calculations are estimates (approximations). As such, readers are invited to apply our method and use their own internal numbers.
Common Impairments in Coherent Deployments
Polarization Mode Dispersion
Polarization mode dispersion (PMD) is an impairment due to the fact that the two principal states of polarization in a fiber do not travel at the same speed as a result of environmental considerations (temperature changes, mechanical stresses on the fiber, etc.), or poor fiber geometry (oval fiber core instead of circular fiber core, etc.). Several references explain PMD in details, so it will not be covered herein.
PMD has been a well-known issue in fiber communications since the advent of 10 Gbit/s signals many years ago. Generally speaking, coherent systems can support higher PMD values than 10G noncoherent channels, due to the PMD compensation offered by the DSP. However, coherent systems sometimes fail to compensate for PMD when the differntial group delay (DGD) is very high, or when DGD changes quickly. In a brownfield deployment, the 10G channels, and especially the 40G non-coherent channels, are as likely to suffer from PMD issues as a purely non-coherent system. In fact, EXFO has documented cases of high BER on 40G noncoherent due to PMD, and in many countries, including the United States, Canada, Germany and China. In conclusion, brownfield deployments are more prone to PMD issues than greenfield deployments.
To estimate the costs of PMD problems, it has been assumed that a PMD problem can be fixed within one to six hours (an optimistic hypothesis), and therefore the SLA penalty of a single 10G wavelength is $50,000. According to the ITU data in figure 5, PMD issues occur once a year on a 10G system, in which case it can be surmised that the same would be true on a brownfield coherent deployment. Given that our sample system has three 10G wavelengths, PMD would generate a total cost of $150,000 per year.
Inter-channel crosstalk is the second impairment, and its prevalence can be affected by the choice of greenfield or brownfield technology. Inter-channel crosstalk refers to two neighboring channels that overlap in the spectral domain. Figure 6 clarifies this concept, showing a system of six 100G channels with 50 GHz spacings (0.4 nm), which is a very common configuration. The curve in black shows these six channels turned on, while the curve in grey displays the same system with one channel out of two turned off. In analyzing one specific channel bandwidth (highlighted in blue in the figure), it is clear that when that channel is turned off, its neighbors to the left and right sides extend into that channel bandwidth, which can be seen even more clearly in figure 7. The neighbor represents noise to the channel of interest, which is why this shaded area is referred to as “crosstalk noise.” Therefore, in this case, the total noise present in the system consists of amplified spontaneous emission (ASE) noise, the traditional source of noise coming from optical amplifiers and crosstalk noise.
Inter-channel crosstalk mainly depends on two factors: signal width and channel spacing:
• The tighter the channel spacing, the worse the inter-channel crosstalk will be.
• The larger the signal width, the worse inter-channel crosstalk will be.
Since 100G channels are modulated faster than 10G channels (their baud rate is higher), 100G channels will be spectrally larger than 10G channels. This is a consequence of the Schrodinger uncertainty principle, a key principle in quantum physics. In general, the spectral width of 40G channels is greater than that of 10G channels; however, it depends on the modulation format. Since greenfield systems predominantly include 100G channels, interchannel crosstalk generally occurs more often in greenfield systems than brownfield systems.
To calculate the cost of a crosstalk issue, it will be assumed that it takes more than 16 hours to fix - a easonable assumption, because the technical expertise required to recognize inter-channel crosstalk is high, few test instruments are available to diagnose it, and fixing inter-channel crosstalk is lengthy. This means the SLA penalty for a single 10G wavelength will be $250,000 for our sample brownfield system. According to the ITU data in figure 5, inter-channel crosstalk happens once a year on a 10G system, and therefore it can be concluded that this problem would be more common in coherent systems due to the larger spectral width of 100G signals. For this reason, two events per year will be factored in. If these failures affect two 100G channels, the cost will be 10 times that of a 10G channel for each 100G wavelength, so the total SLA penalty will be:
• Crosstalk SLA penalty = Two events per year x 2 ch. x 10 x $250,000 = $10M per year.
This figure is an approximation, and readers are encouraged to carry out the same calculations using the right data for their companies.
Cross-phase modulation is a more subtle but very important impairment in coherent brownfield systems. Figure 8 depicts the situation: assume that a fiber with some 10G on-off keying signals (amplitude modulation) is traveling alongside 100G phase-modulated coherent channels. The 10G channel will go on and off, locally heating the fiber (shown in blue in the top part of figure 8), which will result in changes in the fiber index of refraction, because the index of refraction depends on the temperature. This means the index of refraction will change as a function of time and space in the fiber. Since the index of refraction is a measurement of the velocity of light in a medium (glass in this case) versus that in vacuum, it follows that the changes in the index of refraction will modify the phase of the signals as they propagate in the fiber (see the bottom section of figure 8). This phenomenon of the 10G signal interfering on the other signals is called cross-phase modulation. For 10G signals, this effect does not matter much, because the receiver detects the signal amplitude to recover the signal, which is not affected by cross-phase modulation. However, cross-phase modulation is bad for 100G signals, which are phase-modulated. Put another way, cross-phase modulation creates phase noise, which leads to higher BER. In summary, cross-phase modulation primarily affects brownfield coherent deployments.
Cross-phase modulation shows dependence on a number of factors. First, it is worse if the channels travel at the same speed, i.e., if their wavelengths are close (the index of refraction depends on the wavelength). This is why guard bands are used in brownfield deployments: to ensure that the 10G channels and the coherent channels do not travel at the same speed, thereby reducing crossphase modulation. Second, cross-phase modulation depends on power, like most other non-linear effects. Alcatel-Lucent, with the collaboration of an Italian university2, has clearly demonstrated this power dependence in an experiment in which they employed 80 channels of 100G, spaced by 50 GHz, and used 15 spans of 100 km of single-mode fiber. They then measured two contributions to the total noise:
• SNRlin: the noise from the optical amplifiers
• SNRNL: the noise from non-linear effects
In figure 9, the x-axis is the channel power, and the y-axis is the percentage of the total noise coming from ASE noise (SNRlin) and from non-linear effects (SNRNL). This figure clearly displays that the higher the signal power, the higher the contribution from non-linear effects to the total noise, up to 70% of total noise coming from non-linear effects at a power of 5 dBm.
In order to calculate the cost generated by cross-phase modulation related failures, the fact that it takes more than 16 hours to fix such failures had to be taken into consideration. Indeed, troubleshooting cross-phase modulation is not easy, and requires highly skilled personnel. Therefore, the SLA penalty for downtimes greater than 16 hours was used, which is $250,000 for a single 10G wavelength. Figure 5 from the ITU® states that cross-phase modulation issues happen once per 10 years for 10G signals. However, as was previously seen, 10G signals are barely affected by cross-phase modulation, whereas 100G signals are affected much more. For this reason, one occurrence per year has been factored in for 100G brownfield systems. In our sample brownfield system, it is assumed to affect two 100G wavelengths. Therefore, the total cost is:
• Cross-phase modulation SLA penalty = 1 event per year x 2 ch. x 10 x $250,000 = $5M per year.
Impairment Diagnosis and Analysis
This article focuses on three impairments, for which the choice of greenfield vs. brownfield has the most impact: PMD, interchannel crosstalk and cross-phase modulation. But, how can these impairments be analyzed? Until the recent introduction of approaches based on optical spectrum analyzers (OSAs), PMD analyses could only be carried out on dark fibers (fiber without any live signals). These approaches present the unique benefit that PMD can now be assessed non intrusively on non-coherent channels. EXFO’s WDM Investigator, an OSA option, does just that with a user-friendly user interface.
Inter-channel crosstalk is a difficult impairment to analyze without turning off channels. It usually entails very close analysis of OSA traces by a highly skilled person. The more practical way to analyze this impairment is EXFO’s WDM Investigator, which to our knowledge, is the only solution on the market.
Finally, cross-phase modulation is usually almost impossible to troubleshoot with standard test instruments. The typical troubleshooting approach usually involves changing channel power of the brownfield systems until the BER decreases. As an alternative, WDM Investigator can also be used by looking at the “non-linear depolarization” impairment.
PMD Mitigation Techniques
The difficult aspect of cross-phase modulation is identifying it. Once that is done, there are a fair number of options available to reduce it. First, a guard band can be used. Figure 11 shows a guard band separating 10G channels (on the left) from 100G channel (on the right). The guard-band width depends on many factors. WDM Investigator can be used to find the minimum guard-band width capable of ensuring smooth system operation. Note that 200 GHz to 300 GHz guard bands are fairly common.
Cross-phase modulation can also be prevented by separating noncoherent channels (10G, on the left side in figure 12) from coherent channels (100G, on the right side in figure 12). These first two mitigation approaches are typically used together. In addition, as previously shown, the impact of cross-phase modulation depends on signal power (figure 9). Therefore, decreasing signal power will reduce cross-phase modulation. Finally, the use of a fiber that has non-zero chromatic dispersion can weaken cross-phase modulation, because this non-linear effect is reduced when the channels do not travel at the same speed.
Inter-Channel Crosstalk Mitigation Techniques
The options for diminishing inter-channel crosstalk all involve working on the spectral planning:
• Decentralize channel wavelength within the channel bandwidth
This means offsetting the channel wavelength from the ITU grid to move it away from its neighbor causing inter-channel crosstalk. • Ensure that neighboring channels do not have the same polarization
Channels that do not have the same polarization do not suffer from inter-channel crosstalk. • Use guard bands or increase the channel spacing
This approach should be used as a last resort, because it reduces the fiber’s maximum bandwidth.
As discussed earlier, one of the key decisions that operators must make when rolling out coherent technologies is whether to use an existing network with existing traffic (brownfield deployments), or to use unused or new fibers (greenfield deployments). Although brownfield is preferred in about two thirds of coherent deployments, this does not mean that the technology only brings benefits. For instance, brownfield deployments are less easily scalable than greenfield technologies. On the other hand, greenfield deployments offer higher maximum bandwidth; however, this benefit comes at a higher price (see figure 3 for the complete story). Also to be taken into consideration are four impairments that occur more or less frequently depending on the choice of greenfield or brownfield technology. While PMD, CD and cross-phase modulation occur more frequently in brownfield rollouts, inter-channel crosstalk is more common in greenfield deployments. Based on a number of assumptions and calculations, these impairments can generate costs in the hundreds, if not millions, of dollars every year. There are few non-intrusive tools on the market capable of identifying these impairments: in fact, EXFO’s WDM Investigator (for PMD, crosstalk and cross-phase modulation), an option on EXFO’s OSA, is the most complete offering currently available. Finally, a number of mitigation techniques have been introduced to reduce these impairments.
1. Infonetics®, 40G/100G/ROADM Strategies, November 2013, www. infonetics.com/research.asp
2. Vacondio et al., Optics Express, 4 Jan 2012, On Non-linear Distortions of Highly Dispersive Optical Coherent Systems
Abbreviations and Acronyms
ASE: Amplified Spontaneous Emission
BER: Bit Error Rate
CD: Chromatic Dispersion
DSP: Digital Signal Processor
ITU: International Telecommunications Union
OSA: Optical Spectrum Analyzer
PMD: Polarization Mode Dispersion
SLA: Service-Level Agreement
WDM: Wavelength-Division Multiplexing