Optical networks are undergoing significant changes, fueled by the exponential growth of traffic due to multimedia services and by the increased uncertainty in predicting the sources of this traffic due to the ever-changing models of content providers over the Internet. The change has already begun: simple on-off modulation of signals, which was adequate for bit rates up to 10 Gb/s, has given way to much more sophisticated modulation schemes for 100 Gb/s and beyond. The next bottleneck is the 10-year-old division of the optical spectrum into a fixed “wavelength grid”, which will no longer work for 400 Gb/s and above, heralding the need for a more flexible grid. Once both transceivers and switches become flexible, a whole new elastic optical networking paradigm is born.

There continues to be ever increasing demand for bandwidth, to the point where, just as 10 Gb/s technology has reached maturity, service providers (SPs) are already installing higher bit rates, including 40 Gb/s and now 100 Gb/s per wavelength; the optical layer typically uses a technology called dense wavelength division multiplexing (DWDM) to carry 40–80 such wavelengths on a fiber pair. The 50 GHz International Telecommunication Union (ITU) wavelength grid

(3)

divides the relevant optical spectrum range of 1530–1565 nm (the so-called C-band) into fixed 50 GHz spectrum slots, but it is likely that bit rates greater than 100 Gb/s will not fit into this scheme. Therefore, there is a growing need to replacing this 10-year-old standard with a more flexible grid paradigm to meet the needs of future bandwidth demands.

Even if sufficiently broad spectrum is available, high-data-rate signals become increasingly difficult to transmit over long distances at high spectral efficiency; hence, it becomes beneficial for transceivers to be able to maximize spectral efficiency by adapting to the actual conditions of the network and data rate for each given traffic demand.

In addition to the need to enhance the spectral efficiency, an increasing influence from large content providers, newly constructed data centers, and evolving peering relationships between providers are propelling the uncertainty and heterogeneity of the demands (the disparity between small and large demands) across the network.

To properly address this challenge, one needs flexible and adaptive networks equipped with flexible transceivers and network elements that can adapt to the actual traffic needs. Fortunately, the same technologies that are being considered for achieving very high bit rates at 100 Gb/s and beyond can also provide this added flexibility. The combination of adaptive transceivers, a flexible grid, and intelligent client nodes enables a new “elastic” networking paradigm, allowing SPs to address the increasing needs of the network without frequently overhauling it. 100-Gb/s-based transmission systems have been commercialized in the last two years. Since they are compatible with the 50 GHz ITU grid already deployed, the need for replacing the grid did not arise. Both the Telecom and Datacom industries are now considering a standard transmission data rate beyond 100 Gb/s, and 400 Gb/s is receiving a lot of attention. Unfortunately, the spectral width occupied by

(4)

400 Gb/s at standard modulation formats is too broad to fit in the 50 GHz ITU grid, and forcing it to fit by adopting a higher spectral efficiency modulation format would only allow short transmission distances.

Figure 1.1 a) Spectral widths of different bit rates relative to the ITU grid (for a given modulation

format); b) three demands of bit rates 40 Gb/s, 1 Tb/s, and 400 Gb/s connect node A to B, C, and D, respectively.

(5)

(top) vs. a flexible grid (bottom). The fixed grid does not support bit rates of 400 Gb/s and 1 Tb/s at standard modulation formats, as they overlap with at least one 50 GHz grid boundary.

Errore. L'origine riferimento non è stata trovata.(b) shows, different bit rate

demands interconnecting node A with B, C, and D. The component that switches the channels arriving at B toward C or D is called a Reconfigurable Optical Add-Drop Multiplexer (ROADM). If this device conformed to the ITU grid, it would not be able to switch the broader spectrum channels; as Errore. L'origine

riferimento non è stata trovata.(a) shows, the optical spectrum coinciding with

an ITU grid boundary (marked in black) will not be transmitted through the ROADM. Therefore, in order to build a flexible network, a new kind of ROADM is required that allows flexible spectrum to be switched from the input to the output ports.

This new approach is called Elastic Optical Networking (EON). The term elastic refers to two key properties:

• The optical spectrum can be divided up flexibly

• The transceivers can generate elastic optical paths (EOPs); that is, paths with variable bit rates

These new transceivers are called Bandwidth Variable Transceivers (BVTs).

The development of EON will require innovations in both hardware and software. New components will need to be developed, and they are often more complex than their fixed grid counterparts. Also challenging will be the control and management of the network, including setting up EOPs. To make this development worthwhile, it is important to quantify the benefits.

(6)

Table 1.1 compares fixed grid and EON for a 300 km point-to-point link, and shows that the advantages of an EON over a fixed grid can be very significant, even up to 150 percent for 1 Tb/s. Note that these results are reach sensitive: higher order modulation formats will not be feasible when the required reach is more than a few hundred kilometers. As transmission techniques improve, it is possible that the channel bandwidths will drop further, thus giving even larger benefits for EONs. EONs always deliver efficiency improvements, but they are significantly different depending on the exact traffic scenario. It should be noted that 200 Gb/s dual polarization 16-Quadrature Amplitude Modulation (QAM) can fit into the 50 GHz grid, and this makes it the most efficient fixed grid alternative (assuming inverse multiplexing). This has the effect of increasing efficiency gains to well over 50 percent in many cases.

Table 1.1 Efficiency improvement for flexible spectrum over a point-‐to-‐point link, assuming a 50 GHz grid for fixed DWDM and 10 GHz channel guard band and super-‐channels for EONs.

Thus, from a simple efficiency point of view, there is a significant advantage in building EONs, assuming there is sufficient traffic to merit it. Adding this to the other main drivers highlighted above provides sufficient benefits to undertake

(7)

serious research into EONs.

1.1.1 Some Novel

EON Concepts

To gain a better understanding of EON and the flexibilities it provides, let us review several concepts that are not typically part of fixed DWDM networks.

In fixed DWDM networks, there is typically one-way to implement a given demand: the wavelength bit rate is fixed, the optical reach is fixed, and the spectrum is fixed. So the demand can occupy less than a full wavelength (wasting capacity, like demand C in Errore. L'origine riferimento non è stata trovata.(c)) or more than one (requiring multiple wavelengths, as in Errore. L'origine

riferimento non è stata trovata.(a)).

An EON allows for multiple choices when implementing a demand:

• A given demand can be assigned a modulation format that gives sufficient performance to reach the required distance, while minimizing the spectral bandwidth occupied by the optical path; see demands D and E in Errore. L'origine riferimento non è

stata trovata.(d).

Figure 1.2 a) Fixed grid maintains strict guard bands between optical paths

that implement a 300 Gb/s demand; b) the channels for the demand can be grouped tightly into a superchannel and transported as one entity; c) five demands and their spectrum needs on a 100 GHz fixed grid, assuming quaternary phase shift keying (QPSK) modulation; d) the same demands, with adaptive modulation optimized for the required bit rate and reach; e) the same demands, with additional flexible spectrum.

(8)

• Today the ratio between the amount of forward error correction (FEC) and payload is fixed, but it could be made adaptive in EONs to enable greater distances to be reached when the required bandwidth is lower.

• Whenever a connection passes through an ROADM, the ROADM acts as a filter that reduces the optical bandwidth for the channel. When this happens over and over, the resulting bandwidth may be too narrow, affecting the quality of the signal and limiting the reach. With EONs, the spectrum allocated for longer EOPs can be increased to account for such bandwidth narrowing, increasing the number of ROADMs the connection can go through.

If the demand is too large to be handled by a single optical channel, “super-channels” can be constructed; these contain multiple very closely spaced channels, which traverse the network as a single entity, but can be demultiplexed at the receiver. Errore. L'origine riferimento non è stata trovata.(b) illustrates this (compared to the fixed network case shown in Errore. L'origine riferimento non

è stata trovata.(a)).

This concept should not be confused with “virtual concatenation,” which allows a single demand to span several independent channels, in which case guard bands will still be needed. Thus, it is possible that a single demand will be virtually concatenated over multiple super-channels.

This concept is about “slicing” a single BVT into several “virtual transceivers” that serve separate EOPs (in Errore. L'origine riferimento non è stata trovata.(b), a sliceable 400 Gb/s transceiver is sliced into three EOPs: 100 Gb/s, 100 Gb/s and 200 Gb/s). This flexibility is key to the economic justification of EONs since it is hard to justify “wasting” a 400G BVT on, say, a 150 Gb/s EOP alone (as in

(9)

Figure 1.3 Support for three-‐subwavelength demand using fixed or sliceable transceivers (the ROADMs are

assumed to support EON): a) fixed transceivers; b) sliceable transceivers.

One could use instead a standard rigid 400 Gb/s transceiver and electrical sub-wavelength grooming to fill up the remaining 250 Gb/s, but this introduces another layer and eliminates some of the cost gains of EONs. This flexibility seems feasible in next generation BVT designs, as is the ability to use a different modulation format for each virtual transceiver. While such a feature may challenge the transceiver design, the added cost seems fairly modest. Since different EOPs may carry different payload bit rates, one key question is how to connect the client (e.g., an IP router) to a BVT. From a protocol perspective, extensions of the optical transport network (OTN) and Ethernet standards seem appropriate, similar to the concept of ODUflex in OTNs. This client interface is further challenged if the BVT is sliceable, and the bandwidth for each virtual interface must expand/contract over time. In this case, the interface between the client and the

(10)

BVT becomes a flexible pool of channels that can be grouped in different ways. This also affects the client line card architecture. A simpler option is to integrate the BVT into the client box, and avoid the need for a client interface altogether. Since the spectrum slice in EON varies, it will create fragmentation in the spectrum over time, much like a computer hard disk becomes fragmented.

A related but different problem is the need to grow super-channels over time, causing them to use up free spectrum between them and at some point limiting their ability to expand. These two problems will require periodically reallocating the spectrum in a dynamic fashion. The required mechanism will remove stranded fragments of spectrum between EOPs and redistribute them to allow for further growth. In the context of a super-channel comprising different channels, this hitless spectrum shift can be done by adding an adjacent channel to the EOP in the desired direction of the shift and remapping the data sent over the channel that will be released to the new channel using either a transport mechanism called the Link Capacity Adjustment Scheme (LCAS) or an IP layer mechanism called link bundling. Both mechanisms allow for such changes to occur in a hitless manner. Note that this mechanism requires coordinating the changes in spectrum at the client device, the transmitter, and the optical switching device, but this can be done slowly and hence seems achievable.

In fixed DWDM networks, each wavelength must be managed individually: it is switched individually through ROADMs, consumes the same amount of spectrum whether it is part of a larger demand or not, and appears as a separate entity to management systems. If multiple wavelengths are used to realize a larger demand, their bundling occurs at the client layer without really affecting the network. This is not the case with EONs, where most demands will be mapped 1:1 to a single EOP: if the demand is smaller than a BVT, the BVT will be (ideally) sliced to

(11)

carry it independently, and if the demand is larger than a BVT, multiple channels will be bundled into a single super-channel that implements the demand. Either way, the demand will be switched as one entity in ROADMs and managed as a single entity in management systems. A side effect of this is that the number of managed entities is dramatically reduced, at least for large IP networks, in which a large number of channels are routinely bundled. This has a positive impact on the scale of ROADMs: no longer does an ROADM have to add and drop many dozens of separate connections; instead, it will only need to add/drop a handful of EOPs (typically the nodal degree of a large IP core router is just 2–5, which means that the number of EOPs for a dual core router site will be typically less than 10).

1.1.2 From Building Blocks to Architecture

The EON allows for very close mapping between the client and transport layers. This is especially true for a client layer that consumes significant bandwidth like an IP core network. The following description provides a vision for a fully automated EON, based on an intelligent impairment-aware control plane and clients that have the intelligence to exploit a configurable optical network.

• Every client link is mapped to a single EOP, without an extra layer of multiplexing. This mapping is resource-conserving, assuming sliceable BVTs that can serve multiple client demands.

• When a new link is needed, an EON considers the best implementation of the request, in terms of modulation format, FEC, and spectrum, yielding the lowest-cost solution. This decision is based on adaptive behavior that adjusts itself based on actual link conditions and control plane knowledge of impairments in the network, while still taking into account margins for network aging. By contrast, today’s DWDM systems have little knowledge about the actual link characteristics. They rely on laborious manual

(12)

measurements of link characteristics and provisioning this data into offline planning tools. This approach does not allow for adjustments over time and must assume inaccurate data. It therefore requires significant margins in assessing link quality, resulting in more expensive networks.

• When the link requires more (or less) bandwidth, the EOP is extended (or contracted) hitlessly in the most efficient way to accommodate the demand. • The operator need not have good visibility of the traffic forecast and the link

conditions, but just needs to provide sufficient BVTs per site and connect them in a flexible manner to the clients (if they are not already integrated into the clients). Since these transceivers are sliceable, they can be used effectively even for client links that do not require significant amount of bandwidth.

• When a failure occurs, some client links will carry the load from the failed links. Their bandwidth should be extended automatically. Some of the failed links may be rerouted automatically in the EON, by exploiting possibly different modulation, FEC and spectrum to overcome the longer reach of a protection path. This may imply less available bandwidth at the client layer, but if it is intelligent enough, it could choose to drop low-priority traffic or reroute some of the traffic over other paths (via traffic engineering).

• A similar behavior occurs when transmission conditions change due to fiber plant aging. Each EOP will extend itself to accommodate the new conditions. Again, such behavior will benefit greatly from an intelligent client that is able to move traffic over less busy routes.

• Periodically, there may be a need for the above described hitless spectrum reallocation process.

(13)

need to measure the fiber plant, enter the results into the planning tool, provide an accurate forecast on traffic growth, select different transponders to satisfy different connection reaches, and repeat the process when conditions change. Instead, the network will adjust itself to optimally meet the demands with the given transmission conditions, and readjust as needed if demands or conditions change.

1.1.3 Spectrum Allocation in a Static and Dynamic Network

One of the main challenges in an EON is how to determine the minimum necessary spectral resources and adaptively allocate them to an optical channel with minimum guard band assigned between channels. The minimum necessary spectrum is determined by the conditions along the optical path that result in the delivered signal-to- noise ratio (SNR), and linear and nonlinear distortions, while guaranteeing the required data rate and optical reach. This problem is exacerbated in a dynamic environment, where connection must expand/contract without affecting traffic.

The problem of calculating a route and wavelength for an optical channel in conventional transparent wavelength-routed optical networks is called the routing and wavelength assignment (RWA) problem maintaining the same wavelength along the path. Adaptive spectral allocation in EONs introduces another constraint to wavelength-continuity in terms of the available spectrum on each fiber link along the route. This is a more general routing and spectrum assignment (RSA) problem. In order to explain how to determine the route, spectrum range, and modulation format for an optical channel, it is useful to define some terminology. The frequency range an optical channel is allowed to occupy is defined as a frequency slot (FS). The full width of the FS can be flexibly adjusted. From a practical viewpoint to simplify the network design, it is useful to quantize the

(14)

usable fiber spectrum resources into contiguous frequency units with an appropriate width of, for example, 12.5 GHz. This is a building block for the FS, which is called a frequency slot unit (FSU) in this manuscript. The RSA problem can be divided into two stages. The first stage is to create a route list with a sufficient number of free contiguous FSUs from source to destination (corridor

width), and determine the modulation format for each source-destination pair at a

given bit rate; see Errore. L'origine riferimento non è stata trovata.(a) for more details. The second stage allocates contiguous FSUs to the route as shown in

(15)

Figure 1.4 The spectrum assignment process: a) creating a route list with necessary corridor width and

modulation format; b) allocating contiguous FSUs on the route.

There are several proposed algorithms to select contiguous FSUs from an unused spectral resource pool to minimize spectrum fragmentation and to retain as much as possible contiguous spectrum for future utilization. The simplest one is the “First Fit” algorithm, which searches the available contiguous FSUs are from lower to higher-numbered FSU, thus favoring the lowest available contiguous FSUs.

When considering a dynamic network, it is important to leave sufficient available spectrum around a demand, to allow it to grow either up or down in spectrum. However, leaving such room for growth impacts the promised savings of the super-channel concept. While some solutions have been suggested, they only serve to delay the need to reallocate the spectrum at a future point in time.

1.1.4 Enabling Technologies

Key enabling technologies for the EON are flexible bandwidth transmitters and receivers (BTVs) that can scale up to terahertz bandwidth and beyond, and flexible spectrum selective switches (flex Wavelength Selective Switches, flex WSSs) that can multiplex and switch variable spectral bands.

1.1.4.1 Flexible Spectrum Selective Switches

An ROADM is typically constructed out of several interconnected WSS devices and amplifiers (Errore. L'origine riferimento non è stata trovata.(b)). While mature WSS technologies were specific to the ITU grid, newer WSS technologies have been productized recently and allow switching almost arbitrary spectrum slices (in 3.125~6.250 GHz steps), enabling elastic ROADMs. These devices are based on one of several technologies: optical micro electromechanical systems

(16)

(MEMS), liquid crystals on silicon (LICOS), or silica planar light-wave circuits (PLCs). Due to the low-cost overhead for these devices, we expect them to become common in next-generation ROADMs, and to be deployed irrespective of whether the larger EON vision will materialize, simply because they provide assurance that the network will be robust against future uncertainty without paying a significant premium.

1.1.4.2 Transponder (BVT)

Multicarrier solutions such as coherent wavelength-division multiplexing (CoWDM), coherent optical orthogonal frequency-division multiplexing (CO-OFDM), Nyquist-WDM, as well as dynamic optical arbitrary waveform generation (OAWG) have been proposed as possible transponder implementations for EONs. These solutions rely on the generation of many low-speed subcarriers to form broadband data waveforms using lower-speed modulators so that terabit-per-second data can be generated using lower speed electronics at < 40 Gbaud. Despite the similarity of multicarrier signal generation between CO-OFDM, CoWDM, Nyquist-WDM, and OFDM transmitters, there are fundamental differences with respect to their operation principles and capabilities.

Errore. L'origine riferimento non è stata trovata.(a) shows the spectrum of

multibanded CO-OFDM. Here, each modulator generates many low-bandwidth (

Δf ) subcarriers to form each band. Orthogonality is maintained between bands by ensuring a spacing of

Δf

_G

= m

between the outside subcarriers of adjacent bands

for integer values of m. Figure 5b shows a CoWDM spectrum that consists of several subcarriers.

To maintain orthogonality between subcarriers, the CoWDM subcarrier symbol rate is set equal to the subcarrier frequency spacing. This restricts each modulator

(17)

to generating only an integer number of subcarriers. Additionally, for both OFDM and CoWDM, the chromatic dispersion tolerance scales with the subcarrier data rate instead of the total bit rate. However, the use of lower-bandwidth subcarriers (~100 MHz) in OFDM causes high peak-to-average power ratios, which increases susceptibility to nonlinear impairments. The larger subcarrier bandwidths (~10–40 GHz) typically used in CoWDM have a lower peak-to-average-power ratio with comparable performance to isolated single-carrier systems.

Figure 1.5 Comparison between CO-‐OFDM, CoWDM, Nyquist-‐WDM, and OAWG waveform generation: a) CO-‐

OFDM generates many low-‐speed subcarriers using an inverse fast Fourier transform (IFFT) to ensure orthogonality; b) CoWDM operates by combining many orthogonal subcarriers together to form a seemingly random waveform; c) Nyquist-‐WDM combines many independently generated channels together with a minimum guard band; d) OAWG operates over a continuous gapless spectrum, enabling the generation of any combination of single-‐carrier or multicarrier waveforms. The tick marks indicate locations of modulated comb lines. Df: subcarrier frequency spacing; DfG: frequency spacing between CO-‐OFDM bands; MBW: modulator bandwidth.

Nyquist-WDM attempts to minimize the spectral utilization of each channel and reduce the spectral guard bands required between WDM channels generated from independent lasers. Using aggressive optical pre-filtering with spectral shape approaching that of a Nyquist filter with a square spectrum minimizes the channel bandwidth to a value equal to the channel baud rate.

(18)

equal to or slightly larger than the baud rate. However, significant power penalties arise from setting the channel frequency spacing equal to the baud rate, which requires FEC to achieve error-free performance. Hence, a trade-off exists between spectral efficiency and inter-carrier interference.

In OAWG, the coherent combination of many spectral slices generated in parallel enables the creation of a continuous output spectrum. In this case, arbitrary-bandwidth single- and multicarrier channels can be generated, in which each channel can be in a different modulation format (Errore. L'origine riferimento

non è stata trovata.(c)). The versatility of this signal generation technique enables

customization of generated waveforms over the total operational bandwidth of the transmitter. This enables avoiding large peak-to-average-power ratios and incorporating pre-compensation for impairments such as chromatic dispersion. OAWG also removes the restriction that the modulator bandwidth must be a multiple or sub-multiple of any generated channel or subcarrier bandwidth.

CO-OFDM, CoWDM, Nyquist-WDM, and OAWG are all capable of adopting various modulation formats with elastic spectrum assignments as well as generation of Tb/s super-channels. The transmitters for the three technologies utilize a similar structure based on many modulators in parallel at low speeds (< 40 Gbaud) modulating as many optical subcarriers. In case of CO-OFDM, it also utilizes electronic subcarriers. The receivers for CO-OFDM, CoWDM, Nyquist-WDM, and OAWG also require many coherent receivers in parallel at low speeds (< 40 Gbaud) and as many optical subcarriers. While CO-OFDM, CoWDM, and Nyquist-WDM require standard optical subcarrier frequency spacing and fixed symbol rates, OAWG can mix and match variety of optical subcarrier frequency spacings, and the OAWG waveforms can include data waveforms in both single carrier modulation formats and multicarrier modulation formats such as CoWDM,

(19)

CO-OFDM, and Nyquist-WDM. Hence OAWG can interoperate with legacy DWDM as well as CoWDM, CO-OFDM, Nyquist-WDM systems. Such interoperability can help seamless network upgrades from today’s networks to future EON. On the other hand, Nyquist-WDM may be the first technology to be deployed for superchannel applications due to its closest resemblance to WDM.

1.1.5 From Research to Commercialization

In the past two years, EON research has transitioned from theory to experimentation. For example, a recent field trial of EON based OFDM transmission has demonstrated over 620 km distance with 10G/40G/100G/555G with defragmentation, and an EON network tested with real-time automated adaptive control plane has also been demonstrated. We expect more substantial test beds to be built in the near term.

Early standardization initiatives have also started, such as revisiting the notation of frequency resources assigned to an optical channel. The conventional ITU grid, which is a reference set of frequencies used to denote the allowed nominal central frequencies of optical channels, is anchored to 193.1 THz, and supports various channel spacings of 12.5 GHz, 25 GHz, 50 GHz, and 100 GHz. In conventional optical networks, when a nominal central frequency is assigned to an optical channel, a fixed frequency range between plus and minus half the channel spacing from the central frequency is implicitly assigned to the channel. Now, ITU Study Group 15 is introducing a flexible DWDM grid into its Recommendation G.694.1, where the allowed frequency slots have a nominal central frequency (THz) defined by 193.1 + n × 0.00625 (n is a positive or negative integer including 0), and a slot width defined by 12.5 GHz × m (m is a positive integer). Depending on the application, a subset of the possible slot widths and central frequency defined in

(20)

the flexible DWDM grid can be selected.

This is obviously just a starting point: standards are needed to define a flexible client interface into a BVT, control plane standards are needed to signal the setup of an EOP as well as its modification over time due to increased/decreased demands and changing transmission conditions. Network management interfaces should also be standardized. Further research is still needed on more flexible modulation formats and FEC as well as the dynamic behavior of the network, such as allocation of spectrum under changing transmission conditions. EON will also greatly benefit from research and development on BVT and WSS component technologies, particularly from photonic integration of those technologies leading to advanced WSS such as high-port-count flex WSS and > 1 Tb/s BVT integrated circuits

(21)

1.2 Overview of Adaptive Control Planes

The efficient use of network resources is always one of the main concerns of network operators, who face strong pressure for reduction in the per unit bandwidth cost as well as network power consumption. However, network utilization efficiency in current wavelength-routed optical networks is limited due to their rigid nature. One limitation comes from mismatch of granularities between the client layer, which has a broad range of capacity demands with granularities of several gigabits per second to 100 Gb/s or more, and the physical wavelength layer, which has rigid and large granularity of a wavelength. For example, when the end-to-end client traffic is not sufficient to fill the entire capacity of a wavelength, residual bandwidth of that wavelength is wasted. This is the so-called stranded bandwidth issue. In contrast, if the requested end-to-end capacity is higher than that of the wavelength, several wavelengths are grouped and allocated according to the request. The adjacent wavelengths in such groups have to be separated by a buffer in the spectral domain for wavelength demultiplexing, which leads to low spectral efficiency.

Several articles dealing with these problems, and in particular has recently been proposed a novel spectrum-efficient and scalable optical transport network architecture called the Spectrum-sliced Elastic optical path network (SLICE) to address this granularity mismatch issue. In SLICE the necessary spectral resources on a given route are sliced off from the available pool and adaptively allocated to the end-to-end optical path according to the client data rate and the available spectral resources. The ongoing advances in photonic technologies, such as optical multilevel modulation, optical orthogonal frequency-division multiplexing (OFDM), and the seamlessly bandwidth-variable Wavelength Selective Switch (WSS) make it possible to treat optical fiber as a sharable continuous resource pool. Taking advantage of these concepts, a highly efficient spectrum resource allocation scheme adaptive to the client data rate in SLICE

(22)

has been realized using the concepts of subwavelength, superwavelength, and multiple-rate accommodation including future novel bit multiple-rates. In addition, spectrum resource allocation adaptive to the available spectral resources on the route enables highly survivable bandwidth squeezed restoration.

Apart from the granularity mismatch, another limitation of network efficiency in the current wavelength-routed optical network originates from its rigid worst-case design in terms of transmission performance. Such design ensures that the worst-case optical path in the network, usually the longest path with multiple hops of linear optical repeaters, reconfigurable optical add/drop multiplexers (ROADMs), and Wavelength X-Connects (WXCs), can be transmitted with sufficient quality. For example, once the modulation format to transmit the 40 Gb/s client data streams is determined to be, say, differential quadrature phase shift keying (DQPSK), the format is applied to every 40 Gb/s optical path. In such a case every path occupies the same spectral width, regardless of each path’s distance or the number of transmitted hop nodes. As a result, most optical paths whose path lengths are far less than that of the worst case have large transport margins at the receiving end. If, instead, an adaptation mechanism for various physical impairments were introduced into the optical networks, the utilization efficiency could further be enhanced. This approach should be given increased attention now that the physical limit of standard optical fibers in terms of transmission capacity has become a concern.

1.2.1 Link Adaptation Technologies

Link adaptation is a widely used key technology that increases the spectral efficiency of broadband wireless data networks and digital subscriber lines. The basic idea behind link adaptation technologies is to adjust the transmission parameters, such as modulation and coding levels, to take advantage of prevailing channel conditions. In a link adaptation system the most efficient set of transmission parameters under a certain channel condition is selected based on criteria such as maximizing data rate or

(23)

minimizing transmit power. For example, under good channel conditions an efficient set of parameters optimized for spectral efficiency (i.e., more bit loading per symbol using multilevel modulation schemes and less redundant error correction) is used to increase throughput. In contrast, under poor channel conditions a set of parameters optimized for robustness (i.e., less bit loading and more redundant data adding) is used to ensure connectivity. In the case of modulation format adaptation, 16-ary quadrature-amplitude modulation (QAM), despite having worse receiver sensitivity than quadrature phase shift keying (QPSK), provides better spectral efficiency and is able to transmit double the data rate under good channel conditions.

Data rate dynamic adjustment according to the quality of channels was experimentally demonstrated by using bit and power loading of optical OFDM subcarriers without modifying the channel bandwidth and launch power. Network performance of symbol rate tuning in non-return-to-zero (NRZ) modulation formats in an opaque optical network was also studied. It was shown that reach-dependent link capacity adjustment could benefit from the added available capacity for short-distance demands and from the saved optoelectronic interfaces on low-rate long-distance demands. Such studies are based on the idea of increasing data rate for shorter links with large signal-to-noise ratio (SNR) and nonlinear effect (NLE) margins in opaque optical networks based on the fixed International Telecommunication Union, Telecommunication Standardization Sector (ITU-T) frequency grid.

1.2.2 Spectral Resource Allocation

The traditional link adaptation technologies that maximize channel data rates may also be considered in the following manner: the unused SNR and NLE margins for shorter connections can be used for spectrum resource saving, while ensuring a constant data rate. For the same data rate 16- QAM carries twice the number of bits per symbol of QPSK, therefore requiring half the symbol rate and, consequently, half the spectral

(24)

bandwidth. Similarly, 64-QAM carries three times the number of bits per symbol of QPSK, and requires 1/3 the spectral bandwidth. Thus, reducing the symbol rate and increasing the number of bits per symbol to transmit the same data rate can save spectral bandwidth. Since higher-level bit loading decreases the distance between the two closest constellation points, 16-QAM and 64-QAM suffer from SNR penalty per bit of 4 dB and 8.5 dB, respectively, when compared with QPSK. Such penalties may be acceptable for shorter optical paths with unused SNR and NLE margins.

The implementation of spectral width tunable modulation with constant data rate may employ either a single-carrier or multicarrier approach. Sets of parameters to adjust the spectral bandwidth for various modulation levels of QPSK, 16-QAM, and 64-QAM are summarized in Table 1 together with schematic diagrams of spectral profiles for various modulation levels. In the single-carrier modulation approach, the symbol rate is reduced to obtain a narrower spectral width while increasing the number of bits per symbol to keep the data rate constant. Conversely, in the multicarrier approach the number of subcarriers, with uniform symbol rate and bits per symbol, is changed to adjust the spectral bandwidth.

When the scope of discussion is extended from point-to-point to wavelength-routed transparent optical networks, it is necessary to consider the effect of waveform distortion due to cascaded optical filters in ROADMs and WXCs. Since cascaded optical filters introduce significant narrowing of the pass-band, the filtered optical signal suffers distortion due to unwanted spectral clipping. The effect is stronger for optical paths experiencing larger numbers of node hops. The current design of filters in optical nodes allows for the accumulated filter narrowing effect by assuming wide filter bandwidth to accommodate the worst case (i.e., the paths with the largest numbers of node hops). As a result, most optical paths have large spectral clipping margins and are assigned redundant spectral resources. Spectral resource saving is achieved by adaptively

(25)

choosing filter bandwidth according to the numbers of node hops. The necessary minimum bandwidth of optical filters for an optical path is determined to ensure that the effective pass-band of cascaded filters measured at the end of the optical path maintains acceptable performance.

This functionality can be realized using bandwidth-variable optical filters employing a spatial light modulator, such as liquid crystal on silicon (LCoS), configured with a dispersive element to separate WDM signals.

The concepts of adjusting the number of bits per symbol and the filter bandwidth are motivation to consider a novel adaptation technology of spectrum resource allocation, where only the necessary minimum spectral resource is adaptively allocated to an optical path. The allocated spectrum is adapted to the end-to-end physical conditions while ensuring a constant data rate. In general, available spectral resource in an optical fiber is restricted by the gain bandwidth of the optical amplifier used.

When Erbium-doped fiber amplifiers for C-band or L-band are used, the available spectral width ranges from about 4 to 5 THz. The Adaptive Control Plans objective is to make better use of network spectral resources and maximize the number of optical paths within the available spectral constraint: the most efficient set of transport parameters is chosen to minimize the allocated spectrum resources under a certain optical path condition while keeping the data rate unchanged.

The parameters to be adapted include modulation level and optical filter bandwidth. Imagine that we have a network composed of four nodes (A, B, C and D); flexible spectral resource allocation can be done as follows. Since the shortest path, i.e. the path A, experiences the smallest optical SNR degradation and filter narrowing effect, the most spectrally efficient set of parameters (e.g., 16-QAM and filter width of 37.5 GHz) is selected.

(26)

Table 1.2. Parameters in tunable spectral width modulation format: a) single-‐carrier modulation with variable bit and

symbol rate; b) multicarrier modulation with variable bit and subcarrier.

For path C or D having a larger number of node hops, a more robust set of parameters (e.g., QPSK and 50 GHz) is utilized. Since the filter narrowing effect is most crucial for the longest path, B, the broadest filter bandwidth (e.g., 62.5 GHz) should be assigned to ensure an acceptable pass-band at the egress optical node. In a system with spectral allocation as that stated above, the ITU-T frequency grid can no longer effectively specify flexible and scalable spectral resource allocation.

(27)

Figure 1.6 Spectrum resource allocation in distance-‐adaptive SLICE: signal and filter passband arrangement.

In contrast to networks that have an Adaptive Control Plane, current optical networks require the fixed channel spacing determined to transport the worst-case optical path for all optical paths. The channel spacing is standardized by the ITU- T frequency grid with granularities of 12.5, 25, 50, or 100 GHz. This constraint in terms of channel allocation significantly limits spectral resource utilization efficiency.

1.2.3 Impairments Monitoring Technique Based on Overmodulation

Flexible bandwidth networking provides an effective way of scaling existing networks to more efficiently utilize available spectral resources. Specifically, flexible bandwidth networking enables multiple variable bandwidth traffic demands to be satisfied simultaneously using variable bandwidth channels that can each use one of many possible modulation formats. In this way allocated bandwidth is efficiently matched to traffic demand while also overcoming spectral efficiency limitations caused by spectral gaps between WDM channels on the ITU grid. In some cases, the arbitrary bandwidth channels occupy large bandwidths (> 100 GHz), which increases sensitivity to Physical Layer Impairments (PLIs). For example, the Bit-Error Rate (BER) of a channel is sensitive to impairments such as Optical Signal-to-Noise Ratio (OSNR), Chromatic Dispersion (CD), and Polarization Mode Dispersion (PMD) on all traversed links.

(28)

Typical network control and management planes of today’s optical networks assign network resources based on static data and adhere to established specifications. However, many PLIs are time varying as a result of temperature variations, component degradations, and network maintenance activities. Furthermore, dynamic channel bandwidth allocation and release processes can potentially change network conditions. A network capable of customizing the instantiation of new adaptive connections (flexpaths) to account for expected PLIs, and adapt as necessary to time varying PLIs, improves efficiency while optimizing individual connections for changes in network conditions. Quality of Transmission (QoT) monitoring and impairment aware routing and wavelength assignment (IA-RWA) algorithms in traditional WDM networks have already attracted research interest. However, researchers now are facing the challenge of adapting these concepts to flexible bandwidth networking scenarios with potentially large bandwidth channels.

Optical performance monitoring of flexible bandwidth networks coupled with an adaptive control plane can provide a means to improve performance against time varying PLIs through impairment aware networking. In other words, the joint optimization of the Routing and Spectrum Assignment (RSA) by the Path Computing Element (PCE) nodes in such networks should consider the factor of time-varying impairments. For example, by incorporating impairment awareness using a simple and effective performance monitoring technique, flexible bandwidth networks can react to changes in the quality of transmission (QoT) of each channel. This requires an adaptive control plane that receives input from nodes regarding each channel’s QoT. If the QoT for a particular channel degrades below an acceptable threshold, the change in modulation format to a less spectrally efficient format would use more bandwidth, but has an increased resistance to the signal degradation.

(29)

use of network resources (e.g., total spectral bandwidth) while meeting Quality of Service (QoS) requirements under dynamically changing traffic and physical layer impairment conditions. Here, flexible bandwidth networks leverage transmission systems capable of operating with very large bandwidth flexpaths to dynamically adjust the spectral utilization of individual flexpaths. This allows the preservation of a constant flexpath data rate under conditions of varying link impairments by adjusting the flexpath modulation format (i.e., spectral utilization). In this fashion, the spectral efficiency is lowered in situations of increased link impairment (or increased QoS requirement) or increased in situations of decreased link impairments (or decreased QoS requirement) in order to minimize the spectral resource utilized. This is in contrast to conventional WDM networks that are limited to a fixed bandwidth per channel due to the ITU grid spacing (G.649.1), which requires a change in data rate for conditions of varying link impairments. However, successful implementation of wide-scale impairment awareness in flexible bandwidth networks requires improvements to RWA used in conventional impairment-aware networks.

1.2.3.1 Flexible bandwidth networking with an adaptive control plane

The key to implementing impairment awareness in flexible bandwidth networks is the use of an adaptive control plane that is distributed yet centralized and can react to dynamic network situations such as varying flexpath QoT. Successful implementation of an adaptive control plane requires bidirectional communication between the control plane and each node. This allows the control plane to inform the nodes (network elements) of the spectral allocation used in the flexpaths and for the nodes to provide information to the control plane regarding the QoT of each flexpath. Figure 1.7(a) shows an example four-node network with two instantiated flexpaths. Here, impairment on link

2, shown as OSNR impairment, causes variations in the QoT of flexpath A. Figure 1.7_(b)

shows three configurations for implementing the 360-Gb/s flexpath A in a scenario with 360 GHz of available bandwidth. In an ideal situation, flexpath A is implemented using

(30)

8PSK (state A), which ensures the highest spectral efficiency and lowest spectral utilization. In this example, the control plane monitors the OSNR on flexpath A with feedback from Nodes 2 and 3. If the BER of flexpath A degrades above the predetermined Bit-Error Rate (BER) threshold, the control plane reconfigures the flexpath modulation format. Here, QPSK (state B) and BPSK (state C) are alternative modulation formats that less spectrally efficient, but more resistant to OSNR impairments. In principle, this technique can be extended to other higher order single- or multi-carrier modulation formats.

Figure 1.7 (a) Flexible bandwidth networking scenario including time-‐varying OSNR on Link 2. Possible control plane

adaptations to OSNR variations on link 2 include: (b) a single flexpath scenario in which flexpath (A) changes between 8PSK, QPSK, and BPSK to minimize spectral usage, or (c) a two flexpath scenario in which flexpath (A) changes between QPSK and BPSK to minimize spectral usage while flexpath (B) is spectrally shifted to maintain a defragmented spectrum.

In most networking situations, multiple flexpaths will be present simultaneously. Figure 1.7(c) shows possible state configurations for a two-flexpath scenario. In this example, flexpath A experiences a time varying OSNR degradation and changes between BPSK and QPSK, depending on its QoT. In order to accommodate this change, the spectral location of flexpath B is adjusted to maintain a defragmented spectrum. To this extent, the adaptive control plane can adjust both the modulation format and spectral location of each of the flexpaths individually.

In order to react to changes in QoT of each flexpath, the control plane needs an efficient means of monitoring the potentially broadband flexpaths. One method involves using a

(31)

low speed Monitor Signal to monitor the high speed flexpaths, which enables the control plane to cheaply and efficiently react to network changes. In this way, the adaptive control plane maintains flexpath data rate by increasing or decreasing the flexpath spectral efficiency under conditions of decreasing or increasing QoT, respectively.

Figure 1.8(a) details the Flexible Bandwidth Wavelength Cross Connect (FB-WXC) node architecture. At each input to the node is a QoT monitor followed by an N × N Wavelength-Selective Switch (WSS) to route each flexpath to the desired output.

Figure 1.8 (a) Flexible bandwidth wavelength cross connect (WXC) node detail. (b) Quality of transmission (QoT)

monitor detail.

The control plane configures the QoT monitors, WSS and Optical TransPonder (OTP) according to the current spectrum allocation map. Information sent from the QoT monitors to the control plane provides updates of the QoT of each flexpath allowing updates to the spectrum allocation map, if necessary. Figure 1.8(b) shows detail of a possible implementation of a QoT monitor based on a 1 × M WSS, performance monitors and a Field-Programmable Gate Array (FPGA). By detecting a low speed Monitor Signal on up to M flexpaths on an incoming link, the QoT monitor provides QoT information of high-speed flexpaths through known correlation data.

An effective means of implementing the OTP necessary for add/drop operations in each node (Figure 1.8(a)) is to use dynamic Optical Arbitrary Waveform Generation (OAWG)

(32)

and Optical Arbitrary Waveform Measurement (OAWM). This is a bandwidth scalable technique capable of generating the variable bandwidth and arbitrary modulation format flexpaths necessary for flexible bandwidth transmissions. Dynamic OAWG and OAWM operates over broad bandwidth by relying on the parallel processing of many lower bandwidth spectral slices that are manageable with currently available electronics. This technique allows the generation of both single- and multi-carrier flexpaths over its operation bandwidth. In comparison, other techniques are restricted to only using many low speed orthogonal subcarriers to generate broadband flexpaths.

1.2.3.2 RWA considerations for implementing impairment aware flexible bandwidth networks

Effective IA-RWA requires accounting for transmission impairments during path computation. Previous work has shown the feasibility of implementing IA-RWA using Path Computation Elements (PCEs). Moreover, the Internet Engineering Task Force (IETF) has defined two main IA-RWA PCE architectures for Wavelength Switched Optical Networks (WSONs): impairment validation and RWA (IV&RWA), and IV-candidate + RWA. However, there is a lack of implementation detail provided on the required specific PCE Protocol (PCEP) extensions, which rely on a centralized control plane without any distribution of signaling and control. A properly designed combination of distributed and central control planes in a network can potentially support optimal network functionality through timely and well-coordinated control of the network. This architecture is similar to the human nervous system in which the reflex (i.e., the local control at the nodes) and the brain (i.e., the Network Control and Management System (NC&M)) cooperate to support a rapidly responsive, yet well-coordinated, adaptive network. Such a control plane combines both centralized and distributed signaling by employing the centralized Data Control Network (DCN) between the network elements forming the communication for NC&M and the distributed in-band signaling embedded in the datagram to provide local and rapid

(33)

control decisions. The distributed control system can then provide statistical summary of the in-band signaling information to the NC&M, and the NC&M can see the “big picture” based on the summary collected from various parts of the network to instruct distributed control system to take an action. In addition, the IA-RWA PCE architecture needs improvement to consider the inherently new capabilities of flexible bandwidth optical networks to enable assignment of arbitrary bandwidth channels with arbitrary modulation formats. Bringing impairment awareness to flexible bandwidth networking also raises similar challenges. In this case, extension of the IA-RWA algorithms to include impairment aware routing and spectrum allocation (IA-RSA) will enable the implementation of variable bandwidth flexpaths.

An adaptive control plane with impairment awareness has the ability to adjust flexpath modulation format to maintain data rate under conditions of time varying link impairments (see Figure 1.9)

Figure 1.9 (a) Experimental arrangement. Detail of (b) the receiver and (c) the performance monitor (PM). OFCG:

optical frequency comb generator. OAWG: optical arbitrary waveform generation. MZM: Mach-‐Zehnder modulator.

(34)

1.3 Overview of Optical Modulation Formats

1.3.1 Introduction

Today’s information society relies to an unprecedented extent on broadband communication solutions, with applications such as high-speed Internet access, mobile voice and data services, multimedia broadcast systems, and high-capacity data networking for grid computing and remote storage. In order to most cost-effectively meet the widely differing bandwidth demands of various communication applications, several communication technologies are being used, each with their very own characteristics and advantages. One-way of comparing these technologies is to look at the maximum data rates they support for a given, regeneration-free transmission distance.

The regeneration-free transmission distance is defined as the distance that can be bridged without detecting and retransmitting the digital information anywhere along the propagation path. Figure 1.10 compares the most widely used RF and optical communication technologies for

wireline and wireless

communications, based on this metric. The boundary lines represent approximate guides, which are constantly shifting out subject to technological improvements. Figure 1.10 clearly shows that optical communication systems can support Tb/s capacities over many thousand kilometers, which makes them the ideal technology base for high-capacity wireline networking. With unrivaled optical fiber attenuation coefficients below 0.2 dB/km across several THz of bandwidth,

Figure 1.10 Regeneration-‐free transmission distance versus data

(35)

transmission distances exceeding 10 000 km, capacities of more than 10 Tb/s, and correspondingly high capacity × distance products of 36Pb / s⋅ Km have been reported over a single optical fiber (125µm diameter). Table 1.3 shows a selection of record numbers achieved in research laboratories, with up to 256 Wavelength Division Multiplexing (WDM) channels (“ch”) at per-channel data rates of 40 Gb/s. The need for system margins to accommodate carrier-class quality of service in non-ideal networks throughout the lifetime of aging system components typically lets commercial system capacities fall short of these record numbers by up to an order of magnitude. The driver behind the enormous interest in increasing system reach and aggregate WDM transport capacity is the steadily growing volume and bandwidth demand of data services and the desire to meet this demand by simultaneously reducing the cost per transported information bit.

Capacity Distance Capacity × Distance

10.9Tb / s 273ch× 40Gb / s

(

)

117Km 1.3Pb / s⋅ Km 10.2Tb / s 256 ch× 42.7Gb / s

(

)

300Km 3.1Pb / s⋅ Km 6.0Tb / s 149ch× 42.7Gb / s

(

)

6100Km 36Pb / s⋅ Km 2.6Tb / s 64ch× 42.7Gb / s

(

)

4000Km 10Pb / s⋅ Km 1.6Tb / s 40ch× 42.7Gb / s

(

)

10000Km 16Pb / s⋅ Km

Table 1.3. Some Recent Record Fiber-‐Optic Transmission Results

In WDM systems, this cost reduction is achieved by sharing optical components among many (or all) WDM channels; the most prominent examples for shared optical

(36)

components are the optical fiber, used both for transport and dispersion management, and optical amplifiers, placed periodically along the transmission path to re-amplify WDM channels. Since all shared optical components operate within limited wavelength windows, it is beneficial to space WDM channels as closely together as possible, with the ratio of net per-channel information data rate to the WDM channel spacing being referred to as a system’s spectral efficiency (b/s/Hz), sometimes also called information spectral density. For example, transmitting 40-Gb/s data information per WDM channel on the 100-GHz ITU frequency grid, one arrives at an SE of 0.4 b/s/Hz, irrespective of whether one directly transmits at 40.0 Gb/s, or, e.g., at 42.7 Gb/s to include a 7% overhead for FEC.

Another way of lowering the cost per information bit is to increase per-channel data rates. Once sufficiently mature optoelectronic technology at reasonable production volumes can be used, a fourfold increase in per-channel data rate typically implies a 2.5-fold increase in transponder cost. Therefore, quadrupling per-channel data rates eventually leads to 40% transponder cost savings. These considerations, among others, have led to the development of technologies for 40-Gb/s per-channel data rates. Together with advances in narrowband optical filtering technology, now enabling wavelength multiplexing filters on a 50-GHz optical frequency grid, WDM systems with an SE of 0.8 b/s/Hz and beyond have been demonstrated. Systems operating at 40 Gb/s are commercially available since early 2002.

In addition to increasing the point-to-point capabilities of optical transmission systems, and in order to further leverage broadband optical technologies, expanding networking functionality into the optical domain is yet another important aspect of reducing the cost per end-to-end transported information bit: (Reconfigurable) Optical Add/Drop Multiplexers allow adding and dropping wave-length channels at nodes as needed, while Optical Cross Connects (OXCs) allow switching wavelength channels between different

(37)

optical fibers; both are becoming major building blocks of optically routed networks.

High-Spectral Efficiency (SE), high capacity optically routed transport networks, which drive fiber-optic communications research today, are enabled by several key technologies.

• Low-loss optical components (including transmission fiber, dispersion-compensating devices, and optical switching/routing elements) minimize the need for optical amplification and reduce the associated amplification noise.

• Low-noise optical amplifiers (such as distributed Raman amplifiers) lower the noise accumulated along transmission lines.

• Advanced optical fibers reduce nonlinear signal distortions and enable higher signal launch powers.

• Forward Error Correction (FEC) allows for operation at poorer channel bit- error ratio, (BER) which relaxes the requirements on the Optical Signal-to-Noise Ratio (OSNR) at the receiver.

• Advanced modulation formats are used to trade off noise resilience, fiber propagation characteristics, and resilience to narrowband optical filtering due to multiple passes through OADMs.

In this paragraph, we focus on how advanced optical modulation formats enable high capacity optically routed transport networks. Being aware of the sophisticated modulation and detection schemes and algorithms employed in RF engineering, we first want to put the word advanced used in conjunction with optical modulation formats in proper context. communication transponders require broadband RF electronics and optoelectronics, spanning the entire baseband frequency spectrum from a few kHz to several tens of GHz. Up to a few years ago, technological difficulties in cost-effectively manufacturing

Table of Contents