I. Introduction

Flexible optical networks are investigated to be the next candidates for future backbone infrastructures [1–4]. Flexible optical networks offer high spectrum efficiency thanks to the use of coherent transmission and digital signal processing (DSP), bandwidth variable wavelength selective switches (BV-WSSs), and the adoption of the ITU-T flex grid [5], permitting us to reduce the spectral separation among channels in the network. Connections at different bit rates and exploiting different modulation formats may coexist occupying different portions and amounts of the spectrum. Transponders have been proposed supporting multiple bit-rate values, thus serving source–destination pairs with the requested rate on demand (e.g., 100 or $200\mathrm{Gb}/\mathrm{s}$) [6–8]. An important role is played by the reach-adaptive transponders. In this case, transmission characteristics can be set based on the path, with the objective of optimizing spectral efficiency while satisfying the required all-optical reach with an adequate quality of transmission (e.g., error free after forward error correction). Several works [1,8–11] have proposed to select the proper modulation format based on the physical characteristics of the path, e.g., using more spectrally efficient modulation formats when supported by the path [e.g., polarization-multiplexing 16-quadrature amplitude modulation (PM-16QAM) along short distances, typically few hundreds of kilometers], while using more robust formats [e.g., polarization-multiplexing quadrature phase-shift keying (PM-QPSK)] for longer distances [9].

Recently, besides multirate and reach-adaptive capabilities, a multiflow transponder has been proposed as a novel technology capable of handling traffic with high flexibility and reconfigurability [4,7,12]. A multiflow transponder generates multiple optical flows that can be flexibly associated with the traffic coming from the upper layers (e.g., IP), based on traffic requirements. Thus, optical flows can be aggregated (i.e., forming super-channels) or can be “sliced,” i.e., separated and directed toward different paths (as shown in Fig. 1) according to the traffic needs. We refer to the subcarrier as the minimum flow granularity. Multiflow transponders supporting sliceability are also called sliceable transponders or sliceable bandwidth variable transponders (SBVTs).

 

Fig. 1. Subcarriers generated by a single sliceable transponder, which are directed toward different destinations.

Download Full Size | PDF

An experimental implementation of a multiflow transponder has been presented in [7]. Such a transponder exploits a laser source for each subcarrier, so that each subcarrier may be independent from the others and used for specific traffic. Moreover, in [13], the same authors presented a first study on how to interface the optical transport network (OTN) layer to the multiflow transponder. However, SBVTs still require deep investigation, especially considering cost-saving solutions, such as the use of multiwavelength sources [14–16] to reduce the number of lasers.

The main contributions of this paper consist of the proposal and the design of a SBVT architecture, enabling format adaptation while preserving bit rate and implementing the time frequency packing technique [17]. First, a node architecture supporting SBVT in flex-grid optical networks is shown also with add-and-drop modules. Then, the proposed SBVT architecture is presented. The multiple subcarriers of SBVT are obtained by multiple laser sources (i.e., a laser per subcarrier) or by using a novel solution based on a multiwavelength source [15,16] and micro-ring resonators (MRRs) [18]. The multiwavelength source is capable of generating $N$ subcarriers by using a single laser. We propose to use cascaded tunable MRR optical filters to extract subcarriers from the comb generated by the multiwavelength source. Thus, the extracted subcarriers can be flexibly associated and modulated by specific traffic flows. A design of MRRs and its performance in terms of MRR transmission characteristics are presented in this paper. Then, selected subcarriers may form a single super-channel (as in the output of the SBVT in Fig. 1) or they may be sliced, thus independently used and directed toward different node output ports and destinations (as actually done in Fig. 1). The proposed SBVT architecture supports multirate and multimodulation formats (PM-QPSK and PM-16QAM). Differently from other papers in the literature (e.g., [9,10]), the proposed SBVT supports subcarrier format adaptation while preserving the bit rate thanks to the use of a specifically introduced switching matrix. The proposed SBVT also supports the time frequency packing transmission technique, which can achieve a spectral efficiency larger than the Nyquist limit [17]. This transmission technique consists of sending pulses that strongly overlap in time and/or frequency creating inter-symbol and inter-carrier interference. Coding [low density parity check (LDPC)] and detection are properly designed to account for the introduced interference [19]. Time frequency packing achieves a spectral efficiency larger than Nyquist (e.g., $5.1\mathrm{b}/\mathrm{s}/\mathrm{Hz}$ instead of $4\mathrm{b}/\mathrm{s}/\mathrm{Hz}$ achievable with PM-QPSK and Nyquist transmission [19]). Such SBVT architecture also supports code-rate adaptation to increase or reduce redundancy depending on the physical characteristics of the path. At the receiver side, coherent detection with DSP, including a decoder for time frequency packing transmission, is applied. A specific implementation of an SBVT considering standard 100 GbE interfaces and OTN, and supporting a rate of $400\mathrm{Gb}/\mathrm{s}$, is presented.

The proposed SBVT architecture achieves high spectral efficiency (e.g., given time frequency packing technique) and flexibility in terms of the rate, modulation format, reach adaptation, and possibility to generate a super-channel or exploit sliceability. When based on a multiwavelength source (and cheap MRRs), the proposed architecture permits us to reduce costs. The multiwavelength source enables the saving of ($N-1$) laser sources with respect to multiple lasers. If we assume a transceiver equipped with a tunable laser, an optical Mach–Zehnder modulator driven by an integrated electrical driver amplifier, a pin photodiode followed by a transimpedance amplifier, a clock, and a data recovery unit, the tunable laser may constitute around 60%–70% of the transceiver cost. If a multiwavelength source is used, the generated subcarriers are contiguous in the spectrum. This implies a constraint in the routing and spectrum assignment (RSA) when sliceability is exploited. Such a constraint is taken into account through simulations in this paper, and its impact is shown in terms of blocking probability. In particular, the sliceability performance (in terms of blocking probability and number of lasers) achieved by the multilaser and multiwavelength solutions is compared. The latter enables a cost saving due to the reduction of the number of lasers, and the former may experience a lower blocking probability at the expense of higher costs.

II. Requirements of the Flex-Grid Node and of the Sliceable Transponder

In the following sections, a node supporting SBVTs and the proposed SBVT architecture will be presented. Both the node and the transponder architecture are selected and proposed to satisfy the following requirements.

Node requirements:

SBVT requirements:

III. Reference Flex-Grid Node Supporting SBVT

A. Flex-Grid Node Architecture

In order to provide a comprehensive study of SBVT, a possible architecture of the node supporting the aforementioned requirements is considered. Figure 2 shows the possible node architecture, which also supports multicast [21]. The node is composed of splitters and BV-WSSs, and it is based on the broadcast and select technique [21]. The node is shown for a nodal degree of 3; thus it includes three ports and an add&drop module, shared by each port. The incoming signal at each port is split and broadcasted to all the ports, and then is selected or blocked by the BV-WSSs at ecah port. As an example, if we consider a pass-through from port A to port C, the signal coming from port A is split and broadcast to all the ports and to the drop module (which includes a BV-WSS). The drop module blocks such a signal, as well as BV-WSS at port B, while the BV-WSS at port C filters the signal. Such a node architecture, although cheap since it includes splitters, can be suitable for low-degree nodes. Indeed, for higher degree values, splitters would introduce larger losses, that could be unacceptable.¹

 

Fig. 2. Example of node architecture for a low-degree node in a flex-grid network.

Download Full Size | PDF

The architecture is based on BV-WSSs; thus it satisfies requirement 1. Such a node architecture also supports the sliceability (requirement 2) of carriers generated by an SBVT. As in the example in Fig. 1, a SBVT in the add module generates five subcarriers: two of them are directed toward node A for destination E, one to destination B, and two toward node C. It is assumed that port A of Fig. 2 is connected to node A of Fig. 1, as well as port B to node B and port C to node C. In such a scenario, the BV-WSS in port A in Fig. 2 accepts the two subcarriers directed to node A, while it blocks the others. Similarly, port B accepts the subcarrier directed to node B and blocks the others, and port C accepts the two subcarriers directed toward node C, while blocking the others. Requirement 3 will be clarified in the following subsections, when add&drop will be described.

B. Add Module

In this subsection, an architecture for the add module is considered by referring to the node in Fig. 2. The add module is derived from [22,23] and consists of a set of splitters, as shown in Fig. 3. It is assumed that the nodal degree is $n_d$ ($n_d=3$ in Fig. 2) and the add module consists of a number $N_t$ of SBVTs. Subcarriers generated by each SBVT are split and broadcasted toward all the output ports. This way, any subcarrier can be added into a specific port according to the routing specifications.

 

Fig. 3. Add module.

Download Full Size | PDF

Consequently, considering that the BV-WSS at each port must also accept the pass-through for all the other incoming ports (see Fig. 2), the BV-WSS at each port must have the dimensions of $(n_d+N_t)\times 1$. Considering $n_d=3$ and a $20\times 1$ BV-WSS [24], up to 17 SBVTs can be handled per port. The add module is contentionless (requirement 3) because any added signal at a generic $f_i$ toward any port of the node does not collide with any other signal at $f_i$ that has to be selected by a different port.

C. Drop Module

The drop module is derived from the architecture in [22] and is shown in Fig. 4. Each node port is connected to a BV-WSS, which is connected to $N_t·N_{rx}$ switches with dimension $n_d\times 1$, with $N_{rx}$ the number of coherent receivers in each SBVT. Each switch brings the dropped signal to a receiver (detailed in Subsection IV.C) of the SBVT (which, at the receiver side, is composed of multiple coherent receivers). Such architecture permits us to flexibly choose which receiver has to be used for any signal dropped from any port without any contention.

 

Fig. 4. Drop module.

Download Full Size | PDF

IV. Multiformat, Multirate, Code-Adaptive SBVT

A. Transmitter Side

The proposed SBVT architecture is shown in Fig. 5, and it mainly consists of a source of $N$ equally spaced subcarriers, a module for electronic processing, an electronic switch, a set of $N$ photonic integrated circuits (PICs), and a multiplexer or a coupler. The figure shows the architecture when the $N$ subcarriers are generated by a single multiwavelength source; however, such a source may be replaced with $N$ lasers, one per subcarrier.

 

Fig. 5. SBVT architecture.

Download Full Size | PDF

Each client is processed in the electronic domain (e.g., for filtering) and then is routed to a specific PIC through the switching matrix. The generated carriers are equally spaced according to the spectral requirements and adopted transmission technique (e.g., time frequency packing [20]). Generated subcarriers must be selected and routed to the proper PIC. Each PIC is utilized as a single-carrier transponder generating a subcarrier at frequency $f_i$. Each PIC is able to generate PM-16QAM and PM-QPSK signals, satisfying requirement d. A PM-16QAM signal is built within the PIC by combining eight clients, each operating at bit rate $r$. Figure 6 shows the internal PIC architecture enabling a single-carrier PM-16QAM transmission at rate $8\times r$. Traffic clients are provided as input to each PIC to modulate the related subcarrier. Four of eight electrical signals are attenuated to achieve the four-level electrical signals, required to generate 16QAM. Electrical signals are amplified by a linear amplifier and used to drive the in-phase (I) and the quadrature (Q) branches of an IQ modulator. A polarization beam combiner (PBC) finally provides polarization multiplexing. The PIC shown in Fig. 6 can also be flexibly adapted for PM-QPSK transmission at rate $4\times r$ (thus also supporting requirement c). This is achieved by switching off four ports (2, 4, 6, and 8 in Fig. 6) and providing clients through the other ports (1, 3, 5, and 7 in Fig. 6) to the IQ modulator. This way, just two electrical levels enter each branch of the IQ modulator, thus realizing the PM-QPSK transmission. Subcarriers are aggregated by the multiplexer or the coupler. Subcarriers may form a super-channel or can be sliced (requirement a), i.e., separated and directed to specific output ports, thanks to the use of BV-WSSs present at each node port as in Fig. 2. For typical transmission performance of PM-16QAM and PM-QPSK, the reader may refer to [9].

 

Fig. 6. Photonic integrated circuit (PIC) enabling PM-16QAM and PM-QPSK.

Download Full Size | PDF

The considered SBVT generates $N$ subcarriers, each one modulated by a PIC. A maximum rate of $R=N\times 8r$ is achieved. Multiple rate (requirement c) can also be achieved by using different sets of the $N$ subcarriers.

Figure 7 shows a specific architecture for a multirate, multiformat, code-adaptive SBVT able to serve clients from four 100 GbE interfaces. The proposed architecture enables an information rate of $400\mathrm{Gb}/\mathrm{s}$. The figure shows the architecture when a multiwavelength source is used; instead, $N$ lasers could be used. As shown in Fig. 7(a), the traffic coming from the 100 GbE ports can be encapsulated in OTN frames (e.g., for monitoring purposes), in particular OTL4.4 (i.e., four OTU4 physical lines). Then, clients enter in the SBVT and are processed in the electronic domain. This stage may consist of electronic filtering or adding redundancy (through the adaptive-rate encoder) for the time frequency packing technique (adding LDPC code [20]). Requirement e is electronically supported by the adaptive-rate encoder changing the amount of redundancy in the transmission, as done in [20]. Then the encoded clients enter, through the switching matrix, to the proper PICs to modulate subcarriers. The proposed $400\mathrm{Gb}/\mathrm{s}$ architecture is composed of four PICs. Then, generated subcarriers are coupled. The dotted line portion of the SBVT architecture is expanded in Fig. 7(b). Each 100 GbE interface provides four on–off-keying optical lines at $25\mathrm{Gb}/\mathrm{s}$ [25], each encapsulated in one of the four OTL4.4 lines. Note that the four $25\mathrm{Gb}/\mathrm{s}$ lines coming from a 100 GbE interface are not independent; thus they must modulate the same subcarrier. Then, OTN clients are encoded and filtered in the electronic domain. In case the two 100 GbE interfaces are served with a $200\mathrm{Gb}/\mathrm{s}$ (plus overhead) PM-16QAM, only ${\mathrm{PIC}}_{\mathrm{A}}$ is used and encoded/filtered clients are provided to ${\mathrm{PIC}}_{\mathrm{A}}$ inputs. For this purpose, the $1\times 2$ electronic switches are set to switch the electronic signal to their output port 1. This way a $200\mathrm{Gb}/\mathrm{s}$ PM-16QAM is generated at $f_i$, from ${\mathrm{PIC}}_{\mathrm{A}}$. In case PM-QPSK has to be used (e.g., to have a more robust transmission), the two 100 GbE interfaces are served with two $100\mathrm{Gb}/\mathrm{s}$ (plus overhead) PM-QPSK signals. In this case ${\mathrm{PIC}}_{\mathrm{B}}$ is also used. For this purpose, the electronic switches are set to switch the clients toward port 2 of each switch, i.e., the ones connected with ${\mathrm{PIC}}_{\mathrm{B}}$. This way, clients coming from a 100 GbE are directed to ${\mathrm{PIC}}_{\mathrm{A}}$, while clients coming from the other 100 GbE to ${\mathrm{PIC}}_{\mathrm{B}}$. Thus, the proposed SBVT supports format adaptation (PM-16QAM/PM-QPSK) while preserving the bit rate. In the case of PM-QPSK transmission, the encoder can use LDPC code for time frequency packing to increase the spectrum efficiency (e.g., $5.1\mathrm{b}/\mathrm{s}/\mathrm{Hz}$ instead of $4\mathrm{b}/\mathrm{s}/\mathrm{Hz}$ achievable with Nyquist transmission and PM-QPSK). For transmission performance of PM-QPSK based on time frequency packing, the reader may refer to [19,20].

 

Fig. 7. SBVT architecture, when 100 GbE interfaces are connected to SBVT.

Download Full Size | PDF

Some considerations on the electronics after the encoder are reported here. 100 GbE composed of four lines at $25\mathrm{Gb}/\mathrm{s}$ and OTN frames in the form of OTL4.4 are assumed. This implies that with the overhead, the speed of the four OTL lines is around $28\mathrm{Gb}/\mathrm{s}$, thus resulting in a $112\mathrm{Gb}/\mathrm{s}$ subcarrier, if PM-QPSK is adopted. To evaluate the impact of LDPC overhead in the modulator, spectral efficiency has to be considered. The code rate (thus spectral efficiency) varies with the optical signal-to-noise ratio. By referring to [19], for a 3000 km path the spectral efficiency achieved by time frequency packing is $5.16\mathrm{b}/\mathrm{s}/\mathrm{Hz}$. This means that a $112\mathrm{Gb}/\mathrm{s}$ subcarrier is transmitted over about 22 GHz (including information and LDPC coding). Without time frequency packing, electronics would be at 28 GHz with Nyquist-based transmission. Considering a path of 5000 km, spectral efficiency is around $4.25\mathrm{b}/\mathrm{s}/\mathrm{Hz}$, thus implying an electronics of around 26 GHz, still below 28 GHz. Finally, the possibility to include LDPC directly in the OTN framing could also be adopted.

Thus, the architecture proposed in Fig. 7 is multirate, multimodulation format, code-rate adaptive. Multirate capability is achieved by providing one to four 100 GbE interfaces. As an example, $100\mathrm{Gb}/\mathrm{s}$ of information rate is obtained by providing clients of a 100 GbE to a single PIC obtaining a PM-QPSK carrier. A $400\mathrm{Gb}/\mathrm{s}$ information rate can be obtained with four PM-QPSK carriers, by connecting clients from four 100 GbE interfaces to all four PICs. Subcarriers may form a super-channel or can be sliced by using BV-WSS at each node output port.

B. Multiwavelength Source and Subcarrier Filtering

The generation of $N$ subcarriers can be achieved through $N$ lasers or through the use of a multiwavelength source and cascaded MRRs. This section investigates the more innovative second solution.

A multiwavelength source should be able to produce from 1 to $N$ subcarriers (e.g., $N=4$). The spacing among subcarriers should be configurable (requirement b) to assure selectable subcarrier spacing depending on the transmission requirements. In addition to selectable carrier spacing, multiwavelength sources should also exhibit narrow linewidth, high spectral flatness, and good side-mode suppression ratio (SMSR). Multiwavelength sources have been extensively explored, and various techniques have been demonstrated based on mode-locked semiconductor and fiber lasers, and electro-optic modulators [14–16,26]. Tens of comb lines can be generated with sufficient quality. The former type suffers from the bottleneck of fixed free spectral range (FSR), hence not being feasible for SBVT because carrier spacing cannot be tuned. The latter type provides tunable carrier spacing [14] but suffers from bias drift in modulators. However, drift limitation can be overcome by using bias controllers, a standard practice in commercial applications. Moreover, a multiwavelength source for the sliceable transponder should also offer an integrated design comprising an integrated tunable laser and an integrated multicarrier generator. Hybrid III-V and Si tunable integrated lasers have also been demonstrated recently [27]. Regarding the multicarrier generator part, modulator-based designs can provide the flexibility and performance required for SBVT since designs involving fibers are not integrable. A block diagram of the schematic for the multiwavelength source based on the design in [15] is shown in Fig. 8.

A tunable integrated laser provides the parent carrier source, which is coupled with the dual drive Mach–Zehnder modulator DD-MZM. A sinusoidal RF signal is fed to one arm and its double frequency on the second arm by use of a frequency doubler. This design has the capability to generate an arbitrary number of lines in the range 3, 4, 5, and 9 by simply adjusting the RF drive signals’ amplitudes meeting requirement c of SBVT. The carrier spacing can be adjusted by changing the RF frequency (requirement b of SBVT), and the whole comb can also be tuned by simply tuning the parent laser. This design has spectral flatness of $<1.8\mathrm{dB}$ over the whole C band. By referring to the architecture in Fig. 7, the generation of two (if PM-16QAM is used) or four carriers (if PM-QPSK is used) enables an information rate of $400\mathrm{Gb}/\mathrm{s}$.

 

Fig. 8. Suitable multiwavelength source for SBVT.

Download Full Size | PDF

Once subcarriers are generated, the proper subcarrier has to be selected and directed to the corresponding PIC [see Fig. 7(a)]. As schematically shown in Fig. 9, optical filters realized by means of tunable MRRs [18] can operate the subcarrier selection to route each spectral component to the proper PIC. Note that, instead of MRRs, BV-WSSs can be used to route subcarriers toward specific PICs; however, MRRs permit a relevant cost saving. Assuming a multiwavelength source of $N$ subcarriers ($N=4$ in the example), $N-1$ cascaded MRRs are placed in front of each PIC (Fig. 9) to select the proper subcarrier for the PIC. In particular, each of the MRRs is used as a notch filter (i.e., band-rejecting filter) to suppress a not desiderated subcarrier (e.g., from $f_2$ to $f_4$ in ${\mathrm{PIC}}_{\mathrm{A}}$ of Fig. 9), finally leaving only a single subcarrier at the PIC input (e.g., $f_1$ in ${\mathrm{PIC}}_{\mathrm{A}}$ of Fig. 9). By tuning the resonant frequency of the MRRs by means of the thermo-optic effect, the survived subcarrier at the PIC can be flexibly selected. Section V will show the design and transmission performance of MRRs for subcarrier selection.

 

Fig. 9. Multiwavelength source and carriers selected by MRRs.

Download Full Size | PDF

C. Receiver Side

The same coherent receiver shown in Fig. 10 can be used for PM-16QAM and PM-QPSK. A tunable laser used as a local oscillator (LO) must be adjusted in order to match the incoming signal wavelength. The two optical signals beat into the opto/electronic (O/E) conversion module providing four analog electrical signals where information is completely mixed in terms of phase, amplitude, and polarization. Analog-to-digital (A/D) conversion at high sampling rate, such as $50\mathrm{Gs}/\mathrm{s}$, is performed, and real-time DSP is used in order to recover for the data. DSP, implementing a two-dimensional adaptive feed forward equalizer [9], fully compensates for linear and partially for nonlinear fiber transmission impairments and provides feedback to the LO in order to lock its central frequency to the incoming signal [28]. In the case of time frequency packing, the symbol detector iteratively exchanges information with a LDPC decoder according to the turbo principle [19]. Inter-symbol interference (due to time frequency packing) imposes a receiver based on sequence detection, as a Bahl–Cocke–Jelinek–Raviv (BCJR) detector [29]. Regarding BCJR complexity, a four-stage BCJR detector has been demonstrated for time frequency packing (e.g., on 5000 km path) in [19]. Its complexity is comparable with the one of the equalizer. If PM-16QAM is received, the eight output lines are simultaneously active, while just four of them will bring traffic in the case of PM-QPSK.

 

Fig. 10. Coherent receiver.

Download Full Size | PDF

V. Performance Evaluation on the Feasibility of MRRs for Subcarrier Selection

This subsection investigates the design and feasibility of MRRs for the selection of a single subcarrier for a PIC. Numerical analysis for the design of MRRs has been carried out by using a commercial software based on the solution of Maxwell equations.

The tunable MRR optical filters can be conveniently implemented on a silicon over insulator (SOI) platform. SOI is a mature, CMOS-compatible, viable technology for the implementation of compact, cost-effective, versatile MRR-based optical filters for a wide range of applications, including telecom/datacom [30], microwave photonics [18], and sensing [31]. SOI-based MRR can be designed to operate as either a band-pass or a band-rejecting filter. Filters operating in notch configuration (i.e., band-rejecting) with very high suppression ratio can be designed in a simple layout. Thanks to the strong thermo-optic effect, responsible for a change in the material refractive index with temperature, the central frequency of the notch filer can be tuned over a relatively wide bandwidth with low power consumption, for practical and flexible subcarrier rejection operation. A basic configuration of a MRR filter is represented by two silicon wires forming the input/output bus waveguide, which are coupled to a ring cavity, similarly implemented with a circular silicon waveguide, as schematized in Fig. 11(a). Typically, a strip waveguide is formed by deep-etching a layer of silicon, which is deposited over an oxide substrate layer, by using either e-beam or optical lithography or reactive ion etching. Light propagates in the waveguide with an effective refractive index $n_{\mathrm{eff}}$, which depends on the waveguide geometry. Due to the high refractive index jump between the core and cladding layers, strong lateral and vertical confinement in the waveguide is achieved, which allows us to keep the radiation loss due to bends in the waveguide very low, enabling us to realize low-loss, high-quality, compact ring resonators. The MRR-based filter depicted in Fig. 11(a) presents an input port and two output ports, namely the through and drop ports. Due to the interaction of the optical field propagating in the straight and ring waveguides, and the resonant behavior of the ring cavity, light coupled into the ring from the input port that matches a resonant wavelength of the ring cavity is transferred to the drop port. Correspondingly, a notch (i.e., a reduction in the transmission) appears in the transmitted spectrum at the through port. The field suppression at the through port in this resonant case can be made very high by realizing the so-called critical-coupling condition [32]. This allows strong inter-subcarrier cross-talk reduction, which is the main advantage of the through port with respect to the drop port. Several periodic notches are present in the transmitted spectrum at the through port, in correspondence with the ring resonant frequencies. At frequencies away from resonance, most of the signal is let pass at the through port. The main parameters of the MRR filter operating in band-rejection mode are thus the width and depth of the notches at resonances, and the frequency separation between adjacent notches, i.e., the FSR. In particular, the spectral properties of the MRR-based filter are defined by its physical length and the overall loss mechanisms in the ring waveguide (which includes the propagation loss, the coupling to the bus waveguides, and the additional radiation losses in the bend waveguide).

 

Fig. 11. (a) MRR filter schematic. (b) Three MRRs cascade performing subcarrier selection when $N=4$.

Download Full Size | PDF

A possible design for the subcarrier selection with a cascade of three MRRs operating as notch (band-rejecting) optical filters when $N=4$ (e.g., for the $400\mathrm{Gb}/\mathrm{s}$ SBVT) is reported in this section. The corresponding layout is shown in Fig. 11(b). The proposed configuration avoids any residual coupling between subsequent rings that could give rise to undesired resonances. The transmission at the through port of a single ring structure designed to operate in the 1.55 nm spectral region is shown in Fig. 12, where deep periodic notches, each in correspondence with the ring resonances, can be appreciated. The ring has been dimensioned such that the FSR of the cavity matches $(N-1)·{dv}_{\max }$, $N$ being the number of subcarriers ($N=4$ in this case) and $d{v}_{\text{max}}$ the maximum foreseen subcarriers’ frequency separation (37.5 GHz in our design). The rejection bandwidth of a single notch (as shown in detail in Fig. 13) has been designed to provide negligible loss for the selected subcarrier while ensuring strong suppression for the adjacent ones when the frequency separation between subcarriers is $d{v}_{\min }$ (the minimum foreseen subcarriers’ frequency separation, 12.5 GHz in our design). As shown in Fig. 13, the attenuation at 12.5 GHz from the resonant frequency is about 0.3 dB. Considering that the selected subcarrier experiences attenuation due to two adjacent notches (apart from the outermost subcarriers) and neglecting the attenuation due to the furthest resonances, the additional loss for the selected subcarrier would be less than 1 dB. For the design parameters leading to the spectra of Fig. 12, we have used a ring radius of 95 μm and power coupling coefficients with top and bottom straight waveguides of 0.13 and 0.06, respectively. A propagation loss coefficient of ${\alpha }_{\mathrm{prop}}=2.7\mathrm{dB}/\mathrm{cm}$ has been considered, whereas additional losses due to bends and the coupler have been set to 0.2 dB. The values of losses are compliant with those reported for available silicon platforms for the realization of integrated photonic circuits [33]. A single-mode rectangular waveguide at 1.55 μm, with 450 nm of width and 220 nm height, has been considered in the simulation. The value of $n_{\mathrm{eff}}$ has been computed using the commercial software based on the solution of Maxwell’s equations; the same program can be used to determine the distance between ring and bus waveguides providing the optimal coupling coefficients.

 

Fig. 12. Transmission characteristic of a single MRR (through port) versus normalized frequency.

Download Full Size | PDF

 

Fig. 13. Details of transmission characteristic of a single MRR (through port) versus normalized frequency in proximity of ring resonance.

Download Full Size | PDF

The tunability of the device via the thermo-optic effect has also been investigated in order to provide an estimate of the expected power consumption. Thermal tuning of the central frequency of each filter in the rings cascade can be independently achieved by placing a metallic heater above the device and making current flow through the heater contacts to produce a local change of temperature and a corresponding variation of the effective length of the ring (through a change of its refractive index). An increase in the temperature produces thus a shift toward longer wavelengths of the resonant wavelengths. The change of the effective refractive index due to the temperature change has been taken into account by using the thermo-optic coefficient of silicon. The results of the simulation are reported in Fig. 14, where the through port output spectrum for different values of temperature variations is shown. A full resonance shift over one FSR can be realized with a temperature change of about 15°C, which can be achieved with a power consumption of a few milliwatts with an optimized heater design [34]. Thermal heaters can also be used to slightly adjust the effective coupling coefficient for ensuring the strong field suppression at resonance provided by the critical-coupling condition [35]. Thanks to the periodicity of the device response (see Fig. 12), MRRs can be used to select subcarriers arbitrarily tuned in the C band.

 

Fig. 14. Tunability within a FSR versus temperature.

Download Full Size | PDF

VI. Performance Evaluation of Sliceability in Flex-Grid Networks Equipped With SBVTs Based on a Multiwavelength Source

This section analyzes the impact of sliceability with the proposed SBVT (as in Fig. 7), considering lightpath blocking probability and the number of laser sources used at the transmitter side. As anticipated earlier, it is assumed that the multiwavelength source generates subcarriers that are contiguous in frequency. This impacts RSA, especially in the case of independent subcarriers associated with different $s–d$ pairs.

The Spanish network topology in [36] with 30 nodes and 55 bidirectional links is considered. The inter-arrival process of one subcarrier request (of 37.5 GHz) is Poissonian; the holding time follows a negative exponential distribution with mean $5·{10}^4\mathrm{s}$, with requests uniformly distributed among all node pairs. RSA considering SBVTs based on a multiwavelength source is taken from the work we presented in [37]. Briefly, once a first subcarrier of an SBVT is activated (e.g., to connect $s$ and $E$ in Fig. 1), a system of weights discourages other connections (with a source different than $s$) to use frequencies that are contiguous to the ones in use by the considered SBVT. This way, the same SBVT is likely used for future connections with the same source (e.g., $s\text{-}\mathrm{B}$ in Fig. 1) because such contiguous frequencies are found as available. The RSA for multilaser SBVTs is the one proposed in [36]. The performance (in terms of connection blocking probability and number of lasers) obtained by SBVT based on a multiwavelength source and by SBVT composed of $N$ laser sources is compared.

Figure 15 shows blocking probability versus traffic load by varying the inter-arrival time. First, a comparison fixing the same number of SBVTs is done; i.e., 17 SBVTs per node (as the number derived in Subsection III.B) are considered, which corresponds to 510 SBVTs in the network. Then, another comparison is done by fixing the same number of laser sources, which corresponds to 510 laser sources in the case of a multilaser. When the number of SBVTs is the same, multilaser technology experiences a lower blocking probability than multiwavelength technology, which imposes constraints on subcarrier contiguity during RSA. If a blocking of ${10}^{-3}$ is considered, multiwavelength reduces traffic of 15% with respect to multilaser (1100 Erlang versus 1300 Erlang, respectively). However, multilaser technology requires 2040 ($510\times 4$ subcarriers) lasers, while multiwavelength only 510. The benefits of the multiwavelength source are clear when the comparison is done by fixing the same number of lasers, i.e., 510 for both multilaser and multiwavelength. Indeed, the multilaser experiences the highest blocking because each laser generates only a subcarrier, while, with a multiwavelength source, a single laser is able to serve up to four connections.

 

Fig. 15. Blocking probability versus offered traffic load.

Download Full Size | PDF

Figure 16 shows the average number of used lasers, for the same scenarios of Fig. 15. Given that multiwavelength achieves the lowest number of used lasers, Fig. 16 confirms that SBVTs based on the multiwavelength source are very promising solutions to reduce the number of installed lasers (thus, costs), while offering a high level of sliceability.

 

Fig. 16. Average number of used laser versus traffic load.

Download Full Size | PDF

Finally, some considerations on costs are provided here. The cost of the multiwavelength source depicted in Fig. 8 is mainly given by the used laser source and the RF oscillator. For the laser, a distributed Bragg reflector (DBR) laser tunable in the overall C band can be assumed. The cost of the local RF oscillator can be assumed as half of a DBR laser cost. Alternatively, a multilaser SBVT, with tunable flows in the C band, requires four DBRs. This provides the relative costs between multiwavelength SBVT and multilaser SBVT.

VII. Conclusions

In this paper, an SBVT architecture for flex-grid optical networks has been proposed. The presented SBVT enables sliceability and multirate capability and supports optical reach adaptation while preserving the bit rate. Adaptation of the optical reach can be achieved through modulation format selection or by varying the rate of the LDPC code used when time frequency packing transmission is assumed. The introduction of a switching matrix permits us to preserve the same bit rate when PM-16QAM or PM-QPSK is used. A specific implementation enabling an information rate of $400\mathrm{Gb}/\mathrm{s}$ has been presented. The subcarriers generated by the SBVT are obtained by multiple laser sources (i.e., one per subcarrier) or by exploiting a multiwavelength source (i.e., a source able to create several optical subcarriers from a single laser source) and MRRs. The design and performance analysis of cheap MRRs (combined with the multiwavelength source) used for subcarrier selection is specifically carried out showing their feasibility.

The proposed SBVT supports both super-channel generation and sliceability. Multiwavelength-based SBVT achieves a reduction of the number of installed laser sources. In this case, RSA is constrained to the generation of carriers that are contiguous in frequency, because of the multiwavelength source. Simulation results have been presented to show the impact of such a constraint in terms of blocking probability, by comparing multiwavelength and multilaser sources. The multiwavelength source reduces the number of installed laser sources, but if a lower blocking is required in the network, this can be achieved by increasing costs and using multilaser SBVTs. Future works will investigate other potential benefits of multiwavelength-based SBVTs in terms of the frequency stability of optical subcarriers (especially in the case of super-channel) and power consumption.