Sliceable BVT Evolution Towards Programmable Multi-Tb / s Networking

: The sliceable bandwidth variable transceiver (S-BVT) is a key element in addressing the challenges and evolution of optical networks, and supporting the ever-increasing tra ﬃ c volume, speed, and dynamicity driven by novel and broadband services and applications. Multiple designs and conﬁgurations are possible and are evolving towards supporting multi-Tb / s networking, thanks to the adoption of advanced and more mature photonic technologies. In this work, we review and analyze alternative S-BVT design architecture options that target di ﬀ erent network segments and applications. We speciﬁcally focus on S-BVTs based on multicarrier modulation (MCM), which provide a wide range of granularity and more ﬂexible spectral manipulation. A detailed description of the main elements in an S-BVT and their characteristics is provided in order to give design guidelines. The performance in a real testbed network is also reported, comparing a set of S-BVT conﬁgurations that adopt di ﬀ erent technologies. Finally, an extensive discussion of the described architecture, functionalities, and results, including programmability aspects, is provided in view of S-BVT evolution towards future optical network requirements and needs.


Introduction
Recent years have witnessed rapid growth of the speed and volume of traffic driven by novel and broadband (5G) services. In order to accommodate these changes, optical networks are evolving towards a more dynamic, flexible, programmable, and open paradigm, while novel and suitable technologies are being explored and exploited, improving their readiness towards supporting this new paradigm and providing solutions to related challenges [1][2][3][4][5]. In this context, the bandwidth variable transceiver (BVT), with its advanced functionalities and the ability of being sliceable (S-BVT), has assumed an especially relevant role in the research community over the last few years, following the roadmap of optical networks [5][6][7][8][9]. A BVT is a transceiver able to dynamically vary the optical bandwidth and/or bitrate and adapt to the condition of the established path by selecting specific parameters, such as the modulation format, operating wavelength, target rate/performance, and forward error correction (FEC) coding. The set parameter configuration is delegated to software control as per the traffic demand and targeted reach. A BVT can address a single traffic demand on a single portion of the optical spectrum over a specific network path between the source node and the destination node (also referred to as media-channel). The S-BVT is an evolution of BVT enabling multiple independent flows to be routed into different media-channels towards the same or multiple destination nodes (inverse multiplexing) [6]. Accordingly, the S-BVT supports flexible and programmable multi-flow, multi-rate, multi-format, and multi-reach transmission [7]. It can be seen as

S-BVT Architecture: Design and Options
General architecture of the S-BVT is depicted in Figure 1. It is worth noting the modular scheme, based on multiple BVT modules, which ease the grow/pay-as-needed approach, facilitating optical network disaggregation and fostering interoperability [9,19]. When photonic integration is considered, the fundamental/basic module can include multiple BVT front-end modules, which can be activated in a license-based fashion.
In Figure 1, we identify multiple block/elements per BVT at the transmitter (BVTx) and receiver (BVRx), implemented with different technologies and enabling different functionalities:  In each BVT module, the DSP supports the MCM, either DMT or OFDM. DMT represents a more cost-effective solution based on real-valued signal flows. It is typically used for intensity modulation in the optical domain, to be easily recovered with DD. However, this scheme is severely affected by chromatic dispersion (CD), resulting in a solution suitable for short distances. OFDM enables the generation of complex-valued signals and can also be combined with linear field (amplitude) and single-sideband (SSB) modulation, to be recovered either with direct detection (DD) or coherent receiver (CO-Rx). The mapper enables uniform loading (UL) of the digital/electrical DMT/OFDM subcarriers. There is a trade-off between spectral efficiency (SE) and transmission reach; since, for short-reach connections, modulation formats with higher SE can be adopted, while in the case of long-haul links, more robust formats (with less SE) should be used. In case of adaptive DSP, a mapper supporting multiple formats is included. This allows adoption of the most suitable modulation format per subcarrier in order to maximize BVT performance, according to the channel state information (CSI), as shown in the inset of Figure 1. Sub-wavelength granularity is defined as the ratio between the slice/flow bandwidth (B) and the number of electrical/digital DMT/OFDM subcarriers (Nsc). For example, considering a B = 20 GHz and 512 = Nsc, the subcarrier spacing and thus the minimum sub-wavelength granularity is 39 MHz.
The DAC converts the digital signal into the analogue domain; the sample rate can vary according to the adopted device, trading (similarly to ADC) high performance in terms of speed and bandwidth with cost and complexity [20]. After the DAC, the flow/slice is converted to the optical domain by means of an optoelectronic front-end, which can be based on direct laser modulation (DML) or external modulation. The flows are combined by the MF aggregator and transmitted over the network. At the receiver side, the MF after being distributed is received at the specific BVRx frontend, which can be based either on DD or CO-Rx. The ADC converts the flow to the digital domain to be post-processed at the BVRx DSP.
The S-BVT can be programmable: an SDN controller by means of an S-BVT agent (or more specifically S-BVTx and S-BVRx agents) configures the programmable/variable parameters in both the digital and analogue domains. For different slices/flows to be suitably enabled/disabled, as well as for the frequency slots (i.e. the bandwidth occupancy) and optical carrier (operating wavelength) to be occupied, the target rate or FEC is selected, etc. [7,9,19].
In the following subsections, we provide further details on specific S-BVT elements and functionalities.  In each BVT module, the DSP supports the MCM, either DMT or OFDM. DMT represents a more cost-effective solution based on real-valued signal flows. It is typically used for intensity modulation in the optical domain, to be easily recovered with DD. However, this scheme is severely affected by chromatic dispersion (CD), resulting in a solution suitable for short distances. OFDM enables the generation of complex-valued signals and can also be combined with linear field (amplitude) and single-sideband (SSB) modulation, to be recovered either with direct detection (DD) or coherent receiver (CO-Rx). The mapper enables uniform loading (UL) of the digital/electrical DMT/OFDM subcarriers. There is a trade-off between spectral efficiency (SE) and transmission reach; since, for short-reach connections, modulation formats with higher SE can be adopted, while in the case of long-haul links, more robust formats (with less SE) should be used. In case of adaptive DSP, a mapper supporting multiple formats is included. This allows adoption of the most suitable modulation format per subcarrier in order to maximize BVT performance, according to the channel state information (CSI), as shown in the inset of Figure 1. Sub-wavelength granularity is defined as the ratio between the slice/flow bandwidth (B) and the number of electrical/digital DMT/OFDM subcarriers (Nsc). For example, considering a B = 20 GHz and 512 = Nsc, the subcarrier spacing and thus the minimum sub-wavelength granularity is 39 MHz.
The DAC converts the digital signal into the analogue domain; the sample rate can vary according to the adopted device, trading (similarly to ADC) high performance in terms of speed and bandwidth with cost and complexity [20]. After the DAC, the flow/slice is converted to the optical domain by means of an optoelectronic front-end, which can be based on direct laser modulation (DML) or external modulation. The flows are combined by the MF aggregator and transmitted over the network. At the receiver side, the MF after being distributed is received at the specific BVRx front-end, which can be based either on DD or CO-Rx. The ADC converts the flow to the digital domain to be post-processed at the BVRx DSP.
The S-BVT can be programmable: an SDN controller by means of an S-BVT agent (or more specifically S-BVTx and S-BVRx agents) configures the programmable/variable parameters in both the digital and analogue domains. For different slices/flows to be suitably enabled/disabled, as well as for the frequency slots (i.e. the bandwidth occupancy) and optical carrier (operating wavelength) to be occupied, the target rate or FEC is selected, etc. [7,9,19].
In the following subsections, we provide further details on specific S-BVT elements and functionalities.

DSP
In order to generate a real-valued signal for intensity modulation, after serial to parallel (S/P) conversion of the input data and suitable mapping, either the real-valued fast Fourier transform (FFT) or the fast Hartley transform (FHT) can be applied, as shown in Figure 2. The latter algorithm enables the use of one-dimensional constellations, specifically M-ary pulse-amplitude modulation (M-PAM) format, and simplified channel estimation, to obtain the same spectral efficiency and performance as FFT processing with M 2 -ary quadrature amplitude modulation (M 2 -QAM) format [21]. Thanks to the FHT inherent symmetry and self-inverse property, a low-complexity fast processor can be applied, providing an attractive alternative for actual implementation of MCM-based transceivers. Finally, cyclic prefix (CP) is appended before parallel to serial (P/S) conversion.

DSP
In order to generate a real-valued signal for intensity modulation, after serial to parallel (S/P) conversion of the input data and suitable mapping, either the real-valued fast Fourier transform (FFT) or the fast Hartley transform (FHT) can be applied, as shown in Figure 2. The latter algorithm enables the use of one-dimensional constellations, specifically M-ary pulse-amplitude modulation (M-PAM) format, and simplified channel estimation, to obtain the same spectral efficiency and performance as FFT processing with M 2 -ary quadrature amplitude modulation (M 2 -QAM) format [21]. Thanks to the FHT inherent symmetry and self-inverse property, a low-complexity fast processor can be applied, providing an attractive alternative for actual implementation of MCM-based transceivers. Finally, cyclic prefix (CP) is appended before parallel to serial (P/S) conversion. The individual subcarrier can be adaptively modulated by applying a bit loading (BL) algorithm, using adaptive modulation formats (e.g. binary phase shift keying, BPSK, and L-QAM, with L = 2 n and 2 ≤ n ≤ 8) at the DSP mapper, for achieving multiple rate/reach. For subcarrier with low signalto-noise ratio (SNR), more robust and less efficient modulation formats (less number of loaded bits) are used, and vice versa in case of high SNR. Similarly, the power value of each subcarrier can be suitably assigned by applying power loading (PL).
Optimal or suboptimal algorithms (e.g. Levin-Campello or Chow-Cioffi-Bingham [22]) can be adopted for the BL/PL assignment to individual subcarriers. The algorithms can be rate adaptive (RA) and margin adaptive (MA). The former maximizes the data rate for a fixed S-BVT performance, while the latter maximizes the S-BVT performance at a given data rate. Loading strategies can be similarly applied to FFT-based or FHT-based S-BVT DSP; in case of FHT, the mirror symmetric subcarrier property of this alternative transform should be considered [23]. In case of adaptive DSP, by selecting the estimation mode, UL (e.g. using 4-QAM format) is applied to retrieve the CSI by means of the SNR profile (see top left inset in Figure 1). With this information, the data flow mode can be activated for optimal BL/PL assignment, according to the target rate/performance over the established connection.
For successful performance, the bit error rate (BER) threshold (e.g. 10 −3 , 4.62 × 10 −3 , or 2 × 10 −2 ) is set considering the applied FEC coding. To limit the overhead to 7%, hard decision (HD) FEC is adopted; increasing the FEC overhead up to 20%, applying a soft decision (SD) FEC, either a higher bit rate can be transmitted or a longer reach can be covered.
In order to mitigate the high peak to average power ratio (PAPR) of the OFDM signal, a clipping DSP module can be applied. To mitigate the signal distortion due to the clipping noise, an optimal clipping level is selected according to the modulation formats assigned by the mapper. Distortionless PAPR reduction techniques can also be eventually implemented at the BVTx DSP [24].
At the DSP, digital up-and down-conversion at an intermediate radio frequency can also be applied according to the selected modulation scheme; this element is particularly relevant in case of multi-band OFDM [8].
In case of CO-Rx, the DSP at the receiver side requires additional modules (e.g. carrier recovery) [8].  The individual subcarrier can be adaptively modulated by applying a bit loading (BL) algorithm, using adaptive modulation formats (e.g. binary phase shift keying, BPSK, and L-QAM, with L = 2 n and 2 ≤ n ≤ 8) at the DSP mapper, for achieving multiple rate/reach. For subcarrier with low signal-to-noise ratio (SNR), more robust and less efficient modulation formats (less number of loaded bits) are used, and vice versa in case of high SNR. Similarly, the power value of each subcarrier can be suitably assigned by applying power loading (PL).
Optimal or suboptimal algorithms (e.g. Levin-Campello or Chow-Cioffi-Bingham [22]) can be adopted for the BL/PL assignment to individual subcarriers. The algorithms can be rate adaptive (RA) and margin adaptive (MA). The former maximizes the data rate for a fixed S-BVT performance, while the latter maximizes the S-BVT performance at a given data rate. Loading strategies can be similarly applied to FFT-based or FHT-based S-BVT DSP; in case of FHT, the mirror symmetric subcarrier property of this alternative transform should be considered [23]. In case of adaptive DSP, by selecting the estimation mode, UL (e.g. using 4-QAM format) is applied to retrieve the CSI by means of the SNR profile (see top left inset in Figure 1). With this information, the data flow mode can be activated for optimal BL/PL assignment, according to the target rate/performance over the established connection.
For successful performance, the bit error rate (BER) threshold (e.g. 10 −3 , 4.62 × 10 −3 , or 2 × 10 −2 ) is set considering the applied FEC coding. To limit the overhead to 7%, hard decision (HD) FEC is adopted; increasing the FEC overhead up to 20%, applying a soft decision (SD) FEC, either a higher bit rate can be transmitted or a longer reach can be covered.
In order to mitigate the high peak to average power ratio (PAPR) of the OFDM signal, a clipping DSP module can be applied. To mitigate the signal distortion due to the clipping noise, an optimal clipping level is selected according to the modulation formats assigned by the mapper. Distortion-less PAPR reduction techniques can also be eventually implemented at the BVTx DSP [24].
At the DSP, digital up-and down-conversion at an intermediate radio frequency can also be applied according to the selected modulation scheme; this element is particularly relevant in case of multi-band OFDM [8].
In case of CO-Rx, the DSP at the receiver side requires additional modules (e.g. carrier recovery) [8].

S-BVT Front-End Options
Alternative S-BVT optoelectronic front-end module options can be adopted, according to the specific target network segment and application, as shown in Figure 3.
Specifically, at the S-BVTx, external (intensity or amplitude) modulation or a directly modulated laser (DML) can be used. In case of external modulation, a Mach Zehnder modulator (MZM) driven by a tunable laser source (TLS) is used to flexibly adapt the MZM bias point and the operating wavelength, respectively. For intensity modulation (IM), the MZM is biased at the quadrature point; amplitude modulation (AM) is obtained by biasing the MZM near the null point. The S-BVT can also be designed to have a single multi-wavelength source to drive a set of MZMs, each belonging to a different BVT module. This solution is more compact and cost-effective than using an array of TLS; however, it has limited tune-ability [25].

S-BVT Front-End Options
Alternative S-BVT optoelectronic front-end module options can be adopted, according to the specific target network segment and application, as shown in Figure 3.
Specifically, at the S-BVTx, external (intensity or amplitude) modulation or a directly modulated laser (DML) can be used. In case of external modulation, a Mach Zehnder modulator (MZM) driven by a tunable laser source (TLS) is used to flexibly adapt the MZM bias point and the operating wavelength, respectively. For intensity modulation (IM), the MZM is biased at the quadrature point; amplitude modulation (AM) is obtained by biasing the MZM near the null point. The S-BVT can also be designed to have a single multi-wavelength source to drive a set of MZMs, each belonging to a different BVT module. This solution is more compact and cost-effective than using an array of TLS; however, it has limited tune-ability [25]. Alternatively, intensity modulation can be obtained with DML, which allows to further reduce the S-BVTx module cost. Particularly, the use of vertical cavity surface emitting laser (VCSEL) technology can be considered to dramatically reduce costs, power consumption, and footprint [5]. Different VCSEL options can be envisioned to target specific applications, including widely tunable VCSELs based on a micro-electro-mechanical system (MEMS), or short-cavity (SC) VCSELs at long wavelengths characterized by large bandwidths (>18GHz) [26][27][28].
At the S-BVRx, either cost-effective DD, with a simple PIN photodetector (PD) followed by a transimpedance amplifier (TIA), or a more complex CO-Rx (see Figure 3b) can be adopted, including a local laser (local oscillator LO) and a suitable optical hybrid.

MF Aggregator/Distributor
The aggregator/distributor can be a programmable and bandwidth-variable (BV) wavelength selective switch (WSS) implemented in liquid crystal on silicon (LCoS) technology. This S-BVT element can be also realized by using photonic integrated circuit (PIC) technology to be a simple and cost-effective passive element. This option limits the flexibility and programmability of the transceiver architecture, even though with the advent of programmable photonics, it is possible to envision future attractive solutions [29]. A hybrid and electro-optical MCM scheme exploiting both super-and sub-wavelength granularity is proposed and assessed in [30], where the multiple flows are packed into super-channels by optically implementing the discrete wavelet packet transform (DWPT) and its inverse. This allows to increase the S-BVT spectral efficiency and can be either implemented by suitably programming an LCoS-based BV-WSS or using Mach-Zehnder interferometers integrated with silicon-on-insulator (SOI) technology.

Functionalities
Multiple advanced features and functionalities are enabled by adopting a MCM-based S-BVT, eased by its integration in an SDN control plane. Particularly, MCM and adaptive DSP enable rate/distance adaptability for an optimal spectrum usage. Furthermore, unique granularity and grid adaptation is obtained thanks to the hybrid tune-ability of optical carriers (for example, adopting a TLS) and adaptive subcarrier loading/allocation. Slice-ability, inverse multiplexing, and  Alternatively, intensity modulation can be obtained with DML, which allows to further reduce the S-BVTx module cost. Particularly, the use of vertical cavity surface emitting laser (VCSEL) technology can be considered to dramatically reduce costs, power consumption, and footprint [5]. Different VCSEL options can be envisioned to target specific applications, including widely tunable VCSELs based on a micro-electro-mechanical system (MEMS), or short-cavity (SC) VCSELs at long wavelengths characterized by large bandwidths (>18 GHz) [26][27][28].
At the S-BVRx, either cost-effective DD, with a simple PIN photodetector (PD) followed by a transimpedance amplifier (TIA), or a more complex CO-Rx (see Figure 3b) can be adopted, including a local laser (local oscillator LO) and a suitable optical hybrid.

MF Aggregator/Distributor
The aggregator/distributor can be a programmable and bandwidth-variable (BV) wavelength selective switch (WSS) implemented in liquid crystal on silicon (LCoS) technology. This S-BVT element can be also realized by using photonic integrated circuit (PIC) technology to be a simple and cost-effective passive element. This option limits the flexibility and programmability of the transceiver architecture, even though with the advent of programmable photonics, it is possible to envision future attractive solutions [29]. A hybrid and electro-optical MCM scheme exploiting both super-and sub-wavelength granularity is proposed and assessed in [30], where the multiple flows are packed into super-channels by optically implementing the discrete wavelet packet transform (DWPT) and its inverse. This allows to increase the S-BVT spectral efficiency and can be either implemented by suitably programming an LCoS-based BV-WSS or using Mach-Zehnder interferometers integrated with silicon-on-insulator (SOI) technology.

Functionalities
Multiple advanced features and functionalities are enabled by adopting a MCM-based S-BVT, eased by its integration in an SDN control plane. Particularly, MCM and adaptive DSP enable rate/distance adaptability for an optimal spectrum usage. Furthermore, unique granularity and grid adaptation is obtained thanks to the hybrid tune-ability of optical carriers (for example, adopting a TLS) and adaptive subcarrier loading/allocation. Slice-ability, inverse multiplexing, and defragmentation are also supported, together with an SDN-enabled MF generation and routing/switching of the multiple slices on the network. Furthermore, SDN-enabled S-BVT also facilitates a soft migration of fixed grid networks towards a flexi-grid paradigm. In [7], these functionalities are described in detail and their assessment provided for an S-BVT with multiple modules and technologies, supporting a total capacity above 200 Gb/s.

Performance and Programmability
Taking into account the different elements and options described in Section 2, the suitable S-BVT design depends on the targeted network segment and application. The fundamental S-BVT elements can be selected according to their cost/complexity and expected performance, identifying different flexibility level and requirements, as summarized in Table 1. In Reference [7] and [19], more exhaustive descriptions and discussions on the S-BVT elements and characteristics reported in Table 1 are provided. For example, transceivers adopting CO-Rx provide high flexibility and ultimate performance (high capacity for an extended achievable reach), but are more complex and costly than the options based on DD. Similarly, high-speed and large bandwidth DAC/ADC facilitates the improvement of S-BVT capacity and performance. Thus, for cost-sensitive applications, the S-BVT design should take these aspects and the trade-offs related with cost, achievable performance, and flexibility into account. Table 1. S-BVT elements and characteristics [7,19]; SE = spectral efficiency; L = low; M = medium; H = high. According to the different options, alternative S-BVT configurations are possible. Either DMT or OFDM (with or without SSB), implemented with DML (IM) or external modulation (IM or AM) can be combined with DD or CO-Rx. We identify three main configurations: (i) DMT with DD, (ii) SSB-OFDM with DD and (iii) OFDM with CO-Rx. The first option is the simplest with the lowest cost/complexity at the expense of the performance and SE. In fact, the optical B occupation is double compared to the other configurations and it is more affected by CD at increasing reach. The achievable reach can be extended by adopting the second option, since SSB-OFDM is more robust against CD. The ultimate performance can be achieved by adopting CO-Rx (with a tunable LO, TLO, for improved flexibility and adaptability) at the expense of higher cost and complexity at the S-BVRx, with respect to the other options. In addition, hybrid configurations can be envisioned combining different module options to match the specific S-BVT element properties and expected performance with targeted application or use. For example, DMT or SSB-OFDM can also be combined with CO-Rx, for an S-BVT design with simpler and cost-effective S-BVTx able to cope with longer paths and/or more degraded channels. Table 2 shows different S-BVT configurations and the latest experimental results achieved in a real testbed network, namely the ADRENALINE testbed [31]. This is to provide a performance comparison of the alternative proposed S-BVT schemes on a same experimental platform. It can be observed that an S-BVT based on DMT with DD supports high capacity over short paths. Nevertheless, by adopting a suitable dispersion compensator, the achievable reach can also be extended for high-capacity DMT systems, as demonstrated for a programmable 400 Gb/s DMT transceiver in the same testbed [32].

S-BVT Element
SSB-OFDM is more robust against CD over multi-hop paths and offers better performance also with the S-BVRx module based on DM-VCSEL (the table reports results for widely tunable MEMS-VCSELs) [33]. Finally, as shown in Reference [30], CO-Rx allows to achieve high robustness against accumulated dispersion.  [30] [8] 1 optical bandwidth occupation in terms of flexi-grid slots of 12.5 GHz; 2 with 50-GHz WSS; 3 with 10 cascading nodes including 100-GHz AWGs and 50-GHz WSS; 4 10 Gb/s net target bitrate (UL) connection.
As described in Figure 4, the ADRENALINE testbed network consists of four nodes: two optical cross-connects (OXCs) and two reconfigurable optical add-drop multiplexers (ROADMs); two of them (OXC-2 and ROADM-1) are equipped with programmable BV-WSS modules, enabling flexi-grid connections. The fixed-grid connections are based on 100-GHz and 50-GHz arrayed waveguide gratings (AWGs). The mesh network has five links ranging from 35 km to 150 km, for a total of 600 km of standard single mode fibers (G.652 and G.655). The network links are amplified by erbium-doped fiber amplifiers (EDFAs).
As shown in Table 2, the effect of traversing multiple nodes and, thus, multiple filtering stages, degrades the performance (filter narrowing effect) and should be taken into account for an optimal design [9, 34,35]. This effect is particularly detrimental when the filter bandwidth is narrow (<25 GHz) and, therefore, most prominent in flexi-grid networks.  Table 2 shows different S-BVT configurations and the latest experimental results achieved in a real testbed network, namely the ADRENALINE testbed [31]. This is to provide a performance comparison of the alternative proposed S-BVT schemes on a same experimental platform. It can be observed that an S-BVT based on DMT with DD supports high capacity over short paths. Nevertheless, by adopting a suitable dispersion compensator, the achievable reach can also be extended for high-capacity DMT systems, as demonstrated for a programmable 400 Gb/s DMT transceiver in the same testbed [32].
SSB-OFDM is more robust against CD over multi-hop paths and offers better performance also with the S-BVRx module based on DM-VCSEL (the table reports results for widely tunable MEMS-VCSELs) [33]. Finally, as shown in Reference [30], CO-Rx allows to achieve high robustness against accumulated dispersion. As described in Figure 4, the ADRENALINE testbed network consists of four nodes: two optical cross-connects (OXCs) and two reconfigurable optical add-drop multiplexers (ROADMs); two of them (OXC-2 and ROADM-1) are equipped with programmable BV-WSS modules, enabling flexigrid connections. The fixed-grid connections are based on 100-GHz and 50-GHz arrayed waveguide gratings (AWGs). The mesh network has five links ranging from 35 km to 150 km, for a total of 600 km of standard single mode fibers (G.652 and G.655). The network links are amplified by erbiumdoped fiber amplifiers (EDFAs).
As shown in Table 2, the effect of traversing multiple nodes and, thus, multiple filtering stages, degrades the performance (filter narrowing effect) and should be taken into account for an optimal design [9, 34,35]. This effect is particularly detrimental when the filter bandwidth is narrow (<25 GHz) and, therefore, most prominent in flexi-grid networks.  Table 2 (left) and picture indicating the testbed network elements (right).
Remote configuration of the S-BVT is performed by means of specific agents (see Figure 1), which are software component, enabling to map high-level operation requests from the SDN controller to hardware-dependent operations. The SDN controller configures the S-BVT over an established network connection, according to the received request and based on the elements and features described in Section 2.
Depending on the S-BVT design and specific configuration, multiple parameters can be selected. Examples of S-BVT parameters to be set are the enabled flow (S-BVT module and/or module element), operating wavelength, DSP algorithm (RA/MA), target rate or performance, FEC, etc. [7].
Indeed, the specificities of the adopted configuration and/or technologies (e.g. the use of VCSEL or TLS with external modulation, DD or CO-Rx) as well as the limited tune-ability and/or flexibility of the considered solution, should be taken into account for the S-BVT modeling and programmability handled by the SDN controller [19]. Other next generation (YANG) models are used to adopt a common language to describe the network elements to be controlled and managed, easing migration towards disaggregated optical networks and a multi-vendor scenario [3,9]. Examples of YANG models including proof of concept validations of S-BVT programmability considering different S-BVT configurations, based on alternative technologies, can be found in [19,30,36].

Conclusions and Future Perspective
In this work, we reviewed S-BVT architecture based on MCM. We propose the adoption of a modular design that can provide a more flexible and scalable solution in view of the evolutionary path of optical networks towards a more dynamic, disaggregated and ultra-high capacity networking. Different module options are presented, emphasizing on the pros and cons. For a suitable design, the specific network segment or targeted application should be taken into account in order to select the best option for each element and the related trade-offs, considering costs and performance.
The efficient exploitation of multiple dimensions with the softwarization and intelligent control of data plane allows optimal usage of the available resources in the design of high capacity and dynamic as well as highly scalable next generation optical networks.
We propose to adopt adaptive DSP, implementing BL and PL to improve the achievable performance and flexibility, thanks to the wide range tune-ability and spectral manipulation at sub-wavelength level. UL is used to estimate the CSI, which represents valuable information for performance monitoring and adaptation to the traffic demand and channel state. This allows allocating and managing of the network resources, while coping with signal degradation to ensure quality of service. Different S-BVT configurations have been presented and latest experimental results, obtained in a real testbed network, reported, in order to compare the different approaches, with the aim of providing design guidelines and suitably selecting the S-BVT elements. Dense photonic integration and the use of novel, advanced, and more mature photonic devices/technologies allow us to enable multi-Tb/s networking with reduced costs, power consumption, and footprint. Particularly, promising results have been obtained by adopting modular S-BVT based on DM SC-VCSEL at long wavelengths.
In view of the network evolution, the integration of data and control plane facilitates the implementation of novel advanced functionalities and elements towards the achievement of flexible and highly scalable S-BVT. The S-BVT programmability and modularity foster a migration towards flexi-grid and disaggregated networks. Particularly, this eases a grow-as-needed approach as well as smooth migration from chassis-based (proprietary) network elements to white boxes, enabling interoperability.
With a future perspective, we believe that the next-generation S-BVT will be based on advanced programmable modular photonic transceivers and that this S-BVT evolution will pave the way for the convergence of optical, inter-, and intra-data center networks.