Power Consumption Considerations of Coherent Transceivers in Filterless Point-to-Multipoint Metro-Aggregation Networks with Digital Subcarrier Multiplexing

Research Article Journal of Optical Communications and Networking 1 Power Consumption Considerations of Coherent Transceivers in Filterless Point-to-Multipoint Metro-Aggregation Networks with Digital Subcarrier Multiplexing CARLOS CASTRO 1,* , PABLO TORRES-FERRERA 1 , M. SEZER ERKILINÇ 2 , JACQUELINE SIME 3 , GIUSEPPE PARISI1, JOÃO PEDRO4,5, MARCO QUAGLIOTTI6, MARIO PORREGA7, DAVID HILLERKUSS1, CHRIS FLUDGER3, EMILIO RICCARDI6,**,AND ANTONIO NAPOLI1 1Infinera, Munich, Germany 2Infinera, UK 3Infinera, Nuremberg, Germany 4Infinera Unipessoal Lda, Carnaxide, Portugal 5Instituto de Telecomunicações, IST, Portugal 6Telecom Italia Mobile, Turin, Italy 7Infinera, Rome, Italy **Now with FiberCop, Italy *Corresponding author: [email protected] Compiled May 20, 2025 With metro-aggregation being a domain between the core and the access network segments, where the main innovation drivers and requirements are high capacity and low cost architectures, respectively, it finds itself in a position where it is necessary to address both aspects simultaneously. Coherent Digital Subcarrier Multiplexing (DSCM) in combination with a Point-to-Multipoint (P2MP) communication scheme offers the possibility to significantly extend the transceiver solutions sustainably. To demonstrate the potential of this approach, we carry out a techno-economic analysis on the basis of realistic, randomlydesigned, and flexibly configurable filterless metro-aggregation networks under different traffic growth conditions. Our investigation establishes a relation between the design characteristics of a network, its operating conditions, and performance. By evaluating the deployed hardware (i.e., transceivers) in terms of their power consumption, we demonstrate the advantages of optical aggregation and flexible allocation of bandwidth resources compared to full-capacity operation of DSCM-capable modules. Finally, a comparison to the operating conditions of traditional single-carrier coherent pluggables allow us to determine target specifications to manufacture competitive transceivers with DSCM technology. © 2025 Optica Publishing Group http://dx.doi.org/10.1364/ao.XX.XXXXXX 1. INTRODUCTION In optical communication systems, with Internet Protocol (IP) traffic experiencing exponential growth in recent decades [ 1 – 3 ], operators have been actively seeking solutions and devising strategies to meet these increasing demands. In the context of our scientific community, this is carried out by close collaboration between vendors, research institutions, and/or operators in order to shape future technological innovations [ 4 – 8 ], by conducting in-depth investigations on how to modify existing systems (e.g., network convergence and simplification) [ 9 – 11 ], and by—but not limited to—deploying new hardware according to the needs of Content Service Providers (CSPs). While different from other portions of the network (e.g., core or access) in terms of scope and requirements, the metroaggregation segment is still affected by them with respect to technological and economic facets. For instance, despite being based on a traditional Intensity Modulation Direct-Detection (IM/DD) network design, traffic growth due to modern applications and services has motivated researchers to consider coherent transceivers [ 12 , 13 ]. However, as the volume of transceivers is significant in relation to the core segment, aspects that usually take center stage in the access domain, such as low-energy consumption and overall cost, must be also addressed. Research Article Journal of Optical Communications and Networking 2 Historically, standard single-carrier IM/DD transceivers, which operate according to a Point-to-Point (P2P) scheme, have covered the low-cost/low-consumption part of the equation. Nowadays, with rapidly increasing demands for data traffic, coherent solutions, while effective in ensuring high-speed connectivity, could change this delicate balance. To keep costs and power consumption manageable, vendors could opt to simplify the optoelectronic design of the transceiver, lower the complexity of the Digital Signal Processing (DSP) architecture [ 14 ], downscale the symbol rate and/or modulation format [ 15 ], or exploit different transmission techniques [16], among others. Digital Subcarrier Multiplexing (DSCM), a communication technique based on creating a broadband signal from a collection of lower-speed Digital Subcarriers (DSCs) [ 17 , 18 ], which are generated and multiplexed within the DSP at the transmitter, might offer an alternative to control the costs of coherent technologies in the metro-aggregation domain due to its inherent flexibility and scalability. Not only can DSCM-based transmission systems operate in P2P or Point-to-Multipoint (P2MP) conditions, but by being able to treat each DSC as an independent entity/channel, we have precise control over the bandwidth allocated across the network, creating opportunities to save on Capital Expenditure (CAPEX) and Operational Expenditure (OPEX). Our previous work has focused on understanding the technoeconomic aspects of filterless P2MP DSCM metro-aggregation networks. In particular, we studied (i) how they relate to the optical performance of the system [ 9 , 19 , 20 ]; and (ii) how, by exploiting optical aggregation and flexible configuration, we can deliver savings through the reduction in the number of necessary transceiver units or, more efficiently, by optimizing the power consumption by transmitting only the necessary data [ 16 ]. These studies were compared with more traditional P2P network implementations, but still using the same DSCM-capable transceivers. This article presents an extension of our investigations by comprehensively describing the methodology behind our performance-based power consumption analysis of randomly-designed metro-aggregation networks. Herein, we explore and quantify the implications of resource sharing and control over bandwidth resources on system performance and power consumption of DSCM-capable transceivers. We also determine manufacturing goals for competitive solutions powered by DSCM with respect to single-carrier pluggable modules. Additionally, by presenting our findings based on various traffic growth scenarios, our aim is to provide a comprehensive guide for planning networks and devices, thus emphasizing key and significant elements related to the operation of filterless networks and DSCM. This manuscript is organized as follows. Section 2 introduces the concept of DSCM and filterless optical networks, as well as general definitions and considerations for the next part of the study. Section 3 defines the framework for the numerical analysis of horseshoe-based filterless metro-aggregation networks, explaining in detail our methodology, hardware assumptions, and the reasoning behind our approach. Furthermore, we show the impact that network upgrades (i.e., additional hardware to cope with increasing traffic according to a ‘Pay-As-You-Grow’ model) have on the optical performance of the network at the Optical Signal-to-Noise-Ratio (OSNR) level. Section 4 describes the process for carrying out a techno-economic evaluation focused on the power consumption of DSCM-based optical transceivers, optical aggregation, and the concept of flexible adjustment and configuration of DSCs. In this section of our study, we explore the relation between P2P and P2MP as communication schemes and how this, affected by the growing traffic demands, is reflected in the overall consumption of the transceivers in the network. Then, by considering the power consumption of 100 Gb/s coherent single-carrier P2P transceivers [ 21 ], we express the necessary relative consumption of DSCM alternative solutions for this type of technology to be competitive—and even bring advantages—from a transceiver-centric perspective regarding power consumption. Finally, Section 5 presents our conclusions. 2. TRANSMISSION SYSTEMS AND NETWORK ARCHITECTURE A. Digital subcarrier multiplexing In DSCM, where the bandwidth of an otherwise single-carrier optical signal is divided among multiple digital subchannels (i.e., DSCs) [ 17 , 18 ], the individual portions of the signal can be independently modulated, adjusted, and processed. Here, we achieve the same kind of flexibility of a Wavelength Division Multiplexing (WDM) system, but at a smaller scale and with a single laser. In case of DSCM-based P2MP coherent transceivers, as proposed in [ 17 ], these devices can take advantage of DSCs to enable and support operational improvements such as asymmetric traffic, dynamic capacity allocation, spectral optimization, and bidirectional data transmission [ 22 , 23 ]. For a metro-aggregation network scenario, DSCM-based pluggables offer various solutions based on a 25-Gb/s traffic granularity for the DSCs and a higher/lower-speed interoperability of the transceiver modules for P2MP network implementations [ 24 ]. In case of 25Gb/s transmission per DSC, DSCs are constructed from a twopolarization signal using 16-QAM as modulation format and operating at a symbol rate of 4 GBd. Following this, a high-speed version of the pluggable uses 16 DSCs to compose a 400 Gb/s optical signal, while lower-speed ones would work with only eight or four DSCs, depending on whether we require a 200 Gb/s or a 100 Gb/s unit, respectively. −32 −24 −16 −8 0 8 16 24 32 Frequency [GHz] Spectrum dB [a.u.] 400 Gb/s Single-carrier 400 Gb/s DSCM – Standard 400 Gb/s DSCM – Boosted 200 Gb/s DSCM – BiDi Fig. 1. Simulated spectra corresponding to a coherent singlechannel 400 Gb/s optical signal, to a 400 Gb/s DSCM optical signal with and without power adjustment at a DSC level (‘Standard’ and ‘Boosted’, respectively), and to a 200 Gb/s DSCM signal suitable for bi-directional transmission over a single fiber (BiDi). Figure 1 presents a simulated spectrum for comparing transmission schemes at a high level. It includes a 400 Gb/s singlecarrier signal (dual polarization, 64 GBd 16-QAM; orange) and three variations of a sixteen-subcarrier spectrum: a standard 400 Gb/s signal (blue), a 400 Gb/s signal with amplified sub- Research Article Journal of Optical Communications and Networking 3 carriers (red), and a 200 Gb/s signal with alternate DSCs deactivated (green). The ‘Standard’ plot in Fig. 1 shows an ideal sixteen-subcarrier signal for 400 Gb/s transmission. The Nyquist DSCs require in this configuration a 64-GHz-wide bandwidth. Now, due to hardware limitations [ 25 , 26 ] or cascading filters [ 27 , 28 ] or even different link distances for different subsets of DSCs, some portions of the multi-carrier signal’s spectrum might require to be amplified to optimize the system performance. This scenario is exemplified by the ‘Boosted’ plot. The fourth illustration represents a configuration for a Bi-directional (Bi-Di) transmission over a single fiber. Transmitting Uplink (UL) and Downlink (DL) optical channels at the same wavelengths over the same medium is a challenging, non-trivial undertaking, since the physical interaction of the signals will result in a penalty due to Rayleigh Back-Scattering (RBS) [ 23 ] and due to so-called discrete reflections (e.g., those caused by connectors). Depending on the power ratio between the transmitted signal and its counter-propagating attenuated replica, this might become an insurmountable obstacle; however, reducing the signal power to minimize penalties may not be feasible, as it is contingent on both link distances and the receiver’s sensitivity demands. In the end, a power optimization procedure will have to be performed to ensure connectivity for a given network topology. For this reason, typical solutions to this problem are to use a dual-laser architecture at the transceivers to use different wavelengths or a dual-fiber network architecture [ 9 ] and keep the UL/DL channels separate. In this regard, DSCM enables a third approach: free some spectral slots in one of the two directions, UL or DL, to allow for Bi-Di transmission, while simultaneously keeping RBS-caused penalties to a minimum [ 23 ]. Nevertheless, this technique reduces the overall capacity of the system. For example, Bi-Di transmission using a 16-DSC signal, where 8 DSCs are used for UL and 8 DSCs for the DL, results in an effective data rate of 200 Gb/s for each direction. B. Point-to-multipoint filterless optical networks The scalability of DSCM subsystems and the independent functionality of the DSCs enable the technical implementation of broadcast communication systems, where a single higher-speed pluggable can slice its entire spectrum according to a given set of conditions and distribute the resources among a selection of lower-speed modules. In the context of DSCM technology, this is what we call a P2MP communication scheme. In the metro-aggregation segment, incidentally, traffic follows a pattern that resembles a P2MP behavior: data streams are generated at access nodes, which in turn combine traffic from far-end network sites (e.g., business customers, base stations, residential areas, etc.). These access nodes connect to aggregation sites, called hub nodes, which act as domain interfaces; they combine traffic and funnel it to higher layers of the operator’s network. This pattern, however, is not exclusive to a single type of network; multiple physical topologies can be constructed/deployed as metro-aggregation networks, e.g. Hub-and-Spoke (H&S) [ 29 ], ring [ 30 ], and horseshoe architectures [ 20 , 30 ]. So far, this segment of the network has been served using standard IM/DD-based transceivers, which implies carrying out Opto-Electro-Optical (OEO) conversions in all the nodes regardless of their role. This, of course, has a direct impact on the number of transceiver units needed to guarantee connectivity, on the number of ports on the routers and Reconfigurable Optical Add and Drop Multiplexerss (ROADMs), and on the overall latency for the end users. (a) Point-to-point concept Main hub node 100 Gb/s 100 Gb/s 100 Gb/s 100 Gb/s ... ... Access node1 Access node2 Access node3 Access node4 ... 100 Gb/s 100 Gb/s 100 Gb/s 100 Gb/s (b) Point-to-multipoint concept Main hub node 16 DSCs (400 Gb/s) ... ... Access node1 Access node2 Access node3 Access node4 ... 3 DSCs 4 DSCs 4 DSCs 1 DSC Fig. 2. Exemplifying schematic of a (a) P2P network architecture, where all transceivers have a bookended counterpart, and of a (b) P2MP network architecture, where a single transceiver at the hub can communicate simultaneously with several lowerspeed units simultaneously, as long as there are available DSCs. In these diagrams, we assume a granularity of 25 Gb/s per DSC.  represents either a coupler/splitter with some defined ratio among its inputs/outputs. Figure 2 (a) and (b) show simplified conceptual illustrations of P2P and P2MP communications over a horseshoe, respectively. A horseshoe denotes a collection of fiber segments that physically connect the access nodes to each other and to the hub node(s). Throughout this article, we will also refer to them as ‘sublinks’ or ‘branches’. In P2P configuration, all transceivers in the network have a counterpart, with which they are constantly communicating, maintaining a constant transmission speed (e.g., 100 Gb/s), regardless whether that capacity is needed. On the other hand, P2MP places a single transceiver at the hub, which is technically capable of establishing communication with multiple units simultaneously as long as there are available DSCs. Consequently, modules at access nodes can also adapt their resources according to local data demands, allowing for a 400 Gb/s DSCM-capable unit to communicate with more than four 100 Gb/s DSCM pluggables. In the example depicted in Fig. 2 (b), the hub transceiver is explicitly serving a total of 12 DSCs, divided among four access nodes, which still leaves us with 4 available DSCs. Initially, it may seem that the primary change is only in the way traffic is handled and in the (re-)assignment of DSCs. However, when we migrate to a DSCM-based P2MP network architecture, this will necessitate structural changes in the network’s configuration, particularly by updating the node architecture, while retaining functionality. The architecture of a traditional metro-aggregation access node can be seen in Fig. 3 (a). Here, even though the depiction of a ROADM paints a rather straightforward and simple element with only the express and add/drop paths, the internal structure of the network element is much more complicated with its in-line amplifiers, wavelength selective switches, splitters, its input/outputs (i.e., degrees), and grid characteristics [ 31 , 32 ]. This last aspect reveals why a new type of architecture is necessary for DSCM. Flexible-grid ROADMs, while capable of narrow filter Research Article Journal of Optical Communications and Networking 4 (a) Two-degree ROADM Transceivers West East West ROADM (b) Filterless access node Transceivers CAD CAD CAD CAD UL DL Fig. 3. Block diagrams for an access node for a (a) ROADMbased implementation and (b) using a filterless approach.  represents either a coupler/splitter with some defined ratio among its inputs/outputs; and CAD stands for colorless adddrop. Amplifiers are symbolized by a triangle (). bandwidths and steep roll-off values, are not suitable for 4-GHzwide Nyquist DSCs. For this reason, to exploit the flexibility of DSCM at subcarrier granularity, we have been considering and investigating filterless networks [ 9 , 20 ]. That is, networks constructed using passive optical elements such as couplers and splitters, and optical amplifiers. An example of such a node architecture for filterless DSCM networks is depicted in Fig. 3 (b). In this schematic, similar to the ROADM scenario, we have presented a dual-fiber architecture, where one fiber carries the DL signal and the other provides the UL path. To explain the block diagram of the node, we consider the perspective of an incoming signal. Internally, the access node is composed of three different optical elements: an amplifier, a splitter, and a coupler. The optical amplifier (e.g., Erbium Doped Fiber Amplifier (EDFA) or Semiconductor Optical Amplifier (SOA)) ensures meeting the requirements for receiver sensitivity and operating at the desired launch power at the node’s output. In certain conditions, where the overall losses (i.e., fiber loss and the passive optical elements) exceed the amplifier gain, a booster amplifier might be placed right before the output of the access node. The splitters and combiners act as a pair; while the former ‘downloads’ a copy of the optical signal at every stage, the latter ‘uploads’ the locally generated DSCs to the propagating multi-subcarrier signal. Finally, on the line side of the access nodes, there are Colorless Add Drops (CADs), which are passive splitters and combiners, whose only functions are to distribute and collect signals from multiple DSCM-capable pluggable modules, respectively, to receive and transmit the optical data signals simultaneously. 3. NUMERICAL ANALYSIS OF HORSESHOE-BASED NETWORK TOPOLOGIES A. Methodology To investigate the behavior, operation, and elements required to build a metro-aggregation network, we have implemented a numerical tool that creates random horseshoe topologies inspired by examples of typical aggregation networks provided by TIM [ 16 , 33 ]. For the sake of simplicity, the simulated networks are only composed of horseshoes without any spur-arc extensions. Figure 4 illustrates the general network topology considered throughout this publication, where ‘ s ’ and ‘ N ’ denote the total number of sublinks present in the network and the corresponding number of leaf nodes of a given sublink, respectively, and ‘ d ’ represents the distance of the fiber segment between nodes. The network scenarios depict a series of horseshoes connecting two hub nodes (A and B), where they represent the main path and the protection, respectively. A B ... ... ... ... ... Sublink1 Sublink2 Sublinks Leaf(1, 1) Leaf(1, 2) Leaf(1, N1) Leaf(2, 1) Leaf(2, 2) Leaf(2, N2) Leaf(s, 1) Leaf(s, 2) Leaf(s, Ns) d(1, 1) d(1, 2) d(1, N1)d(1, N1+1) d(2, 1) d(2, 2) d(2, N2)d(2, N2+1) d(s, 1) d(s, 2) d(s, Ns)d(s, Ns+1) Fig. 4. General topology of the networks in this study. In this investigation, we will consider that 100 Gb/s and 400 Gb/s DSCM-capable coherent transceiver units are available, since these are the capacity values being considered for potential future P2MP optical networks [ 17 , 24 ]. However, we adhere to certain constraints when placing the transceiver units in the network: for a network operating in a P2P communication scheme, only 100 Gb/s transceivers can be used; while for P2MP, even though the situation at the leaf nodes remains unchanged with the 100 Gb/s units, we will be using exclusively 400 Gb/s transceivers at the hub side. This common reference base of 100 Gb/s modules will allow us to directly compare P2P and P2MP network deployments not only in terms of optical performance but also regarding their operating conditions. All the pluggable devices are assumed to have the capacity to exploit the spectral aspect of DSCM by dynamically adjusting the composite broadband signal at the subcarrier level. In other words, with a granularity of 1 DSC (i.e., 25 Gb/s), we are able to adjust the operating conditions of the transceivers to better and more precisely adapt to the traffic conditions of the network, which are considered to be time-dependent [ 34 , 35 ]. This study presumes that a network will be operational for a given number of years, with the assumption that traffic demands at leaf nodes will increase according to a year-over-year compound annual growth rate (CAGR). Considering the possibility of a ‘Pay-AsYou-Grow’ model, we will analyze the traffic demands of the leaf nodes yearly and consider the deployment of additional transceivers whenever these are required. By extension, this logic applies to the hub node as well. However, in the case of a P2MP scenario, the number of transceivers at the hub node will directly depend on the hardware at the leaf nodes; more specifically, on the number of DSCs that are required to cover the traffic demands at the leaf nodes. To illustrate this concept, let us define a representative network with random characteristics (e.g., number of sublinks, number of leaf nodes in each sublink, and amount of data traffic). Without loss of generality, we can ignore the protection path and just focus on the main connection to hub node A. An example of this network is depicted in Fig. 5, where the values corresponding to the traffic demands and required number of 25-Gb/s DSCs are placed on top of the leaf nodes. Based on our previous description, each leaf node would require a single 100Gb/s pluggable unit. This leads to a situation at the start of the analysis (i.e., operative year 1), in which the hub node must have the right hardware to address a total capacity of 8 DSCs divided among eight leaf nodes. Following this, a single high-speed 400 Gb/s transceiver with their 16 × 25 Gb/s DSCs is enough to guarantee connectivity from a perspective focusing solely on capacity. Research Article Journal of Optical Communications and Networking 5 A4 Gb/s (1) 2 Gb/s (1) 9 Gb/s (1) 3 Gb/s (1) 5 Gb/s (1) 1 Gb/s (1) 13 Gb/s (1) 4 Gb/s (1) Fig. 5. Model network to illustrate the hardware estimation approach. The text on top of the nodes indicates their traffic demands and the number in parentheses, the required DSCs. Year Traffic – (DSCs) Leaf node transceiver(s) Hub transceiver(s) 1 41 Gb/s – (8) 8 ×100 Gb/s 1 ×400 Gb/s 2 57 Gb/s – (8) 8 ×100 Gb/s 1 ×400 Gb/s 3 80 Gb/s – (9) 8 ×100 Gb/s 1 ×400 Gb/s 4 113 Gb/s – (9) 8 ×100 Gb/s 1 ×400 Gb/s 5 158 Gb/s – (10) 8 ×100 Gb/s 1 ×400 Gb/s 6 221 Gb/s – (12) 8 ×100 Gb/s 1 ×400 Gb/s 7 309 Gb/s – (16) 8 ×100 Gb/s 1 ×400 Gb/s 8 432 Gb/s – (21) 9 ×100 Gb/s 2 ×400 Gb/s 9 605 Gb/s – (28) 10 ×100 Gb/s 2 ×400 Gb/s 10 847 Gb/s – (38) 12 ×100 Gb/s 3 ×400 Gb/s Table 1. Summary of the hardware estimation for the model network in Fig. 5. Table 1 expands on the previous explanation by showing the evolution of the network traffic over a period of 10 years with a yearly CAGR of 40%. Here, it is possible to observe how, as traffic increases, additional resources have to be enabled/deployed at the different nodes in the network. For example, in the third year of operation, an additional DSC must be activated on one of the DSCM capable transceivers to match the growing traffic. This, however, does not require adding additional hardware, since the traffic increase can be still be managed by the existing pluggable units. In year 8 and year 10, this is no longer the case; here, the growth in traffic causes the DSCs to exceed the maximum capacity of the 100 Gb/s and 400 Gb/s modules. For this reason, additional transceivers have to be deployed at the corresponding nodes. In the eight year, two units: one 100 Gb/s and one 400 Gb/s; and by the tenth year, compared to year 8, four additional units: three 100 Gb/s and one 400 Gb/s. After determining the required transceiver units and defining the network’s and nodes’ operational conditions (e.g., active DSCs, total optical output power, OSNR, coupling/splitting ratio to combine multiple transceivers, among others), we perform a numerical analysis focusing on the power budget and OSNR changes across the filterless synthetic network. At this stage, to simplify the analysis and without loss of generalization, we focus on the connection between the leaf nodes and a single hub node—for example, hub node ‘A’ in Fig. 4. With regards to the node architecture (Fig. 3 (b)), we have assumed passive optical components with a 75/25 ratio between the pass-through port and the add/drop port, respectively, and equalizers that ensure a total optical launch power of 0 dBm at the output of each leaf node. For the OSNR calculation, we analyze the DL and UL scenarios based on the hardware at that particular point in time and the conditions regarding the optical level aggregation, which is a metric used to conceptualize and quantify how much hardware resources (e.g, transceivers) are shared among multiple sublinks/leaf nodes in a network [ 9 , 19 ]. Should additional transceiver units require to be deployed—for example, for year 10 in Table 1—, couplers/splitters with a uniform ratio will also be added. For example, if two transceivers are needed, the ratio for components will be 50/50; for three, the ratio will be 33/33/33; and so on. The estimated OSNR values are then compared with each other, and the lowest value is deemed to be the limiting performance level of the system. Afterwards, this representative value for a network’s performance will be compared with the specifications of the DSCM-capable pluggable transceivers. By defining the performance threshold to be tied to the minimum required OSNR for error-free operation, we can calculate a so-called OSNR margin, a metric that represents how tightly we configure the operating conditions of the overall optical system in terms of power/performance budget. A discussion of these results in the context of the network scenarios’ characteristics and transmission requirements can be found in subsection 3 E. B. DSCM-powered pluggable transceivers Deploying DSCM-capable devices in the network allows for the possibility to adjust their operation depending on the instantaneous conditions of the network. In real scenarios, data traffic is not constant; not only does it grow over time as new applications and technologies are available, but it also fluctuates during the day [ 16 , 34 ] depending on the habits and needs of the general population. 12345678910 0 300 600 900 1,200 Year Overall capacityLeaf [Gb/s] Single-carrier 100 Gb/s transceivers (40%) Single-carrier 100 Gb/s transceivers (10%) DSCM-capable 100 Gb/s transceivers (40%) DSCM-capable 100 Gb/s transceivers (10%) Dynamic traffic demands (40%) Dynamic traffic demands (10%) Fig. 6. Overall leaf nodes’ capacity as a function of the type of transceivers used and of the network traffic’s CAGR. The provisioning follows a ‘Pay-As-You-Grow’ model, where new 100 Gb/s transceivers are deployed once the maximum capacity of a leaf node is being exceeded. Since DSCM slices the spectrum of a broadband signal into narrower spectral pieces (e.g., the 64 GHz-wide spectrum of the 400 Gb/s signal is divided among 16 DSCs), we can think of and operate the individual DSCs as independent ‘channels’ by activating or deactivating them based on the traffic conditions of the network. The traffic conditions stated in Table 1 (40% CAGR) and those corresponding to a 10% CAGR are depicted in Fig. 6 using solid lines and dashed lines, respectively. In the graph, a dark color indicates the total capacity that the hardware at the leaf nodes can handle, whereas a lighter color represents an Research Article Journal of Optical Communications and Networking 6 adjusted capacity through the dynamic control of DSCs (i.e., an on-demand activation/deactivation of DSCs). To compare and understand the operating conditions of single-carrier and DSCMbased pluggables, let us analyze the 40%-CAGR scenario. Considering the network’s topology and traffic characteristics in Fig. 5, we can observe how the single-carrier 100-Gb/s units, while able to handle relatively large amounts of data, ‘waste’ resources by operating at an overdimensioned capacity in relation to the actual data requirements for most of the duration of the analysis. Once additional units are required (year 8), we start to see a progressively better utilization of the available bandwidth, since the increasing demands require a considerable bandwidth in relation to the nominal hardware capacity. On the other hand, by deploying a DSCM-capable solution, we enable an additional degree of flexibility, as it is technically possible to dynamically adjust the number of active DSCs according to the evolving traffic conditions at a given point in time. This is represented in Fig. 6 by the lighter-color bar following closely the growing traffic. In conclusion, regarding the operating conditions of the transceiver pluggables, we have two options: full-capacity and adjusted operation. Whereas driving the module at full capacity refers to activating all possible DSCs of a transceiver, for the adjusted operation, we only use the necessary bandwidth resources (i.e., DSCs) that are required to address specific traffic demands. When comparing both approaches, the latter could allow for some degree of savings compared to the former regarding the power consumption of the units. 012 3 45678910 11 12 13 14 15 16 0.5 0.6 0.7 0.8 0.9 1 Number of active DSCs Norm. power consumption (ASIC) [a.u.] 400 Gb/s 200 Gb/s 100 Gb/s Fig. 7. Measurements corresponding to the power consumption of the transceiver’s Application-specific Integrated Circuits (ASIC) for different operating conditions. For this reason, we measured the power consumption of the 400 Gb/s transceiver module’s ASIC as a function of the number of active DSCs for various operating conditions. The measured results can be seen in Fig. 7, where the data have been normalized according to the full capacity conditions of the ASIC (i.e., 16 DSCs at 25 Gb/s for a total data rate of 400 Gb/s). The measurements are carried out using a 16-QAM modulation format where the change in capacity indicates a change in the maximum number of available DSCs. For instance, while the 400-Gb/s case with 8 DSCs represents 16 available DSCs, where 8 of them have had their Forward Error Correction (FEC) engine turned off; in the 200-Gb/s case with 8 DSCs, there are only 8 DSCs available, since the other 8 have been turned off completely. One very noteworthy aspect from the ASIC’s measurements is that the reduction of the number of active DSCs does not cause the consumed power to decrease in the same proportion. In other words, 50% reduction in terms of DSCs does not equal power savings of 50%. The reason for this is that while there are elements that could be turned off or that could be working at a limited capacity (e.g., FEC, equalizers, clock recovery, among others), there are elements that are shared among all of the DSCs, such as the Digital-to-Analog Converters (DAC) and Analog-to-Digital Converters (ADC). Furthermore, under close inspection, we can observe that the 400-Gb/s and the 200Gb/s cases with 8 DSCs do not match, nor do the 200-Gb/s and 100-Gb/s cases with 4 DSCs. This apparent discrepancy is due to the additional elements being completely turned off in the lower-capacity operation. Despite the fact that the limited capacity modes (i.e., 200 Gb/s and 100 Gb/s) consume less power than the operation at 400 Gb/s, it is nevertheless important to mention that in the case of high traffic demands (e.g. 16 DSCs), it is preferable to use the highest-capacity modules instead of multiple lower-speed ones. For instance, a single 400-Gb/s module is both more efficient than 2 × 200 Gb/s units and 4 × 100 Gb/s. However, when considering the overall power consumption of the pluggable, we need to highlight that the ASIC is just part of it; the total power is the combination of ASIC and the optical components (i.e., the Transmitter-Receiver Optical Sub-Assembly (TROSA)). 65 70 75 0.75 1 1.25 400 Gb/s 200 Gb/s 100 Gb/s Temperature [°C] Norm. power consumption [a.u.] Fig. 8. Measurements from manufactured DSCM-capable coherent transceivers for various operating conditions as a function of the temperature of their cases. Figure 8 depicts measurements corresponding to the power consumption of manufactured DSCM-capable 400 Gb/s coherent pluggable transceivers as a function of the case’s temperature and operating capacity, when all of the available DSCs are active. In this normalized graph, the colored cloud of data points represents individual measurements of the transceiver units (circular markers), while the highlighted marker denotes the mean consumption value at a case temperature of 70°C. These measurements reveal, similar to Fig. 7, a disproportionate relation between the operating conditions of the transceiver and capacity (i.e., number of available DSCs). For the study presented in this article, we assume that the power consumption of the TROSA is independent of the number of DSCs as we consider the parameters for the optical operation of the transceiver unit to remain unchanged. The only value that changes the modules’ power consumption is the number of required DSCs and the type of transceiver. Following this, we can combine the measured data from Fig. 7 and from Fig. 8 to estimate a potential relation between the overall power consumption of a DSCM-capable transceiver for different capacities as a function of the operating conditions. The resulting data has been summarized in Fig. 9. Here, it is important to remark on the fact that, as the optics of the transceivers remain unchanged Research Article Journal of Optical Communications and Networking 7 for the different operating capacities, the relative reduction in terms of power consumption is lower than the reference values shown in Fig. 7. 012 3 45 6 7 8 9 10 11 12 13 14 15 16 0.6 0.7 0.8 0.9 1 1.1 Number of active DSCs Norm. power consumption [a.u.] 100 Gb/s 200 Gb/s 400 Gb/s Fig. 9. Estimated power consumption of a DSCM-capable transceiver for different operating conditions. In our calculations, we also assume that the total output optical power of the module is fixed to 0 dBm regardless of the number of DSCs. For the subsequent OSNR analysis, this implies that the signal-to-noise conditions across the network will vary, since the power of the individual DSCs might have to be adjusted to maintain the overall value at the desired level. C. Sample-based horseshoe design Telecom Italia Mobile (TIM) provided us with a series of realistic network topologies, from which we have extensively used a subset of scenarios in our previous work [ 9 , 16 , 19 , 33 ]. These examples of metro-aggregation networks, despite their similarities, have slightly different characteristics in terms of number of sublinks, number of transit/leaf nodes, traffic demands, node placement, link distances, etc. For this reason, to provide a more generalized insight into the operation and expectations of horseshoe-like network topologies, it is necessary to look into scenarios with different design parameters. Consequently, we analyzed the available examples to summarize their characteristics. Our detailed inspection revealed that, in general, the fifteen metro-aggregation networks had a total of 46 sublinks and 187 leaf nodes. Moreover, the topologies are composed of multiple sublinks (2 – 4 branches) that connect to at least one of the hub nodes. To indicate the number of leaf nodes that are expected to be found in any given sublink of a network, we have summarized our observations in Fig. 10. In our numerical analyses, we will base the creation of network scenarios on this probability distribution. 12345 6 0 5 10 15 20 25 30 Number of leaf nodes/sublink Probability [%] Fig. 10. Probability of generating sublinks with a particular number of leaf nodes. Similarly, we can determine the probabilistic distribution for the fiber segments in the network. Here, we define a fiber segment as the fiber link that directly connects two nodes–either leaf node to leaf node or hub node to leaf node. Figure 11 shows the normalized probability density of the fiber segments’ distances along with the observations from the network data of TIM. For the fitting, we used a Gamma distribution [ 36 ] with the following parameters: α=2.9 and β=0.3 . In the observed data, there were data points beyond 80 km for the fiber segments; however, since we focus on constructing networks without the need for additional amplification—other than the stage at the leaf nodes according to the node architecture design in Fig. 3 (b)—, the maximum distance for a fiber segment has been capped to a value of 80 km in our simulations. 010 20 30 40 50 60 70 80 0 0.5 1 Distance [km] Norm. probability densityFiber Network observations Gamma distribution Fig. 11. Normalized probability density corresponding to the fiber segments in an illustrative metro-aggregation network topology. For the traffic characteristics of the network, we follow the same approach of fitting a Gamma distribution to the observed data (Fig. 12). In this instance, the α and β parameters are equal to 2.4 and 0.3, respectively. 010 20 30 40 50 0 0.5 1 Data rate [Gb/s] Norm. probability densityTraffic Network observations Gamma distribution Fig. 12. Normalized probability density corresponding to the leaf nodes’ base traffic demands in an illustrative metroaggregation network topology. We utilize the probability distributions of the fiber segments and of the traffic demands to generate random values that are representative of the different ‘realistic’ and valid network scenarios that might arise in the metro-aggregation domain. Furthermore, this is complemented by the previous descriptions regarding the number of horseshoes in a network and the corresponding number of leaf nodes that are to be expected. Due to the random-oriented nature of our analysis, it is necessary to simulate a large number of network scenarios to provide results that are significant for the types of networks we consider. In this study, we have considered 1000 different networks. D. Optical aggregation in filterless networks One very attractive aspect of any P2MP implementation is the possibility to consolidate elements to operate with multiple connections in mind. For filterless metro-aggregation networks, Research Article Journal of Optical Communications and Networking 8 DSCM enables a realization that is simultaneously able to deliver high bandwidth promises and to simplify provisioning by subsisting multiple elements at the hub node(s) for a single piece of hardware, which is capable of connecting to multiple endpoints at once. A(a) Agg. Level 1 A(b) Agg. Level 1.33 A(c) Agg. Level 2 A(d) Agg. Level 4 Fig. 13. Example of metro-aggregation network topologies with four sublinks, illustrating how they are connected to the hub node: (a) individually, (b) two stand-alone sublinks and one set of two, (c) two sets of two sublinks, and (d) all simultaneously.  represents a coupler/splitter with equal ratio among its inputs/outputs, respectively. This concept of a common source at the hub node that connects to multiple leaf nodes across one or multiple sublinks is what we call ‘optical aggregation’, since the connections must be physically possible through the use of couplers and splitters. If the capacity of the DSCM-based transceivers and the OSNR requirements allow, the units at the hub can even communicate with endpoints across multiple sublinks. By this account, we could try to serve as many sublinks—and hence, leaf nodes— with the least amount of hardware at the hub to save on the total number of transceivers that would otherwise be necessary to provide full connectivity to the network. We have, therefore, defined a metric to convey the degree of re-utilization of the pluggable unit(s) at the hub with respect to the sublinks in a given network; we refer to it as ‘aggregation level’, and we define it as follows [9]: Agg. Level =1 Cs · Cs ∑ i=1 Aggs;(1) where Aggs denotes how many sublinks are combined (i.e. branches that are simultaneously connected to the same hardware at the hub node), and Cs is the number of sublink combinations necessary to ensure that the entire network is served by the transceiver unit(s) at the hub. As a metric, the lowest value for the level of aggregation is 1, while the highest value depends on the number of sublinks in a network. The former describes a scenario in which the pluggables at the hub node perform optical aggregation functions exclusively for individual sublinks. Conversely, the higher the value, the more horseshoes are served by the hardware at the hub node simultaneously, relying on the filterless architecture. As an example, let us consider a network topology with four sublinks and the various possible aggregation levels: 1 (i.e., minimum aggregation), 1.33 (i.e., low degree of aggregation), 2 (i.e., moderate aggregation), and 4 (i.e., full aggregation). A high-level description of all these scenarios is illustrated in Fig. 13. E. OSNR-based performance evaluation Due to the statistical nature of this analysis and the conditions regarding the aggregation level at an optical level, each simulation of the metro-aggregation networks will produce different OSNR margin values depending on the characteristics of that particular instance: number of sublinks, number of leaf nodes, distance of the fiber segments, amount (and type) of transceiver units, level of aggregation, among others. Moreover, increasing traffic over time might require the operator to deploy additional transceivers at particular nodes in the network. This comes with the additional requirement of adding new couplers and splitters to ensure full connectivity in the filterless metro-aggregation network, which in turn modify the power levels across certain connections and, by extension, the OSNR conditions across the whole network. (a) Agg. Level 1 Downlink 12345678910 6 10 14 18 22 26 Year OSNR margin [dB] (b) Agg. Level 4 12345678910 6 10 14 18 22 26 Year OSNR margin [dB] Min. Max. Fig. 14. OSNR margin of the four-sublink network scenarios as a function of the year-in-operation and level of aggregation for a traffic CAGR of 20%. The color mapping has been normalized to the maximum value of occurring instances and the solid line highlights the mean OSNR margin value. Among the 1000 simulated different networks, the OSNR margin calculations for the four-sublink network scenarios presented in Fig. 13 (a) and (d), which correspond to the lower and upper limits with respect to optical aggregation, respectively, have been summarized in Fig. 14. The first aspect to notice is the time-dependency of the overall performance: the longer a filterless network remains in operation, due to the increasing traffic requirements and the subsequent deployment of new elements such as transceivers and optical couplers/splitters, the more the overall OSNR of the network tends to decrease. This, however, does not imply that all individual leaf nodes will experience the same degradation, but rather than as more connections experience lower OSNR levels, it is more likely that the limiting OSNR value (i.e., the worst-performing signal in terms of OSNR) is worse than that of a previous point in time. Secondly, by comparing the performance according to the aggregation level of the four-sublink networks, we observe how, by combining more sublinks and sharing hardware resources, the OSNR performance is affected (i.e. ,degraded). The reason behind this is, in principle, the same as the previous one: passive optical components are introduced to enable for more simultaneous connections. Figure 15 shows the average OSNR margin values across a variety of conditions over the years for different CAGR settings. These graphs allow us to observe not only the degradation of the Research Article Journal of Optical Communications and Networking 9 12 3 45678910 0 5 10 15 20 25 10% CAGR Year Avg. OSNR margin [dB] 12345678910 0 5 10 15 20 25 20% CAGR Year Avg. OSNR margin [dB] 12345678910 0 5 10 15 20 25 30% CAGR Year Avg. OSNR margin [dB] 12 3 45678910 0 5 10 15 20 25 40% CAGR Year Avg. OSNR margin [dB] 2H(1) 2H(2) 3H(1) 3H(1.5) 3H(3) 4H(1) 4H(1.33) 4H(2) 4H(4) Fig. 15. Average OSNR margin value as a function of the type of network and degree of resource sharing over the years according to the simulated traffic growth. In the legend, the numbers that precede and follow the letter ‘H’ indicate the number of horseshoes in the network scenarios and the level of aggregation, respectively. network’s performance over time, but also how it is affected— and even determined—by the rate at which traffic grows. Here, for instance, a 10% year-over-year CAGR shows an almost negligible variation over the years, indicating that there has been no need to upgrade the network considerably in terms of hardware (i.e., passive optical components or transceivers); whereas for more dire traffic conditions such as a 30% or 40% CAGR, it is clear that major network upgrades are needed to stay ahead of the increasing data demands. In filterless networks, being planned according to a ‘Pay-As-You-Grow’ approach, we must be aware of an eventual tightening of the performance margins, if the goal is to ensure a minimum-required transmission quality as time goes on. Condition Min. OSNR [dB] DSCs 400 Gb/s 24 16 200 Gb/s 20 8 100 Gb/s 17 4 50 Gb/s 14*2 25 Gb/s 11*1 Table 2. Required receiver OSNR for an optical transceiver in a P2MP edge aggregation scenario according to specifications by Open XR Optics Forum [ 37 ]. While values marked by * are not defined explicitly in this document, they can be linearly extrapolated from the stated data as an approximate value to investigate a fully-flexible operation of the transceivers. An interesting aspect of the results shown in Fig. 15 is the absolute value of the performance margin, especially in the early years. While an average OSNR value between 15 dB and 20 dB might appear large at first glimpse, it is necessary to put this into context: at start-of-life, the high-speed transceiver will most likely not be operating at full capacity and will therefore have a larger power budget than scenarios with more stringent traffic requirements. In this regard, Table 2 summarizes the OSNR specifications of DSCM-capable hardware to guarantee a post-FEC Bit-Error Rate (BER) performance of 10−15 as defined by the Open XR Optics Forum [37]. To support the previous claim of a scaled-down operation of the transceivers, their percentage of usage at the hub node can be used as a proxy to roughly identify the OSNR requirements of the pluggable modules. In this context, we define ‘usage percentage’ as the ratio between active DSCs (according to traffic demands) and the number of available DSCs. For example, if (a) CAGR 10% 12345678910 0 20 40 60 80 100 Year Transceivers’ usage Hub [%] (b) CAGR 40% 12 3 45678910 0 20 40 60 80 100 Year Transceivers’ usage Hub [%] 2H(1) 2H(2) 3H(1) 3H(1.5) 3H(3) 4H(1) 4H(1.33) 4H(2) 4H(4) Fig. 16. Mean usage of the bandwidth resources (i.e., DSCs) at the hub site as a function of time for a year-over-year CAGR of (a) 10% and (b) 40%, for different level of aggregation. a 400 Gb/s transceiver is only using four DSCs of the available sixteen, it exhibits a usage percentage of 25%. Now, should multiple transceivers be required at the node of the hub, the mean value among all available units will be calculated. The usage estimations in relation to the characteristics of the networks, the year of operation and the traffic growth conditions are displayed in Fig. 16. In these graphs, as expected, we can see an inverse behavior of the curves compared to Fig. 15: data corresponding to higher average OSNR margin values correlate with lower usage of the bandwidth resources of the transceivers. Furthermore, the evolution rates regarding the usage of transceivers, as years go by, showcase the activation of DSC and the deployment of new pluggable modules. F. Distance-performance relation By establishing a relation between the fiber segments of the networks and the corresponding optical performance, we can better understand the links between traffic, reach, and OSNR conditions. In this context, we shall understand ‘reach’ D as the mean value of the overall sum of all fiber segments between the hub node and the last leaf node among all possible sublinks. Using the terminology used in Fig. 5, we can express this value as follows: D=1 s· s ∑ j=1 Ns ∑ i=1 di;(2) where ‘s’ indicates the available sublinks in the network and Ns , the number of leaf nodes in a given sublink. In Fig.17, we have summarized the data from all simulated scenarios that concern a topology with three horseshoes to exemplify how traffic growth and optical aggregation conditions Research Article Journal of Optical Communications and Networking 16 J. Light. Technol. 38, 4744–4756 (2020). 19. C. Castro, A. Napoli, M. Quagliotti, J. Pedro, M. Porrega, and E. Riccardi, “OSNR-Based Hardware Optimization of a Filterless Point-toMultipoint Network Using Digital Subcarrier Multiplexing,” in 2024 International Conference on Optical Network Design and Modeling (ONDM), (Madrid, 2024), pp. 1–6. 20. C. Castro, S. Xia, A. Napoli, J. Pedro, Y. Rumaldo, N. Costa, N. Calabretta, B. Spinnler, and A. Rafel, “A Low-Cost Network Architecture Enabled by SOA-Based Filter-less OADMs and Digital Subcarrier Multiplexing,” in 2024 Optical Fiber Communications Conference and Exhibition (OFC), (Optical Society of America, San Diego, 2024), W2A. 21. “Coherent 100ZR,” https://www.adva.com/ (2022). 22. “Infinera’s ICE-X Portfolio of Intelligent Coherent Pluggables,” https://www.infinera.com/ (2022). 23. P. Torres-Ferrera, J. Sime, T. Duthel, E. Virgilito, V. Curri, R. Gaudino, C. Fludger, and A. Napoli, “Single-Fiber Bidirectional Transmission using 400G Coherent Digital Subcarrier Transceivers,” in Optical Fiber Communication Conference, (Optical Society of America, San Diego, 2024), Tu3E5. 24. A. Rashidinejad, A. Yekani, T. A. Eriksson, A. Napoli, R. Maher, A. Kakkar, V. Dominic, T. Duthel, M. Missey, P. Samra, D. Pavinski, P. Evans, W. Sande, M. Torbatian, C. Fludger, H. Sun, M. Ziari, F. Masoud, A. Somani, and D. Welch, “Real-Time Point-to-Multipoint for Coherent Optical Broadcast and Aggregation – Enabled by Digital Subcarrier Multiplexing,” in Optical Fiber Communication Conference (OFC) 2023, (Optical Society of America, San Diego, 2023), W3H1. 25. A. Napoli, M. M. Mezghanni, S. Calabro, R. Palmer, G. Saathoff, and B. Spinnler, “Digital predistortion techniques for finite extinction ratio IQ Mach–Zehnder modulators,” J. Light. Technol. 35 (19), 4289–4296 (2017). 26. P. W. Berenguer, M. Nölle, L. Molle, T. Rahman, A. Napoli, C. Schubert, and J. K. Fischer, “Nonlinear digital pre-distortion of transmitter components,” J. lightwave technology 34 (8), 1739–1745 (2015). 27. G. Rizzelli, P. Torres-Ferrera, A. Napoli, and R. Gaudino, “Filtering Effects Characterization in Metro-Access Networks and Future HighSpeed Coherent Optical Communication Schemes,” J. Light. Technol. 43 (3), 1114–1122 (2025). 28. P. Torres-Ferrera, G. Parisi, G. Rizzelli, R. Gaudino, and A. Napoli, “Analytical Model for Performance Estimation of Coherent Digital Subcarrier Multiplexing Systems Affected by Filtering, Optical and Electrical Noise,” in 2024 International Conference on Transparent Optical Networks (ICTON), (IEEE, Bari, 2024), pp. 1–5. 29. J. Bäck, J. Pedro, T. Schaich, A. Napoli, P. Wright, A. Chase, D. Welch, and A. Lord, “Hubbedness: a Metric to Describe Traffic Flows in Optical Networks and an Analysis of its Impact on Efficiency of Pointto-Multipoint Coherent Transceiver Architectures,” in 2021 European Conference on Optical Communication (ECOC), (IEEE, Bordeaux, 2021), Th2A1, pp. 1–4. 30. F. Masoud, A. Napoli, C. Castro, J. Pedro, B. Spinnler, A. Chase, D. Hillerkuss, and D. Welch, “Environmental Impact of Coherent Pointto-Multipoint Pluggables in Metro Aggregation Optical Networks,” in 2023 International Conference on Optical Network Design and Modeling (ONDM), (Coimbra, 2023), pp. 1–6. 31. “The Seven Vectors of ROADM Evolution,” https://www.infinera.com/. 32. “Open ROADM Network Model,” http://openroadm.org/ (2020). 33. J. Bäck, A. Napoli, E. Riccardi, M. Quagliotti, M. Porrega, J. Pedro, T. A. Eriksson, F. Masoud, A. Mathur, and D. Welch, “A filterless design with point-to-multipoint transceivers for cost-effective and challenging metro/regional aggregation topologies,” in 2022 International Conference on Optical Network Design and Modeling (ONDM), (IEEE, 2022), pp. 1–6. 34. J. A. Hernández, F. Arpanaei, A. Napoli, C. Castro, Óscar González de Dios, and J. P. Fernández-Palacios, “On Clustering Coherent Optics Point-to-multipoint Trees for Cost-effective Bandwidth Assignment in MANs,” J. Opt. Commun. Netw. 15 (12), 999–1007 (2023). 35. M. Balanici, P. Safari, B. Shariati, A. Jafari, J. K. Fischer, and R. Freund, “Live Demonstration of Autonomous Link-Capacity Adjustment in Optical Metro-Aggregation Networks,” in 2024 Optical Fiber Communications Conference and Exhibition (OFC), (Optical Society of America, San Diego, 2024), M3Z. 36. E. W. Stacy, “A generalization of the gamma distribution,” The Annals mathematical statistics pp. 1187–1192 (1962). 37. “Open XR Optics Transceiver Optical and Client Interface Specification OXR.TCVROPTCLI.01.0,” https://www.openxrforum.org/ (2024). 38. “Cisco QSFP28 100ZR,” https://www.cisco.com/ (2024). 39. “Nokia 100G ZR Coherent Module,” https://www.nokia.com/ (2024).