Evaluation of the INSPACE system benefits - CORDIS · Figure 3.10 Two M×N WSS-based A/D module...

FP7-ICT-GA 619732

`SPATIAL-SPECTRAL FLEXIBLE OPTICAL NETWORKING ENABLING SOLUTIONS FOR A SIMPLIFIED AND EFFICIENT

SDM

SPECIFIC TARGETED RESEARCH PROJECT (STREP) INFORMATION & COMMUNICATION TECHNOLOGIES (ICT)

Evaluation of the INSPACE system benefits

D2.3

Document Type : Deliverable

Dissemination Level : PU

Lead Beneficiary : AIT

Contact Person : Ioannis Tomkos

[email protected]

Delivery Due Date : 31/01/2017

Submission date : 21/03/2017

Contributing institutes : AIT

Authors : Behnam Shariati (AIT)

This deliverable reports on the benefits of introducing of the INSPACE solutions in an optical network. In this

deliverable we provide extensive studies and analysis with the purpose to: a) identify the system performance in

different scenarios, b) calculate the expected cost benefits from the adoption of the INSPACE solutions, and c)

evaluate the energy consumption of the nodes in networks. In all cases, comparisons are made with legacy

technologies with the main purpose to identify the expected economic benefits and the improvements in power

consumption.

Ref. Ares(2017)3331579 - 03/07/2017

INSPACE D2.3 Evaluation of the INSPACE system benefits Version 3.0

FP7-ICT-GA Dissemination level: PU 2 | 33

Revision History

No. Version Author(s) Date

1 1.0 Behnam Shariati (AIT) 16/02/2017

Comments: Initial Release, ToC


Comments: Submitted for internal review


Comments: Final submitted version

4 4.0

Comments:

5 5.0

Comments:

6 6.0

Comments:

7 7.0

Comments:

Comments:

Participants

The INSPACE Project Consortium groups the following Organizations:

No Partner Name Short Name Country

1 Optronics Technologies S.A. OPT Greece

2 Telefonica Investigation y Desarrollo TID Spain

3 The Hebrew University of Jerusalem HUJI Israel

4 Research and Education Laboratory in Information Technologies AIT Greece

5 Optoscribe Ltd. OPTOSCRIBE United Kingdom

6 Center for Research and Telecommunication Experimentation for

Networked Communities CN Italy

7 Aston University ASTON United Kingdom

8 Opsys Tech Ltd. OPSYS Israel

10 W-Onesys, S.L. WONE Spain



List of Tables

Table 2.1 Values of selected ChBW with the amount of corresponding spectral contents supported by two

WSS technologies ............................................................................................................................................................. 12

Table 2.2 Maximum point-to-point transmission reach (km) with the indicated baud rate, ChBW, and

modulation format ........................................................................................................................................................... 12

Table 3.1 Estimated cost of HPC WSSs ....................................................................................................................... 18

Table 3.2 Relationship between maximum number of ports and number of LCoS SLM pixels. N_px is the

total number of pixels in the steering direction. Px_port is the number of pixels per port................................. 19

Table 3.3 Loss of splitters and switches with different splitting ratios to be used in realizing MCS with various

port counts ........................................................................................................................................................................ 20

Table 3.4 The required number of WSSs and their port count ................................................................................ 23

Table 3.5 Cost of the A/D modules presented in Figure 3.9 to Figure 3.11 ......................................................... 24

Table 3.6 CMOS power dissipation based on Fig. 2 of [7] ....................................................................................... 26

Table 3.7 Power-limited reach based on total consumption per 600 Gb/s module ............................................. 27

Table 3.8 Total power consumption per 600 Gb/s module. .................................................................................... 27



List of Figures

Figure 2.1 Cumulative density function (CDFs) of the assumed traffic profiles with (a) fixed σ=200 Gbps and

µ=700, 1150, and 1600 Gbps, (b) fixed µ=1248 Gbps, and σ=96, 192, 384, 768 Gbps. ..................................... 10

Figure 2.2 Blocking probability in terms of average number of live connections per A/D node for three

profiles of traffic forming of small, large and medium size demands which their distributions are plotted in

Fig. 13-5: a) µ=1600, b) µ=1150, and c) µ=700 Gbps. .............................................................................................. 11

Figure 2.3 BP in terms of standard deviation when µ=1248 Gbps and average number connection per A/D

node is 112. ........................................................................................................................................................................ 11

Figure 2.4 Average spectrum utilization per link per fiber under different spatial Sp-Ch switching paradigms

considering current WSS technology for (a) 50 GHz, (b) 37.5 GHz, and (c) 25 GHz ChBWs at a baud rate of

32, 19.5 and 7 Gbaud, respectively in terms of load. .................................................................................................. 13

Figure 2.5 Average data occupancy in percentage considering current WSS technology for (a) 50 GHz, (b)

37.5 GHz, and (c) 25 GHz ChBWs at a baud rate of 32, 19.5 and 7 Gbaud, respectively in terms of load. .... 13

Figure 2.6 Average spectrum utilization per link per fiber under different spatial Sp-Ch switching paradigms

considering improved resolution WSS technology for (a) 50 GHz, (b) 37.5 GHz, and (c) 25 GHz ChBWs at a

baud rate of 41, 28.5 and 16 Gbaud, respectively in terms of load. ......................................................................... 15

Figure 2.7 Average data occupancy in percentage considering finer resolution WSS technology for (a) 50

GHz, (b) 37.5 GHz, and (c) 25 GHz ChBWs at a baud rate of 41, 28.5 and 16 Gbaud, respectively in terms of

load. .................................................................................................................................................................................... 15

Figure 2.8 Average spectrum utilization per link per fiber for J-Sw considering: (a) the current WSS

technology which requires 18 GHz for guard band and (b) finer resolution WSS which requires 9 GHz for

guard band. Five values of ChBWs are assumed for the simulations. The amount of spectral contents that can

be utilized considering different WSS technologies is provided in Table I. ........................................................... 16

Figure 3.1 single-carrier transceiver ............................................................................................................................... 17

Figure 3.2 Integrated spatial super-channel transceiver with seven integrated sub-channels .............................. 17

Figure 3.3 Relative cost/price of WSS required for a given number of ports........................................................ 18

Figure 3.4 Port-reconfigurable MN WSS ................................................................................................................... 19

Figure 3.5 A possible design of MC-EDFA ................................................................................................................ 20

Figure 3.6 Average spectrum utilization per link per fiber in terms of total offered load to the network in

Pbps .................................................................................................................................................................................... 21

Figure 3.7 Total relative cost of the network considering 4 different offered load for 5 SDM deployments

compare to an equivalent parallel system using 100G TRx. ...................................................................................... 21

Figure 3.8 Route and Select ROADM architectures for (a) Ind-Sw and (b) J-Sw in an SDM scenario with S=2.

In (a) we show all internal connectivity from the ‘East’ ingress WSS. LCs are supported if all solid, dashed and

dotted connections exist. Ind-Sw without LCs requires solid and dashed lines. ................................................... 23

Figure 3.9 Two A/D module architectures for ROADM CD operation. Only ‘Drop’ WSS are shown. ......... 23

Figure 3.10 Two M×N WSS-based A/D module architectures for ROADM CDC operation. Only ‘Drop’

WSS are shown. ................................................................................................................................................................ 23

Figure 3.11 Three MCS-based A/D module architectures for ROADM CDC operation. Only ‘Drop’ MCS

are shown. .......................................................................................................................................................................... 24

Figure 3.12 Relative cost of the components wrt 1×9 WSS ..................................................................................... 24

Figure 3.13 Total relative cost (TRC) of different ROADM implementations. .................................................... 25

Figure 3.14 MIMO-DSP complexity of FDE ............................................................................................................. 26

Figure 3.15 Predicted (a) power consumption per mode and (b) total consumption per 6×100Gb/s module,

for 15 nm CMOS.............................................................................................................................................................. 27

Figure 3.16 Predicted (a) power consumption per mode and (b) total consumption per 6×100Gb/s module,

for 15 nm CMOS.............................................................................................................................................................. 27

Figure 3.17 Power consumption of control unit of WSSs needed for realizing SDM nodes .............................. 28





Table of contents

Revision History ................................................................................................................................................ 2

Participants ......................................................................................................................................................... 2

List of Tables ..................................................................................................................................................... 3

List of Figures .................................................................................................................................................... 4

Table of contents ............................................................................................................................................... 6

Executive summary ........................................................................................................................................... 7

1. Introduction.......................................................................................................................................... 8

2. Networking level performance evaluations ................................................................................... 10

2.1. Impact of traffic profile on the performance of SDM switching schemes ............................... 10

Simulation environment and assumptions ........................................................................................ 10

Results and discussions ......................................................................................................................... 10

2.2. Impact of spectral and spatial granularity on the performance of INSPACE proposed

switching schemes ............................................................................................................................................. 12

Simulation environment and assumptions ........................................................................................ 12

Results and discussions ......................................................................................................................... 13

3. Cost analysis and power consumption evaluations ...................................................................... 17

3.1. Cost benefit quantification of INSPACE proposed solutions ................................................... 17

Cost model .............................................................................................................................................. 17

Network-level cost analysis .................................................................................................................. 20

3.2. Comparison of CD(C) ROADM architectures for SDM networks........................................... 22

Motivation ............................................................................................................................................... 22

ROADM architectures and cost analysis ........................................................................................... 22

Discussion on the total relative cost of ROADM implementations ............................................ 25

3.3. Power consumption of MIMO processing and its impact on the performance of SDM

networks .............................................................................................................................................................. 25

Motivation ............................................................................................................................................... 25

Power consumption of MIMO processing for SDM ...................................................................... 26

Network-wide performance evaluation of integrated transceiver utilizing different MIMO-

DSP modules ......................................................................................................................................................... 28

3.4. Control unit power consumption of WSSs used for realizing SDM nodes .............................. 28

Conclusions ...................................................................................................................................................... 29

Annex – abbreviations and acronyms .......................................................................................................... 30

References ........................................................................................................................................................ 32



Executive summary

The INSPACE project aims at extending the established spectral flexibility concept to the space domain, by

proposing novel switching solutions that simplify the Super-Channel (Sp-Ch) allocation mechanisms and route them

transparently in an optical network. Three switching paradigms have been identified and developed in INSPACE,

which guarantee various levels of spatial granularity (i.e. grouping of the spatial channels) exploiting wavelength

selective switches (WSSs) of different port counts. The core novelty is the development of a high port count (HPC)

WSS capable of switching all spatial dimensions jointly with the expected benefit of reducing the complexity/cost as

well as providing appropriate switching element when highly coupled transmission media (i.e. few mode fibers

(FMFs)) are in place. The proposed switching element trades off some level of flexibility for the sake of

architectural simplicity and satisfying, at the same time the major switching requirements. Therefore, it is very

important to investigate it in different scenarios showcasing its expected benefits in terms of performance, cost, and

power consumption as is promised in T2.3 of INSPACE.

In this deliverable we thoroughly investigate SDM networks utilizing the proposed switching schemes and

compare it with the conventional schemes with the purpose of showing their benefits/drawbacks in different

networks and under various traffic profiles. We show that the performance of Joint Switching (J-Sw) paradigm

converges to that of Independent Switching (Ind-Sw) as traffic increases. In other words, we show that the J-Sw is a

favorable switching scheme for networks with large demands, where the size of the demands is comparable with the

capacity of spatial Sp-Chs, while Ind-Sw and Fractional Joint Switching (FrJ-Sw) cases are favorable options for

networks with high levels of traffic diversity. We further show that the performance of J-Sw can be improved when

WSSs capable of switching the spectral content with finer spectral granularity are in place. For this reason, we report

extensive computer simulations comparing SDM proposed switching schemes with conventional approaches. We

take into account WSS capable of switching at various spectral granularity as well as WSSs with two levels of

spectral resolution (i.e. the current technology and an improved resolution WSS technology).

In addition, we perform a network wide cost analysis considering spatially integrated elements to showcase the

cost benefit of SDM solutions. For this purpose, we develop a cost model based on the current available technology

and complementary estimations provided by INSPACE industrial partners. We show that the INSPACE switching

solution can bring up to 50% cost savings in the development of reconfigurable optical add drop multiplexer

(ROADM) line side. We further show that utilizing J-Sw case, cost benefits can be expected from other components

like spatially integrated TRx and amplifiers. We also conclude that exploiting J-Sw scheme can bring huge cost

savings for the add/drop unit of ROADMs. We propose several colorless directionless (CD) and CD contention-less

(CDC) architectures where J-Sw case can contribute to extra cost savings. Finally, we present results on the power

consumption benefits of SDM proposed nodes, more pronounced for the realization of the control unit of the HPC

WSSs.



1. Introduction

Traffic in telecommunication networks is growing at an annual rate of 20-60% and is approaching the capacity

limits of the single-mode fiber (SMF) [1]. Space division multiplexing (SDM) over multi-core fibers (MCF), multi-

mode fibers (MMF), few-mode multi-core fibers (FM-MCF), or even bundles of SMFs would allow the network

capacity to scale to orders of magnitude higher than what can be achieved with an SMF-based network

infrastructure.

SDM systems would also reduce the cost per bit delivered to the end users, compared with parallel-fiber systems, by

sharing network elements, taking advantage of dense optical integration (SDM transceivers (TRx), switching

elements and in-line amplifiers) among different spatial dimensions. However, the denser the integration of the

spatial channels it becomes, the more significant is the crosstalk (XT) interactions among them, as in the case for

strongly-coupled MCFs and FMFs. Such spatial XT is expected to be mitigated by multiple-input multiple-output

(MIMO) digital signal processing (DSP) at the TRx side at the expense of higher complexity and power

consumption.

SDM TRx integration allows generating spatial super-channels (Sp-Ch), which can be defined, similar to

spectral Sp-Chs, as the aggregation of signals modulated on a certain optical carrier across a number or all of the

spatial dimensions of an SDM transmission medium [2]. Spectral and spatial Sp-Ch allocation policies were

compared in [2] for single and multiple modulation format transmission, showing that even though spectral Sp-Ch

transmission generally leads to superior network performance, resulting from increased flexibility, networks based

on spatial Sp-Ch allocation can benefit from cost savings related to the possibility of sharing switching elements

among spatial dimensions.

Three switching strategies were proposed in INSPACE: (a) independent switching (Ind-Sw): all spectral slices

and spatial dimensions can be independently directed to any output port; (b) joint switching (J-Sw): all spatial

dimensions are treated as a single entity, while spectral slices can be freely switched by the WSS; and (c) fractional

joint switching (FrJ-Sw): a hybrid approach in which a number of subgroups of G spatial dimensions, as well as all

spectral slices, can be independently switched to all output ports. The last two paradigms can be categorized as

spatial group switching (SG-Sw) solutions since the spatial resources are switched in groups rather than

independently, as in the case of Ind-Sw. Note that several spatial switching granularities result from different levels

of grouping of the spatial dimensions: from G=1, which assumes individual fibers, thus corresponding to the Ind-Sw

case and offering the finest spatial granularity, all the way through to G = S, which considers all spatial dimensions

as one spatial group, thus corresponding to the J-Sw case and offering the coarsest spatial granularity. In planning an

SDM network, the choice of one of the above SDM switching paradigms has a considerable impact on both the

flexibility of the implemented resource allocation (RA) policies and the switching infrastructure deployment cost.

SDM amplifier integration has also been investigated. Various schemes for pumping of Erbium Doped Fibers

(EDF) have been explored: i) core pumping using individual single-mode pump diodes, ii) shared pump, and iii)

cladding-pumping using a single high power (over 1 W) multimode pump laser diode [3]. Cladding pumping

requires fewer optical components and has the potential to allow the use of low-cost, energy efficient multimode

diodes. Bundled EDFs, consisting of bundles of identical single-core EDFs, are one candidate amplification medium

for SDM EDF amplifiers (EDFA). Few-mode and multi-core EDFAs enable cladding pumping with a single laser

diode. The non-uniformity of the modal gain and the noise figure (NF) between modes/cores are issues that need to

be further improved, but FM- and MC-EDFAs are indeed promising solutions for cost-effective SDM network

deployments since mode mixing can be made negligible [3].

In this deliverable, we present a comparative performance evaluations of SDM networks based on INSPACE

proposed switching solutions and the legacy approaches. In particular, we investigate the impact of spatial switching

granularity and spectral switching granularity on the performance and the implementation cost of SDM switching

schemes. The spectral switching granularity is related to the capabilities of wavelength selective switches (WSSs).

In order to reflect the impact of different technologies on the performance of INSPACE proposed switching

schemes, we consider two WSS technologies for handling of the SDM switching paradigms: 1) the current WSS

realization, 2) WSS technology with a factor-two resolution improvement. Moreover, with the aim of exploring the

performance of SDM switching schemes in different parts of networks (i.e. national, regional, or inter-data center

networks), we investigate the impact of different traffic profiles on their performance. Then, after deriving a cost

model for all network elements, we present cost analysis of SDM networks with the main purpose of showcasing the

benefits brought about by INSPACE proposed switching nodes. For the cost analysis, we consider spatial and

spectral Sp-Ch integrated TRx, amplifier and node implementations for Ind-Sw (spectral and spatial Sp-Ch), J-Sw

and FJ-Sw (spatial Sp-Ch). We also compare them to a parallel-fiber system based on 100G transmission. We

quantify the values of cost savings, under given assumptions, for several cases of parallel-fiber system deployments,

and discuss how the cost of the system can be further reduced. Furthermore, some efforts are made to estimate the

power consumption of different SDM approaches as well as the power consumption due to the control unit of

INSPACE proposed WSSs.

The rest of the deliverable is structured as follows:



In chapter 2, we present extensive computer simulation results evaluating the performance of INSPACE

proposed switching paradigms under different traffic profiles and considering various spectral and

spatial granularity as well as two different WSS technologies.

In chapter 3, we present cost and power consumption analysis showcasing the benefits of INSPACE

proposed schemes in comparison with the conventional approaches.

Finally, we present the conclusions of the deliverable.



2. Networking level performance evaluations

2.1. Impact of traffic profile on the performance of SDM switching schemes

A detailed comparison of SDM switching schemes was reported in D5.2 of INSPACE for an offline network

planning scenario. The study considered a specific traffic matrix with a limited number of demands. The comparison

revealed that under high traffic load the networking performance in terms of spectrum utilization for the case of FrJ-

Sw and J-Sw (i.e. spatial group switching options) almost converges to that of Ind-Sw.

The assumed traffic profile (i.e. large aggregated demands) is the most typical traffic profile in core networks

today, as we have traffic aggregation at the edge of the network. However, due to the introduction of application-

centric services and dependency of traffic increase on i) the type of network (access/metro/core), ii) the kind of

connectivity services offered by the network (e.g. based on 4G/5G radio, FTTX, etc.), and iii) the type of offered

applications (e.g. file sharing, video conferencing), traffic de-aggregation will be a possible networking policy.

Therefore, in this section, we want to study the cases in which lightpaths might be established based on i) a large

number of small demands (typically seen in regional part of networks), ii) a small number of large demands

(applicable for example in inter-datacenter communications), and iii) any combinations of last two options (found

e.g. in national scale networks serving a heterogeneous type of traffic demands).

In this section, we thoroughly evaluate the impact of various traffic profiles and dimensioning approaches on the

performance of SDM switching paradigms, in an online operation scenario. We show that the performance of the

three SDM switching paradigms is highly dependent on the traffic profile. While J-Sw shows similar performance as

Ind-Sw for large demands, it presents a reduced performance when network is fed by a large number of small

demands. However, we show that the performance of the J-Sw can improve substantially when the spectral

granularity of the switching paradigm is reduced.

Simulation environment and assumptions

In this study we use the Spanish backbone model of Telefónica which is the reference network considered for

INSPACE performance evaluations studies. It comprises 30 nodes (average nodal degree 3.7, max. 5), 14 of which

with add/drop capability (A/D), as well as 56 links with an average length of 148 km. In order to have a fair

comparison among the three SDM switching paradigms, regardless of any transmission medium related performance

constraints, bundles of 12 SMFs were considered for all links in the network.

Moreover, based on the network characteristics and the related performance evaluation studies [1], DP-8QAM at

32-Gbaud was chosen as the modulation format offering the best compromise between transmission reach and

spectral efficiency. A 50-GHz channel spacing is used.

According to the above and considering an available spectrum per fibre equal to 4.8 THz (C-band) on the ITU-T

12.5-GHz grid, discrete event simulation studies were carried out for the purpose of performance evaluation

exploiting the online simulator reported in D5.2 of INSPACE. In the simulator, the routing, space, and spectrum

allocation (RSSA) problem is solved with a k-shortest path (k = 3) and using a spatial and spectral resource

allocation algorithm, that follows a first-fit strategy, starting with the shortest computed path. The load generation

followed a Poisson distribution process. Traffic demands for each source-destination pair were generated randomly

following a normal distribution with mean µ and standard deviation σ over the range of study, namely 50 Gbps to

2.25 Tbps. Blocking Probability (BP) was used as a quantitative performance measure.

Figure 2.1 Cumulative density function (CDFs) of the assumed traffic profiles with (a) fixed σ=200 Gbps and µ=700, 1150, and 1600 Gbps, (b) fixed µ=1248 Gbps, and σ=96, 192, 384, 768 Gbps.

Results and discussions

In the first part of the study, we consider three traffic profiles corresponding to the three previously mentioned

cases: small, large and medium-size demands. The distribution of demands for the 3 different mean demand values

and fixed σ is shown in Figure 2.1(a), while the effect of the deviated demands over a fixed mean value is shown in

0%

20%

40%

60%

80%

100%

0 576 1152 1728 2304

Perc

enta

ge o

f C

DF

Average Bit Rate per Connection (Gbps)

σ=96 Gb/s

σ=192 Gb/s

σ=384 Gb/s

σ=768 Gb/s

(b)



Figure 2.1(b). The lower and upper mean values were chosen according to the following: a) for µ=700 Gbps, 98% of

demands requires less than half of the 12 spatial dimensions (i.e. bundles of SMFs (BuSMFs) in our case study) to

be allocated, b) for µ=1600 Gbps, we have large aggregated demands that result in more than 98% of them requiring

more than half of the 12 spatial dimensions to be allocated. The three traffic profiles of Figure 2.1(a) are used to

obtain the results shown in Figure 2.2. It is noted that in all cases traffic dimensioning is realized by varying the

number of live connections per A/D node.

Figure 2.2 Blocking probability in terms of average number of live connections per A/D node for three profiles of traffic forming of small, large and medium size demands which their distributions are plotted in Fig. 13-5: a) µ=1600, b) µ=1150, and c)

µ=700 Gbps.

For high mean traffic demands (Figure 2.2(a)), the three switching paradigms show the same performance. Since

most demands require more than half of the spatial resources, the unutilized resources of FrJ-Sw and Ind-Sw cases

cannot be allocated thus leading to the same results as the J-Sw case. For a traffic profile with diverse and relatively

medium-size demands (Figure 2.2(b)), FrJ-Sw and Ind-Sw start performing better than J-Sw in terms of BP, since

now part of the incoming demands that require less than 6 spatial dimensions to be allocated can fit within the free

spatial resources that FrJ-Sw and Ind-Sw enable them to use. For small mean traffic demands (Figure 2.2(c)) the

performance difference between J-Sw and FrJ-/Ind-Sw is more pronounced, since the allocation options in space

dimension are increased and small demands can fit in available spatial slots.

Previous discussion on Figure 2.2 strongly suggests that the optimal switching paradigm for an SDM network, in

fact, depends on the nature of its traffic, specifically whether there is a prevalence of relatively small or large

demands. However, since most of the demands in the core networks are aggregated traffics, J-Sw would be a

suitable choice, considering its cost benefit which will be shown in the next chapters.

In order to see the impact of traffic diversity on the performance of SDM switching paradigms, a complementary

set of simulations is carried out, where the traffic dimensioning is done by varying σ and keeping µ and the number

of live connections fixed (Figure 2.1(b)). Note, in this study, the total offered load to the network during the whole

range of the simulation is fixed to 14*112*1248 Gbps = ~1.95 Pbps.

Figure 2.3 BP in terms of standard deviation when µ=1248 Gbps and average number connection per A/D node is 112.

Results plotted in Figure 2.3 show that at the beginning the performance of three SDM switching is the same

(similar to Figure 2.2(a)), and by increasing σ which is equivalent to the diversity level of the traffic profile, Ind-Sw

and FrJ-Sw show a remarkable improvement which justifies their suitability for networks with high level of

diversity in their traffic profile. In conclusion, the performance of the three SDM switching paradigms is highly

dependent on the traffic profile. While J-Sw shows similar performance as Ind-Sw for large demands, it presents a

reduced performance when network is fed by a large number of small demands. J-Sw shows better performance for

large demands, because when the load increases the capacity of a spatial Sp-Ch becomes comparable to the demand

that has to be served. Therefore, if we can form spatial Sp-Chs with lower capacity spreading across all spatial

dimensions, smaller demands can fill most of Sp-Ch container and, thus, reduce the unutilized spectral and spatial

resources compared to the spatial Sp-Chs with higher capacity. One of the required technology to realize narrower

spatial Sp-Chs is the WSSs with finer spatial granularity capable of switching narrower spectral bands. In next

0.0001

0.001

0.01

0.1

1

75 100 125 150 175 200 225 250 275

BP

Average Number of Live Connections per A/D Node

J-Sw

FrJ-Sw

Ind-Sw(a)0.0001

0.001

0.01

0.1

1

75 100 125 150 175 200 225 250 275

BP


J-Sw FrJ-Sw Ind-Sw(b)

0.0001

0.001

0.01

0.1

1

75 100 125 150 175 200 225 250 275

BP


J-Sw

FrJ-Sw

Ind-Sw (c)

0.0001

0.001

0.01

0.1

1

24 48 96 192 384 768

BP

Standard Deviation of the Traffic Profile (Gbps)

J-Sw

FrJ-Sw

Ind-Sw



section, we investigate the impact of spatial and spectral granularity on the performance of SDM networks based on

spatial Sp-Ch switching.

2.2. Impact of spectral and spatial granularity on the performance of INSPACE proposed switching schemes

Assuming bundles of 12 SMFs, we showed in D5.2 of INSPACE that the performance of SG-Sw cases becomes

similar to that of Ind-Sw as the total offered load to the network (hereinafter referred to as load) increases. However,

SG-Sw cases showed a reduced performance for low values of load. In the previous section, we further showed that

SG-Sw paradigms with lower G values perform well for networks with high level of traffic diversity, while J-Sw is

favorable for networks with large demands. Note that in the previous studies, Ind-Sw, J-Sw and FrJ-Sw with G = 3

were only examined with a fixed spectral channel width (ChBW) —defined as the spectrum over each of the spatial

dimensions used to allocate the spatial Sp-Ch constituents— equal to 50 GHz. In this section, in addition to the three

levels of spatial switching granularity studied previously, we examine FrJ-Sw with G equal to 2, 4, and 6. The

spectral switching granularity is related with the capabilities of WSSs. Current WSS technology allows occupying

32-GHz on a 50 GHz grid due to channel pass bandwidth imposed by the WSS resolution. The same WSS resolution

can allocate finer channels, typically according to a 6.25 GHz grid, i.e., 25.75 GHz can be provisioned on 43.75

GHz, 19.5 GHz on 37.5 GHz, 13.25 GHz on 31.25, or 7 GHz on 25 GHz. In this section, we investigate the impact

of the spectral switching granularity on the performance and cost of spatial Sp-Ch switches based on two practical

WSS technologies: 1) the current generation WSS realization, 2) a WSS technology with a factor-two resolution

improvement (i.e. requiring 9 GHz for guard band instead of the 18 GHz considered above). A summary of the

ChBW values selected for this study and the corresponding clear channel bandwidth that can be allocated with data

and switched by current and future (factor-two resolution improvement) WSS realizations is provided in Table 2.1.

Table 2.1 Values of selected ChBW with the amount of corresponding spectral contents supported by two WSS technologies

Simulation environment and assumptions

The network simulations were carried out over the Telefónica Spain backbone network model. Even though

MCFs/MMFs/FM-MCFs are the ultimate transmission media for SDM networks, to ensure a smooth migration from

currently deployed networks to SDM, network operators will seek to leverage their current infrastructure by

exploiting the capacity increase enabled by parallel transmission through BuSMFs. These have the advantage that

the transmission is not affected by XT between spatial dimensions (fibers), and SDM multiplexers/demultiplexers

are not required for component interconnection. Assuming BuSMFs also allows us to make a fair comparison

between different spatial Sp-Ch switching paradigms, due to the fact that BuSMFs are the only type of SDM

transmission medium compatible with all switching paradigms. We therefore limited the network performance study

to the case of BuSMFs and use the offline planning platform presented in D5.2 of INSPACE to perform the

simulations. Regarding the transmission technology, single carrier (SC) multi-channel (MC) multi-line rate (MLR)

systems were considered, in which the MLR behavior is achieved by changing the number of spatial channels and/or

by employing different modulation formats. The choice of modulation format is limited to dual-polarization (DP) –

BPSK, QPSK, 8QAM and 16QAM, with maximum transmission reach calculated by means of the Gaussian Noise

(GN) model of nonlinear interference in coherent (Nyquist) WDM systems proposed in 32. The obtained reach

values are presented in Table 2.2.

Table 2.2 Maximum point-to-point transmission reach (km) with the indicated baud rate, ChBW, and modulation format

Baud rate in GSamp/s ChBW in GHz DP-BPSK DP-QPSK DP-8QAM DP-16QAM

32 50 9800 4900 1900 900

25.75 43.75 10800 5400 2100 1000

19.5 37.5 12300 6100 2400 1200

13.25 31.25 15400 7700 3000 1500

7 25 16650 8300 3250 1650

41 50 7300 3500 1200 600

34.75 43.75 7800 3700 1400 700

28.5 37.5 8300 4000 1500 800

22.25 31.25 9200 4500 1800 900

16 25 10500 5200 2000 1000

Spectral Channel plan [GHz] 50 43.75 37.5 31.25 25

Current generation WSS resolution

(clear channel BW and % utilization) 32 25.75 19.5 13.25 7

64% 59% 52% 42% 28%

Improved resolution WSS technology (clear channel BW and % utilization)

41 34.75 28.5 22.25 16

82% 79% 76% 71% 64%



Results and discussions

In this section, we first compare the performance of spatial Sp-Ch switching paradigms under different spatial and

spectral granularities in a network planning scenario for the Telefónica Spain backbone network model. We assume

bundles of 12 SMFs with 4.8 THz available spectrum per fiber across all links as a near-term SDM solution. The

performance evaluation is done in terms of load and its growth. Since there is a fixed number of demands in the

traffic matrix (84 demands), traffic growth is achieved by increasing the size of the demands. In order to perform the

studies, we scale the load (i.e. total offered load to the network) up to 1 Pbps, which is equivalent to 10-year total

traffic growth, assuming 45% annual traffic increase. For the spatial switching granularity, we consider groups (G)

of 1, 2, 3, 4, 6 and 12 fibers out of 12 fibers in BuSMFs, where G = 1 and 12 correspond to the cases of Ind-Sw and

J-Sw, respectively, which offer the finest (Ind-Sw) and the coarsest (J-Sw) spatial granularities. Intermediate values

represent FrJ-Sw with spatial groups formed of 2-6 SMFs.

Figure 2.4 Average spectrum utilization per link per fiber under different spatial Sp-Ch switching paradigms considering current WSS technology for (a) 50 GHz, (b) 37.5 GHz, and (c) 25 GHz ChBWs at a baud rate of 32, 19.5 and 7 Gbaud,

respectively in terms of load.

Figure 2.5 Average data occupancy in percentage considering current WSS technology for (a) 50 GHz, (b) 37.5 GHz, and (c) 25 GHz ChBWs at a baud rate of 32, 19.5 and 7 Gbaud, respectively in terms of load.

Figure 2.4 shows the results considering present-day WSS technology requiring 18 GHz guard band. Figure 2.4(a)

presents the results for the case of fixed-grid 50 GHz WDM ChBW, in which 32 Gbaud is selected for the contents

0

20

40

60

80

100

0 0.2 0.4 0.6 0.8 1

Av

erag

e dat

a occ

upan

cy w

ith

in

the

allo

cate

d w

avel

ength

chan

nel

s

Load in Pbps

G=12 G=6

G=4 G=3

G=2 G=1

(a)

0

20

40

60

80

100

0 0.2 0.4 0.6 0.8 1

Av

erag

e d

ata

occ

up

ancy

wit

hin

the

allo

cate

d w

avel

eng

th c

han

nel

s

Load in Pbps

G=12 G=6

G=4 G=3

G=2 G=1

(b)

0

20

40

60

80

100

0 0.2 0.4 0.6 0.8 1

Av

erag

e d

ata

occ

up

ancy

wit

hin

the

allo

cate

d w

avel

eng

th c

han

nel

s

Load in Pbps

G=12 G=6

G=4 G=3

G=2 G=1

(c)



of each ChBW as it is the maximum baud rate supported by present-day WSS resolution. The average spectrum

utilization per link per fiber is used as a quantitative network performance metric. Thus, for example, J-Sw with the

coarsest granularity in our studies corresponds to G = 12 and ChBW = 50 GHz, allows for 32GHz×12=384 GHz for

data loading, corresponding to 3072 Gb/s assuming DP-16QAM format. Note that, DP-16QAM is the most used

modulation format, as most of the lightpaths in the Telefónica network can be established within its optical reach

limit. Demands smaller than that will simply occupy the whole spatial-spectral slot, resulting in low data occupancy

within the available bandwidth. At the finest granularity of G = 1 and ChBW = 50 GHz, the equivalent minimum

spatial-spectral bandwidth slot amounts to 256 Gb/s.

In Figure 2.4(a), Ind-Sw shows the best performance for all loads, i.e, lowest utilization for the given traffic load,

since it offers the finest granularity. We use it as the benchmark to estimate the unutilized bandwidth due to

grouping of spatial dimensions. The performance of the rest of spatial Sp-Ch switching paradigms is seen to

converge to that of Ind-Sw as load increases. This is due to the fact that when the load increases the spectral-spatial

slot becomes comparable to the demand that has to be served and therefore the amount of unutilized bandwidth due

to SG-Sw reduces. Additionally, we observe that, independently of the load, but more noticeably for smaller loads,

the curves for the SG-Sw cases with lower values of G (i.e. finer spatial granularity) are closer to the Ind-Sw curve.

This is a consequence of the higher flexibility that SG-Sw with low G offers to allocate smaller demands in the

space dimension. However, finer spatial granularity results in higher switching infrastructure cost. Current WSS

technology with 6.25-GHz assignable spectral slots enables the switching of smaller ChBWs, which allows us to

evaluate the impact of spectral switching granularity on the performance of spatial switching paradigms. To carry

out this investigation, we repeated the above simulations for 37.5-GHz ChBW at a baud rate of 19.5 Gbaud (Figure

2.4(b)), and 25-GHz ChBW at 7 Gbaud (Figure 2.4(c)). As observed in Figure 2.4(b) and (c), for small values of

load, all curves show improved performance compared to the 50-GHz ChBW case since finer data loads can be

accommodated per channel. It is noteworthy that the performance of J-Sw (switching paradigm with the coarsest

spatial granularity) converges to that of Ind-Sw for smaller load values, as the ChBW decreases, compared to the

case of 50-GHz ChBW. Yet for high loads the spectrum utilization is higher for the case of finer DWDM channel

grid, as the finite channel guard bands exhibit lower spectral utilization and more channels have to be provisioned to

carry the data.

Another way to quantify the impact of the switching group size is to consider the average ‘data occupancy’

within an allocated wavelength channel. The data occupancy metric is defined by the bandwidth required to support

the data (i.e. equivalent to the baud rate yet measured in GHz) divided by the available bandwidth for data (which is

the clear channel bandwidth multiplied by the group size). Ind-Sw always satisfies 100% data occupancy, whereas J-

Sw will have the lowest data occupancy (as low as 1/G). As shown in Figure 2.5, the amount of data occupancy

increases for switching paradigms with finer spatial and/or spectral granularity. Additionally, inflection points in the

performance of SG-Sw cases are observed at loads of 0.2 and 0.7 Pb/s, respectively, from where the data occupancy

of SG-Sw cases increases significantly. In particular, as observed in Figure 2.5(b), the data occupancy in the case of

J-Sw goes above 90% for loads higher than ~0.65 Pb/s with 37.5 GHz ChBW, instead of for loads above ~1 Pb/s

with 50 GHz ChBW. Finally, Figure 2.5(c) shows a further improvement of the performance of different SG-Sw

paradigms in comparison with the previous two cases (e.g. the data occupancy of J-Sw increases up to 90% at ~0.2

Pb/s).

Therefore, we can conclude that, for small load values, the utilization of WSSs with finer spectral switching

granularity can compensate for the spatial granularity rigidity of SG-Sws. For larger load values, on the other hand,

the performance of all switching paradigms is degraded (i.e. the average spectrum utilization increases) as ChBW is

decreased. This is due to a less efficient utilization of the spectrum arising from a lower amount of occupied

spectrum containing actual traffic compared to the required guard band for the WSSs (i.e. 32/50=64% vs. 7/25=28%

for ChBW equal to 50 and 25 GHz, respectively).

In order to evaluate the improvement of the SG-Sw performance resulting from the utilization of WSSs with

improved resolution, we repeat the above studies for a WSS technology with a factor-two resolution improvement

(i.e. requiring a 9-GHz guard band instead of the 18 GHz considered previously). Figure 2.6 attest that the

performance of spatial Sp-Ch switching paradigms can be improved by using spatial switches with finer spatial

granularity. Likewise, similar to Figure 2.4, the use of smaller ChBWs results in performance improvement. Note

that, comparing Figure 2.6 with Figure 2.4, due to the lower guard band required by the WSS with finer resolution,

the performance of switching paradigms is not degraded for large load values, in contrast to the case of current WSS

technology with coarser resolution. However, by comparing Figure 2.7 with Figure 2.5, we observe that the amount

of data occupancy is less significant when the improved resolution WSSs are in place. This is due to the fact that one

spectral-spatial slot can accommodate more traffic when WSS technology with factor-two resolution improvement

is in place compared to the case of present-day WSS technology. For example, assuming ChBW equal to 50 GHz

and DP-16QAM, the capacity of one spatial-spectral is 72 Gb/s higher (i.e. 328 Gb/s – 256 Gb/s) in the case of

improved resolution WSS compared to the current WSS technology. Note that, even though the data occupancy

(which is a normalized metric) is lower for WSSs with finer spectral resolution, refining the WSS spectral resolution

results in a globally better performance, as is shown in Figure 2.8.



Figure 2.6 Average spectrum utilization per link per fiber under different spatial Sp-Ch switching paradigms considering improved resolution WSS technology for (a) 50 GHz, (b) 37.5 GHz, and (c) 25 GHz ChBWs at a baud rate of 41, 28.5

and 16 Gbaud, respectively in terms of load.

Figure 2.7 Average data occupancy in percentage considering finer resolution WSS technology for (a) 50 GHz, (b) 37.5 GHz, and (c) 25 GHz ChBWs at a baud rate of 41, 28.5 and 16 Gbaud, respectively in terms of load.

Figure 2.8 presents the results of a more detailed performance evaluation of SG-Sw paradigms for the values of

ChBWs indicated in Table 2.1 and for two WSS technologies with coarser and finer resolutions. For the sake of

clarity, the results are only shown for J-Sw. Figure 2.8(a) shows the average spectrum utilization with the current

WSS technology. For small loads (<80 Tb/s), as shown previously, smaller ChBW values lead to better J-Sw

performance. However, as traffic increases, smaller ChBW values result in significant performance degradation.

Figure 2.8(b) shows the results when the finer resolution WSS is used. Due to the more efficient utilization of the

optical spectrum, smaller ChBW values lead to better performance for loads lower than 800 Tb/s. Even if the

performance of J-Sw with smaller values of ChBW reduces for loads larger than 800 Tb/s, this is remarkably better

than in the case of WSSs with coarser resolution. Another important finding is that, for small and large loads, the

best J-Sw performance is obtained for the lowest and highest values of ChBWs, respectively. Consequently, ChBW

must be adaptable to the load level in order to achieve a globally optimum spectrum utilization in an SDM network.

This highlights the importance of utilizing flex-grid transmission enabled by spectrally flexible ROADMs and

bandwidth variable transceivers when SG-Sw paradigms are considered.

0

20

40

60

80

100

0 0.2 0.4 0.6 0.8 1

Av

erag

e d

ata

occ

up

ancy

wit

hin

the

allo

cate

d w

avel

eng

th c

han

nel

s

Load in Pbps

G=12 G=6

G=4 G=3

G=2 G=1

(a)

0

20

40

60

80

100

0 0.2 0.4 0.6 0.8 1

Aver

age

dat

a occ

upan

cy w

ithin

the

allo

cate

d w

avel

ength

chan

nel

s

Load in Pbps

G=12 G=6

G=4 G=3

G=2 G=1

(b)

0

20

40

60

80

100

0 0.2 0.4 0.6 0.8 1

Av

erag

e d

ata

occ

up

ancy

wit

hin

the

allo

cate

d w

avel

eng

th c

han

nel

s

Load in Pbps

G=12 G=6

G=4 G=3

G=2 G=1

(c)



Figure 2.8 Average spectrum utilization per link per fiber for J-Sw considering: (a) the current WSS technology which requires 18 GHz for guard band and (b) finer resolution WSS which requires 9 GHz for guard band. Five values of ChBWs are assumed for the simulations. The amount of spectral contents that can be utilized considering different WSS technologies is

provided in Table I.

0

400

800

1200

1600

0.01 0.1 1

Av

erag

e S

pec

tru

m U

tili

zati

on

per

Lin

k p

er F

iber

(in

GH

z)

Load in Pbps

7Gbaud on 25GHz

13.25Gbaud on 31.25GHz



32Gbaud on 50GHz

(a)

0

400

800

1200

1600

0.01 0.1 1

Av

erag

e S

pec

tru

m U

tili

zati

on

per

Lin

k p

er F

iber

(in

GH

z)

Load in Pbps

16Gbaud on 25GHz




41Gbaud on 50GHz

(b)

0

400

800

1200

1600

0.01 0.1 1

Av

erag

e S

pec

tru

m U

tili

zati

on

per

Lin

k p

er F

iber

(in

GH

z)

Load in Pbps

7Gbaud on 25GHz




32Gbaud on 50GHz

(a)

0

400

800

1200

1600

0.01 0.1 1

Av

erag

e S

pec

tru

m U

tili

zati

on

per

Lin

k p

er F

iber

(in

GH

z)

Load in Pbps

16Gbaud on 25GHz




41Gbaud on 50GHz

(b)



3. Cost analysis and power consumption evaluations

In this chapter of the deliverable, we present the results showcasing the cost and power consumption benefits of

exploiting INSPACE proposed solutions with respect to the legacy approaches.

3.1. Cost benefit quantification of INSPACE proposed solutions

In this section, we present a cost analysis of SDM networks based on spatial and spectral Sp-Ch allocation

considering integrated TRx, amplifier and node implementations for Ind-Sw (spectral and spatial Sp-Ch), J-Sw and

FJ-Sw (spatial Sp-Ch), compared to a parallel-fiber system based on 100G transmission. We quantify the values of

cost savings, under given assumptions, for several cases of parallel-fiber system deployments, and discuss how the

cost of the system can be further reduced.

Cost model

Super-channel transceiver cost model

The cost of the Sp-Ch TRxs is based on the cost model presented in [5]. In contrast to the case of parallel fiber

systems employing single-carrier 100G OIF MSA TRxs (shown in Figure 3.1), the spectral Sp-Ch TRx employs two

comb generator modulators and drivers, two arrayed waveguide gratings (AWG), and a variable gain dual-stage

amplifier, in order to avoid the use of two lasers per sub-channel (Sb-Ch). The spatial Sp-Ch TRx (shown in Figure

3.2), since all Sb-Chs are transmitted at the same frequency, does not require frequency combs or AWGs, and can

bring the cost down by 5-20% for integrated spatial Sp-Ch TRx with 2-10 Sb-Chs vs. spectral Sp-Ch TRx.

Figure 3.1 single-carrier transceiver

Figure 3.2 Integrated spatial super-channel transceiver with seven integrated sub-channels



Cost model for switching solutions

The implementation of switching solutions for Ind-Sw, J-Sw and FJ-Sw requires the design of new SDM

switching nodes. A potential realization for OE node architecture based on currently available WSS technology is

composed of a number of WSSs equal to 2·D·ceil(S/G) with port count D·G, where D is the number of nodal

degrees, S is the number of spatial dimensions and G is the number of spatial dimensions (out of the total number S)

that are jointly switched. Therefore, G equals 1 for Ind-Sw, S for J-Sw and any intermediate number for FrJ-Sw.

FrJ-Sw and J-Sw make necessary a redesign of the WSSs. Joint WSSs compatible with BuSMF, FMF or MCFs

were reported in [8]. They are configured to operate as S×(I×O) WSSs, i.e. they direct I input ports, each carrying S

spatial modes/cores, toward O output ports. In the cases of FMFs, MCFs or FM-MCFs, SDM interfaces capable of

converting the MCF/FMF inputs/outputs into BuSMFs, such as photonic lanterns [9] or MCF fan-ins/fan-outs, need

to be interposed between the fiber and the WSS. Alternative WSS implementations for FMF transmission replace

the WSS I/O SMF array with an FMF array14, with a 3×(1×9) configuration having been demonstrated. The

considered cost model for various component are presented in the next sections.

To compare the three SDM switching paradigms, we must take into account how the architectural complexity of

the switching solutions affects, not only the network performance, but also the equipment cost. Cost differences

between node architectures arise from the fact that the required number of WSSs per node and the WSS port count

differ, as indicated above, according to the chosen switching strategy. In the rest of this chapter, we consider route

and select node architectures for BuSMFs, thus focusing on WSS realizations with I/O SMF arrays and obviating the

need of SDM interfaces.

HPC 1N WSS cost model

The HPC 1×N WSS cost model uses the cost of commercial LCoS-based 1×9 WSSs as a reference (cost = 1) and

follows the rule that an increase of 4 in the number of ports results in a 2.5 increase in cost, based on a WSS

design analysis performed in the framework of the EU project INSPACE. Node architectures requiring WSSs with

less than 3, 6, 10 and 21 ports (i.e. 1×2, 1×5, 1×9, and 1×20 WSSs, respectively; all commercially available) have a

cost per WSS of 0.4, 0.63, 1 and 1.58, respectively. The cost of HPC WSSs with up to 40, 80 and 160 ports (1×40,

1×80, 1×160 WSSs), assuming technology maturity and mass production, was estimated to be 2.50, 3.95, and 6.25,

respectively. In Table 3.1 and Figure 3.3, we show the relative cost for WSS with port counts from 2 to 160.

Table 3.1 Estimated cost of HPC WSSs WSS port count Estimated cost of HPC WSS

12 0.40

15 0.63

19 1

120 1.58

140 2.50

160

180 3.95

1160 6.25

Figure 3.3 Relative cost/price of WSS required for a given number of ports



MN WSS cost

MN WSS are not commercially available. We have estimated the cost of an 824 WSS to be the same as that of a

180 WSS, i.e. 3.95 times the cost of a 19 WSS, as indicated. According to [6], a ballpark estimate for the

maximum number of ports supported by an LCoS SLM is given by the formulas presented in Table 3.2.

Table 3.2 Relationship between maximum number of ports and number of LCoS SLM pixels. N_px is the total number of pixels in the steering direction. Px_port is the number of pixels per port.

WSS port configuration

General formula Formula assuming a number of pixels per

port (Px_port) equal to 5

Maximum number of ports for N_px = 1200

pixels

1N N+1 = 0.7*N_px/Px_port N+1 = 0.7*N_px/5 N = 168

MN M·N = 0.7 N_px/Px_port M·N = 0.7*N_px/5 M·N = 168 (if M = 8, then N = 20)

Twin 1N N+1 = 0.7 N_px/(Px_port·2) N+1 = 0.7*N_px/(5·2) N = 80

Quad 1N N+1 = 0.7 N_px/(Px_port·4) N+1 = 0.7*N_px/(5·4) N = 40

From Table 3.2, we can see that a 19201200 pixel LCoS panel could support a 1160 WSS or an 820 WSS.

An 824 WSS (design based on a two column arrangement, as shown in Figure 3.4, where I/O fibers were arranged

in two columns, thus doubling the addressable port count) can be supported with such a conventional LCOS panel.

Our assumption is that an 824 WSS is just as complex as a 180 WSS (four times a 120 WSS). There is a

secondary switching plane (and technology) required, accounting for a doubling in the cost for the switching

technology. Additionally, a 2D fiber and collimator array is more expensive to realize than the same number of

fibers in a 1D array. Finally, testing time is significantly longer for such a switch, accounting for extra cost.

Figure 3.4 Port-reconfigurable MN WSS

ROADM cost model

For the A/D nodes, we assume colorless, directionless and contentionless (CDC) reconfigurable optical A/D

multiplexer (ROADM) operation based on multicast switches (MCS). The A/D module cost (CA/D) is estimated as

𝐶𝐴/𝐷 = min {2𝑁 (𝐷 · 𝑆 · 𝐶𝑠𝑝 + ⌈𝑇

𝑁⌉ 𝑆 · 𝐶𝑠𝑤) + 𝐶𝑎𝑚𝑝}

where N is the number of Twin MCSs, T is the number of TRxs that add/extract traffic to/from the node; Csp is

the cost of the splitters (with splitting ratio 1:(T·S/N)) and Csw is the cost of the opto-mechanical switches (with

port count (D·S)×1) forming the MCS, obtained by averaging costs from different vendors, and Camp is the cost of

the required amplifiers per MCS, given by 2·D·S·N·CEDFA, where CEDFA is the cost of an SMF EDFA. We

assume that amplifiers are only used if the power at the receiver (without amplification) is below -20 dBm. In the

equation for CA/D, the value of N minimises CA/D.

The A/D nodes are composed of 2·D·ceil(S/G) WSSs with port count (D+N)·G. The total cost of a ROADM is

estimated to be

𝐶𝑅𝑂𝐴𝐷𝑀 = 𝐶𝐴/𝐷 + 2𝐷 ⌈𝑆

𝐺⌉ 𝐶𝑊𝑆𝑆



where CWSS is the cost of the WSS, which is estimated as indicated above. The cost of an OE node is given by the

above equation with CA/D=0 and N=0.

MCS specifications

MCS use low-cost single stage amplifiers with gain of typically ~17 dB. The losses of an integrated MCS include

coupling in and out of a PLC (-0.75 dB), the splitter intrinsic power loss (1/N) plus some excess loss. If we assume

the split is performed by cascaded 1:2 splitters, then the loss in dB will be approximately -3.25* (log2N). Say for 16

way split, this will be -13dB (instead of intrinsically -12dB). At the output there is a selector switch, which is based

on a cascade of 12 switches. Each switch will have a typical loss of -0.35dB, hence a selector switch of 4 inputs

will have another -0.7 dB of loss. Finally, there will be many waveguide crossings on the PLC. Those will add some

excess loss which scales linearly with the number of inputs and outputs (in dB scale). With 17dB amplifiers you can

support MCS 8×16 at most.

Table 3.3 Loss of splitters and switches with different splitting ratios to be used in realizing MCS with various port counts

Amplifier cost model

Regarding the in-line amplifiers, MC-EDFAs with cladding pumping can be used for SDM systems using

BuSMFs, provided that proper fan-in/fan-out devices are used. Therefore, we consider two scenarios: (a) SMF

EDFAs (one per fiber) and (b) a realization composed of a fan-in (which converts the output of the SMF bundle into

an S-core MCF), an MC-EDFA and a fan-out as shown in Figure 3.5. The MC-EDFA cost was estimated based on

[7], which, for S = 9, is 3.3× the cost of a conventional SMF EDFA. The cost of the fan-in/fan-out was estimated in

the framework of INSPACE and to be proportional to the number of channels plus a fixed cost for packaging,

material, etc.

Figure 3.5 A possible design of MC-EDFA

Network-level cost analysis

Telefónica’s Spanish national network model was considered for the studies. As baseline for comparison we

consider two cases with 9 parallel SMFs, each having a different line system without spatial multiplexing of

networks elements: one considering conventional single-carrier 100 Gbps TRx (100G case) and the other an SDM

network based on spectral Sp-Ch TRx (C1 case). In addition, we consider four SDM deployment cases, with

spatially integrated network elements, based on bundles of 9 SMFs (because it allows us to make a fair comparison

among different cases regardless of any transmission medium related physical constraints): C2) SDM employing

individual EDFAs, Ind-Sw and integrated spatial TRx; and SDM network employing integrated spatial Sp-Ch TRx

and SDM EDFAs using C3) Ind-Sw, C4) FrJ-Sw(G=3), and C5) J-Sw.

To route the demands, we use the offline planning platform presented in D5.2 of INSPACE. Figure 3.6 shows

the average spectrum utilization for scenarios C1-C5. SDM network based on spectral Sp-Ch benefits from higher

spectral efficiency compared to spatial Sp-Ch one, as is also shown in D5.2 of INSPACE. SDM networks based on

Ind-Sw, FrJ-Sw and J-Sw show almost similar performance, particularly when the total offered load to the network

is large.

inputs outputs splitter switch PLC coupling WG crossings Total

4 8 -9.75 -0.7 -0.75 -0.64 -11.84

4 16 -13 -0.7 -0.75 -1.28 -15.73

4 32 -16.25 -0.7 -0.75 -2.56 -20.26

8 8 -9.75 -1.05 -0.75 -1.28 -12.83

8 16 -13 -1.05 -0.75 -2.56 -17.36

8 32 -16.25 -1.05 -0.75 -5.12 -23.17

Losses



In order to present the total deployment cost of networks, we use the cost model described in the previous

section and, considering the cost of commercial LCoS-based 1×9 WSSs as 1 unit, we report all cost values with

respect to this value. Figure 3.7 shows the total relative cost (TRC) of different cases under study. Total cost is the

summation of the cost of TRxs, A/D nodes, amplifiers (Amp in Figure 3.7), and OE nodes.

Figure 3.6 Average spectrum utilization per link per fiber in terms of total offered load to the network in Pbps

Figure 3.7 Total relative cost of the network considering 4 different offered load for 5 SDM deployments compare to an equivalent parallel system using 100G TRx.

As depicted in Figure 3.7, going from single-carrier (i.e. 100G case) to spectral and spatial Sp-Ch based

approaches results in huge total cost savings (50%-60%) due to the reduction of the TRx cost (enabled by photonic

integration and component sharing), which dominates the total cost of the network.

After the TRx cost, whose contribution to the total cost ranges from 45% to 58%, the second most costly element

in all cases is the A/D nodes (consisting of MCSs and WSSs), with 36%-40% of the total cost of the network. The

four SDM cases (C2-C5) offer a total cost benefit ranging from only 4% to 15% compared to C1. Excluding TRxs

and A/D nodes, the cost of the rest of the elements, with 50%-55% and ~63% savings for OE nodes and EDFAs

respectively, demonstrate the benefits of joint switching and amplifier integration. The relatively small overall cost

savings due to the high cost of TRxs and A/D nodes mean that, for an SDM solution to offer higher cost savings, the

focus should be on the reduction of the cost of TRxs and A/D nodes. To reduce the cost of TRxs, we can consider

the use of common DSP chips (as proposed in [10]). To reduce the cost of A/D units, alternative designs need to be

considered. In the next section, we present a comprehensive cost analysis of A/D units revealing the benefits of

INSPACE proposed J-Sw scheme in reducing the overall cost of ROADMs including both OE and A/D units.

250

500

1000

2000

0.5 1 1.5 2Avera

ge S

pectr

um

Utiliz

ation per

Lin

k p

er

Fib

re (

GH

z)

Total Offered Load to the Network (Pbps)

C1C2, C3C4C5

0

20

40

60

80

To

tal R

ela

tive

Co

st

(in

th

ou

sa

nd

s)

Total Offered Load to the Network (Pbps) 0.5 1 1.5 2

100G C1 C2 C3 C4 C5OE Node

Amp

A/D Node

TRxs



3.2. Comparison of CD(C) ROADM architectures for SDM networks

Motivation

CDC ROADMs has attracted significant interest from the optical networking industry. CDC ROADMs offer

architectural flexibility and operational efficiencies leading to reduced cost, while they also support enhanced

capabilities for optical layer restoration and re-optimization in the case of need for dynamic capacity re-allocation

[11]. For enhanced performance, colorless-directionless (CD) and CDC ROADMs use a common optical core based

on twin-WSS modules in a route-and-select (R&S) architecture, and only differ in the way that the A/D side is

implemented.

Such ROADMs are becoming a commercial reality for conventional flexible WDM networks. Moreover, the

existing architectures can be upgraded to address the challenges introduced by spatial division multiplexing (SDM),

which will arguably be the next step in the evolution of optical networks. Not only does SDM promise higher

capacities, but it also hints at a reduction of the cost per transmitted/switched bit through the utilization of spatial

integration of networks elements. As we have also seen in the previous section, A/D unit of a ROADM node is one

of the dominant contributor to the cost of an SDM network. Despite the increasing interest in SDM-based optical

networking studies, a detailed study summarizing the architectures of CD(C) ROADMs for SDM spatial super-

channel routing, while comparing their scaling potential and associated implementation costs, is missing. In this

section, we propose CD(C) ROADM architectures enabling Ind-Sw, FrJ-Sw and J-Sw of spatial SChs and we

present a cost model based on the premises presented in the previous section for comparing the proposed

architectures. We find that a CDC ROADM design which maximizes the number of A/D ports, and yet keeps the

port count of pass-through WSSs low, is the most cost-effective solution.

ROADM architectures and cost analysis

We consider the ROADM architectures illustrated in Figure 3.8, as well as the case (not shown) of FrJ-Sw, where

each group of G spatial dimensions is jointly routed by a G×(I×O) WSS. Lane changes (LCs) between groups,

defined as the possibility of routing a spatial SCh from a given set of spatial dimensions to a different one, can also

be supported in the case of FrJ-Sw with the appropriate connectivity and port count. The required number of WSSs

and their port count is shown in Table 3.4 as a function of G (thus providing a common formula valid for all cases)

for routing options with/without LC. The evaluation of the ROADM cost (CROADM) considers the cost of the

WSSs (CWSS) and an A/D module (CA/D) composed of K elements, such that K minimizes the overall cost.

Tx3

‘West’

WS

SW

SS

‘East’

WS

SW

SS

WSSWSS

‘North’

WSS WSS

WS

SW

SS

WS

SW

SS

Rx1

Rx2

Rx3

Rx4

12

2

1

12

1

2

1st spatial

dimension

2nd spatial

dimension

A/D module (Drop)

(a)

Tx1

Tx2 Tx4

12

2

1

12

1

2

A/D module (Add)

To ‘North’

To ‘West’



Figure 3.8 Route and Select ROADM architectures for (a) Ind-Sw and (b) J-Sw in an SDM scenario with S=2. In (a) we show all internal connectivity from the ‘East’ ingress WSS. LCs are supported if all solid, dashed and dotted

connections exist. Ind-Sw without LCs requires solid and dashed lines.

Table 3.4 The required number of WSSs and their port count

Pass-through WSS Total ROADM cost

(CROADM) # WSS Port count

Without LC

2 · 𝐷 ·

𝑆/G

𝐺 × [1 × (𝐷 + 𝐾 − 1)] min𝐾∈ℤ+

{𝐶𝐴/𝐷 + 2𝐷 · (𝑆/𝐺) · 𝐶𝑊𝑆𝑆𝐺×[1×(𝐷+𝐾−1)]}

With LC 𝐺 × {1 × [(𝑆/𝐺) · (𝐷 − 1) + 𝐾]} min𝐾∈ℤ+

{𝐶𝐴/𝐷 + 2𝐷 · (𝑆/𝐺) · 𝐶𝑊𝑆𝑆𝐺×{1×[(𝑆/𝐺)·(𝐷−1)+𝐾]}}

Figure 3.9 Two A/D module architectures for ROADM CD operation. Only ‘Drop’ WSS are shown.

Figure 3.10 Two M×N WSS-based A/D module architectures for ROADM CDC operation. Only ‘Drop’ WSS are shown.

‘West’

WS

S

‘East’

WS

S

WSS

‘North’

WSS

WS

S

WS

S

1st spatial dim.

2nd spatial dim.

(b)

Tx3Rx1

Rx2

Rx3

Rx4

12

2

1

12

1

2

A/D module (Drop)

Tx1

Tx2 Tx4

12

2

1

12

1

2

A/D module (Add)

Rx1

S (D 1)

S (1 T/K )

Rx2

Rx3

Rx4

12

2

1

121

2

S (1 N) WSS

S (M 1) WSS

(a) 1st spatial dimension 2nd spatial dimension

Rx1

Rx2

Rx3

Rx4

12

2

1

121

2

D 1

1 T/K 1 N WSS

M 1 WSS

1 N WSS

M 1 WSS

(b)

Rx1

S (D T/K )

Rx2

Rx3

Rx4

12

2

1

121

2

S (M N) WSS(a)

1st spatial dimension 2nd spatial dimension

Rx1

Rx2

Rx3

Rx4

12

21

121

2

M N WSS M N WSSD T/K

(b)



Figure 3.11 Three MCS-based A/D module architectures for ROADM CDC operation. Only ‘Drop’ MCS are shown.

For CWSS we use as a baseline the cost of commercial LCoS-based 1×9 WSSs (cost = 1). Node architectures

requiring WSSs with port counts 1×5, 1×9, and 1×20, all commercially available, have a cost per WSS of 0.63, 1

and 1.58, respectively. The cost of WSSs with port counts 1×40, 1×80 and 1×160 was extrapolated to be 2.50, 3.95,

and 6.25, respectively, according to 58% premium for a fiber port doubling.

Table 3.5 Cost of the A/D modules presented in Figure 3.9 to Figure 3.11

The A/D modules are implemented as shown in Figure 3.9 for CD operation and Figure 3.10 and Figure 3.11 for

CDC operation (only the drop module is shown). In Figure 3.9 to Figure 3.11, T is the number of transceivers (TRx)

connected to the A/D ports. Table 3.5 show our A/D module cost estimation. In the case of CDC operation

implemented with multicast switches (MCS), Figure 3.11, the cost model is based on the estimation of the number

of discrete components (splitters and opto-mechanical switches), whose costs were obtained by averaging costs from

different vendors, and it does not include the cost of packaging and electronics. We consider low-cost single-stage

amplifiers with ~17 dB gain (which limits the splitting ratio to about 1:16) and cost CEDFA ~ 0.17. The total cost

of amplifiers per A/D module is estimated as C_amp=2·K·S·D·C_EDFA. The architecture shown in Figure 3.11(a)

supports full CDC switching between all common ports and A/D ports (case investigated in the previous section)

and is required in the case of single-channel TRxs. In Figure 3.11(b) connectivity is simplified to provide CDC

operation per spatial dimension, which is sufficient for integrated SCh TRxs. In the case of CDC operation with

M×N WSS, Figure 3.10, we assume WSS configurations with complexity similar to that of a 1×80 WSS, e.g. 8×24.

If we decide to use a higher number of common ports M, the number of A/D ports N has to decrease in proportion

so that their product remains constant.

Figure 3.12 Relative cost of the components wrt 1×9 WSS

(S·D) 1 Switches

1: S·T/KSplitters

Rx1

(S·D) S·T/K

Rx2

Rx3

Rx4

12

2

1

121

2

(a)

D 1 Switches

1: T/KSplitters

Rx1

Rx2

Rx3

Rx4

12

2

1

121

2

(S·D) S·T/K

(b)

D 1 Switches

1: T/KSplitters

MCS for 1st spatial dimension MCS for 2nd spatial dimension

Rx1

Rx2

Rx3

Rx4

12

2

1

121

2

D 1 Switches

1: T/KSplitters

D T/K

(c)

0 5 10 15 20 25 30

1×2 WSS1×5 WSS1×9 WSS

1×20 WSS1×40 WSS1×80 WSS

1×160 WSSTwin 3×52 WSSTwin 9×18 WSSTwin 18×9 WSSTwin 36×4 WSSTwin 3×16 MCSTwin 9×16 MCS

Twin 18×16 MCSTwin 36×16 MCS

Twin 9×48 MCSTwin 18×96 MCS

Twin 36×192 MCS

Relative Cost of the Components wrt

1×9 WSS



Figure 3.13 Total relative cost (TRC) of different ROADM implementations.

Discussion on the total relative cost of ROADM implementations

Based on the cost models described in the previous section, we calculate the total implementation cost of each

ROADM architecture. A degree-3 ROADM for an SDM network with 6 spatial dimensions is considered for the

calculations. For the realizations of the designs, we use the components with port count and relative cost provided in

Figure 3.12 and calculate the total relative cost (TRC) of each design for the three SDM switching paradigms. In

order to evaluate the scaling of their implementation cost with the number of A/D ports, we repeat the calculations

for 20, 40, and 60 spatial SCh TRx, each comprising 6 spatial sub-channels, resulting in 120, 240, and 360 required

add and drop ports. To minimize the A/D module cost, we consider a number K of A/D elements in parallel, whose

impact on the port count of pass-through WSSs is also taken into account. Figure 3.13 shows the resulting TRC

values. C1-C7 represent the ROADM architectures illustrated in Figure 3.9(a-b), Figure 3.10(a-b), and Figure

3.11(a-c), respectively.

It is found that the most cost-effective architecture is the one which i) maximizes the number of available A/D

ports, and ii) does not heavily increase the port count of pass-through WSSs. Design C7, which utilizes 3×16, 9×48,

and 18×96 MCSs (with internal connectivity as shown in Figure 3.11 (c)) for Ind-Sw, FrJ-Sw with G=3, and J-Sw,

respectively, is the one that best meets the above conditions and therefore turns out to be the most cost-effective

solution. Surprisingly, its implementation cost is very similar to the implementation cost of CD ROADMs (design

C2). The study also reveals that J-Sw based ROADMs are more cost effective than those implementing Ind-Sw and

FrJ-Sw, except for design C4, where the Ind-Sw based realization, by taking advantage of the fact that M×N WSSs

with port count 3×52 maximize the available A/D ports with very small increase in the port count of pass-through

WSSs, outperforms the other two switching strategies. Design C3, based on a single M×N WSS with common ports

connected to all spatial dimensions coming from all pass-through WSSs, shows the highest implementation cost,

because of the imposed restriction that the M×N WSS complexity should not be higher than that of a 1×80 WSS.

Therefore, the only possible M×N WSS realization supporting a ROADM of degree 3 in an SDM system with 6

spatial dimensions is using WSSs with port count 18×9. The very low number of A/D ports supported by this

component means that a large number of them is required to support the same number of TRxs that could be

supported with fewer components by other architectures (e.g. C4, which can be implemented with M×N WSSs of

port count 3×52). Additionally, we have seen that when the number of parallel A/D elements increases, some of the

designs have a huge impact on the port count of pass-through WSSs. These designs turn out to be too costly or

impractical to be estimated (e.g. J-Sw based realizations of C3 and C4 when the number of TRxs scales beyond 40)

due to the need for WSSs with extremely large port counts.

3.3. Power consumption of MIMO processing and its impact on the performance of SDM networks

Motivation

SDM has been proposed as a solution for increasing the capacity of fiber-optic networks with a reduced cost-per-bit

through dense optical parallelism. The denser the integration of the spatial channels it becomes, however, the more

significant is the XT interactions among them, e.g. in strongly-coupled MCFs and FMFs. Such spatial XT is

expected to be mitigated by MIMO DSP at the TRx side, with a complexity for real-time implementation

determined by the number of spatial channels and their corresponding delay spread.

Current C-form pluggable (CFP)-based analog and digital coherent optics modules (i.e. CFPx-ACO and CFPx-

DCO) are focusing on decreasing the cost and volume, as well as the power consumption of the DSP by separating

the DSP chip from the module, while utilizing smaller complementary metal-oxide-semiconductor (CMOS)

platforms. While commercial CFP2-ACO DSP modules have been produced utilizing down to 16 nm CMOS

platforms [12], it will be extremely difficult to support further power reduction with sub-10 nm CMOS-based DSP

0

100

200

300

400

500

To

tal

Rel

ati

ve

Co

st w

rt 1

9 W

SS

s

Number of Spatial Superchannel Transceivers with 6 Spatial Channels

(C1) Ind-Sw (C1) FrJ-Sw (G=3) (C1) J-Sw







C1 C2 C3 C4 C5 C6 C7 C1 C2 C3 C4 C5 C6 C7 C1 C2 C3 C4 C5 C6 C7

20 40 60



modules due to heat dissipation issues [13]. Therefore, even though MIMO-DSP can ideally compensate all the

linear impairments in SDM systems, the power consumption of the MIMO-DSP and of the overall SDM-TRx can be

a limited factor in the maximum capacity and achievable reach of SDM networks.

In this section, an estimation for the power consumption of real-time MIMO-DSP considering the current CMOS

technology is presented based on the computation complexity of frequency-domain equalizers (FDEs). In addition, a

reach/power consumption tradeoff for MIMO-DSP modules is investigated for FMFs with up to 6 spatial modes.

Finally, the performance of FMF-based SDM networks with 3 and 6 modes is compared with single-mode fiber

(SMF)-based SDM networks, revealing the impact of power-limited reach on the overall network performance.

Power consumption of MIMO processing for SDM

Considering MIMO-FDE with 2S spatial and polarization channels, based on training sequence and without any

cyclic-prefix, the computational complexity in terms of complex multiplications/bit can be given by [14][15]:

𝐶𝐶𝐹𝐷𝐸 = 𝑟𝑠

2𝑆 𝑁𝑓𝑓𝑡 log2 𝑁𝑓𝑓𝑡 + 2(2𝑆)2𝑁𝑓𝑓𝑡 + (2𝑆)3𝑁𝑓𝑓𝑡

2𝑆 (𝑁𝑓𝑓𝑡 − ⌈𝑇𝑇𝑠

⌉ + 1) log2 𝑀= 𝑟𝑠

𝑁𝑓𝑓𝑡 log2 𝑁𝑓𝑓𝑡 + 4𝑆 𝑁𝑓𝑓𝑡 + (2𝑆)2𝑁𝑓𝑓𝑡

(𝑁𝑓𝑓𝑡 − 𝑁 + 1) log2 𝑀

where, r_s is the oversampling ratio, M the modulation order, T_s the symbol duration, T the channel delay in

[sec], N=⌈T/T_s ⌉ the total channel delay in [symbols], and N_fft=2^d≥N the fast-Fourier transform (FFT) size with

d an integer number. In the above equation, the first term of the numerator accounts for the complexity of 2×FFTs,

the second term for the calculation the MIMO channel matrix and the multiplication of its inverse with the received

frame, and the third term for the calculation of the inverse of the MIMO channel matrix. Note, that even though the

MIMO matrix estimation and inversion is expected to take place every several frames [15], since here we are

focusing on the total power consumption, the worst case was assumed in which all the functions of channel

estimation and equalization are performed together. The resulting complexities for 32 GBaud, 100 Gb/s signals over

linear polarized (LP) FMFs with 3- and 6-spatial modes are compared to the SMF case in Figure 3.14 for different

channel delay spread values. This figure depicts the complexity increment based on the number of spatial channels

and the delay spread of the MIMO channel and how for larger delays, larger FFT sizes and number of taps are

required for equalization.

Figure 3.14 MIMO-DSP complexity of FDE

Table 3.6 CMOS power dissipation based on Fig. 2 of [7]

Moore’ law

(norm.)

Actual

(norm.)

Deviation from

Moore’s law

90 nm 0.129 0.153 15.7 %

45 nm 0.03 0.09 66.7 %

22 nm 0.008 0.07 88.6 %

15 nm 0.004 0.065 93.9 %

In [16], using a reversed-engineering approach, the energy dissipation per real multiplication and per real

addition for 90 nm CMOS platforms, was calculated to be 1.5 pJ and 0.5 pJ, respectively. Therefore, considering

that a single complex multiplication can be described by 4 real multiplications and 2 real additions, an estimation of

the power consumption from the computational complexity can be drawn. In addition, to account for current CFP2-

ACO DSP implementations based on sub-20 nm CMOS [13], a power reduction from the 90 nm platforms can be

drawn based on dissipation values for different CMOS sizes. Here, we based our analysis on the published results of

[13], summarized in Figure 3.6. In the first column, the theoretical predicted values based on the Moore’s law are

shown, while the second column provides values for actual deployed systems. Since the deviation of the actual



values from the Moore’s law is increasing for smaller sizes, here only the actual values were considered with a

deviation of ±5 %.

The resulting power consumptions for 15 nm CMOS, based on the computational complexities of Figure 3.14,

are depicted in Figure 3.15. Here an FFT size of 1024 was considered. The x-axis, which describes the reach, was

converted from the delay spread of Figure 3.14 considering a chromatic dispersion coefficient of 20 ps/nm/km and a

differential mode group delay (DMGD) of 0 ps/km, 1 ps/km, and 6 ps/km for the SMF, 2-LP FMF, and 4-LP FMF

case, respectively. Particularly, the values for the FMF cases were considered based on recent fabricated fiber links

incorporating DMGD compensation [17][18].

(a) (b)

Figure 3.15 Predicted (a) power consumption per mode and (b) total consumption per 6×100Gb/s module, for 15 nm CMOS.

Table 3.7 Power-limited reach based on total consumption per 600 Gb/s module

100 W 200 W 300 W

(50 W/mode)

600 W

(100 W/mode)

SMF 2104 km 2609 km 2780 km 2950 km

2-LP FMF 1699 km 2152 km 2306 km 2459 km

4-LP FMF 666 km 1050 km 1180 km 1309 km

Table 3.8 Total power consumption per 600 Gb/s module. 750 km 1000 km 1500 km fibers modes

SMF 42.8 W 47.9 W 62.7 W 6 × 1

2-LP FMF 49.1 W 56.7 W 82.2 W 2 × 3

4-LP FMF 112.2 W 176.4 W - 1 × 6

Particularly, in Figure 3.15(a) the calculated power consumption per mode is depicted, while in Figure 3.15(b)

the total consumption per 600 Gb/s is depicted based on 6-spatial/parallel channels carrying 100 Gb/s each (i.e.

6×SMFs, 2×3-mode FMFs, and 1×6-mode FMF). Considering that current real-time DSP for long haul applications

consumes 50~100 W per 400~500 Gb/s [19], and that current Intel’s processors consume powers around 200 W

[20], a limit on the maximum reach can be drawn. The resulting power-limited reaches for different power

thresholds and fibers are summarized in Table 3.7 for clarity. In addition, the total power consumption for 750-km,

1000-km, and 1500-km are summarized in Table 3.8. Such estimated distances are of great interest since they are

covering most of the deployed European-scale core networks. From this we can conclude that by shifting from

parallel SMFs implementations to FMFs with 3- and 6-modes, the realistic power-limited reaches are expected to be

reduced by 18~19 % and 60~68 %, respectively, for total consumptions of 100~200 W. Alternatively, the overall

consumption increases by 15~18 % and 168~268 %, respectively for the 3- and 6-mode upgrade, between 750-km

and 1000-km.

Figure 3.16 Predicted (a) power consumption per mode and (b) total consumption per 6×100Gb/s module, for 15 nm CMOS.

1.E-05

1.E-04

1.E-03

1.E-02

1.E-01

1.E+00

612.5 735 857.5 980 1102.5 1225

Blo

ckin

g P

rob

abil

ity

Total Offered Load to the Network (in Tb/s)

SMF, 100W

2LP-FMF, 100W

4LP-FMF, 100W

(a)

1.E-05

1.E-04

1.E-03

1.E-02

1.E-01

1.E+00

612.5 735 857.5 980 1102.5 1225

Blo

ckin

g P

rob

abil

ity

Total Offered Load to the Network (in Tb/s)

SMF, 200W

2LP-FMF, 200W

4LP-FMF, 200W

(b)



Network-wide performance evaluation of integrated transceiver utilizing different MIMO-DSP modules

The impact of power consumption on the achievable reach affects the successful establishment of incoming

connections. Therefore, to investigate the overall performance reduction due to the power-limited reaches of

different MIMO-DSP modules, a networking-wide simulation is performed. We use the Spanish national network of

Telefónica. For the SDM network evaluations, 6 spatial dimensions are assumed per link, which can be realized by

utilizing either bundles of 6×SMFs, bundles of 2×FMFs with 3 spatial modes, or FMFs with 6 spatial modes.

Available spectrum equal to 4.8 THz (C-band) on the ITU-T WDM 50-GHz grid per spatial dimension is assumed.

Regarding transmission, single-carrier spatial superchannel TRxs with 6 spatial channels, each carrying 32 GBaud

DP-QPSK signals have been considered. TRxs with different MIMO-DSP modules for the different fiber types are

assumed, resulting in various achievable distances as provided by Table 3.7. The blocking probability (BP) was used

as a qualitative performance measure. Note that in our simulator, a blocked connection occurs due to i) transmission

distance longer than the optical reach of the signal and ii) unavailability of resources (spectrum, fiber) to establish

the connection. The results are summarized in Figure 3.16. Each data point was obtained by simulating 3×105

connection requests and has a confidence interval of 95%. As shown in Figure 3.16 (a), for a consumption of 100 W

per TRx module, BuSMFs and 2LP-FMF show similar performance due to the medium-size topology considered, in

which most of the connections can be established by the achievable reaches. However, 4LP-FMF shows

unacceptable performance resulting in rejecting more than 30% of the incoming connection requests. By increasing

the power per module, as shown in Figure 3.16 (b), the performance of 4LP-FMF improves significantly and gets

closer to the other cases. The performance for BuSMFs and 2LP-FMFs do not improve much, because the blocking

occurs due to the lack of available resource (even when considering 100 W per module). Therefore, it can be

concluded that, for a MIMO-DSP module for a 4LP-FMF based SDM network to perform close enough to SDM

networks utilizing BuSMFs or 2LP-FMFs, more than 2 times the power is required.

3.4. Control unit power consumption of WSSs used for realizing SDM nodes

As we have seen in the previous sections, the three SDM switching paradigms require different number of WSSs of

various port counts for the realization of a node. However, regardless of the port count of WSSs, each WSS is

equipped with a control unit which consumes, according to internal partners data, a fixed 10 watts of power to

operate properly. Therefore, as the number of required WSSs for realizing the three switching schemes scales

differently, the overall power consumption of the node due to the control unit scales in a different way. Figure 3.17

(a) shows the power consumption of a single degree of a ROADM for the three switching schemes in terms of the

number of spatial dimensions. As is shown, while the power consumption increase slightly in the case of J-Sw, it

increases sharply when Ind-Sw (i.e. the switching schemes which requires lots of WSSs) is in place. Figure 3.17 (b)

and (c) show the results in terms of number of ROADM degrees when 6 and 12 spatial dimensions are considered,

respectively. The power consumption of Ind-Sw case has a significant different compared to FrJ-Sw and J-Sw case.

This amount of power consumption, which is just due to control unit of WSSs, will be more pronounced and an

important contributor to the operational expenditure of a network owning many nodes. Therefore, utilizing the

INSPACE proposed switching scheme (i.e. J-Sw) will lead to significant power savings compared to the Ind-Sw.

Figure 3.17 Power consumption of control unit of WSSs needed for realizing SDM nodes



Conclusions

In this deliverable, we evaluated the benefits of INSPACE proposed switching solutions in terms of networking

level performance, deployment cost, and power consumption. The main findings are listed below.

We investigated the suitability of three SDM switching paradigms for different traffic profiles in various parts of

a network (i.e. core segment, regional segment, or datacenter interconnection). We found that, for the case of

BuSMFs, Ind-Sw and FrJ-Sw perform well for networks with high level of traffic diversity, while J-Sw is a

favourable option for networks with large demands, considering also its cost benefit. However, J-Sw can perform

significantly better in high diverse traffic scenarios, if spatial Sp-Chs occupy smaller spectral width which can be

switched by WSSs with finer granularity. Therefore, we further investigate the performance of different switching

schemes under various spectral and spatial granularity.

We compared the performance of different SDM switching paradigms, in terms of spectral utilization and data

occupancy under various practically feasible spatial and spectral switching granularities considering the current

WSS realization and an improved resolution WSS technology over a network based on BuSMFs. The network-level

simulation results showed that the performance of all switching paradigms converges as traffic increases while the

switching-related infrastructure cost can be reduced up to 52% in the SG-Sw cases. Additionally, it was shown that,

considering the current WSS technology, the utilization of finer spectral switching granularity significantly

improves the performance of SG-Sw paradigms for small values of traffic, which correspond to small demands in

the traffic matrix. However, as the load increases, the performance of all switching paradigms reduces due to the

less efficient utilization of spectrum arising from a lower amount of occupied spectrum containing actual traffic

compared to the required guard band for the WSSs. We also showed that, by utilizing WSSs with improved

resolution which require 50% less guard band, the performance of switching paradigms in the case of large values of

traffic can be improved by a factor of two. Having said that, it results from our study that, irrespective of the WSS

resolution, large values of ChBW are more beneficial for large values of traffic, and consequently spectral switching

granularity must be adaptable to the traffic size in order to achieve a globally optimum spectrum utilization in an

SDM network, for which spectrally flex-grid ROADMs and bandwidth-variable transceivers are a requirement.

Additionally, we examine the cost benefits of an SDM networks utilizing spatially integrated components. We

showed that going from single-carrier (i.e. 100G case) to spectral and spatial Sp-Ch based approaches results in

huge total cost savings (50%-60%) due to the reduction of the TRx cost (enabled by photonic integration and

component sharing), which dominates the total cost of the network. After the TRx cost, whose contribution to the

total cost ranges from 45% to 58%, the second most costly element is the A/D nodes (consisting of MCSs and

WSSs), with 36%-40% of the total cost of the network. The SDM cases under study exploiting spatial Sp-Ch TRx

offer a total cost benefit ranging from only 4% to 15% compared to the case based on spectral Sp-Ch TRx.

Excluding TRxs and A/D nodes, the cost of the rest of the elements, with 50%-55% and ~63% savings for OE nodes

and EDFAs respectively, demonstrate the benefits of joint switching and amplifier integration. All in all, the

relatively small overall cost savings due to the high cost of TRxs and A/D nodes mean that, for an SDM solution to

offer higher cost savings, the focus should be on the reduction of the cost of TRxs and A/D nodes. In order to find a

cost-efficient A/D architecture we further investigate various CD(C) ROADM architectures. It is shown that an

architecture that maximizes the number of A/D ports, while keeping the pass-through WSSs port count low,

achieves the best cost performance. The study also reveals that J-Sw based ROADMs are more cost effective than

those implementing Ind-Sw and FrJ-Sw, except for one of the designs, where the Ind-Sw based realization, by

taking advantage of the fact that M×N WSSs with port count 3×52 maximize the available A/D ports with very

small increase in the port count of pass-through WSSs, outperforms the other two switching strategies.

Finally, we made some efforts to estimate the power consumption of SDM network based on alternative

transmission media which requires MIMO-DSP modules with different levels of complexity. In particular, the

reach/power consumption tradeoff of SDM networks based on the power consumption of MIMO-DSP has been

investigated considering up to 6-spatial channels. It has been shown that the MIMO-DSP for a 4LP-FMF network

requires more than twice the power required for SMF and 2LP-FMF based networks to achieve similar performance.

Furthermore, we estimate the power consumption scaling of the control unit of WSSs required for the realization of

SDM nodes. We showed that utilizing the INSPACE proposed switching scheme (i.e. J-Sw) will lead to significant

power savings compared to the Ind-Sw.



Annex – abbreviations and acronyms

A/D: add/drop

AWG: arrayed waveguide grating

BP: blocking probability

BuSMF: bundle of single mode fibers

CD: colorless/directionless

CDC: colorless/directionless/contention-less

CDF: cumulative density function

ChBW: spectral channel width

DMGD: differential mode group delay

DSPaF: degenerate space-first

EDF: erbium-doped fiber

EDFA: erbium-doped fiber amplifier

FFT: fast Fourier transform

FMF: few-mode fibre

FM-MCF: few mode multi core fiber

FrJ-Sw: fractional-joint switching

GN: Gaussian noise

HPC: high port count

Ind-Sw: independent switching

J-Sw: joint switching

KSP: k-shortest path

LC: lane change

LP: linearly polarized

MC: multi-carrier

MCF: multi-core fibre

MC-FMF: multi-core few-mode fibre

MIMO: multiple-input multiple-output

MLR: multi line rate

MMF: multi-mode fibre

RA: resource allocation

ROADM: reconfigurable optical add drop multiplexer

RSMSA: routing, space, modulation level, and spectrum allocation

RSSA: routing, spectrum and space allocation

SA: spectrum assignment

SC: single-carrier

SDM: space division multiplexing

SE: spectral efficiency

SMF: single-mode fiber

SP: spectral penalty

Sp-Ch: super-channel



TRC: total relative cost

TRx: transceiver

WSS: wavelength selective switch

XT: crosstalk



References

[1] P. J. Winzer, D. T. Neilson, “From scaling disparities to integrated parallelism: A decathlon for a decade,” J. Lightw.

Technol., DOI: 10.1109/JLT.2017.2662082.

[2] P. S. Khodashenas, J. M. Rivas-Moscoso, D. Siracusa, F. Pederzolli, B. Shariati, D. Klonidis, E. Salvadori, and I.

Tomkos, “Comparison of spectral and spatial super-channel allocation schemes for SDM networks”, J. Lightw.

Technol., vol. 34, no. 11, pp. 2710-2716, June 2016.

[3] P. M. Krummrich, S. Akhtari, “Selection of energy optimized pump concepts for multi core and multi-mode erbium

doped fiber amplifiers,” J. Opt. Exp. Vol. 22, no. 24, pp. 30267-30280, 2014.

[4] E. Palkopoulou, G. Bosco, A. Carena, D. Klonidis, P. Poggiolini, I. Tomkos, “Nyquist-WDM-based flexible optical

networks: exploring physical layer design parameters”, J. Lightw. Technol., vol. 31, no. 14, pp. 2332-2339, Jul 2013.

[5] J. M. Rivas-Moscoso, S. Ben-Ezra, P. S. Khodashenas, D. M. Marom, D. Klonidis, P. Zakynthinos and I. Tomkos,

"Cost and Power Consumption Model for Flexible Super-channel Transmission with All-Optical Sub-channel

Add/Drop Capability" (invited) ICTON 2015, paper Th.B2.5.

[6] D. M. Marom, P. D. Colbourne, A. D’Errico, N. K. Fontaine, Y. Ikuma, R. Proietti, L. Zong, J. M. Rivas-Moscoso,

I. Tomkos, “Survey of photonic switching architectures and technologies in support of spatially and spectrally

flexible optical networking [Invited],” J. Opt. Commun. Netw., vol. 9, no. 1, pp. 1-26, Jan., 2017.

[7] Y. Jung, et al., “Will SDM Systems ever make sufficient financial sense to justify extensive commercial deployment

in terrestrial networks? If so, when, where and why?” OFC 2016, Workshop M1E.

[8] L. E. Nelson, M. D. Feuer, K. Abedin, X. Zhou, T. F. Taunay, J. M. Fini, B. Zhu, R. Isaac, R. Harel, G. Cohen, and

D. M. Marom, “Spatial superchannel routing in a two-span ROADM system for space division multiplexing,” J.

Lightw. Technol., vol. 32, no. 4, pp. 783-789, 2014.

[9] B. Huang, N. K. Fontaine, R. Ryf, B. Guan, S. G. Leon-Saval, R. Shubochkin, Y. Sun, R. Lingle, and G. Li, “All-

fiber mode-group-selective photonic lantern using graded-index multimode fibers,” Opt. Express, vol. 23, no. 1, pp.

224-234, 2015.

[10] M. Feuer, et al., PTL (2012), 24(21), pp.1957-1960.

[11] Y. Li, L. Gao, G. Shen, L. Peng, “Impact of ROADM colorless, directionless, and contentionless (CDC) features on

optical network performance,” J. Opt. Commun. Netw. vol. 4, no. 11, pp. B58-B67, 2012.

[12] http://www.ntt-electronics.com/en/news/2016/3/industry-first-16nm-100g-200g-coherent-dsp.html, accessed

Oct. 2016.

[13] D. C. Kilper, G. Atkinson, S. K. Korotky, S. Goyal, P. Vetter, D. Suvakovic, O. Blume, “Power Trends in

Communication Networks,” J. Sel. Topics Quantum Electron., vol. 17, no. 2, pp. 275-284, 2011.

[14] S. Randel, P. J. Winzer, M. Montoliu, R. Ryf, “Complexity analysis of adaptive frequency-domain equalization for

MIMO-SDM transmission,” in ECOC 2013, paper. Th.2.C.4.

[15] B. Inan, B. Spinnler, F. Ferreira, D. Borne, A. Lobato, S. Adhikari, V. A. J. M. Sleiffer, M. Kuschnerov, N. Hanik, S.

L. Jansen, “DSP complexity of mode-division multiplexed receivers,” Opt. Express, vol. 20, no. 9, pp. 10859-10869,

2012.

[16] S. J. Savory, “Digital signal processing for coherent systems,” in OFC 2012, paper. OTh3C.7.

[17] R. Maruyama, N. Kuwaki, S. Matsuo, M. Ohashiet, “Two mode optical fibers with low and flattened differential

modal delay suitable for WDM-MIMO combined system,” Opt. Express, vol. 22, no. 12, pp. 14311-14321, 2014.

[18] R. Ryf, S. Randel, N. K. Fontaine, X. Palou, E. Burrows, S. Corteselli, S. Chandrasekhar, A. H. Gnauck, C. Xie, R-J.

Essiambre, P. J. Winzer, R. Delbue, P. Pupalaikis, A. Sureka, Y. Sun, L. Grüner-Nielsen, R. V. Jensen, R. Lingle,

“708-km combined WDM/SDM transmission over few-mode fiber supporting 12 spatial and polarization modes,”

in ECOC 2013, paper. We.2.D.1.

[19] S. J. Savory, “Digital signal processing for multilevel modulation formats,”in ECOC 2016, paper. Tu.2.A.1.

[20] http://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html, accessed Oct. 2016.

http://www.ntt-electronics.com/en/news/2016/3/industry-first-16nm-100g-200g-coherent-dsp.html

http://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html



[21] Tiejun J. Xia, Herve Fevrier, Ting Wang, and Toshio Morioka, “Introduction of spectrally and spatially flexible

optical networks” IEEE Commun. Mag., vol. 53, no. 2, pp. 24-33, Feb. 2015.

[22] P. Khodashenas, J. M. Rivas-Moscoso, B. Shariati, D. Marom, D. Klonidis, I. Tomkos, “Investigation of spectrum

granularity for performance optimization of flexible Nyquist-WDM-based optical networks”, IEEE J. Lightw.

Technol., vol. 33, no. 23, pp. 4767-4774, Dec. 2015.

Date post:	16-Apr-2018
Category:	Documents
Upload:	hanhi
View:	217 times
Download:	2 times

Evaluation of the INSPACE system benefits - CORDIS · Figure 3.10 Two M×N WSS-based A/D module...

Documents