Impact of Physical Layer Impairments on C+L-band Metro Networks

Filipa F. Gomes

1

, Lu

´

ıs G. Cancela

1,2

and Jo

˜

ao L. Rebola

1,2

work, we study the filterless drop and waste amplified and unamplified solutions used in metro networks with

L-band tributary nodes. The cost and power consumption of these nodes is evaluated, as well as the impact of

several physical layer impairments on the network performance. In amplified solutions, we show that L-band

networks with 9 tributary nodes are feasible, being the optical signal-to-noise ratio achieved when considering

amplifiers noise and fiber non-linear interference above 14 dB. In unamplified solutions, to fullfil the power

budget, frequency reuse should be used between spans. We show that the in-band crosstalk caused by this

frequency reuse leads to 1.5 dB power budget degradation, for a crosstalk level of -20 dB.

1 INTRODUCTION

To offer the best Internet experience, nowadays, it is

necessary to resort to cloud services which, in the

opens the opportunity for cheaper solutions than plac-

ing several fibres in parallel.

Filterless solutions, both amplified and unampli-

fied, are being proposed to reduce even more the cost

of upgrading the optical networks to the L-band (Em-

merich et al., 2020b), (Paolucci et al., 2020). In

particular, the filterless solution is usually applied to

networks with horseshoe topology (Emmerich et al.,

2020b). In (Paolucci et al., 2020), the L-band filter-

less unamplified solution is proposed, but due to fre-

neous emission (ASE) noise and fiber non-linear in-

terference (NLI) become the main impairments that

degrade the network performance.

In all these works, regarding filterless L-band

transmission, we have seen that they resort on ex-

perimental results and simulations to study the up-

grade of the metro area networks. However, an an-

alytical study of the filterless solutions has not been

performed in these works.

In this work, we analyse the node components and

architectures for the C+L-band, and their cost and

tio (OSNR) of the horsehoe topology is studied for

filterless L-band amplified solutions and the influence

of in-band crosstalk on the electrical signal-to-noise

ratio (SNR) is assessed for unamplified solutions.

This work is organized as follows. In section 2,

we review the L-band node architectures amplified

and unamplified solutions for the C+L transmission.

In section 3, we perform an analysis regarding the

L-band hardware cost and power consumption. In

section 4, the several PLIs that impair the L-band

2 NODE ARCHITECTURE FOR

C+L-BAND

Several solutions regarding metro networks with C+L

band transmission rely on the horseshoe network

tributary nodes, and perform the interface with ac-

cess metro networks) (Paolucci et al., 2020), (Em-

merich et al., 2020a). As shown in Fig. 1, a C+L

tributary node architecture has two independent struc-

tures, one for C-band signals and other for L-band sig-

nals (Emmerich et al., 2020a). These two bands are

separated/aggregated by a C/L band filter at the in-

put/output of the node, which typically provides high

bands isolation and low insertion losses, introducing a

300 to 500 GHz gap between the C and L bands. The

such as filterless drop and waste (FD&W), fixed op-

tical add/drop multiplexers (FOADM) and reconfig-

urable optical add/drop multiplexer (ROADM) solu-

tions (Emmerich et al., 2020a).

Figure 1: Tributary C+L-band node configuration.

has no filtering in the tributary nodes, which means

that when a wavelength is added to the network, it re-

mains in the network until the hub node removes it

(amplified solution) or until it is reused once again in

the network (unamplified solution) (Emmerich et al.,

2020b), (Paolucci et al., 2020). The hub node L-band

structure typically uses ROADMs as in the C-band

structure counterpart (Hernandez et al., 2020).

shoe topology, where each tributary node has a spe-

cific wavelength to communicate with the hub node.

This solution does not allow the reuse of frequencies

inside the network since each wavelength can only

be extracted by the hub node (represented in pink in

Fig. 2). The tributary nodes (represented in green in

Fig. 2) do not have such capacity since they are based

signed to the connection between the hub node 1 and

reuse for the horseshoe topology. Only the L-band wave-

length assignment for the downlink transmission is de-

picted.

The main goal of the L-band unamplified solution is

to reduce the nodes cost by not using OAs. This so-

lution limits the L-band transmission distance to ad-

jacent nodes or near adjacent nodes (Emmerich et al.,

2020a).

ment for the unamplified solution, to illustrate a wave-

length reuse possibility. As shown in Fig. 3, the wave-

after one hop. As the WDM channel is not amplified,

it arrives to the next node with very low power, mean-

ing that a new channel with the same wavelength can

be used. However, the (even weak) power leakage on

node H can impair the new channel leading to in-band

In Fig. 5, we consider that each 5-degree hub node

is connected to 3 different hub nodes and two tribu-

tary nodes. So, for an A/D ratio of 20%, a total of

112 wavelengths must be added/dropped in each hub

node. As wavelength selective switches (WSSs) with

maximum dimension of 1:20 are considered (Hernan-

dez et al., 2020), it is necessary to use 6 cards in

the A/D structure to achieve this number of wave-

and hub nodes in the L-band is, respectively, 7 Tbps

and 28 Tbps, determined from the number of WDM

channels and the transponder capacity, which is, re-

spectively, 70 channels and 100 Gbps for the tributary

node and, respectively, 140 channels and 200 Gbps

between hub nodes. The total capacity per node pre-

sented corresponds to the capacity per A/D structure.

So, the capacity per tributary node is 2.8 Tbps and the

4 PHYSICAL LAYER

IMPAIRMENTS IN THE

where p

is the average signal power at the optical

receiver input, p

power and p

is the NLI power at the receiver in-

dB/km, slightly higher than in the C-band. The chan-

nel bandwidth of 26.75 GHz is equal to the symbol

rate, R

, and the WDM channel spacing is 100 GHz.

Fig. 6 represents the spectral occupancy of the

70 channels belonging to the L-band and of the

40 C-band channels, considering the International

Telecommunication Union (ITU)-T 100 GHz channel

grid (Telecommunication Standardization Sector of

ering 40 channels for C-band and 70 channels for L-band.

Regarding the optical filtering effect in the L-

band, only the WSSs in the hub nodes contribute to

this effect. Hence, the effect of the optical filtering is

neglected in the following analysis.

In this section, we present the insertion losses of the

amplified and unamplified solutions. The losses of

each hardware element are those considered in (Her-

nandez et al., 2020). The WSS losses, which are in-

dependent of the WSS size are 7 dB (Sequeira et al.,

and express paths of the tributary and hub nodes are

shown. The fiber losses after 10 km are 2.5 dB. The

hub node leads to the highest losses, 25 dB. For the

tributary node, the add and drop paths losses are the

same, 19.6 dB.

Table 2: Insertion losses in the L-band, of the tributary node

the hub node.

Tributary node

Add

path

S/C 16:1+S/C 2:1+C/L filter=19.6 dB

path

Drop

The amplifier gains are designed to perfectly com-

pensate the insertion losses of the worst path inside

the node, the add path. For tributary nodes, only pre-

amplification is used and the amplifier gain is 22.1 dB

1) channel insertion at the hub node 1 and drop at

node I in Fig. 2, and 2) channel insertion at node

A and drop at hub node 2. The accumulation of ASE

noise is quantified by the OSNR using (1), firstly, con-

sidering the absence of NLI.

The accumulated ASE noise power is given by:

where n

is the number of spans, p

ASE noise power per polarization generated at each i-

th pre-amplifier output and p

is the ASE noise

power per polarization generated at each i-th post-

2

where m=pre or pos, h is the Planck’s constant and

is the optical frequency. The ASE noise power as a

function of the channel frequency (in lowpass equiv-

alent definition) after nine spans of the horseshoe net-

work for the cases of channel insertion at tributary

node A and hub node 1 is shown in Fig. 7. The ASE

Figure 7: Accumulated ASE noise power as a function of

the channel frequency after 9 spans of the horseshoe net-

work for channel insertion in hub 1 and tributary A nodes.

Fig. 8 shows the OSNR evolution along the net-

work, for both cases of signal insertion, for the chan-

nel frequency of 188 THz. When signal insertion oc-

curs at node 1, the OSNR is significantly degraded

as the signal crosses the nodes, decreasing 9.3 dB.

For signal insertion at node A, the OSNR decreases

7.1 dB. The minimum reference OSNR is 8.5 dB to

In this section, the NLI noise power and the associated

OSNRs are studied for the amplified solution. To per-

Figure 8: OSNR in the horseshoe topology, for signal inser-

tion at the tributary node A and signal insertion at the hub

node 1 for p

Rx

=0 dBm.

form these studies, the NLI calculation and the OSNR

formulation must take into account the presence of

inter-channel Stimulated Raman Scattering (SRS). In

this section, the Gaussian noise model presented in

T

C

f

is the WDM signal

launch power, B

and C

is the slope of the linear regression of the nor-

malized Raman gain spectrum (Semrau et al., 2019).

The total XPM contribution is given by (Semrau et al.,

2019)

SRS.

In Fig. 9 (b), the corresponding NLI power as a

function of the WDM channels frequency is shown.

The NLI power is determined by p

et al., 2019), As can be observed in Fig. 9 (b), the

power, the NLI power remains practically constant for

all frequency channels due to the reduced SRS. Notice

that the results presented for the NLI power are inde-

pendent on where the signal is inserted in the network.

The OSNR as a function of the channel frequency

after 9 spans considering the ASE noise power and

NLI is shown in Fig. 10, for several input chan-

nel powers, including the optimal power per chan-

nel, which leads to the best OSNR, and the maximum

power per channel to comply the minimum OSNR,

As can be seen in Fig. 10, for the input power

per channel of -2 and 0 dBm, the NLI power tilt

has practically no influence, so an almost constant

nels frequency after nine 10 km spans considering coherent

NLI accumulation, for several input powers per channel.

OSNR along all channels is observed. For input pow-

ers higher than 0 dBm, we observe that the OSNR

increases as the channel frequency increases due to

the higher NLI power contribution to the OSNR slope

variation. So, we can state that there is an input chan-

nel power that gives an optimum OSNR. This opti-

mum OSNR is obtained by evaluating the best possi-

ble OSNR, with the highest flatness along the WDM

ing the same received quality for all WDM channels.

When the signal is inserted in the hub node, the op-

timum OSNRs are around 0.5 dB lower than in the

case of channel insertion in the tributary node due to

the higher ASE noise power added by the hub node.

Finally, it can be observed that all channel powers

considered in Fig. 10 comply with the 8.5 dB mini-

Figure 10: OSNR as a function of the channel frequency

after 9 spans considering the ASE noise power and NLI, for

several input channel powers and insertion channel in the

(a) tributary node A and (b) hub node 1.

mum OSNR required, except for the input power of

the minimum OSNR is at least 5.5 dB higher, which

ensures enough margin to include other PLIs. For ex-

ample, if we consider a minimum system margin of

3 dB and an extra power penalty due to other trans-

mission effects of 1.5 dB, like components aging, the

minimum OSNR is still accomplished.

In this subsection, the analysis of PLIs, considering

the signal transmission in the L-band in a horseshoe

topology based on tributary nodes with the FD&W

unamplified solution is performed. In this scenario,

the main PLI is the in-band crosstalk that arises due

to the wavelength reuse in these networks. First, we

present the analytical formalism used to assess the

duce the in-band crosstalk in our formalism and as-

sess the respective degradation in terms of the number

where P

the local oscillator (LO), CMRR is the Common Mode

Rejection Ratio of the balanced photodetector, SNR

is an additional term to include several implementa-

are given, respectively, by (Rizzelli et al., 2021)

, where

bandwidth, q is the electron charge and RIN is the LO

RIN parameter. The parameters used to calculate the

SNR are based on the fitting to experimental results

and are given in (Rizzelli et al., 2021). Only a differ-

ent symbol rate of 26.75 Gbaud is considered.

In Fig. 11, the BER as a function of the received

signal power for different local oscillator powers is

presented, considering QPSK modulation, and ob-

tained from the SNR calculated from Eq. 7. The re-

quired target BER assumed for the system is 4 ×10

leads to at least 2.2 dB of power budget improvement

for SD-FEC in relation to HD-FEC.

From Fig. 11, it can also be observed that the

LO power of 14 dBm leads to the minimum received

Figure 11: BER as a function of the received signal power

for different local oscillator powers P

dBm leads to the higher and most demanding received

power to reach the same target BER. Hence, in the

following results, we consider P

the optimum LO power that leads to the target BER.

So, knowing the minimum received power, the main

goal is to compute the maximum power budget and

In Tab. 3 are shown, the insertion losses consid-

not sufficient to accommodate these insertion losses.

A possible solution to decrease the insertion losses

per span is to reduce the S/C dimension in the A/D

structures. So, in the tributary nodes shown in Fig.

4, the S/Cs 1:16 are replaced by S/Cs with smaller

size, as considered already in some works (Emmerich

et al., 2020b), (Emmerich et al., 2020a), for the un-

amplified solution. The associated insertion losses are

A/D structure of the tributary nodes, and transmission

should not exceed one fiber span. This allows fre-

quency reuse in the horseshoe network, leading to in-

band crosstalk. For the crosstalk level of -20 dB, a 1.5

degradation on the optical power budget is observed.

ACKNOWLEDGEMENTS

This work was supported under the project of Instituto

de Telecomunicac¸