You are on page 1of 13

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224090334

Transmission Impairments in DWDM Networks With Reconfigurable Optical


Add-Drop Multiplexers

Article in Journal of Lightwave Technology · March 2010


DOI: 10.1109/JLT.2009.2037832 · Source: IEEE Xplore

CITATIONS READS
103 4,421

2 authors:

Sorin Tibuleac Mark Filer


Adtran Microsoft
110 PUBLICATIONS 1,802 CITATIONS 63 PUBLICATIONS 925 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optical telecommunications, DWDM transmission View project

Diffractive Optics View project

All content following this page was uploaded by Sorin Tibuleac on 09 December 2014.

The user has requested enhancement of the downloaded file.


JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010 557

Transmission Impairments in DWDM Networks With


Reconfigurable Optical Add-Drop Multiplexers
Sorin Tibuleac, Member, IEEE, and Mark Filer, Member, IEEE

(Invited Paper)

Abstract—Reconfigurable optical add/drop multiplexers for processing and switching. Such network nodes, known as
(ROADMs) based on 1 N wavelength-selective switches (WSS) optical add/drop multiplexer (OADM) nodes use optical filters
are evolving to support DWDM networks with higher capacity to add/drop the desired wavelengths, while passing through all
and increased flexibility in wavelength routing. Different WSS
technologies can be employed to provide colorless and steerable other wavelengths towards other OADM nodes in the network.
functionality for ring, or meshed architectures. Improvements in Over the past few years, there has been a growing trend to
specifications of WSS modules operating on the 50 GHz wave- add wavelength reconfigurability to OADM nodes (ROADMs),
length grid have enabled 40 Gb/s transmission rates through thereby allowing remote changes to the add/drop traffic pattern
extensive ROADM networks. The same ROADMs are also ex- across the network [1].
pected to support 100 Gb/s transmission in the near future. In
parallel, development of lower-cost WSS technologies is allowing
This versatility of ROADM-enabled DWDM systems trans-
ROADMs to expand into edge networks. In all these network lates to several benefits for the network operator including
applications, propagation through multiple ROADMs generates simpler network planning, installation and turn-up, and a
transmission penalties for the DWDM channels, which need to simplified process of adding new wavelengths to the network.
be factored into the network design. Such OSNR or Q factor The elimination of manual fiber connections upon changes in
penalties can be induced by passband narrowing, imperfect iso-
wavelength add/drop pattern also reduces the time required to
lation across the signal bandwidth, insertion loss, PDL, and other
effects. The impact of these impairments depend on the trans- implement these changes and increases network availability. All
mitter and receiver types (e.g., data rate and modulation format), these benefits contribute to a reduction in operating expenses of
and on the WSS characteristics (e.g., insertion loss, passband the DWDM network. Capital expenses can also be reduced in
width, shape, isolation magnitude and isolation stopband). Key ROADM networks by eliminating opto-electronic conversion
transmission impairments such as bandpass narrowing, crosstalk, and switching of each wavelength at fiber junction nodes, and
insertion loss, and PDL are estimated based on experiments and
numerical simulations for common data rates and modulation accomplishing the same function with multi-degree ROADM
formats. Implications of temporal fluctuations during power nodes [2]. Wavelength power equalization, typically available
setting throughout a ROADM network are also discussed. at ROADM nodes, can also increase the optical transmission
Index Terms—Bandwidth limitations, crosstalk, optical fil- distance, eliminating potential opto-electronic regeneration
tering, reconfigurable optical add/drop multiplexer (ROADM), points and their associated cost. Equally important, ROADMs
wavelength-selective switch (WSS). represent a key element required for DWDM networks to
evolve towards new network topologies, from the prevalent
point-to-point and ring based fiber connectivity to meshed
I. INTRODUCTION networks [3]–[5]. New types of telecommunication services,
ENSE WAVELENGTH division multiplexed (DWDM) whereby wavelength connections are scheduled for shorter time
D systems have become the standard technology in high
capacity telecommunication networks ranging in distance
intervals than in today’s networks, are also made possible by
the remote and automated wavelength switching capability of
from a few kilometers to thousands of kilometers. Commer- ROADMs [6], [7].
cial DWDM systems use wavelengths modulated at 2.5, 10, ROADM deployments began in long-haul networks, where
or 40 Gb/s covering the C and L band spectral range with the added cost of reconfigurability represented a smaller
typical frequency spacing between channels of 100 GHz or fraction of the total network cost. Technical innovations, im-
50 GHz. These wavelengths are added and dropped from provements in manufacturing yield, and increasing deployment
the multi-wavelength network fiber at geographical locations volume have reduced the cost of ROADMs and enabled them
where the optical information is converted to electronic format to penetrate into regional and metropolitan DWDM networks
[8], [9]. Multiple types of ROADMs have been proposed and
implemented to provide different combinations of functionality,
Manuscript received November 10, 2009. First published December 15, 2009;
performance, and cost adapted to each network application
current version published February 10, 2010. [3]. The paper reviews some key ROADM architectural op-
The authors are with ADVA Optical Networking, Norcross, GA 30092 USA tions from the simplest 2-degree ROADM node with fixed
(e-mail: stibuleac@advaoptical.com).
Color versions of one or more of the figures in this paper are available online
add/drop wavelengths, to the more flexible colorless or steer-
at http://ieeexplore.ieee.org. able wavelength ROADMs, and more complex multi-degree
Digital Object Identifier 10.1109/JLT.2009.2037832 node architectures required at junction nodes. The following
0733-8724/$26.00 © 2010 IEEE

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
558 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

Fig. 1. ROADM node architectures: 2-degree ROADM with fixed wavelength per port (a), colorless 2-degree ROADM (b), colorless 2-degree ROADM with
power equalization (c) steerable 2-degree ROADM (d), colorless and steerable 2-degree ROADM (e), and 4-degree ROADM (f).

section presents the technologies currently employed in design channels during switching, and the ability to switch wave-
of wavelength-selective switches and their most relevant speci- lengths from other DWDM equipment. Multiple ROADM
fications impacting transmission performance. The paper then implementations are possible [3], with some relevant examples
focuses on the key ROADM-induced transmission impairments: shown in Fig. 1. The key element of a ROADM node is a
insertion loss, polarization-dependent loss, passband effects, wavelength-selective switch (WSS), capable of demultiplexing,
in-band crosstalk, and transient power effects. Measurements multiplexing, and switching DWDM wavelengths between
and simulation results quantify the performance impact of each one, or multiple input fiber ports, and one or more output fiber
impairment type independent of the others. These studies use ports. Common WSS subsystems provide 1 input (or output)
conventional ROADMs operating in systems with typical data port, and N output (or input) ports where N can be up to 9
rates (10 Gb/s and 40 Gb/s) and modulation formats of the at present, but increasing port counts up to 20–23 ports are
DWDM wavelengths (OOK, DPSK, DQPSK). expected in the near future. Prototypes with multiple input and
output ports (NxM) have been reported [10], [11], and may
II. ROADM NODE CONFIGURATIONS find applications in the future. The wavelength selective switch
Throughout this paper, ROADMs are referred to as network is typically also capable of adjusting the attenuation for each
nodes capable of selective and automatic addition and removal wavelength on the output port(s), or blocking it altogether. This
of individual wavelengths from an optical fiber. Beyond this requires additional components to monitor the output power
basic property, there are a number of features typically as- per channel. Other optical components, which may be found
sociated with ROADM nodes, for instance: channel power within a ROADM node are: 1:N power splitters or couplers,
monitoring, the addition and removal of any number and multiplexers and demultiplexers for local wavelengths, taps for
combination of wavelengths, the support of a wide variety power monitoring, amplifiers for pass-through and/or add and
of channel bandwidths, the absence of any impact on other drop channels.

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
TIBULEAC AND FILER: TRANSMISSION IMPAIRMENTS IN DWDM NETWORKS 559

Fig. 1(a) shows a common ROADM node configuration using a free-space grating, or an integrated optical grating (arrayed-
a power splitter to transmit all channels towards the drop and waveguide grating—AWG). A similar diffractive device, often
pass-through paths. Only one direction of propagation through the same component, is used for multiplexing wavelengths into
the ROADM node is shown. The wavelength selective switch a single data stream emerging from an output port. The wave-
with two input ports and one output port selects wavelengths length routing function is accomplished by a number of different
coming from either the multiplexer or the splitter while blocking implementations which include the following.
the same wavelength from the other port. It becomes apparent • 2-degree MEMS mirror array, with one mirror per wave-
that any residual power that is not blocked on the pass-through length, where one rotation axis provides wavelength
path becomes in-band crosstalk for the add channel [12], [13]. routing, while the second offsets the beam from the fiber
This potential problem can be alleviated by replacing the power
core thereby controlling attenuation [14].
splitter with another WSS operating with one input and 2 output
• Liquid Crystals (LC) arrays, each array using polarization
ports, thereby improving the isolation between the add and pass-
through channels. However, the crosstalk penalties in a single control to provide 1 2 WSS functionality, including at-
WSS ROADM are usually low and can be accommodated in tenuation control [15], [16].
the system OSNR or Q factor budget, while the increased cost • Combined 1-degree MEMS and Liquid Crystals, where the
of the second WSS makes this solution less desirable in practical MEMSs achieve wavelength switching, while the LC’s set
applications. the desired attenuation [15].
Although this ROADM node offers the choice of passing • Liquid Crystal on Silicon (LCOS), which consists of a
through or adding any combination of wavelengths from the net- large liquid crystal array, similar to one that would be used
work fiber, the wavelength-specific connections between the de- in video display applications, with a multitude of LC’s per
multiplexer and the receiver prevents a transponder from being wavelength fulfilling both switching and attenuation func-
used on any other wavelengths without manual fiber reconnec- tions [17], [18].
tion. Additional flexibility is provided by the wavelength-inde- • Digital MEMS using a different video display technology
pendent (colorless) ROADM configuration employing a 1 N consisting of an array of MEMS mirrors with two possible
WSS and an N:1 coupler (Fig. 1(b)). This ROADM type used in angular positions, and multiple mirrors assigned to a single
conjunction with transponders employing tunable lasers allows wavelength [5].
any wavelength to be terminated at a given transponder [3]. Note • Integrated planar lightwave circuit (iPLC) devices fabri-
that in this case, power equalization cannot be accomplished be- cated in Silica or polymer waveguides which include an
tween the channels on the add and express paths. An alternative array of 2 1 switches and variable optical attenuators
colorless ROADM design with full wavelength equalization ca-
[19]–[21].
pability is achieved (Fig. 1(c)) by combining the 2 1 WSS and
A different category of ROADMs can be implemented with
1:2 coupler of Fig. 1(a) with the 1 N WSS and N:1 coupler of
Fig. 1(b). tunable filters using one of several technologies [5], [22]. A tun-
Another avenue for increasing the functionality of a ROADM able wavelength filter can separate a single wavelength from a
node is to allow, in addition to an express path through the DWDM spectrum. Multiple filters in sequence can be used to
node, multiple add/drop paths for a given transponder. Fig. 1(d) add/drop a corresponding number of wavelengths at a ROADM
illustrates such a steerable (or directionless) ROADM where node. Alternatively, filters with wavelength and bandwidth tun-
a 2 1 WSS on the drop path selects a wavelength from ei- ability have also been proposed as a solution for add/drop of 4–8
ther east or west direction of propagation through the trans- DWDM channels with a single device [5].
mission fiber. These wavelength and direction independent add/
drop switching features can be combined into more complex B. WSS Specifications Impacting System Performance
node architectures, with an example shown in Fig. 1(e) [4]. Wavelength-selective switches are characterized by a set
The number of network interfaces defines the degree of optical, electrical, and mechanical parameters specified in
of a ROADM. The ROADM configurations presented in product data sheets. A subset of optical parameters which
Figs. 1(a)–(e) are all 2-degree ROADMs. A multi-degree impact the transmission performance of the DWDM system
ROADM is achieved using several 1:N splitters and N 1 is presented in Table I, along with typically-specified values
WSS modules, where the N output ports on each splitter are
for each parameter across the different technologies used for
connected to an input port on each of the N 1 WSS modules
WSS implementation. These values represent the worst-case
(Fig. 1(f)) [3]. This results in the equivalent of an N N WSS
with the advantages over the single N N WSS offered by over specified temperature range, wavelengths, and across all
the modular implementation. The node can evolve towards polarization states, and may include margin for ageing. Av-
higher degrees of connectivity with incremental cost and with erage values measured at room temperature can be significantly
no impact on existing traffic through the node. The modular better. For instance, average values for insertion loss at room
design also ensures higher equipment reliability and network temperature for an 8 1 WSS operating on the 100 GHz grid
availability. can be in the 3–4 dB range, whereas the specifications are
1–2 dB higher. Higher loss is expected on WSS subsystems
III. ROADM TECHNOLOGIES AND SPECIFICATIONS designed for 50 GHz channel spacing.
The transmission bandwidth of the WSS has a significant im-
A. Technologies for Wavelength-Selective Switches pact on the transmission performance, particularly for wide-
Wavelength-selective switching is achieved with parallel de- bandwidth (high-data rate) signals [23], [24]. Since different
multiplexing of all wavelengths on the incident fiber port using WSS implementations may yield different passband shapes, it is

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
560 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

TABLE I
KEY SPECIFICATIONS OF WSS SUBSYSTEMS FOR 50 GHZ OR 100 GHZ
CHANNEL SPACING UNDER ALL OPERATING CONDITIONS

Fig. 2. Decrease in OSNR with number of 20 dB or 25 dB spans for 2 cases of


ROADM insertion loss: 12 dB and 20 dB.

common to define the bandwidth at two different values below


the peak, e.g., at 0.5 dB and at 3 dB below the transmission loss of the power splitter, the ROADM node loss varies with
maximum. A third characteristic of the passband is the offset the loss specification of the ROADM’s main building block,
with respect to the wavelength specified for the DWDM sig- the WSS. The power levels entering the post-amplifier also
nals [25]. The passband and the offset can be combined into a depend on the power variation across the spectrum resulting
single passband specification defined as twice the minimum be- from propagation through upstream fiber and other network
tween the two half-bandwidths at 0.5, or 3 dB below the trans- elements. The ROADM performs wavelength equalization by
mittance at the reference (ITU-T) wavelength [26]. The isolation reducing the power on all wavelengths to the lowest channel
is a critical parameter, which determines the in-band crosstalk power within the spectrum. Therefore, the difference in power
in a DWDM system. However, a simple definition of minimum between the average channel power and the lowest channel
isolation over a fixed bandwidth is not adequate to characterize power, and the accuracy in power setting also need to be
the impact of this parameter on the system performance [12], considered as contributors to the total ROADM loss [27], [26].
[13], as discussed in more detail in Section 6. As an example, the impact of ROADM insertion loss values
The next sections are dedicated to the ROADM-induced im- on the transmission distance of an 80-channel long-haul DWDM
pairments which have the highest impact on the transmission network with identical ROADM nodes is analyzed for two cases
performance of the DWDM system: insertion loss and polariza- of total ROADM loss: 12 dB and 20 dB. The 12 dB node can
tion-dependent loss (PDL), passband, and isolation. Other rel- correspond to a 2-degree ROADM (Fig. 1(a)) while the 20 dB
evant specifications with potential impact on channel BER, but loss may represent an 8-degree ROADM (similar to Fig. 1(f)).
not discussed are polarization-mode dispersion (PMD), chro- Each node is assumed to have an EDFA pre-amplifier and an
matic dispersion and phase ripple, accuracy of power setting, EDFA post-amplifier to compensate fiber span and node loss,
and power stability. respectively. The reduction in OSNR with span number is illus-
trated in Fig. 2 for span loss values of 20 dB and 25 dB. The
IV. LOSS IMPACT amplifiers are assumed to have noise figure values:
dB at 12 dB gain, dB at 20 dB gain and
dB at 25 dB gain. The relatively minor contributions to OSNR
A. Insertion Loss and Optical Transmission Distance
degradation of EDFA gain ripple, and uncompensated fiber tilt
The reduction of insertion loss is a primary goal of any are also included. The maximum propagation distance is deter-
DWDM network design. The attenuation of signals passing mined by the intersection of the link OSNR curve with the line
through a ROADM node is compensated by amplifiers, which representing the minimum OSNR required at the receiver for a
increase signal power at the cost of added noise. The accu- specified BER. The additional number of spans supported with
mulated noise throughout the optical link and the associated the lower loss ROADM is higher for 20 dB spans than for 25
reduction in optical signal to noise ratio (OSNR) limits the dB spans. For a minimum required OSNR at the receiver of 16
maximum distance for propagation of optical signals with dB, the lower loss ROADM offers an increase in transmission
acceptable bit error rate (BER). Hence, there is a desire to distance by 4 spans.
reduce the loss and the amplification requirements throughout a In metropolitan networks characterized by shorter span
DWDM network. The insertion loss of a ROADM node depends lengths, lower loss ROADMs offer the possibility of elimi-
largely on the node architecture and the type of components that nating the post-amplifier, thereby simplifying the network, and
are required to support it. For example, multi-degree ROADMs reducing its cost. For example, a network with 10 dB spans
employing 1:8 power splitters and 8 1 WSS modules on the requires post-amplification in the case of the 20 dB ROADM
pass-through path incur higher loss than two-degree ROADMs loss. By reducing the ROADM loss to 12 dB, the network can
using a 1:2 splitter and a 2 1 WSS per direction of propa- be supported without post-amplifiers at approximately the same
gation. For a particular node architecture, and given the fixed OSNR values across the link.

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
TIBULEAC AND FILER: TRANSMISSION IMPAIRMENTS IN DWDM NETWORKS 561

verified in a recirculating loop [33] experiment performed with


a DWDM signal generated by a 10 Gb/s NRZ-OOK transmitter
using a Lithium Niobate modulator with zero chirp and 12
dB extinction ratio. The signal was transmitted multiple times
through the same sequence of three 50 GHz ROADM nodes.
Each ROADM node consisted of a 100 GHz WSS, which in-
cludes a pair of arrayed waveguide gratings to demultiplex and
multiplex 100 GHz spaced channels, a planar lightwave circuit
(PLC) array of 2 1 switches, and a pair of interleaver filters
to separate and recombine the 50 GHz spectrum into, and from
two sets of 100 GHz-spaced DWDM channels. The receiver
Fig. 3. OSNR evolution across multiple 20 dB or 25 dB spans with PDL values
of 0, 0.5, and 1.0 dB, assuming fixed ROADM loss of 12 dB.
was based on a PiN photodiode and a limiting amplifier with
optimized decision threshold. Controlled amounts of ASE were
added at the receiver to replicate the lower OSNR that may be
B. Polarization-Dependent Loss obtained in a real network. The loop included an EDFA for
A similar degradation in OSNR results from the polariza- each ROADM node and an additional amplifier to compensate
tion-dependent loss (PDL) of the ROADMs [28]. Although most the loss of loop components (switch and attenuators). A short
ROADMs have the capability to adjust power levels, and re- length of standard single-mode fiber (10 km) with matching
duce any slow variations in power originating from upstream DCM was employed to minimize any effects from fiber and iso-
nodes, or from the same node, PDL-induced power changes can late only the passband effects. Fig. 4(a) shows the transmittance
occur at a faster rate than the typical response time of a ROADM on two DWDM channels of the individual interleavers and 100
system (Table I). This power variation increases in a statistical GHz PLC WSS with 3 dB bandwidths in the range 38–42 GHz,
manner with number of ROADM nodes present in the optical and 62–65 GHz, respectively. The measured transmittance
link [29]. A network designer therefore needs to estimate the after propagation eight times through the same sequence of
impact of PDL when calculating the lowest OSNR across all three PLC ROADMs and three pairs of interleaver filters is
channels propagating on an optical path. On most WSS sub- shown in Fig. 4(b). An insignificantly small, but systematic dip
systems PDL increases with attenuation setting. Since it is dif- (0.1–0.2 dB) in the center of the transmittance function of each
ficult to associate specific attenuation values required on each interleaver is enhanced by the 48 traversals, generating a notch
wavelength for equalization of the DWDM spectrum, a statis- around the transmitter wavelengths of up to 5 dB. The impact
tical worst-case assumption for PDL value has to be made for of the recirculating loop set-up itself is verified with Q factor
reliable network design. Fig. 3 shows the impact of PDL on the measurements performed on the same experimental configura-
OSNR and the optical transmission distance for optical links tion with attenuators replacing the ROADMs. No measureable
with identical spans and ROADM nodes for span loss values of difference is recorded between the Q factors of the signals
20 dB and 25 dB, and PDL values for each ROADM of 0 dB, bypassing the recirculating loop or traversing it 8 times without
0.5 dB, and 1 dB. The ROADM node loss is assumed to be 12 the ROADMs. In spite of the severe passband distortion of the
dB in all cases. For a minimum OSNR required at the receiver cascaded ROADMs, the Q factor values resulting from BER
equal to 16 dB, an increase in PDL from 0.5 dB to 1 dB reduces measurements at an dB are between 13.5–14.5
the maximum number of spans from 23 to 18. dB for 38 wavelengths across the spectrum (Fig. 5). These Q
factor values are well above the FEC correction limit of 9.5
V. PASSBAND EFFECTS dB, demonstrating robust performance of the DWDM system
with 24 50 GHz ROADMs. The experiment was repeated with
A. Performance With 10 Gb/s NRZ-OOK Modulation the identical set-up except for a different receiver which has
The transmission passband of the ROADM can be the a fixed decision threshold. In this case, Q penalties between
most significant contributor to performance degradation in a 0.5–1.5 dB were measured across different wavelengths passed
DWDM network, and this topic has been the subject of several through the 24 50 GHz ROADMs. A clear dependence of the
research studies published to date [23], [24], [30]–[32]. The passband penalties on the receiver design emerges from these
transmission penalties incurred by ROADMs can be attributed experimental studies.
to filtering/shaping of the frequency spectrum of the optical
communication signals and distortion of the optical pulses as B. ROADMs in Access Networks
they propagate through a sequence of ROADM nodes. This can The impact of the ROADM bandwidth is aggravated by rela-
be generated by the limited bandwidth of the ROADM in com- tive detuning between the center of the passband and the trans-
parison to the signal bandwidth, or by the amplitude and phase mitter wavelength, which may occur due to limited manufac-
variations within the signal bandwidth accumulated with prop- turing tolerances, temperature, and ageing effects. The spectral
agation through multiple ROADMs. Signals with modulation shift of the ROADM passband from the ITU-T grid can be up
rates of 10 Gb/s and lower using standard non-return-to-zero to nm, while the wavelength offset of the laser depends
on/off keying (NRZ-OOK) modulation format are robust to on the transmitter design. Core networks typically use wave-
passband narrowing and shaping induced by a large sequence length-locked lasers with wavelength setting accuracy specified
of typical ROADM filter transmittance functions. This was over the lifetime of the network to be within nm from

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
562 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

Fig. 6. OSNR penalty generated by propagation of 10 Gb/s NRZ-OOK chan-


nels up to 6 times through the loop (6 TF-ROADMs per loop) with different
laser detuning values from the ITU-T grid.

wavelength. This generates a much larger relative detuning be-


tween the ROADM passband and the signal wavelength, and po-
tentially higher transmission penalties. The impact of cascaded
bandwidth under wavelength drift conditions was studied [30]
for a metro-access ROADM, comprised of a 3-port tunable filter
able to drop (or add) between 1–8 wavelengths on a 100 GHz
ITU grid on one port, and pass-through the remaining wave-
lengths on the other output (or input) port of this tunable-filter
ROADM (TF-ROADM).
A tunable laser was modulated with NRZ-OOK by a Lithium
Niobate modulator at 10.7 Gb/s, and transmitted through the
recirculating loop with six TF-ROADMs within the loop. The
signal was detected with a PiN receiver with optimized deci-
sion threshold. Optical filtering of the signal passing through the
TF-ROADMs occurs only in the case when an adjacent channel
in the spectrum is added or dropped. Therefore, in an effort to
Fig. 4. Transmittance spectrum of 3 individual PLC ROADMs and 6 inter- measure potential worst-case filtering, the devices were set to
leavers used in recirculating loop experiments (a), and of the three 50 GHz pass the test channel and drop the adjacent channel, alternating
ROADMs cascaded 8 times in the recirculating loop (b).
between the lower and higher frequency neighbor from one de-
vice to the next. An ASE source with variable output power cou-
pled to the test wavelength in front of the receiver was used to
add controlled amounts of noise. The BER was measured as
a function of the OSNR at the receiver for one to six signal
passes through the loop, corresponding to propagation through
6–36 devices. The measurements were then repeated with the
TF-ROADMs replaced by an increased loss in the loop VOA to
determine the OSNR penalty incurred by transmission through
the cascaded TF-ROADM devices. The OSNR penalties were
repeated for different wavelength detuning values from the ITU
grid. Fig. 6 summarizes the results across different detuning
values and number of passes through the recirculating loop. For
locked lasers, the penalties generated by propagation through up
Fig. 5. Q factor measurements across different 10 Gb/s NRZ-OOK channels to 36 devices are negligible, and lie within the margin of mea-
transmitted 8 times through the recirculating loop with three 50 GHz ROADMS surement error. The OSNR penalties increase for large detuning
in the loop. Reference measurements at 3 wavelengths with ROADMs replaced values, but remain below 0.8 dB in the range corresponding to
by attenuators in the loop, and in back-to-back configuration are also shown;
OSNR = 15 dB for all measurements. typical wavelength drift of unlocked lasers (up to nm).

C. Core Networks With High Spectral Density DWDM Systems


the ITU-T grid. The cost sensitivity of DWDM access networks Core metro and long-haul networks continue to expand
require ROADMs to operate with low-cost transmitters with un- transmission capacity by increasing the modulation rate of
locked lasers which may drift up to nm from the nominal the DWDM wavelengths (at 40 Gb/s and 100 Gb/s) while

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
TIBULEAC AND FILER: TRANSMISSION IMPAIRMENTS IN DWDM NETWORKS 563

Fig. 7. Bandwidth narrowing with increasing number of filters in the optical


path, for different Gaussian orders, and for typical bandwidths of 100 GHz
ROADMs (70 GHz) and 50 GHz ROADMs (40 GHz).
Fig. 8. Spectrum of 40 Gb/s NRZ-DPSK and RZ-DQPSK relative to typical
passbands of 100 GHz and 50 GHz ROADMs.
maintaining the channel spacing at 50 GHz. Under these condi-
tions of high spectral density, passband narrowing in ROADM
networks can become the limiting factor, which determines the
maximum number of ROADM nodes allowed in an all-optical
transmission link. The gradual narrowing of the passband with
increasing number of ROADMs on the optical path is depicted
in Fig. 7 for two 3 dB bandwidths, 70 GHz, and 40 GHz, corre-
sponding to typical values for 100 GHz and 50 GHz ROADMs,
respectively. The shape of the passband, represented in this
example by Gaussian-type filters of different orders, plays an
important role in the rate of bandwidth decrease with number
of ROADM units. For example, the 3-dB bandwidth resulting
from a sequence of 16 ROADMs, where each ROADM has
a 2nd order Gaussian filter shape with 40 GHz bandwidth
is 19 GHz. The same number of ROADMs with the same
3 dB bandwidth but with a 4th order Gaussian filter shape
yields a cumulative 3-dB bandwidth of 27 GHz. First order Fig. 9. Required OSNR at BER 1E-5 for 40 Gb/s RZ-DPSK transmission
through 100 GHz and 50 GHz ROADM nodes as a function of net bandwidth
Gaussian bandwidths are not shown since they generate a more of the cascaded ROADMs. Results with tunable bandwidth filter are also
pronounced bandwidth decrease, which severely limits any presented.
network application.
This practical challenge of transmitting high data rate sig-
nals through increasing number of ROADMs with tighter fre- OSNR penalties can be tolerated on transmission distances up
quency spacing has spurred numerous studies reported to date to 1500 km due to the low OSNR requirements of DPSK re-
on passband effects [23], [24], [32], [34]. Among several mod- ceiver (13.5 dB at the FEC BER threshold of 1E-3).
ulation formats which have been proposed for 40 Gb/s trans- A study of passband effects on 40 Gb/s RZ-DPSK transmis-
mission [35], differential phase-shift keying (DPSK) [23], and sion was conducted in a linear link with separate 100 GHz or
differential quadrature phase-shift keying (DQPSK) [36] have 50 GHz ROADM devices. The DPSK signal was transmitted
become the most widely used in commercial networks due to through up to ten 100 GHz PLC ROADMs with AWG multi-
the low required OSNR and high tolerance to nonlinear effects plexers and demultiplexers characterized by a 3 dB bandwidth
and passband narrowing. However, as illustrated in Fig. 8, the of 67 GHz and a passband shape approximated by a Gaussian
spectra of the RZ-DPSK and NRZ-DPSK modulation formats order 1.8. No significant length of fiber was used in order to iso-
at the 40 Gb/s data rate are relatively wide compared to a typ- late the passband impact from fiber-induced effects. The BER
ical 3-dB bandwidth of a ROADM passband for 100 GHz (70 was measured as a function of OSNR, which varied with addi-
GHz) and even more so relative to the 3-dB bandwidth of a 50 tional ASE inserted in front of the receiver. The required OSNR
GHz ROADM (40 GHz). In spite of the severe clipping of the for BER 1E-5 was determined after each ROADM and plotted
high frequencies in the signal, transmission through up to 16 as a function of the accumulated 3 dB bandwidth. The results
ROADMs has been reported with an OSNR penalty of 1.5 dB of Fig. 9 indicate a slight improvement in performance after 10
using NRZ-DPSK with free spectral range (FSR) of the receiver ROADMs resulting in a net bandwidth of GHz. Inserting
delay-line interferometer (DLI) of 66 GHz [23], [37]. Simul- 50/100 GHz interleavers before, and after each PLC ROADM
taneous optimization of the power imbalance between the two to form a sequence of 50 GHz ROADM induces higher required
DLI outputs and of the FSR can further improve the tolerance OSNR for the same BER with a 3 dB increase between the 1st
to optical filtering of DPSK transmission allowing up to 20 cas- and 7th ROADM. Moreover, the required OSNR is higher for
caded ROADMS with less than 1 dB penalty [34]. Such low the 50 GHz ROADMs than for the 100 GHz case even for the

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
564 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

Fig. 10. Measured OSNR penalty at BER 1E-5 as a function of 3 dB bandwidth Fig. 11. Numerical simulations of 40 Gb/s DQPSK eye opening penalty (EOP)
of variable bandwidth filter for 40 Gb/s NRZ-DPSK and RZ-DQPSK. after 4 passes through identical filters as a function of the filter bandwidth and
for filter shapes of different Gaussian orders.

same net bandwidth. This can be attributed to the different pass-


band shape resulting from the two ROADM sequences.
This sensitivity to passband shape is reflected by additional
measurements on a DPSK transponder with a tunable band-
width filter. While some discrepancies can be observed at wider
bandwidths, at narrower bandwidths, where the penalties are
higher, the required OSNR values for the tunable bandwidth
filter and the sequence of interleaver ROADMs converge to the
same values. These results indicate that DPSK offers adequate
performance at 40 Gb/s modulation rates with net bandwidth
higher than GHz. The tolerance to passband narrowing
is improved with the more complex RZ-DQPSK modulation
Fig. 12. Similar 40 G DQPSK simulation results as shown in Fig. 11 but with
format due to the narrower signal bandwidth achieved with the 16 filters instead of 4.
encoding of 2 bits per symbol. Fig. 10 shows a direct compar-
ison of bandwidth-induced OSNR penalties at BER 1E-5 be-
tween NRZ-DPSK and RZ-DQPSK. The DPSK and DQPSK
signals at the wavelength of 1546.12 nm were passed through
a VOA and combined with ASE noise for OSNR adjustment.
The noise-loaded signal was then transmitted through a tunable-
bandwidth optical filter with a constant 3 dB-to-25 dB band-
width difference of 35 GHz. The bandwidth was tuned over a
range of 22–80 GHz to measure the impact of optical filtering on
the DPSK and DQPSK signals. For each modulation format, the
OSNR penalty is calculated with respect to the lowest required
OSNR with optimum filtering. At the minimum bandwidth set-
ting of 22.6 GHz the penalty generated on RZ-DQPSK is 0.7
Fig. 13. Illustration of wide bandwidth signal (solid line) filtered by a narrow
dB, while NRZ-DPSK has a penalty above 3 dB. filter stop-band (dotted-dashed), resulting in a spectrally-shaped crosstalk
The impact of transmittance filter shape on 40 Gb/s DPSK (dashed), and the resulting weighted crosstalk (dotted).
and DQPSK signal distortions for different bandwidth values
can be distinguished more clearly with numerical simulations
of the modulated signals passing through cascaded filter ele- at narrow bandwidths. For a low number of ROADM nodes, a
ments in the absence of any other signal degradation effects low-order Gaussian filter (1 or 2) may yield similar or slightly
between the transmitter and the receiver. The optical filter was lower EOP than a higher order Gaussian filter (4 or 6). How-
modeled with a Gaussian transfer function of different orders ever, at a higher number of ROADM nodes, filters with steeper
and bandwidths and placed in a loop to simulate bandpass nar- edges yield better performance. The same trends are observed
rowing through the same filter. The DPSK receiver had a DLI for NRZ-DPSK, but the penalties corresponding to the same
FSR of 66 GHz for enhanced optical filtering tolerance, whereas bandwidth, Gaussian filter order and number of filters is con-
the DQPSK receiver utilized a classical 1-symbol (2-bit) delay. sistently higher than for RZ-DQPSK.
The transmission performance impact for RZ-DQPSK quanti-
fied in terms of eye opening penalty (EOP) relative to the case VI. IN-BAND CROSSTALK
of no optical filtering is presented in Figs. 11 and 12 for 4 and Crosstalk can be caused either by imperfect isolation from
16 filter transitions, respectively. These figures demonstrate that an adjacent channel in the spectrum (out-of-band or hetero-
the transmittance function shape has a more significant impact dyne crosstalk) or by imperfect isolation from another signal
on performance at higher number of filters in sequence, and transmitted at the same wavelength (in-band or homodyne). The

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
TIBULEAC AND FILER: TRANSMISSION IMPAIRMENTS IN DWDM NETWORKS 565

Fig. 14. OSNR penalty as a function of unweighted (f) and weighted (g) crosstalk with crosstalk values calculated for different isolation filters: no filtering (a),
shaped isolation level (b), frequency offset (c), filter bandwidth (d), and filter shape (e).

out-of-band crosstalk has a minor effect since it is typically re- power spectral density of the input signal to yield the weighted
jected by the optical demultiplexing filter in front of the re- crosstalk function (Fig. 13).
ceiver. The in-band crosstalk, which can potentially be gener- The weighted crosstalk approach was applied to a set of
ated for example in the ROADM configuration of Fig. 1(a) by simulations employing isolation filters of varying properties
the drop channel interfering with the add channel in the WSS, to demonstrate its usefulness in predicting the impact of the
can have a significant impact on the signal BER [38]. Wave- filter isolation profile on transmission performance [13]. For
length-selective switches typically have an isolation around the the simulation examples, 2 1 WSS isolation filters were
passband center in excess of 35 dB. This isolation level ensures generated with the following variable parameters as depicted in
a low crosstalk for narrow-linewidth signals. However, when the Fig. 14: (a) uniform isolation level, (b) shaped isolation level,
bandwidth of the signal approaches the channel spacing (e.g., 43 (c) filter center frequency offset, (d) filter bandwidth, and (e)
Gb/s DPSK at 50 GHz channel spacing), the interfering signal filter shape.
may be shaped by the block port of the WSS. In this case, Fig. 14(f) and (g) illustrate the resulting OSNR penalties
the crosstalk can vary significantly across the signal bandwidth as a function of the integrated crosstalk without, and with the
reaching high peak crosstalk values. The traditional method for weighting applied, respectively. It is evident that unlike the
computing crosstalk as the ratio of the total power in the primary (unweighted) crosstalk, the weighted crosstalk is a reliable pa-
signal to the total power of the interfering signal is inadequate. rameter to use in predicting the performance impact of shaped
This is due to its failure to take into account the higher impact crosstalk signals. By accounting for the shape of the signal
on the signal BER of the interfering signal energy closest to the and appropriately weighting the crosstalk function, there is
center frequency of the primary signal compared to the energy very high correlation between OSNR penalty and the weighted
towards the edges. Addressing this challenge, several authors crosstalk value in all filter cases studied. Furthermore, simu-
proposed integrated and weighted crosstalk parameters to char- lations demonstrate that the same OSNR penalty is obtained
acterize the crosstalk regardless of its spectral content and cor- for a given crosstalk value obtained with interfering signals
relate it with the effect it has on system performance [12]–[13], of different data rates (10 and 40 Gb/s), modulation formats
[39]–[41]. (OOK, DPSK, and DQPSK) and detuning values from the
For an N 1 WSS, the weighted crosstalk for a given port primary signal (up to 40 GHz).
may be calculated based on the spectral characteristics of the Experiments were also performed in order to validate the nu-
primary signal and blocked signal(s), and on the port isolation merical simulation results, as well as to demonstrate the feasi-
profile of the block port(s). For the specific case of a 2 1 WSS, bility of using the weighted crosstalk approach on real-world
the crosstalk signal may be computed by simply multiplying devices. In this case, a 2 1 WSS prototype was used to mea-
the blocked signal spectrum by the isolation profile [13]. The sure the performance impact of the crosstalk caused by a 43
same general principle applies to the calculation of the weighted Gb/s NRZ-DPSK interfering (block) signal on an identical pri-
crosstalk when arbitrary modulation formats and data rates are mary (add) signal. Three channels were chosen on the WSS,
present on each of the blocking ports. In all cases, once the each of which exhibited different bandwidths, center frequency
crosstalk function is known, it can then be weighted by the offsets, and peak isolation levels. In each case, OSNR penalties

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
566 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

were measured as a function of increasing crosstalk power. Ex-


pressing the crosstalk in terms of the weighted crosstalk leads
to a common OSNR penalty curve, as predicted by numerical
simulations. Good agreement was seen between measured and
simulated performance, with matching within 0.2 dB for OSNR
penalties up to 2 dB [13].

VII. TRANSIENT POWER VARIATIONS


Channel power levels in DWDM networks are subject to tem-
poral fluctuations, which have an adverse impact on the signal
BER through reduction in OSNR, increase in nonlinear effects,
power drop below the receiver threshold, or increase above re-
Fig. 15. Increase in power oscillations after 5 and 10 ROADMs with input vari-
ceiver saturation limit. A variety of sources of power varia- 6
ations of 3 dB.
tions with different time scales may occur in a fiber optic net-
work [1]. These include slow changes caused by temperature
and ageing effects, PDL and amplifier gain instabilities, and fast ROADM is equivalent to a multiplexer/demultiplexer pair and
power changes with rise/fall times in the millisecond range or a VOA. There is no correlation between the ROADM nodes
faster, generated by network failures such as fiber cuts or equip- at the network level, thereby allowing the ROADMs to adjust
ment malfunction. These failure scenarios can induce large tran- the channel power simultaneously (and independently). In this
sient power excursions on surviving channels, which increase example, each ROADM attempts to reach a target channel
with propagation through a sequence of amplifiers. Following power of dBm while the input power varies within a range
the transient power changes, modifications in DWDM channel of dB. The delay between power measurement and power
distribution and power levels also lead to steady-state power setting is 8 ms, the rise time of the power setting VOA is 1
changes on surviving channels due to stimulated Raman scat- ms, and the maximum step in power adjustment is 1 dB. The
tering and amplifier gain ripple and tilt [27], [28], [42]. Such time evolution of the channel power after 10 ROADM nodes
power variations have been studied, and the transient suppres- indicates fluctuations up to 10 dB from the desired power level
sion techniques have been proposed prior to large scale deploy- (Fig. 15). Such power fluctuations cause delays in dynamic
ment of ROADMs. However, ROADMs do enable propagation network restoration, and can lead to traffic impacting power
through a larger number of network nodes, which means that fluctuations on other channels. This example does not include
transient power effects previously associated only with long- several other effects, which can impact the temporal evolution
haul networks can be encountered in metro networks covering of power levels throughout a large ROADM network such as
a smaller geographical area but including a large density of net- error in power setting, potential variations in ROADM attenu-
work nodes. ation, fiber propagation delay, Raman-induced power transfer,
The power adjustment capability typically offered by com- amplifier gain ripple, or gain fluctuations. This emphasizes the
mercial ROADM subsystems allows compensation or reduction need for intelligent control software at the network level to
in amplitude of slower power variations, depending on the re- provide fast power convergence while preventing large power
sponse time of the ROADM node’s power control algorithm and variations.
the network level power control algorithm. At the node level, the
response time results from the addition of the time delays associ- VIII. CONCLUSION
ated with channel power measurement, power setting, and soft- This paper reviews some of the more significant ROADM-in-
ware for communication and control of channel power. A power duced transmission impairments, which are common to multiple
adjustment on all DWDM channels can occur in the timeframe WSS technologies used in commercial ROADM networks. The
of s, which is adequate for compensation of slow effects, insertion loss of the ROADM is largely dependent on the node
but not fast enough for PDL compensation and EDFA transient architecture choice with multi-degree ROADMs typically incur-
control, for instance. ring a penalty compared to lower-loss 2-degree ROADMs. The
The ROADM itself does introduce a potential new source of loss of wavelength selective switches also increases with tighter
power variation to a ROADM-based DWDM network through frequency of the DWDM channels (50 GHz vs. 100 GHz). Po-
the process of addition and deletion of wavelengths. The auto- larization-dependent loss can be significant at high attenuation
matic setting of power levels through a sequence of ROADM settings, and may not be compensated by the power adjustment
nodes can lead to large power excursions towards the end of the ability of the WSS, inducing an OSNR penalty at the receiver.
ROADM chain, caused by the inherent delays between power Faster wavelength power monitoring and power setting can be
measurement and power setting on each node. This can lead to achieved on some WSS technologies, which can open new av-
down-stream ROADM nodes attempting to set the target output enues in the future for reduction of power variations in DWDM
power by increasing or decreasing the power of channels based networks [5].
on measurements taken at a previous time instant, unaware of Passband effects enhanced by propagation through multiple
the power change that has occurred in the meantime. nodes represent the most significant steady-state impairment in-
This increase in power setting error with ROADM node duced by ROADMs. The impact depends on the signal modu-
count can be shown with a simple model, whereby each lation type and data rate, and on the filter transmittance func-

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
TIBULEAC AND FILER: TRANSMISSION IMPAIRMENTS IN DWDM NETWORKS 567

tion of the WSS characterized by the bandwidth, the passband available wavelengths, and channel protection requirements
shape, and the frequency offset. Performance degradation gener- can be extended to include transmission impairments, including
ated by passband effects can occur in networks with a high spec- those induced by ROADMs. Such impairment-aware routing
tral density of DWDM channels and large number of ROADMs. demands simple and fast, yet reliable algorithms for calculation
This is a particular challenge for 100 Gb/s transponders cur- of end-to-end transmission penalties. The amount of signal
rently under development, which are required to operate on an degradation occurring at each node can be quantified through
existing ROADM network infrastructure designed for 10 Gb/s off-line analysis and assumed to be constant in time. Alter-
and 40 Gb/s modulated signals. The solution being actively pur- natively, optimum paths can be established based on actual
sued is the development of transponders with high tolerance to measurements of signal quality obtained from more complex
passband narrowing [35], [43]–[45]. Filtering effects can also optical performance monitoring devices than currently used at
become a major factor limiting the number of nodes for all-op- typical ROADM nodes [47]. The latter approach also includes
tical transmission in DWDM access networks, which employ the temporal variations in component characteristics which
lower-cost transponders with relaxed specifications. Improve- impact the system performance [48]. Impairment-aware routing
ments in design and manufacturing of WSS subsystems have in ROADM networks is an active research topic enabling
led to filter passbands with steeper edges (equivalent to 4th order dynamic traffic reconfiguration and promising to further extend
Gaussian filter shape [23]) even for 50 GHz frequency spacing. the limits of all optical networking.
Other avenues for mitigation of passband effects are offered
by WSS technologies with continuous transmittance function REFERENCES
and software-defined passband [5]. In this case, the WSS would [1] M. D. Feuer, D. C. Kilper, and S. L. Woodward, “ROADMs and their
not create a passband and induce a corresponding penalty on a system applications,” in Optical Fiber Telecommunications V. New
York: Academic, 2008, pp. 293–343.
DWDM signal passing through a ROADM node where adjacent [2] M. Mezhoudi, R. Feldman, R. Goudreault, B. Basch, and V. Poudyal,
channels are not dropped. “The value of multiple degree ROADMs on metropolitan network eco-
In-band crosstalk is generated in 50 GHz ROADMs with non- nomics,” in Proc. OFC/NFOEC, 2006, paper NThA4.
[3] E. B. Basch, R. Egorov, S. Gringeri, and S. Elby, “Architectural trade-
ideal isolation across the wide bandwidth of a signal such as offs for reconfigurable dense wavelength-division multiplexing sys-
DPSK and DQPSK for 40 Gb/s and higher data rates. Accu- tems,” IEEE J. Sel. Top. Quantum Electron., vol. 12, no. 4, pp. 1–12,
rate estimation of crosstalk-induced penalties is required, and Jul./Aug. 2006.
[4] P. Roorda and B. Collings, “Evolution to colorless and directionless
this can be accomplished in simulation and experiments using a ROADM architectures,” in Proc. OFC/NFOEC, 2008, paper NWE2.
weighted crosstalk definition that accounts for the spectral vari- [5] T. Strasser, “ROADM technologies and evolution,” in Proc. IEEE
ation in crosstalk across the signal bandwidth. LEOS, 2008, tutorial TuH1.
[6] A. Sahara et al., “Demonstration of colorless and directed/direction-
The power setting feature of ROADM nodes can reduce less ROADMs in router network,” in Proc. OFC/NFOEC, 2009, paper
some of the power variations occurring in a DWDM network, NMD2.
and their associated transmission penalties. At the network [7] T. Lehman, J. Sobieski, and B. Jabbari, “DRAGON: A framework for
service provisioning in heterogeneous grid networks,” IEEE Commun.
level, however, simultaneous power adjustments on multiple Mag., vol. 44, pp. 84–90, 2006.
ROADMs can lead to power variations high enough to impact [8] B. E. Smith, “AT&T transport services,” in Proc. OFC/NFOEC, 2009,
traffic on other channels. An optimum software algorithm is paper NMB3.
[9] D. R. Templeton, “Impact of FTTP on metro network architecture,” in
required to reach the best trade-off between power stability and Proc. OFC/NFOEC, 2009, paper NWC3.
speed of network-level power equalization. [10] S. Tibuleac et al., “Field trial of metro DWDM system with ROADMs
Design of ROADM-based DWDM networks requires estima- 2 2
based on 5 1 and 5 5 wavelength selective switches,” in Proc.
ECOC, 2006, pp. 401–402.
tion of the performance impact (e.g., OSNR or Q penalties) [11] M. Nagy and S. Tibuleac, “Wavelength selective switches for fiber op-
induced by each ROADM impairment, understanding of the tics telecommunications,” Photonics Spectra, p. 74, Nov. 2006.
scaling rules of these penalties with the number of ROADMs, [12] B. Collings, B. C. Collings, F. Heismann, and C. Reimer, “Dependence
of the transmission impairment on the WSS port isolation spectral pro-
and a methodology to combine these ROADM penalties with file in 50 GHz ROADM networks with 43 Gb/s NRZ-ADPSK signals,”
those generated by other propagation effects (fiber nonlinearity, in Proc. OFC/NFOEC, 2009, paper OThJ3.
PMD, etc.). These engineering rules can become complex since [13] M. Filer and S. Tibuleac, “Impact of ROADM in-band crosstalk on 40
G DPSK signals,” in OFC/NFOEC, 2010, paper NThF5.
they vary with ROADM technology and depend on transponder [14] B. Keyworth, “ROADM subsystems and technologies,” in Proc. OFC/
parameters, as described in previous sections. Such calculations NFOEC, 2005, paper OWB5.
of cumulative OSNR transmission penalties for each path in a [15] J. Homa and K. Bala, “ROADM architectures and their enabling tech-
nologies,” IEEE Commun. Mag., vol. 46, pp. 150–153, Jun. 2008.
network are currently performed by network planning software [16] J. Kelly, “Application of liquid crystal technology to telecommunica-
tools. tion devices,” in Proc. OFC/NFOEC, 2007, paper NThE1.
As WSS modules continue to evolve to higher port count and [17] G. Baxter et al., “Highly programmable wavelength selective switch
based on liquid crystal on silicon switching elements,” in Proc. OFC/
improved optical characteristics, they enable meshed networks NFOEC, 2006, paper OTuF2.
with increased transparency, and more complex topologies. [18] M. A. F. Roelens et al., “Dispersion trimming in a reconfigurable wave-
Planning and operation of extended meshed DWDM networks length selective switch,” J. Lightw. Technol., vol. 26, no. 1, pp. 73–78,
Jan. 2008.
requires new software for on-line computation of transmission [19] R. Narevich et al., “Integrated optical switch, variable attenuator
penalties and optical path validation during provisioning and and power monitor tap chip for 40-channel PLC ROADM,” in Proc.
recovery of optical connections. GMPLS control plane software Photon. Switching, 2007, paper WB3.2.
[20] T. Goh et al., “Port scalable PLC-based wavelength selective switch
[46] currently used for optimum route and wavelength selection with low extension loss for multi-degree ROADM/WXC,” in Proc.
based on physical layer constraints such as network topology, OFC/NFOEC, 2008, paper OWC6.

Authorized licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.
568 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

[21] L. Eldada et al., “40-channel ultra-low-power compact PLC-based [41] X. Liu, Y. H. Kao, M. Movassaghi, and R. C. Giles, “Tolerance to
ROADM subsystem,” in Proc. OFC/NFOEC, 2006, paper NThC4. in-band coherent crosstalk of differential phase-shift-keyed signal with
[22] M. Muha, “MEMS based channelized ROADM platform,” in Proc. balanced detection and FEC,” IEEE Photon. Technol. Lett., vol. 16, no.
OFC/NFOEC, 2008, paper JThA24. 4, pp. 1209–1211, Apr. 2004.
[23] F. Heismann and P. Mamyshev, “43-Gb/s NRZ-PDPSK WDM trans- [42] F. Smyth, D. C. Kilper, S. Chandrasekhar, and L. P. Barry, “Applied
mission with 50-GHz channel spacing in systems with cascaded wave- constant gain amplification in circulating loop experiments,” J. Lightw.
length-selective switches,” in Proc. OFC-NFOEC, 2009, paper OThC1. Technol., vol. 27, pp. 4686–4696, Nov. 2009.
[24] H. Maeda, Y. Tada, M. Sumida, S. Saito, and Y. Hayashi, “Perfor- [43] S. Chandrasekhar and X. Liu, “Experimental investigation of system
mance degradation of long-distance optical WDM networks due to impairments in polarization multiplexed 107-Gb/s RZ-DQPSK,” in
bandwidth limitations of wavelength multiplexers and demultiplexers,” Proc. OFC/NFOEC, 2008, paper OThU7.
IEEE Photon. Technol. Lett., vol. 11, no. 11, pp. 1509–1511, Nov. [44] J. Renaudier, G. Charlet, O. B. Pardo, and S. Bigo, “Long-haul trans-
1999. mission systems involving coherent detection for linear impairments
[25] N. N. Khrais, A. F. Elrefaie, R. E. Wagner, and S. Ahmed, “Perfor- mitigation,” in Proc. IEEE/LEOS Summer Topical Meetings, 2008,
mance degradation of multiwavelength optical networks due to laser paper WD1.1.
and (de)multiplexer misalignments,” IEEE Photon. Technol. Lett., vol. [45] T. Schmidt, C. Malouin, R. Saunders, J. Hong, and , “Mitigating
7, no. 11, pp. 1348–1350, Nov. 1995. channel impairments in high capacity serial 40 G and 100 G DWDM
[26] Transmission characteristics of optical components and subsystems, transmission systems,” in Proc. IEEE/LEOS Summer Topical Meet-
ITU-T Recommendation G.671, 2005. ings, 2008, paper ME2.2.
[27] D. C. Kilper, C. A. White, and S. Chandrasekhar, “Control of channel [46] A. Farrel and I. Bryskin, GMPLS Architecture and Applications. New
power instabilities in constant-gain amplified transparent networks York: Elsevier, 2006.
using scalable mesh scheduling,” J. Lightw. Technol., vol. 26, no. 1, [47] T. Anderson, J. C. Li, D. Hewitt, and O. Jerphagnon, “Optical per-
pp. 108–113, Jan. 2008. formance monitoring for intelligent networks,” in Proc. ECOC, 2009,
[28] M. Lee, N. Antoniades, and A. Boskovic, “PDL-induced channel paper 9.3.3.
power divergence in a metro WDM network,” IEEE Photon. Technol. [48] S. Azodolmolky et al., “A dynamic impairment-aware networking so-
Lett., vol. 14, no. 4, pp. 561–562, Apr. 2002. lution for transparent mesh optical networks,” IEEE Commun. Mag.,
[29] A. Mecozzi and M. Shtaif, “The statistics of polarization-dependent vol. 47, pp. 38–47, May 2009.
loss in optical communication systems,” IEEE Photon. Technol. Lett.,
vol. 14, no. 5, pp. 313–315, Mar. 2002.
[30] M. Filer et al., “System performance of tunable-filter ROADM,” in
Proc. IEEE LEOS, 2008, pp. 202–203.
[31] J. D. Downie and A. B. Ruffin, “Analysis of signal distortion and
crosstalk penalties induced by optical filters in optical networks,” J. Sorin Tibuleac graduated with a Diploma in applied physics from the Univer-
Lightw. Technol., vol. 21, no. 9, pp. 1876–1886, Sep. 2003. sity of Bucharest, Romania, in 1988, and received the M.Sc. and Ph.D. degrees
2
[32] B. Clouet et al., “Cascadability study of 16 1 9 wavelength selective in electrical engineering from The University of Texas at Arlington (UTA), in
2
switches with 5 42.6 Gb/s CS-RZ channels,” in Proc. ECOC, 2005, 1996 and 1999, respectively.
Between 1988 and 1993, he was a Research Scientist at the Lasers Depart-
paper We4.P.117.
[33] N. S. Bergano and C. R. Davidson, “Circulating loop transmission ment of the Institute of Atomic Physics, Bucharest, Romania studying laser-in-
experiments for the study of long-haul transmission systems using duced dynamic gratings, four-wave mixing and phase conjugation. During his
erbium-doped fiber amplifiers,” J. of Lightwave Tech., vol. 13, pp. graduate studies at UTA his research focused on properties and applications of
875–888, May 1995. guided-mode resonances in waveguide grating devices. In 1999 he joined Ge-
[34] S. Chandrasekhar et al., “Improving the filtering tolerance of 42.7-Gb/s nuity Inc. (formerly GTE Internetworking), where he evaluated 10 Gb/s long-
partial DPSK by optimized power imbalance,” in Proc. OFC/NFOEC, haul DWDM transmission systems for nation-wide network deployment. Since
2009, paper OMM1. 2001 he has held key positions in system design and product management at
[35] P. J. Winzer and R. J. Essiambre, “Advanced optical modulation for- Movaz Networks and ADVA Optical Networking in Norcross, Georgia. In his
mats,” in Optical Fiber Telecommunications V B. New York: Aca- current position of Director of Optical Systems, he is responsible for devel-
demic, 2008, pp. 23–93. opment of optical engineering rules for ADVA Optical Networking’s DWDM
[36] C. Fuerst, M. Camera, H. Wernz, and H. Griesser, “Experimental ex- system. His activity involves research in optical system impairments occurring
periences in high speed DQPSK transmission,” in Proc. OFC/NFOEC, in high data rate transmission systems, and evaluation of new optical technolo-
2009, paper OMT5. gies for system performance enhancement.
[37] B. Mikkelsen, C. Rasmussen, P. Mamyshev, and F. Liu, “Partial DPSK
with excellent filter tolerance and OSNR sensitivity,” Electron. Lett.,
vol. 42, no. 23, pp. 1363–1364, Nov. 2006.
[38] E. L. Goldstein, L. Eskildsen, and A. F. Elrefaie, “Performance im- Mark M. Filer (M’08) received the B.Sc. and M.Sc degrees in electrical en-
plications of component crosstalk in transparent lightwave networks,” gineering from Georgia Institute of Technology, Atlanta, in 2000 and 2005,
IEEE Photon. Technol. Lett., vol. 6, 5, no. 5, pp. 657–660, May 1994. respectively.
[39] H. Bissessur and C. Bastide, “Experimental assessment of frequency- Since August 2000, he has been employed as an Optical Systems Engineer
dependent crosstalk penalty with different 43 Gb/s modulation for- with ADVA Optical Networking (formerly Movaz Networks) in Norcross,
mats,” in Proc. ECOC, 2007, paper P083. Georgia. His work has included optical systems studies at 10, 40, and 100 Gb/s,
[40] T. Zami et al., “Comparative study of crosstalk created in 50 GHz- phase ripple effects, ROADM and optical filtering impairments (crosstalk,
spaced wavelength selective switch for various modulation formats at concatenation), FWM optimized channel allocation, dispersion management,
43 Gbit/s,” in Proc. ECOC, 2006, paper We3.P.81. recirculating loop studies, and modeling of system performance.

View publication Authorized


stats licensed use limited to: Sorin Tibuleac. Downloaded on February 10, 2010 at 14:02 from IEEE Xplore. Restrictions apply.

You might also like