Sunteți pe pagina 1din 3

Impact of Channel Spacing on the Design of a Mixed-Line-Rate

Optical Network
Avishek Nag and Massimo Tornatore
University of California, Davis, USA
Email: {anag, mtornatore}@ucdavis.edu
Abstract-Due to the increasing heterogeneity and the growing
volume of traffic, telecom backbone networks are going through
new innovations and paradigm shift. The wavelength-division
multiplexed (WDM) optical networks may cost-effectively sup-
port the growing heterogeneity of traffic demands by having
mixed line rates (MLR) over different wavelength channels.
The coexistence of wavelength channels with different line
rates in the same fiber brings up the important issue of the choice
of the channel spacing that one can have in these MLR networks.
The channel spacing affects the signal quality in terms of bit-
error rate (BER), and hence affects the maximum reach of the
Iightpaths, which is a function of line rates. Various approaches
to set an opportunistic width of the channel spacing can be
considered: i) on a practical side, one may choose uniform fixed
channel spacing specified by the ITU-T grid (typically 50 GHz);
ii) alternatively, to optimize the usage of the fiber spectrum, one
can explore different channel spacing for different line rates, also
referred to as the "one-size-does-not-fit-all" approach; iii) else
an optimal value of channel spacing that leads to minimum cost
can be sought.
In this work, we investigate the third case by evaluating the
cost of a MLR network for different channel spacings. Our results
show that, even under the assumption of uniform channel spacing
for a MLR network, it is possible to identify optimal values of
channel spacing for a minimum-cost MLR network design.
Index Terms-Optical Network, WDM, Network Design,
Mixed Line Rate, Maximum Transmission Range, Bit-Error Rate,
Channel Spacing.
I. INTRODUCTION
Today's backbone telecom networks employing optical
wavelength-division multiplexing (WDM) technology are on
the verge of a paradigm shift. Increasing heterogeneity in
the traffic, and the ever-increasing need for bandwidth, may
make it convenient to set up lightpaths on these networks with
multiple line rates. Such networks can be termed as mixed-
line-rate (MLR) optical networks: in [1], [2] it has been shown
that MLR networks are more cost-effective than a single-line-
rate (SLR) network and design methods for MLR networks
have been presented. Other related works on this topic [3], [4],
[5], focus on how to upgrade the network by putting higher line
rates, e.g., 40-Gbps and 100-Gbps line rates, over an existing
10-Gbps SLR network.
In MLR networks, different wavelength channels can have
different capacities. So, the WDM channel spacing is an
important issue. Most works consider the coexistence of
multiple line rates on the same fiber, assuming a fixed channel
spacing of 50 GHz, which is an ITU-T specification. In
MLR networks that support line rates up to 40 Gbps, this
channel spacing is compatible with comparatively simple mod-
ulation techniques such as on-off keying, optical duobinary,
etc. But, if we want to fit a 100-Gbps signal into a 50-
GHz grid, then the role of modulation formats becomes
very important, because their spectral efficiency needs to
be very high to support high bit rates over narrow channel
spacing. For instance, in [4], [5], the authors propose advanced
modulation formats, such as differential quadrature phase-
shift keying (DQPSK), polarization-multiplexed DQPSK, and
optical orthogonal frequency-division multiplexing (OOFDM),
to fit-in 40/100 Gbps over an existing 10-Gbps infrastructure.
Therefore, if a network operator wants to preserve a fixed
channel spacing of 50 GHz, then it can only resort to advanced
modulation formats to support multiple line rates.
The choice of channel spacing is significant in a MLR
scenario, especially if one needs to have a uniform modulation
format throughout the network'. On one hand, small channel
spacing may bandlimit the high-bit-rate signals and degrade
their BER and, hence, the maximum optical reach becomes
limited. On the other hand, if the channel spacing is high, the
low-bit-rate signals may have poor BER because more optical
noise (amplified spontaneous emission (ASE) noise from the
optical amplifiers) will be added (more discussion in Section
II). Thus, having a fixed channel spacing in a MLR network
leads to penalties, at least for a subset of the line rates. So,
the question is: what should be the optimum channel spacing
so that these penalties are minimized?
So, how the channel spacing affects the MLR network
design is an important problem. In earlier works [1], [2], we
have shown how the different set of available lightpaths over
different line rates significantly affects the total network cost
in terms of transponders (based on the fixed threshold BER,
different line rates will have different maximal reach). Now,
the set of available lightpaths will also be a function of the
channel spacing. Therefore, in this paper, we solve various
instances of the MLR network-design problem for different
values of fixed channel spacing and investigate if there is any
optimum value of channel spacing for a minimum-cost MLR
network design.
II. PROBLEM DESCRIPTION
In MLR network design, as discussed in [1], [2], the BER
of the lightpaths sets the feasibility of a particular line rate
for a given source-destination (s-d) pair. The BER model
consisted of physical impairments such as switch crosstalk,
optical amplifier noise (ASE noise), dispersion, laser center
1This assumption appears to be more practical because optical switches
with uniform filters are cheaper and easier to operate and design. Another
alternative can be to use different channel spacings on the same network. In
[6], the authors propose different channel spacings for different line rates.
V (i ,j)
V (i ,j) (3)
V (m, n), V >. (4)
if s = j
if d = j
otherwise
L L X ijk>. . CXijk>. S; 1
i, j EP",n k
L L rk . X ijk>. . CXijk>' L f&d
x k s,d
(5)
The solution to this ILP is a set of lightpaths between the
s-d pairs over different line rates and wavelengths, which carry
all the demands. These set of lightpaths in the global solution
is a subset of a pre-computed set of feasible lightpaths ; the
feasibility depends on whether the BER of the paths are below
a certain threshold. With variation in channel spacing, this pre-
computed set of lightpaths changes and we obtain a different
solution. In the next section, we present our results.
III. RESULTS AND DISCUSSION
We present our results in the form of total network cost
(total cost of transponders) for different values of channel
spacing and different traffic volumes. The example network
is the 14-node topology shown in Fig. 2; and the nonuniform
traffic matrix is given in Table I. It represents a total traffic
of 1 Tbps, which is multiplied by different factors (e.g., 1,
5, 10, 20) to represent a range of loads. The costs of 10-
Gbps, 40-Gbps, and 100-Gbps transponders are, respectively,
of the MLR network gains an added dimension if we consider
the channel spacing as a parameter.
The design problem formulation thus remains mostly similar
as in [I] . The problem turns out to be an integer linear program
(ILP), which minimizes the overall network cost in terms of
transponders. For completeness, the formulation is provided
below.
Input Parameters:
G(V,E): Physical topology of the network with V nodes and
E links; T = [Asd]: Traffic matrix with aggregate demands
Asd in Gbps between a s-d pair; R = {rl ' r2, . . . , rk} :
Set of available channel rates; Di: Cost of a transponder
with rate n: L
ij
: Length of the lightpath between a s-
d pair (in km) ; lmn: Physical link between nodes m-n;
W : Maximum number of wavelengths supported on a link,
>. E {I, 2, . , . ,W}; B: Threshold BER: a lightpath with
a higher BER will be rejected; BERijk>.: BER for the
lightpath between a s-d pair ij at rate rk and wavelength >.;
_{I if BERijk>. S; B V(' ')k >..
CXijk>' - 0 otherwise 2, J , , ,
P
mn:
Set of lightpaths passing through link lmn.
Variables:
X ijk>.: Number of lightpaths at rate rk and wavelength >.
between nodes ij; f tl Traffic from source s to destination
d routed on lightpath ij .
Problem Formulation:
Minimize : L L L X ijk>. . o, (2)
such that x ij k
------' 40'--------' SO-.....L
aO
- --1...
1
OO- --L 120- j-'--40-j-LS O-1.L..aO----'200
OplicalChannel Spacingin GHz
Fig. I. HER vs. optical channel spacing for 40-Gbps DQPSK system with
a lightpath length of 1000 km.
Now, this variability of BER with B; will be line-rate-
sensitive. Thus, with different values of B
o
, the set of admis-
sible lightpaths over different line rates will change and this
will result in different optimum cost for the network. Thus,
the capacity-vs.-reach tradeoff which affects the overall cost
frequency offset, and optical filter misalignment. The crosstalk
and ASE noise generate beat noise terms in the receiver. Both
the beat noises' power and the received signal power are
strong functions of the optical filter bandwidth B
o
, i.e., of the
channel spacing. Therefore, the BER, which is a function of
the signal-to-noise ratio (see Eqn. (1) below), is also a function
of the channel spacing. The formula for the BER and the noise
powers are reported below for reference :
1 IT lsI - IT
BER = -4 [erfc( M ) +er fc( M )] (1)
y2aO y2al
where IT = O"oI!I+o)I,o and I and I are the sig-
0"0+0"1' s l sO
nal current due to " I" and "0" reception, respectively; and
afi = a;h + a;hO + a;p_sp + a;_x + a ;_sp; ar = a;h +
2+2 +2 +2 +2 +2.2
a sh1 a sp- sp a x-x a s 1- x a s1-sp' a t h
is the thermal-noise variance; a shO is the shot-noise vari-
ance for "0" reception; a;hl is the shot-noise variance for
"I" tion: 2 2 2 2 d 2
recep ion; a sp- sp' a x- x' a
x- sp' a
sl-
x' an a
s1-sp are
the spontaneous-spontaneous, crosstalk-crosstalk, crosstalk-
spontaneous, signal-crosstalk, and signal-spontaneous beat-
noise variances, respectively.
The BER decreases with increasing B; at lower values
of B; and then attains a minimum at an optimum value
of B; (Fig. I). This is because , at lower values of B
o
,
the received optical power gets bandlimited and so does the
crosstalk power, but the BER becomes high as signal is not
strong enough to combat the total noise power including all
of its components (the summation of all the noise variances) .
With increasing B
o
, the signal loss decreases faster than
the reduction in crosstalk power, as the signal-crosstalk beat
noise is distributed over a bandwidth wider than the electrical
signal components. This results in the improvement of BER
with increasing B; for small values of bandwidth. As the
bandwidth increases beyond its optimum value, truncation in
signal spectrum becomes negligible but signal-crosstalk beat
continues to increase and so does the signal-spontaneous beat.
This explains the increase of the BER beyond the optimum
ti:
10.
3
r--- - - - - - - - - - - - - ----,
Fig. 2. I4-node NSF network (link lengths in km) .
TABLE I
TRAFFICMATRIX FORNSF NETWORK(EACHENTRYIS INGBPS).
Node 1 2 3 4 5 6 7 8 9 10 11 12 13 14
1 0 2 I I 1 4 I 1 2 I I I I I
2 2 0 2 I 8 2 I 5 3 5 I 5 I 4
3 1 2 0 2 3 2 11 20 5 2 I I I 2
4 1 I 2 0 1 1 2 1 2 2 I 2 I 2
5 1 8 3 I 0 3 3 7 3 3 I 5 2 5
6 4 2 2 I 3 0 2 1 2 2 I I I 2
7 I I II 2 3 2 0 9 4 20 I 8 I 4
8 I 5 20 I 7 1 9 0 27 7 2 3 2 4
9 2 3 5 2 3 2 4 27 0 75 2 9 3 I
10 1 5 2 2 3 2 20 7 75 0 I 1 2 1
11 1 I 1 1 1 1 1 2 2 1 0 2 1 61
12 1 5 I 2 5 1 8 3 9 I 2 0 I 81
13 1 I I I 2 1 I 2 3 2 I I 0 2
14 I 4 2 2 5 2 4 4 I I 6 1 81 2 0
l x ,2 .5x and 3.75x as in [7]. Thus, higher-rate transponders
provide volume discount. The modulation format is DQPSK,
the threshold BER is 10-
3
, and other physical layer parameters
are identical as in [8]. The number of available wavelengths
W on a fiber is chosen such that the total spectrum usage in
the C-band is 4 THz (e.g., for 50-GHz spacing, W =80; and
for 100-GHz spacing, W=40.).
Figure 3 shows results for non-uniform (using the traffic
matrix in Table I) and uniform (i.e., equal traffic for all the s-
d pairs) traffic demands. For nonuniform traffic, the total cost
first comes down as channel spacing increases. The cost is
minimum at a channel spacing of 80 GHz and then the cost
rises up again slightly. This is because, for smaller channel
spacing, the number of high-bit-rate paths is less due to
bandlimiting. So the volume discount is not fully exploited.
Then, as the channel spacing increases, some higher-bit-rate
paths become feasible and the cost goes down because the
volume discount can be exploited in a better way. At the same
time, due to the increase in channel spacing, the number of
available wavelengths decreases as well as some of the low-bit-
rate paths become infeasible (due to more noise as discussed
in Section II). Both these factors have an increasing effect on
the cost.
If low-bit-rate paths are fewer compared to the high-bit-
rate paths, then for some small demands there would have
been over-provisioning of capacity which tends to increase
the cost. Similarly, if the number of wavelengths are less, the
solution space becomes constrained and the cost tends to get
higher.
Thus, the combined effect of all these factors makes the total
cost a concave function of the channel spacing. This concavity
is more pronounced at higher volumes of traffic.
For uniform traffic, the cost shows the same behavior with
channel spacing. Here, the concavity is more pronounced than
in case of nonuniform traffic. This is because, in case of
nonuniform traffic, most of the bulky demands are along short
distances, in contrast to the uniform traffic case, where all s-d
pairs have equal volume of traffic. Hence, the change in the set
of feasible lightpaths with change in channel spacing affects
the uniform traffic case more.
IV. CONCLUSION
We have investigated the effect of channel spacing on the
design of a MLR network. We have shown that, as channel
spacing increases, the network cost (in terms of transponders)
comes down up to a certain optimum channel spacing, and
beyond the optimum channel spacing, the cost increases. The
behavior of network cost with channel spacing may vary
with network topologies. Larger topologies may exhibit more
sensitivity of the cost with channel spacing. We plan to extend
this work, with different practical topologies, in future.
55 250
REFERENCES
240
5Tbps
[I] A. Nag and M. Tornatore, "Transparent Optical Network Design with
50
_ 230
Mixed Line Rates," in Proc. IEEE Advanced Networks and Telecommu-
;; .
nication Systems (ANTS) , Mumbai , India, Dec. 2008.
0
0
o o 220 [2] A. Nag, M. Tornatore, and B. Mukherjee, "Optical Network Design
45
with Mixed Line Rates and Multipl e Modulation Format s," IEEEIOSA
210
J. Lightwave Technology (In Press) .

[3] C. Meusburger and D. A. Schupke, "Optimizing the Migration of Chan-

60 70 80 90
60 70 80 90 100 nels with Higher Bit-Rates," in Proc. OFCINFOEC Post-Deadline Papers,
Channel Spacing inGHz
Channel SpacinginGHz
San Diego, CA, Mar. 2009.
[4] T. Wuth, M. Chbat , and V. Kamalov, "Multi-rate (l00G/40G/lOG) Trans-
500
port Over Deployed Optical Networks," in Proc. National Fiber Opt ic
10Tbps
20 Tbps
Engineers Conference, OSA Technical Digest (CD) , San Diego, CA, 2008.
480
110
[5] M. Alfiad et aI., " I I I-Gb/s Transmission Over 1040-km Field-Deployed
Fiber With I OG/40G Neighbors," IEEE Photonics Technology Letters,
_ 460
vol. 21, no. 10, May 2009.
00
0
[6] A. Gumaste and N. Ghani , "Reach Opt imized Architecture for Multi-
o 440
rate Transport System (ROAMTS): One Size Does Not Fit All," in Proc.
420
OFCINFOEC, San Diego, CA, Mar. 2009.
[7] J. M. Simmons, Opt ical Network Design and Planning, Spr inger, 2008.

60 70 80 90 100

60 70 80 90 100
[8] B. Ramamurthy, D. Datta, H. Feng, J. Her itage, and B. Mukherjee,
Channel Spacing in GHz
Channel SpacinginGHz
"Impact of Transmission Impairments on Teletraffic Performance of
Wavelength-Routed Opti cal Networks," IEEEIOSA J. Lightwave Technol-
Fig. 3. Normal ized cost vs. channel spacing for four different traffic volumes.
ogy, vol. 17, no. 10, pp. 1713-1723, Oct. 1999.

S-ar putea să vă placă și