01 a review of the polarization nulling technique for monitoring optical-signal-to-noise ratio in dynamic wdm networks

4162 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 11, NOVEMBER 2006
A Review of the Polarization-Nulling Technique for
Monitoring Optical-Signal-to-Noise Ratio in
Dynamic WDM Networks
J. H. Lee, H. Y. Choi, S. K. Shin, and Y. C. Chung, Fellow, IEEE
Abstract—The polarization-nulling technique utilizes the dif-
ferent properties of optical signal and amplified spontaneous
emission (ASE) noise for accurate monitoring of the optical-
signal-to-noise ratio (OSNR) in dynamic optical networks. How-
ever, the performance of this technique is bound to be deteriorated
if the signal is depolarized by polarization-mode dispersion and/or
nonlinear birefringence or the ASE noise is partially polarized
due to polarization-dependent loss (PDL) in the transmission link.
The authors analyze these effects on the performance of the
polarization-nulling technique and introduce several techniques
to overcome these problems. These improved versions of the
polarization-nulling techniques could monitor the OSNR with
accuracy of better than ±1 dB, even when the differential group
delay is as large as 60 ps. These techniques could also negate the
effect of the signal depolarization caused by nonlinear birefrin-
gence in a highly nonlinear transmission link. The effect of the
partially polarized ASE noise due to PDL is found to be not severe
in most cases, as long as the PDL/span is smaller than 0.2 dB. To
verify the possibility of using the polarization-nulling technique
in real systems, the OSNR of the wavelength-division-multiplexed
(WDM) signals transmitted through a 120-km-long aerial fiber
link is measured for one week. No significant degradation in the
monitoring accuracy is observed during this long-term measure-
ment. In addition, the performance of the polarization-nulling
technique in an ultralong-haul transmission link is evaluated by
using a 640-km-long recirculating loop. The results show that this
technique could accurately measure the OSNR in the transmission
link longer than 3200 km. From these results, the authors conclude
that the polarization-nulling technique is well suited for monitor-
ing the OSNR in dynamic WDM networks.
Index Terms—Dynamic optical network, nonlinear birefrin-
gence, optical performance monitoring (OPM), optical-signal-
to-noise ratio (OSNR), polarization-dependent loss (PDL),
polarization fluctuation, polarization-mode dispersion (PMD),
polarization-nulling technique.
I. INTRODUCTION
FOR EFFICIENT operation and maintenance of a modern
dynamic wavelength-division-multiplexed (WDM) net-
work, it is necessary to monitor various optical parameters
such as the channel power, wavelength, and optical-signal-to-
Manuscript received May 2, 2006; revised June 29, 2006.
J. H. Lee, H. Y. Choi, and Y. C. Chung are with the Department of Electrical
Engineering, Korea Advanced Institute of Science and Technology, Yuseong-
gu, Daejeon 305-701, Korea (e-mail: ychung@ee.kaist.ac.kr).
S. K. Shin is with the Teralink Communications, Inc., Daejeon 305-335,
Korea.
Color versions of Figs. 8, 10, 12, and 13 are available online at
http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JLT.2006.883120
noise ratio (OSNR) [1]–[3]. In particular, OSNR is an important
parameter to be monitored for the estimation of the signal’s
quality in the optical layer. In addition, OSNR can be used
for link setup and optimization, root-cause analysis of the
system’s problems, fault detection and localization, early signal
degradation alarm, resilience mechanism activation, correlation
with the end-terminal bit error rate, service level agreement
(SLA) verification, etc.
Previously, OSNR has been measured by using the linear
interpolation technique, in which the amplified spontaneous
emission (ASE) noise was measured in between the WDM
channels and then interpolated into the signal’s wavelength
[4], [5]. However, in a dynamically reconfigurable network,
WDM signals are added/dropped or cross-connected directly
in the optical layer. Thus, each channel could traverse through
different routes and a different number of optical amplifiers.
In addition, the noise spectrum in these networks may not be
uniform due to the optical filtering occurring in various network
elements [6]. As a result, the accumulated noise level could be
quite different from channel to channel. Thus, the ASE noise
located within the signal’s bandwidth (and, consequently, the
true value of OSNR) cannot be measured by the conventional
linear interpolation technique [1], [6].
Recently, there have been many efforts to monitor the true
value of OSNR by utilizing the different polarization properties
of signal and ASE noise [7]–[18]. For example, it has been
demonstrated that the in-band ASE noise could be measured in
the presence of a signal by using the polarization-nulling tech-
nique [7]–[13]. In this technique, the received signal (together
with ASE noise) is split into two orthogonal polarization
components in which one component consists of signal and
polarized ASE noise, while the other has polarized ASE noise
only (assuming that the signal is highly polarized, and the ASE
noise is completely unpolarized). Thus, it is possible to measure
the signal and noise powers right at the signal’s wavelength
since the powers of the polarized ASE noises measured in
these polarization components should be the same (i.e., one-
half of total ASE noise power). However, the performance of
this polarization-nulling technique could be affected by various
polarization effects in the transmission link. For example, it
could be seriously deteriorated if the signal is depolarized by
polarization-mode dispersion (PMD) and nonlinear birefrin-
gence [7]–[9], [19]. The accuracy of this technique could also
be degraded significantly if the ASE noise is partially polarized
due to polarization-dependent loss (PDL) [20]. In addition, for
0733-8724/$20.00 © 2006 IEEE

LEE et al.: REVIEW OF THE POLARIZATION-NULLING TECHNIQUE FOR MONITORING OSNR IN WDM NETWORKS 4163
Fig. 1. Operating principle of the polarization-nulling technique.
use in the transmission link consisting of aerial fibers, this
technique should be able to track the fast fluctuation of the
state-of-polarization (SOP) of optical signal caused by wind
and electric currents in the neighboring power line [21]–[23].
Several techniques have been developed to overcome some
of these problems [10]–[13]. These techniques either calibrate
out the small amount of signal power leaked into the noise
in the orthogonal polarization state (due to PMD or nonlinear
birefringence) by using an additional optical filter or measure
the noise power at the slope of the signal’s spectrum to mitigate
the effect of PMD or nonlinear birefringence.
In this paper, we review the polarization-nulling tech-
nique for monitoring the OSNR in dynamic WDM networks.
In Section II, we describe the operating principle of the
polarization-nulling technique and its performances measured
in various types of fiber links. The results show that this
technique is susceptible to various polarization effects that oc-
curred in the transmission link. Thus, to identify the dominant
source of errors, we investigate the deleterious effects of PMD,
nonlinear birefringence, PDL, and fast fluctuation of SOP in
Section III. We then introduce the techniques developed to
mitigate these problems in Section IV. Finally, this paper is
summarized in Section V.
II. POLARIZATION-NULLING TECHNIQUE
A. Operating Principle
Fig. 1 shows the operating principle of the polarization-
nulling technique. This technique utilizes different polarization
properties of optical signals and ASE noises. The SOP of the
optical-signal incident on the OSNR monitor can be linear,
circular, or elliptical due to the random polarization rotation
along the transmission fiber. However, this arbitrarily polarized
signal can be changed to a linearly polarized signal simply by
using a polarization controller. The linearly polarized signal and
unpolarized ASE noise can then be split into two orthogonal
polarization components (in which one polarization component
consists of the signal and polarized ASE noise, while the other
has polarized ASE noise only) by using two linear polarizers.
Thus, the signal power, together with the polarized ASE noise,
can be measured with the first linear polarizer, which is aligned
Fig. 2. Experimental setup to evaluate the performance of the polarization-
nulling technique in various types of fiber links. (a) Back-to-back, (b) 640 km
of SMF (average PMD = 1.32 ps), (c) 640 km of NZDSF (average
PMD = 1.52 ps), and (d) 640 km of SMF+DCF (average PMD = 3.22 ps).
with the signal’s polarization. The polarized ASE noise (i.e.,
half of the total ASE noises) can be measured by using the
second linear polarizer, which is aligned to be orthogonal from
the signal’s polarization. Thus, the optical powers measured
after the first and second linear polarizers can be expressed as
Pp = Ps + 0.5Pn (1)
Po = 0.5Pn (2)
where Ps and Pn represent the optical powers of the signal
and ASE noise, respectively. Using these measured powers
(i.e., Pp and Po), OSNR can be obtained as
OSNR =
Ps
Pn

Bn
Br

=
Pp − Po
2Po

Bn
Br

(3)
where Bn is the noise equivalent bandwidth, and Br is the
resolution bandwidth. Bn is determined by the passband of
the demultiplexing filter. Thus, we can estimate the OSNR at
the resolution of Br simply by measuring Pp and Po. It should
be noted that both Pp and Po are measured right at the signal’s
wavelength. Thus, it would be possible to monitor the “true”
value of OSNR by using the polarization-nulling technique.
B. Measured Performances in Various WDM Links
To evaluate the performance of the polarization-nulling tech-
nique under the realistic networking environment, we measured
the OSNR in various types of 640-km-long fiber links. Fig. 2
shows the experimental setup. We multiplexed the outputs of
six DFB lasers operating in the range of 192.5–193.5 THz
and modulated with either 2.5- or 10-Gb/s nonreturn-to-zero
(NRZ) signal (pattern length: 231
− 1) by using a LiNbO3
modulator. The channel spacing was 200 GHz. The extinction
ratio of the modulated signal was about 13 dB. The multiplexed
WDM signals were first sent to a 13-km-long single-mode fiber
(SMF) for decorrelation and traversed through eight Erbium-
doped fiber amplifier (EDFA) modules followed by eight
80-km-long SMFs [Fig. 2(b) and (d)] or nonzero dispersion

Fig. 3. Measured OSNR in various types of fiber links. (a) 2.5 Gb/s.
(b) 10 Gb/s.
shifted fibers (NZDSFs) [Fig. 2(c)]. All the EDFA modules
in Fig. 2(d) consisted of a two-stage EDFA and a disper-
sion compensating fiber (DCF). The average PMD for these
640-km-long SMF, NZDSF, and SMF+DCF links were mea-
sured to be 1.32, 1.52, and 3.22 ps, respectively. The input chan-
nel power was set to 0 dBm for both SMF and NZDSF, while it
was reduced to −3 dBm for DCF. We used an additional EDFA
module as a preamplifier in front of the demultiplexer. After
640-km transmission, the signal was demultiplexed by using
an arrayed waveguide grating (AWG) and sent to an optical
attenuator to set the optical-power incident to the monitoring
module to about −20 dBm. The 3-dB channel bandwidth of the
AWG was 0.944 nm.
Fig. 3 shows the measured OSNR by using the polarization-
nulling technique. For each link in Fig. 2, we measured the
OSNR for more than 10 h. The OSNR was set to about 25 dB
in all cases. In the back-to-back experiment shown in Fig. 2(a),
it was necessary to use an additional noise source to set the
OSNR to 25 dB. In this case, the monitoring error was measured
to be less than 0.4 dB for every channel, regardless of the bit
rate. This monitoring error was still maintained to be less than
0.6 dB, even when we transmitted 2.5-Gb/s signals over the
640-km-long SMF link [in Fig. 2(b)], as shown in Fig. 3(a).
However, when we replaced the SMF link with either NZDSF
link in Fig. 2(c) or SMF+DCF link in Fig. 2(d), the maximum
errors were increased up to about 4 and 6 dB, respectively.
On the other hand, when we increased the bit rate to 10 Gb/s,
the maximum errors were somewhat decreased to 1.2 and
3.7 dB for the 640-km-long NZDSF link and SMF+DCF link,
respectively, as shown in Fig. 3(b). We attributed these errors
to PMD and nonlinear birefringence [9]. For example, if a
modulated optical signal is transmitted through the fiber link
with nonnegligible PMD, its spectral components could have
different polarization states (i.e., the signal could be depolar-
ized) [24], [25]. Thus, it would be impossible to make all these
spectral components linearly polarized at the same time, which,
in turn, causes an error in the measured OSNR. The optical
signal could also be depolarized by nonlinear birefringence
[19], [26], [27]. When multiple optical signals are transmit-
ted through optical fiber, the polarization state of one signal
could be modulated by the other intensity-modulated signals.
If this nonlinear polarization modulation is faster than the time
required for the polarization-nulling process, the polarization-
nulling technique could overestimate the ASE noise power due
to the depolarized signal component and become erroneous.
III. PERFORMANCE-LIMITING FACTORS
The polarization-nulling technique utilizes different polariza-
tion properties of the optical signal and ASE noise. As a result,
the performance of this technique is bound to be affected by
various polarization effects that occurred in the transmission
link. For example, as described in the previous section, this
technique becomes erroneous if the signal is depolarized due to
PMD or nonlinear birefringence. The accuracy of this technique
could also be deteriorated if the ASE noise becomes polarized
due to PDL. In addition, the failure in tracking the fast fluc-
tuation of the SOP of optical signal could cause a large error.
We investigated these potential problems to identify the major
limiting factors on the performance of the polarization-nulling
technique.
A. Polarization-Mode Dispersion (PMD)
When the optical signal (broadened by modulation) propa-
gates through the fiber link having nonnegligible PMD, each
spectral component could have different polarization state at
the end of the fiber due to PMD [24]. Thus, it is impossible
to nullify all the spectral components of the signal (by using
the polarizer aligned to the orthogonal state from the signal’s
polarization) at the same time, which, in turn, would result in
the monitoring errors. To include this effect of PMD, we can
rewrite (1) and (2) as
Pp = Ps(1 − εPMD) + 0.5Pn (4)
Po = PsεPMD + 0.5Pn (5)
where εPMD represents the fraction of the optical signal leaked
into the orthogonal polarization state from the signal’s polariza-
tion due to PMD. Assuming that there exists only the first-order
PMD, this term can be estimated as [25]
εPMD =
1
2
−
1
2

cos2 θ+sin2
θ

cos(τ∆ω)P(∆ω)d∆ω
2
(6)
where θ represents the polarization angle between the input
SOP of the optical signal and the PMD vector of the transmis-
sion link, τ is the magnitude of the PMD vector, ∆ω is the
optical angular-frequency offset from the center frequency, and
P (∆ω) is the power spectral density function of the modulated
optical signal defined as

P(∆ω)d∆ω = 1. Thus, by using
(3)–(5), the OSNR error caused by PMD can be described as
Error (dB) = 10 log

OSNRr
OSNRm

= 10 log

1 + 2εPMDOSNRr(Br/Bn)
1 − 2εPMD

(7)

Fig. 4. OSNR errors due to PMD measured in the 640-km-long SMF+DCF
link [9]. The average PMD of this link was 3.22 ps. In this paper, we used
only one channel to measure the OSNR errors without the effect of nonlinear
birefringence.
where OSNRr is the real OSNR, and OSNRm is the measured
OSNR. As expected, the OSNR error caused by PMD increases
with εPMD. Since εPMD is dependent on the magnitude of
the PMD vector and the spectral width of the optical signal,
the OSNR error increases with PMD and transmission speed.
This equation shows that the OSNR error is also dependent on
OSNRr. Thus, if OSNR is high, even a small εPMD could cause
a large monitoring error.
To evaluate the effect of PMD (without the effect of nonlin-
ear birefringence), we transmitted only one channel over the
SMF+DCF link in Fig. 2(d), which had the largest PMD due to
DCF (3.22 ps). The bit rate was changed from 1 to 10 Gb/s, and
the OSNR was set to 25 dB after the transmission. The moni-
toring error fluctuated significantly due to the random nature of
PMD. Thus, we measured the OSNR at each bit rate for more
than 10 h. Fig. 4 shows the average monitoring errors (solid
circles) in comparison with the theoretically calculated curve
(which was obtained by approximating (6) and (7) for small
PMD, and then averaging the results for τ and θ). In this figure,
the error bars represent the fluctuations of the measured OSNR
errors. This figure shows that the monitoring error caused by
PMD was increased with the bit rate. For example, when the
bit rate was 2.5 Gb/s, the maximum error was measured to be
only about 0.6 dB. However, the maximum error was increased
up to about 2 dB at 10 Gb/s. This was because the spectral
bandwidth of the optical signal was broadened as we increased
the bit rate. Thus, the effect of PMD on the performance of
the polarization-nulling technique could not be neglected if the
system is operating at high speed (≥ 10 Gb/s) over the fiber link
with large PMD.
B. Nonlinear Birefringence
When multiple optical signals are transmitted through optical
fiber, the polarization state of one signal (probe) can be affected
by the other signals (pumps) due to nonlinear birefringence
[26], [27]. Thus, the polarization state of the probe signal
could be modulated by the intensity-modulated pump signal.
However, this modulation frequency is typically faster than
the polarization-adjusting time of the polarization-nulling tech-
nique. Thus, it would be difficult to make this polarization-
modulated signal to be completely linearly polarized during
the OSNR measurement. This effect would result in the over-
estimation of the noise power and cause monitoring errors. We
analyzed the effects of nonlinear birefringence in an M-channel
WDM system. Because of the nonlinear birefringence, a small
portion of the signal power could be included in the noise power
measured by using the linear polarizer set in the polarization
state orthogonal to the signal. Neglecting the pulse distortion
caused by chromatic dispersion, this portion of the signal power
for the ith
channel in an M-channel WDM system could be
estimated as (for the worst-case analysis) (8), shown at the
bottom of the page, where ŝi and t̂ are the normalized Stokes
vectors of the ith
channel and polarizer, respectively, · denotes
the time average, ω is the angular frequency, J0 is the Bessel
function of the first kind of order 0, N is the number of spans,
ηij is the link enhancement factor (ηij = N for the dispersion-
compensated link, and ηij = | sin(NωdijL/2)/ sin(ωdijL/2)|
for the link without dispersion compensation) [28], L is the
span length, dij is the group velocity mismatch between chan-
nels i and j, Φij is the ac portion of the nonlinear phase shift of
the ith channel caused by the jth channel, and ϕij is the phase
retardation factor [9], [26]. The monitoring error caused by
nonlinear birefringence could be obtained by replacing εPMD
with εNL in (7) as
Error (dB) = 10 log

1 + 2εNLOSNRr(Br/Bn)
1 − 2εNL

. (9)
To evaluate the effect of nonlinear birefringence, we mea-
sured the monitoring errors in a simple two-channel experi-
ment. We multiplexed a continuous wave (CW) signal (probe)
and a modulated signal (pump) and transmitted the multiplexed
signals over 40-km-long SMF. The total PMD of this link
was only about 0.3 ps. The average optical powers of the
probe and pump signals were measured to 0 and 7 dBm at the
input of SMF, respectively. The channel spacing was 200 GHz,
and the OSNR was set to 25 dB. In this paper, the effect of
PMD could be neglected since we measured the OSNR of
εNL,i =

1
2

1 − ŝi · t̂

P(ω)dω
=

1
2


1 − J0



2
3





M

j=1,j=i
ηijΦij(ω) cos {ϕij(ω)}


2
+


M

j=1,j=i
ηijΦij(ω) sin {ϕij(ω)}


2


1
2





 P(ω)dω (8)

Fig. 5. (a) OSNR errors due to nonlinear birefringence measured in a two-
channel experiment. (b) OSNR errors measured in the various fiber links shown
in Fig. 2 [9]. The lines represent the theoretically calculated maximum errors
by using (8). The symbols are the measured data including the effect of PMD
(maximum values).
the CW signal. Fig. 5(a) shows the maximum errors caused
by the nonlinear birefringence in this system. The measured
data agree well with the calculated curve by using (8) and (9).
When the modulation frequency was smaller than 100 MHz, the
monitoring error was measured to be as large as 1 dB. However,
the error was substantially reduced as the modulation frequency
was increased. This was because the polarization modulation
caused by the nonlinear birefringence was averaged out more
quickly due to the group velocity mismatch between channels
caused by chromatic dispersion as the modulation frequency of
the pump signal was increased. We also evaluated the effect of
nonlinear birefringence using the various fiber links described
in Fig. 2. Fig. 5(b) shows the measured maximum OSNR errors
in Fig. 3 in comparison with the calculated curves by using (8)
and (9). In general, the measured data agreed well with the
calculated values (although the effect of PMD was included
in the measured data). Thus, we attributed the OSNR errors
observed in Fig. 3 mostly to the effect of nonlinear birefrin-
gence. The largest OSNR error was observed in the SMF+DCF
link. This was because the DCF had a small effective area,
and the effect of the nonlinear birefringence generated in each
span could linearly accumulate in the dispersion-compensated
fiber link. In addition, the OSNR error in the NZDSF link was
measured to be larger than that in the SMF link due to its
smaller dispersion and effective area. Fig. 5(b) also shows that
when we transmitted the 10-Gb/s signals over the SMF+DCF
link (which had the largest PMD of 3.22 ps), the measured error
was much larger than the calculated value due to the effect of
PMD. Thus, we concluded that the dominant error source of
the polarization-nulling technique was nonlinear birefringence,
except when PMD was quite large.
C. Polarization-Dependent Loss (PDL)
The polarization-nulling technique estimates the power of
the ASE noise within the signal’s bandwidth by measuring
only the noise power in the polarization state orthogonal to the
signal since the ASE noise is assumed to be fully unpolarized.
However, if the ASE noise is partially polarized due to PDL, the
noise power in the orthogonal polarization state may no longer
be identical to the noise power in the state parallel to the signal.
Fig. 6. Experimental setup to measure the effect of PDL on the polarization-
nulling technique (sw: Optical switch, Att: Optical attenuator, PC: Polarization
controller, and BPF: Bandpass filter).
Thus, it could cause an error in the measured OSNR by using
the polarization-nulling technique. In the case where OSNR is
much higher than 10 dB, this monitoring error (caused by the
partially polarized ASE noise) can be estimated by
Error (dB) = 10 log [1 − DOPASE(ŝ · n̂)] (10)
where DOPASE is the degree of polarization (DOP) of the
ASE noise, ŝ and n̂ represent the normalized Stokes vectors of
the signal and the partially polarized ASE noise, respectively,
and ŝ · n̂ represents their inner product [20], [29]. It has been
reported that DOPASE should have Maxwellian distribution,
since the polarized portion of the ASE noise arisen from each
amplifier due to PDL should be added randomly [30]. Thus,
the average DOP of the ASE noise (after the transmission of N
amplifier spans) can be described as
DOPN 2
=

1 +
2
3
Γ2

1 −
1
N
2
DOPN−12
−
16
9π

1 −
1
N

Γ2
DOPN−1 +
8
3π
Γ2
(11)
for N ≥ 2 and DOP1 = Γ, where the parameter Γ represents
the magnitude of the PDL vector of each span defined as
(1 − 10−PDL_dB/10
)/(1 + 10−PDL_dB/10
), and PDL_dB is
the PDL in decibels [30]. Using this result, the probability that
the error in the measured OSNR using the polarization-nulling
technique becomes larger than x dB can be obtained as
Probability {|error| ≥ x(dB)}
≈
1
2

erfc

2(10x/10
− 1)
√
πDOPN

+ erfc

2(1 − 10−x/10
)
√
πDOPN

(12)
for DOPN

1 and x 3 [29].
Fig. 6 shows the experimental setup used to investigate the
effect of PDL. We transmitted 20 WDM channels (channel
spacing: 50 GHz) through the link consisted of 15 EDFA spans.
The first 10 channels (channel number: 1–10) operated in the
range of 1547.3–1550.9 nm, while the other channels (channel
number: 11–20) operated in the range of 1556.2–1559.8 nm.
An additional laser (reference channel) operating at 1553.8 nm
was used with an optical switch to measure the Stokes para-
meters of the signal and the ASE noise at this wavelength.
Each span consisted of an EDFA, an optical attenuator, a
polarization controller, and a PDL element. The PDL of each

Fig. 7. (a) Cumulative probability of the errors in the measured OSNR’s by
using the polarization-nulling technique (due to the partially polarized ASE
noise) in a transmission link consisted of 15 spans (average PDL/span =
0.57 dB) [29]. The solid curve represents the theoretically calculated values
by using (12). (b) Probability that the error in the measured OSNR by using the
polarization-nulling technique becomes larger than 1 dB (due to the partially
polarized ASE noise caused by PDL) [29].
span was set to about 0.57 dB by using PDL elements. We
first measured the Stokes parameters of the reference signal by
using a polarization analyzer. We then turned off the optical
switch to measure the Stokes parameters of the ASE noise at
the same wavelength. Using these measured Stokes parameters,
we evaluated the accuracy of the polarization-nulling technique
by comparing the total noise power with the noise power in the
polarization state orthogonal to the signal.
Fig. 7(a) shows the cumulative probability of the measured
OSNR errors in comparison with the theoretical curve obtained
by using (12). The measured data agreed well with the the-
oretically calculated curve. For the transmission link used in
this paper (where the average PDL/span was 0.57 dB), the
probability that the errors in the measured OSNR (caused by
the partially polarized ASE noise due to PDL) became larger
than 1 dB was about 10−2
. Fig. 7(b) shows this probability (i.e.,
the probability that the error in the measured OSNR becomes
larger than 1 dB) as a function of PDL/span and the number
of spans. The result shows that if the PDL/span is smaller than
0.2 dB (which is a typical value for current systems [31]), the
effect of PDL on the measured OSNR using the polarization-
nulling technique is very small even for an ultralong-distance
transmission system. For example, the probability that the error
in the measured OSNR becomes larger than 1 dB is less than
10−4
in a 50-span transmission system, as long as the PDL/span
is smaller than 0.2 dB. However, the PDL/span could be raised
above 0.2 dB in the network utilizing reconfigurable optical
add/drop multiplexers (ROADMs). In such cases, the increased
PDL/span would certainly increase the error probability. For
example, if the PDL/span is increased to 0.3 dB, this probability
could be increased to 2 × 10−3
. Thus, in this case, the number
of amplifier spans should be reduced to 23 to maintain the error
probability within 10−4
.
D. Fast Polarization Fluctuation
When the optical signal is transmitted through an installed
fiber (especially aerial fiber), its SOP could fluctuate rapidly in
response to the environmental conditions [21]–[23]. To evaluate
the effect of this polarization fluctuation, we first measured the
Fig. 8. (a) Fourier components of the Stokes parameters of the optical signal
measured in a 120-km aerial fiber link [21], [22]. (b) Measured OSNR by using
the polarization-nulling technique in a 120-km aerial fiber link [21], [22].
SOP fluctuation of an optical signal in a 120-km-long aerial
fiber link installed in the field. Fig. 8(a) shows that the SOP
fluctuation measured in this aerial fiber link had two strong
frequency components at 60 and ∼0.3 Hz. We attributed the
60-Hz peak to the Faraday rotation caused by the current in
the electrical power transmission line, since it was identical
to the current frequency used in the electrical power line. The
other dominant peak at ∼0.3 Hz was caused by the wind and
the pendulum motion of the optical ground wire [21]–[23].
Fig. 8(b) shows the measured OSNR in this aerial link us-
ing the polarization-nulling technique. We intentionally
changed the OSNR from 19 to 28 dB to evaluate the perfor-
mance of the polarization-nulling technique at various OSNR
values. The optical-power incident to the OSNR monitor was
maintained to −22 dBm by using a variable optical attenuator.
The OSNR was measured for 10 min at each OSNR value (19,
23, 26, and 28 dB). The result shows that the polarization-
nulling technique could monitor the OSNR with accuracy better
than ±0.3 dB in most cases. This was because our OSNR
monitoring setup could track the polarization fluctuation up to
∼400 Hz. The accuracy of the measured OSNR was slightly
degraded to ±1 dB when the OSNR was extremely high
(28 dB). However, no significant change was observed in the
measurement accuracy during our long-term experiment, which
lasted one week.
IV. SOLUTIONS FOR IMPROVED PERFORMANCE
The performance of the polarization-nulling technique could
be deteriorated if the signal is depolarized (due to PMD or
nonlinear birefringence), or the ASE noise is polarized (due
to PDL). However, it has been shown in the previous sec-
tion that the effect of PDL could be neglected unless the
PDL/span was very large, and the transmission distance was
transoceanic. Thus, for practical use, it is necessary to improve
the polarization-nulling technique so that it could endure the
effects of PMD and nonlinear birefringence. Several techniques
have been proposed for this purpose [10]–[13]. These tech-
niques either calibrated out the small amount of signal power
leaked into the noise in the orthogonal polarization state (due
to PMD or nonlinear birefringence) by using an additional
optical filter or measured the noise power at the side of the
signal’s spectrum to mitigate the effects of PMD or nonlinear

Fig. 9. Schematic diagram of the polarization-nulling technique improved
by using an additional optical filter (PBS: Polarization beam splitter, BPF:
Bandpass filter, and PD: Photodetector).
birefringence. In this section, we describe these improved ver-
sions of the polarization-nulling techniques in details.
Fig. 9 shows the schematic diagram of the improved
polarization-nulling technique implemented by using an addi-
tional optical filter [10]. The optical signal, which is demulti-
plexed by using the first optical bandpass filter (BPF), is sent to
a polarization beam splitter (PBS) via a polarization controller.
This controller is used to maximize the signal power P1 in one
arm. Thus, the signal and ASE noise are split into two polar-
ization components after the PBS. If the effects of PMD and
nonlinear birefringence could be neglected, one polarization
component should have the polarized ASE noise only, while the
other has the signal and polarized ASE noise. However, if the
signal is depolarized by PMD and/or nonlinear birefringence,
a small amount of signal power could be leaked into the arm
where we intend to measure only the noise power. To calibrate
this out, we split the polarized ASE noise component into two
parts (P2 and P3) by using a 3-dB coupler and then filter one
part by using the second optical BPF. This filter is used to
reduce the bandwidth of the ASE noise. Neglecting the loss of
each path (which can be calibrated, if necessary), the optical
powers detected at photodiodes can be described as
P1 = Ps (1 − ε) + 0.5Pn (13)
P2 = 0.5Psε + 0.25Pn (14)
P3 = 0.5Psε + 0.25αPn (15)
where ε is the ratio of the signal power excluded from P1 (thus,
included in P2 and P3) due to PMD, nonlinear birefringence,
and/or incomplete polarization control, and α is the bandwidth
reduction factor determined by the transmission characteristics
of the first and second BPFs. Using these equations, the OSNR
can be obtained as
OSNR =
Ps
Pn

Bn
Br

=

(P1 + 2P2)
4(P2 − P3)/(1 − α)
− 1

Bn
Br

.
(16)
To evaluate the performance of this improved technique, we
measured the OSNR in the 640-km-long fiber links described
in Fig. 2. The passbands of the first and the second BPFs
were 0.944 and 0.668 nm, respectively. The parameter α was
measured to 0.577. Fig. 10 shows the OSNR measured by this
technique for 10 h. The results show that this improved version
of the polarization-nulling technique could measure the OSNR
Fig. 10. Measured OSNR by using the improved technique illustrated in
Fig. 9 [10]. The OSNR was measured for 10 h in the various fiber links shown
in Fig. 2.
Fig. 11. Schematic diagram of the polarization-nulling technique improved
by using a narrow tunable filter and a PMDC.
with accuracy better than ±1 dB, regardless of the bit rates
and/or types of the fiber links.
Although the technique illustrated in Fig. 9 could effectively
negate the deleterious effects of PMD and nonlinear birefrin-
gence, it requires the precise alignment of two tunable filters.
This may not be an easy task since these filters should be
tuned together rapidly for every WDM channels. To avoid
this complexity, another version of the improved polarization-
nulling technique has been proposed by using only one tunable
filter [11]–[13]. This technique measures the noise power both
at the center and slope of the signal’s spectrum to mitigate the
effect of nonlinear birefringence. Fig. 11 shows the schematic
diagram of the proposed technique. The optical signal is first
sent to a PMD compensator (PMDC) to eliminate the effect of
PMD on the OSNR measurement. After passing through the
PMDC, the signal is sent to a tunable BPF (having bandwidth
much narrower than the signal). This filter is first adjusted to
the center of a specific WDM channel. Then, the power of the
filtered signal (Pλ1) is measured before the linear polarizer. At
the same time, the ASE noise power polarized orthogonal to the
signal (Pλ1,null) is measured by using the polarization-nulling
technique. The tunable BPF is then adjusted again to measure
the powers of signal (Pλ2) and polarized ASE noise (Pλ2,null)
on the slope of the signal’s spectrum (i.e., at the frequency a few
gighertz apart from the center frequency). If the optical signal
is slightly depolarized after the transmission due to nonlinear
birefringence, there can be a small portion of optical signal

Fig. 12. Effects of PMD on the performance of the improved technique
illustrated in Fig. 11 [22]. In this experiment, we changed the DGD value from
0 to 60 ps by using a PMD emulator. OSNROSA represents the OSNR values
measured by using an optical spectrum analyzer.
in addition to the polarized ASE noise in Pλ1,null or Pλ2,null.
Thus, Pλ1,null and Pλ2,null can be expressed as
Pλ1,null = (Pλ1 − Pn)εNL +
1
2
Pn (17)
Pλ2,null = (Pλ2 − Pn)εNL +
1
2
Pn. (18)
Since Pλ2 is smaller than Pλ1, the portion of the signal power
transferred into Pλ2,null (due to nonlinear birefringence) is also
smaller than the portion transferred into Pλ1,null. As a result,
in case the signal is depolarized due to nonlinear birefringence
(i.e., εNL = 0), the OSNR derived from Pλ2,null becomes more
accurate than the OSNR obtained by using Pλ1,null [11], [12].
However, if εNL were large, this method could still suffer from
large errors since the portion of Pλ2 transferred into Pλ2,null
cannot be neglected. This problem can be solved by eliminating
εNL in (17) and (18) [13]. Thus, we can now estimate the
power of ASE noise and OSNR accurately without the effect of
εNL as
Pn =
2(Pλ1Pλ2,null − Pλ1,nullPλ2)
Pλ1 − Pλ2 − 2Pλ1,null + 2Pλ2,null
(19)
OSNR =
Pt − PnBt/Bf
PnBr/Bf
(20)
where Pt is the total power of the optical signal and ASE noise
within the bandwidth Bt (measured by scanning the tunable
filter over the signal’s whole spectrum), and Bf is the bandwidth
of the tunable BPF.
We implemented an OSNR monitor based on this technique
and evaluated the effect of PMD and nonlinear birefringence.
The PMDC was implemented by using a piece of high-
birefringent fiber and a polarization controller. The bandwidth
of the tunable BPF was about 3 GHz. Because of this extremely
narrow bandwidth, this OSNR monitor could be relatively
insensitive to the effect of PMD, even without the PMDC.
However, we used the PMDC to extend the PMD limit to
beyond several tens of picoseconds. We first investigated the
effect of PMD for a 10-Gb/s NRZ signal by using a first-
order PMD emulator. Fig. 12 shows that the OSNR could
Fig. 13. Effect of nonlinear birefringence on the improved technique illus-
trated in Fig. 11. (a) Channel powers incident on the SMF and DCF were set
to −5 dBm. (b) Channel powers incident on the SMF and DCF were increased
to 0 dBm.
be measured with accuracy of better than ±1 dB, even when
the differential group delay (DGD) was as large as 60 ps
(at OSNR = 27 dB). It has been reported that 30-ps DGD
could incur a power penalty of ∼1.4 dB for the 10-Gb/s NRZ
signal (in an optically preamplified system) [32]. Thus, this
technique could measure the OSNR accurately as long as the
PMD-induced power penalty is not extremely large. The effect
of the higher order PMD is not a concern here since the
bandwidth of the BPF is merely 3 GHz. To evaluate the effect
of nonlinear birefringence, we also measured the OSNR for
10 h in the 640-km-long SMF+DCF link described in Fig. 2(d).
In this experiment, we transmitted eight channels operating
at 10 Gb/s and reduced the channel spacing to 100 GHz to
induce large nonlinear birefringence. The channel power was
set to either −5 or 0 dBm at the input of SMF and DCF.
The OSNR was set to 20 dB after transmission. Fig. 13 shows
the performance of the proposed technique in comparison with
that of the conventional polarization-nulling technique. For fair
comparison, a PMDC was also used in front of the conventional
polarization-nulling technique so that we could investigate the
effect of nonlinear birefringence without the influence of PMD
in both cases. Fig. 13(a) shows that the performance of the
conventional technique was seriously affected by nonlinear
birefringence, even when the channel power was set to as low as
−5 dBm at the input of SMF and DCF. The OSNR error caused
by nonlinear birefringence in the conventional technique was
1.7 ± 1.3 dB. However, this version of the improved technique
could measure the OSNR with accuracy of better than ±0.3 dB
under the same condition. There was no significant difference
in the accuracy of the measured OSNR when we changed
the transmission speed to 2.5 Gb/s. Fig. 13(b) shows that this
technique could also measure the OSNR accurately, even when
the channel power was increased up to 0 dBm, while the
OSNR error in the conventional technique was as high as 8 dB.
We also evaluated the performance of this technique in an
ultralong-distance transmission link by using a recirculating-
loop experiment (made of a 640-km-long SMF+DCF link). We
launched eight 10-Gb/s channels in this recirculating loop and
measured OSNR as a function of the transmission distance. The
result in Fig. 14 shows that this technique could measure the
OSNR accurately, even with a transmission distance as long
as 3200 km.

Fig. 14. Measured OSNR in the recirculating loop made of 640-km-long
SMF+DCF link.
V. SUMMARY
For proper operation and maintenance of a dynamic WDM
network, it would be necessary to monitor the OSNR of
each channel. However, the conventional linear interpolation
technique cannot measure the true value of OSNR in such
networks. To overcome this problem, it has been proposed to
utilize the different polarization properties of optical signal and
ASE noise. For example, it has been demonstrated that the
polarization-nulling technique could measure the true value of
OSNR in a dynamic networking environment. In this paper,
we reviewed this polarization-nulling technique and discussed
its performance-limiting factors and possible solutions to over-
come these limitations.
The operating principle of the polarization-nulling technique
is based on the assumptions that the signal is highly polarized
and that the ASE noise is completely unpolarized. However,
in practice, these assumptions are easily violated by various
polarization effects in the transmission link. For example, the
signal could be depolarized by PMD and nonlinear birefrin-
gence, and the ASE noise could be partially polarized due to
PDL. In addition, for use in aerial fiber links, this technique
should be able to track the fast fluctuation of polarization
caused by winds and electric currents in the neighboring power
lines. Several techniques have been developed to overcome
some of these problems. These improved techniques either
calibrated out the small amount of signal power leaked into
the noise in the orthogonal polarization state (due to PMD
or nonlinear birefringence) by using an additional optical fil-
ter or measured the noise power at the slope of the signal’s
spectrum.
To evaluate the possibility of using these improved
polarization-nulling techniques in the real systems, we thor-
oughly investigated the effects of PMD, nonlinear birefrin-
gence, PDL, and fast polarization fluctuation. The results
showed that these techniques could monitor the OSNR with
accuracy of better than ±1 dB, even when the first-order PMD
was as large as 60 ps (at OSNR ≤ 27 dB). The effect of
the signal depolarization caused by nonlinear birefringence
was measured to be negligible, even in a highly nonlinear
transmission link. We also investigated the effect of the partially
polarized ASE noise caused by PDL. The results showed that
as long as the PDL/span was smaller than 0.2 dB (as in most
current systems [31]), the OSNR could be monitored accurately
by using the polarization-nulling technique, even in a long-
distance system. For example, when the PDL/span was 0.2 dB,
the probability that the error in the measured OSNR became
larger than 1 dB was merely 10−4
in the transmission link
made of 50 amplifier spans. To verify the practicality of the
polarization-nulling technique, we measured the OSNR of the
optical signals transmitted through a 120-km-long aerial fiber
link for one week. In this aerial fiber link, the SOP of the
optical signal was measured to be fluctuated at ∼0.3 and 60 Hz
due to winds and electric currents in the neighboring power
lines, respectively. Despite these slow and fast fluctuations,
the polarization-nulling technique could monitor OSNR with
accuracy better than ±1 dB (at OSNR = 19 ∼ 28 dB). No
significant degradation in the monitoring accuracy was ob-
served during this long-term measurement. We also evaluated
the performance of the polarization-nulling technique in an
ultralong-distance transmission link by using a 640-km-long
recirculating loop. The results showed that this technique could
measure the OSNR accurately, even in the transmission link
longer than 3200 km.
REFERENCES
[1] Y. C. Chung, “Optical monitoring techniques for WDM networks,” pre-
sented at the LEOS Summer Topical Meetings, Aventura, FL, 2000,
Paper FC2.2.
[2] D. C. Kilper, R. Bach, D. J. Blumenthal, D. Einstein, T. Landolsi,
L. Ostar, M. Preiss, and A. E. Willner, “Optical performance monitoring,”
J. Lightw. Technol., vol. 22, no. 1, pp. 294–304, Jan. 2004.
[3] Optical Monitoring for DWDM Systems, Jun. 2004. ITU-T Recommen-
dation G.697.
[4] K. Asahi, M. Yamashita, T. Hosoi, K. Nakaya, and C. Konoshi, “Optical
performance monitor built into EDFA repeaters for WDM networks,”
presented at the Optical Fiber Communication Conf., San Jose, CA, 1998,
Paper THO2.
[5] H. Suzuki and N. Takachio, “Optical signal quality monitor built into
WDM linear repeaters using semiconductor arrayed waveguide grating
filter monolithically integrated with eight photodiodes,” Electron. Lett.,
vol. 35, no. 10, pp. 836–837, May 1999.
[6] D. C. Kilper, S. Chandrasekhar, L. Buhl, A. Agarwal, and D. Maywar,
“Spectral monitoring of OSNR in high speed networks,” presented at the
Eur. Conf. Opt. Commun., Copenhagen, Denmark, 2002, Paper 7.4.4.
[7] M. Rasztovits-Wiech, M. Danner, and W. R. Leeb, “Optical signal-
to-noise ratio measurement in WDM networks using polarization
extinction,” in Proc. Eur. Conf. Opt. Commun., Madrid, Spain, 1998,
pp. 549–550.
[8] D. K. Jung, C. H. Kim, and Y. C. Chung, “OSNR monitoring technique
using polarization-nulling method,” presented at the Optical Fiber Com-
munication Conf., Baltimore, MD, 2000, Paper WK4.
[9] J. H. Lee, D. K. Jung, C. H. Kim, and Y. C. Chung, “OSNR monitor-
ing technique using polarization-nulling method,” IEEE Photon. Technol.
Lett., vol. 13, no. 1, pp. 88–90, Jan. 2001.
[10] J. H. Lee and Y. C. Chung, “Improved OSNR monitoring technique
based on polarization-nulling method,” Electron. Lett., vol. 37, no. 15,
pp. 972–973, Jul. 2001.
[11] Y. C. Chung, S. K. Shin, and C. H. Kim, “OSNR monitoring method
and apparatus using tunable optical bandpass filter and polarization-
nulling method,” U.S. Patent 20040114923, June 17, 2004.
[12] M.-H. Cheung, L.-K. Chen, and C.-K. Chan, “A PMD-insensitive OSNR
monitoring scheme based on polarization nulling with off-center narrow-
band filtering,” presented at the Optical Fiber Communication Conf., Los
Angeles, CA, 2004, Paper FF2.
[13] H. Y. Choi, J. H. Lee, S. B. Jun, Y. C. Chung, S. K. Shin, and S. K. Ji,
“Improved polarization-nulling technique for monitoring OSNR in WDM
network,” presented at the Optical Fiber Communication Conf., Anaheim,
CA, 2006, Paper OThP2.

01 a review of the polarization nulling technique for monitoring optical-signal-to-noise ratio in dynamic wdm networks

Recommended

Recommended

More Related Content

What's hot

What's hot (19)

Similar to 01 a review of the polarization nulling technique for monitoring optical-signal-to-noise ratio in dynamic wdm networks

Similar to 01 a review of the polarization nulling technique for monitoring optical-signal-to-noise ratio in dynamic wdm networks (20)

Recently uploaded

Recently uploaded (20)

01 a review of the polarization nulling technique for monitoring optical-signal-to-noise ratio in dynamic wdm networks