Optics Communications 355 (2015) 44–49

Contents lists available at ScienceDirect

Optics Communications
journal homepage: www.elsevier.com/locate/optcom

Adaptive optimization for pilot-tone aided phase noise compensation

Sheng Cui a,b,c, Mengran Xu a,b,c, Wenjuan Xia a,b,c, Chanjian Ke a,b,c,n, Zijie Xia a,b,c,
Deming Liu a,b,c
a
School of Optical and Electronic Information, Huazhong University of Science and Technology, Hubei, China
b
National Engineering Laboratory for Next Generation Internet Access System, Wuhan, Hubei, China
c
Wuhan National Laboratory for Optoelectronics, Wuhan, Hubei, China

art ic l e i nf o a b s t r a c t

Article history: Pilot-tone (PT) aided phase noise compensation algorithm is very simple and effective, especially for
Received 9 March 2015 flexible optical networks, because the phase noise coming from both Tx/Rx lasers and nonlinear cross
Received in revised form phase modulation (XPM) during transmission can be adaptively compensated without high computa-
26 May 2015
tional cost nonlinear operations, or the information of the neighboring channels and the optical link
Accepted 11 June 2015
Available online 15 June 2015
configuration. But to achieve the best performance the two key parameters, i.e. the pilot to signal power
ratio and pilot bandpass filter bandwidth need to be optimized. In this paper it is demonstrated that
Keywords: constellation information can be used to adjust the two parameters adaptively to achieve the minimum
Coherent transmission system BER in both homogenous and hybrid single carrier transmission systems with different LPN, XPM and
Pilot tone
amplified spontaneous emission (ASE) noise distortions.
Phase noise compensation
& 2015 Elsevier B.V. All rights reserved.
Laser phase noise
Cross phase modulation
Digital signal processing

1. Introduction linewidth tolerance for mQAM formats at the expense of im-

plementation complexity. DPLL generally performs poor for high-
With the development of advanced modulation and coherent speed optical transmission systems requiring high-degree of par-
detection techniques, the spectral efficiency and transmission allel processing due to the delay of the feedback loop. ML algo-
distance of optical transmission systems have been greatly im- rithms often require an initial training data sequence at the be-
proved [1,2]. For advanced high order modulation formats, such as ginning of the algorithm or are used in multi-stage hybrid phase
quadrature phase shift keying (QPSK) and m-ary quadrature am- noise compensation. With regard to nonlinear distortion com-
plitude modulation (mQAM), laser phase noise (LPN) is an essen- pensation, backward propagation (BP) algorithms are powerful [7],
tial degrading factor [3–5]. In addition, the optical signal phase is but in order to deal with XPM distortions the coherent receivers
also perturbed by transmission fiber nonlinearity [6]. For high must capture the total electrical fields of all the channels, which is
spectral efficiency, long-haul systems or hybrid systems accom- unfeasible due to the limited bandwidth of the electrical devices.
modating both amplitude and phase shift keying (ASK/PSK) sig- Furthermore, BP algorithms are known to have high computa-
nals, the nonlinear phase noise induced by XPM distortions is tional cost and require optical link configuration information,
another important degrading factor [7–11]. Therefore LPN and which may not be feasible for real time processing and flexible
XPM compensation techniques are very important and have been networks [9,10].
studied intensively recently [4–15]. For LPN compensation the In contrast to the above methods, PT-aided phase noise com-
M-th power algorithms and its variants are often used, but they pensation algorithm is much simpler, because it involves no non-
require nonlinear operations with high computational load, and linear operations, and the LPN and XPM phase noise can be com-
are sensitive to the thresholds of partitioning the symbols into pensated adaptively without the information of the neighboring
QPSK subsets [4,5]. LPN can also also be dealt with blind-phase channels and optical link configuration [8–12]. It can be realized by
search (BPS) [13], digital phase lock loop (DPLL) [14] and max- subtracting the PT phase from that of the main signal because the PT
imum likelihood (ML) approaches [15]. BPS can achieve the best is affected by nearly the same LPN and XPM phase noise as the main
signal [10]. Furthermore compared to the other methods the PT-ai-
n ded method can achieve much larger LPN tolerance. Experiments
Corresponding author at: School of Optical and Electronic Information, Huaz-
hong University of Science and Technology, Hubei, China. Fax: þ 86 27 87556188. reported recently show that vertical cavity surface emitting lasers
E-mail address: [email protected] (C. Ke). (VCSELs) with large linewidth can be used as Tx and Rx LO lasers in

http://dx.doi.org/10.1016/j.optcom.2015.06.030
0030-4018/& 2015 Elsevier B.V. All rights reserved.
 S. Cui et al. / Optics Communications 355 (2015) 44–49 45

10Gbaud PM-QPSK coherent transmission systems by using PT-aided signal, so that the phase information of the PT can be extracted.
phase noise compensation. The laser linewidth tolerance is improved The optimal PSR and BPF-BW depend on the relative magnitude of
from 10 MHz per laser to 300 MHz per laser [12]. the LPN, XPM and ASE noise components and have to be optimized
Although with the above merits, two key parameters, i.e. the according to different systems [8]. So far, there is no effective way
pilot to signal power ratio (PSR) and pilot bandpass filter band- to optimize these parameters adaptively, and almost all of the
width (BPF-BW) need to be optimized to achieve the best perfor- previous works directly used bit error rate (BER) or Q factor
mance. The PSR is defined as 10log10(Ppilot/Psig), where Ppilot and feedback to optimize PSR and BPF-BW [8–10]. While this works
Psig represent the power of the PT and the main signal, respec- well in offline processing, it may be slow and not effective in real
tively. At the transmitter PT with appropriate PSR is inserted into system implementations. In this paper it is demonstrated that the
the signal spectrum by frequency multiplexing. The crosstalk can constellation information can be used to adjust PSR and BPF-BW to
be mitigated by filtering the main signal spectrum with narrow- achieve the best performance. Numerical simulations are then
band Nyquist-filters [10]. At the receiver a digital BPF with ap- carried out to demonstrate the effectiveness and efficiency of this
propriate bandwidth is used to separate the PT from the main method in homogenous and hybrid systems.

Fig. 1. The comparison between the actual phase and estimated phase by PT with (a) too small BPF-BW (b) optimal BPF-BW and (c) too large BPF-BW. The beat linewidth is
200 kHz.
46 S. Cui et al. / Optics Communications 355 (2015) 44–49

2. Operating principles

The capability of PT-aided phase noise compensation depends

on the PSR and BPF-BW. As we know, for a fixed launch power, low
PSR reduces signal to noise ratio (SNR) of the inserted PT, thus
compromise phase noise estimation accuracy, while high PSR de-
creases main signal SNR. Meanwhile, larger BPF-BW enhances ωc
noise but it is beneficial for the compensation of the high fre-
quency component in the phase noise. Such high frequency phase
noise component is dominant when the laser beat linewidth is
large [12] or the XPM distortions are strong. It is noteworthy that
the XPM phase noise usually has several GHz of bandwidth due to
the ultrafast XPM response [16]. Therefore an optimal tradeoff
must be achieved. To get a better insight of the importance of the
optimization of the parameters, Fig.1 shows the real phase noise
and the estimated one from PT phase with different BPF-BWs. The
root mean
square (rms) of the phase estimation errors and the constella-
tions obtained after compensation are also shown in the figure.
The error is smallest at the optimal BPF-BW. For BPF-BW smaller Fig. 2. System setup of the PT-aided Nyquist WDM transmission system.
than the optimum one, the estimation error comes mainly from
the slow phase tracking speed, while for BPF-BW larger than the off factor of 0.1. The PT is then added to the frequency band at-
optimal one, it comes mainly from ASE noise induced phase dis- tenuated by the narrow-band Nyquist-filter. The frequency gap
tortions. In the constellations the noise clouds due to ASE and the between the PT and main signal is 12 GHz and the guard-band
residual phase noise after compensation are clearly visible. The around the PT is 500 MHz, considering the filtering attenuation
noise clouds are more elliptical when non-optimal BPF-BW is imposed by various filters in the systems and the expansion of the
used. This is can be explained by the fact that the ASE noise, which PT linewidth due to XPM induced phase noise. After that, the
follows complex Gaussian distribution, leads to circular noise
electrical QAM signal is pre-equalized and fed to the digital to
clouds, while the phase noise which rotates the symbols along the
analog converters (DACs) for generating the electrical QAM signal
angular direction leads to elliptical ones. Thus the circularity of the
which is used to drive the IQ-modulator to generate the optical
noise clouds can indicate the amount of the residual phase noise
signals. The launch optical signal power is set at 3 dBm. The
and be used to optimize the parameters. Here we use the the
optical signal is then multiplexed with the other eight channels
second moment difference to evaluate the noise cloud circularity
and sent into the transmission fiber. The inset of Fig. 2 shows the
which is defined by
launched signal optical spectrum with two neighboring channels.
E(Im − I¯m)2 − E(Q m − Q¯ m)2 At the coherent receiver, the field of the optical signal along
Ecir = ∑ , with the PT is converted into digital domain through the intradyne
E(Im − I¯m)2 + E(Q m − Q¯ m)2 (1)
m coherent optical front-end and high-speed analog to digital con-
where E represents mathematical expectation, while Im ( Im ) and verters (ADCs). Chromatic dispersion is then compensated by
Q m ( Q m ) represents the value (averaged value) of the real and overlap frequency domain equalization [17], while polarization de-
imaginary parts of the received symbols of the m-th symbol clouds multiplexing and equalization are realized by a Ts/2-spaced (Ts is
in the constellation, respectively. For signals with only ASE noise the symbol duration) adaptive equalizer with butterfly config-
Ecir is close to zero. Although all of the noise clouds can be used in uration [1]. Carrier frequency offset is compensated using the
the calculation, to reduce the computational cost, we only choose modified differential Viterbi algorithm [18]. Then the LPN and XPM
the symbols belong to the outmost ring to calculate Ecir . As can be phase noise is compensated by the method shown in Fig. 3
seen from Fig. 1, with the same residual phase noise the clouds on (b) which is the same as that used in [10]. The pilot tone is se-
the outmost ring are more elliptical and thus more sensitive to the parated from the main signal with a digital BPF and frequency-
residual phase noise. shifted to zero frequency. The extracted pilot tone is frequency-
shifted, complex-conjugated, and then multiplied with the main
signal. After phase noise compensation the signal is fed to the
3. Adaptive optimization in different systems clock recovery and decision block.
Fig. 4 and Fig. 5 show the variations of BER and Ecir versus PSR
In this section numerical simulations are carried out to de- and BPF-BW under different laser beat linewidth Δυ and OSNR
monstrate the effectiveness of the optimization method in systems conditions. 210 symbols are used to calculate Ecir . As expected,
with different LPN, XPM and ASE noise distortions. VPI Trans- there exit an optimal PSR and BPF-BW leading to the minimum
missionMaker V9.0 is used to simulate the coherent transmission BER, and as we can see they are very close to the optimal values
systems, within which the optical signal propagation in optical obtained from the minimum Ecir .
fibers is simulated using nonlinear Schrödinger equations (NLSEs) In order to automatically find the optimal BPF-BW (denoted by
and symmetric split-step Fourier method, and the DSPs in the Δf ) in the receiver, gradient-descent algorithm can be applied in
transmitter and receiver are simulated using Matlab. The setup of the PT-based phase noise compensation module. In every iteration
the PT-aided coherent transmission system is shown in Fig. 2. The Δf is updated by
system is a 25 GHz-grid wavelength division multiplexing (WDM) dEcir
23 Gbaud/channel PM-16QAM system. The transmitter DSP block Δf = Δf − μ ,
dΔf (2)
diagram is shown in Fig. 3(a). First the input data is mapped to
QAM constellation and up-sampled to 2 samples/symbol. Then the where μ is the step size of tuning the BPF-BW. For a preliminary
digital signal is pulse-shaped with Nyquist-filters with a small roll- coarse search we can choose a large step size. When the minimum
 S. Cui et al. / Optics Communications 355 (2015) 44–49 47

f f

Fig. 3. (a) DSP block diagram at the transmitter and (b) the PT-aided phase noise compensation algorithm diagram at receiver.

Fig. 4. The variations of BER versus PSR when TsΔυ and OSNR are equal to (a) 1e-6 and 20 dB, and (b) 1e-5 and 24 dB, respectively. The BPF-BW is set at 300 MHz. The
optimum PSRs obtained from the two curves are close to each other.

Fig. 5. The variations of BER versus BPF-BW when TsΔυ and OSNR are equal to (a) 1e-4 and 20 dB, and (b) 1e-5 and 24 dB, respectively. The PSR is set at  11 dB. The optimum
BPF-BWs obtained from the two curves are close to each other.

Ecir is coarsely localized, the step size can be reduced for better conditions. The solid and dash lines represent the optimal values
performance. For example, Fig. 6 shows the variations of Δf , Ecir obtained from the minimum BER and Ecir , respectively. As can be
and μ versus the number of iterations. Here TsΔυ and OSNR are the seen, these two curves are very close to each other and the dif-
same as those used in Fig. 5(a). The initial values of Δf and μ are set ference is negligible in a very large range of Δυ ¼100 kHz to 5 MHz.
at 300 MHz and 80 MHz, respectively. Every time when dEcir/dΔf Furthermore Fig. 7(a) shows that the optimal PSR increases with
changes its sign, μ is halved for finer tuning, noting that dEcir/dΔf is Δυ , because increasing PT power can raise the more of PT spectrum
equal to zero at the optimal Δf (see Fig. 5). After about 10 above the ASE noise level, thus the high frequency LPN compo-
iterations Δf converges to 670 MHz which is very close to the nents can be extracted out more completely. While when Δυ is
optimal Δf as can be seen from Fig. 5(a). The optimization method fixed, the optimal PSR increases with decreasing OSNR, because
of PSR is similar thus is discussed for concision. We note that larger PT power can maintain the PT SNR and phase noise esti-
optimization of BPF-BW can be realized at the receiver, while mation accuracy, regardless of the increasing ASE noise level. Fig. 7
optimization of PSR is possible for the networks where the feed- (b) shows that the optimal BPF-BW increases with Δυ , because the
back to the transmitter is allowed [19]. high frequency LPN components increase with Δυ . Furthermore it
Fig. 7(a) and (b) shows the variations of the optimal PSR and shows that the optimal BPF-BW also increases with OSNR, because
BPF-BW versus laser beat linewidth Δυ under different OSNR when ASE noise level decreases, using a larger BPF-BW is
48 S. Cui et al. / Optics Communications 355 (2015) 44–49

Fig. 6. The variations of Δf , Ecir and μ versus the number of iterations.

Fig. 7. The variations of the (a) Optimal PSR and (b) Optimal BPF-BW versus linewidth obtained from minimum BER and Ecir .

200 kHz. The chromatic dispersion, attenuation, nonlinearity coeffi-

cients and effective area of the SSMF are set at 16 ps/(nm km), 0.2 dB/
km, 2.6 × 10−20 m2/W, and 80 mm2, respectively. The launched signal
power is varied from  10 to  2 dB m to change the XPM impact. In
the homogenous system nine PM-16QAM channels are transmitted
over 1000 km. In the hybrid systems the PM-16QAM channel is
transmitted with eight 10 Gb/s NRZ channels surrounding it. For the
hybrid system, dispersion compensation modules are added in every
span to compensate the dispersion, and the transmission distance is
set at 500 km as the nonlinear distortions are much larger, thus
limiting the transmission distance.
Fig. 8 shows the variations of the optimal BPF-BW versus the
launched optical signal power. The solid and dash lines represent
the optimal BPF-BWs obtained from the minimum BER and Ecir ,
respectively. These two curves are very close to each other,
showing that the proposed optimization method is effective for
both systems. Furthermore as can be seen, the optimal BPF-BW
increases with launched signal power, because the high frequency
Fig. 8. Optimal BPF-BW versus launch optical power per channel for the long-haul
XPM phase noise components increase with launched signal
homogenous and hybrid transmission systems, respectively. power. While at the same launched signal power, the optimal BPF-
BW of the hybrid system is larger than that of the homogenous
beneficial for fast tracking speed, and thus reducing the residual one, because the nonlinear XPM impact from the NRZ channels is
phase noise, without incurring SNR loss. larger when compared to 16QAM channels, especially in the high
To validate the optimization method in PT-aided homogenous and power region [6]. Finally, it is noteworthy that the PT-aided phase
hybrid WDM systems, extensive numerical simulations are carried noise compensation method is not effective for self-phase mod-
out. The simulation system setup is shown in Fig. 1. It consists of N ulation (SPM) induced phase noise [8]. Thus to improve the per-
spans of 100 km standard single-mode fiber (SSMF). In every span an formance further, a SPM compensator, such as the one based on BP
erbium doped fiber amplifier (EDFA) is used to compensate the span algorithm, can be added before the LPN and XPM compensator
loss. The transmitter and receiver laser beat linewidth Δυ is set at [8,9].
 S. Cui et al. / Optics Communications 355 (2015) 44–49 49

4. Conclusion References

In this paper an adaptive optimization method for the PT-aided [1] S.J. Savory, IEEE J. Sel. Topics Quantum Electron 16 (2010) 1164–1179.
phase noise compensator is proposed. The noise cloud circularity [2] S.ébastien Bigo. Optical Fiber Communication Conference Los Angeles, Cali-
fornia, March 2012 4 8.OTh3A.
in the constellation is used to adjust PSR and BPF-BW to achieve [3] Matthias Seimetz Optical Fiber Communication Conference 2008 San Diego
the minimum BER. Extensive numerical simulations show that this CA.paper OTuM2.
method is effective for systems with different LPN, XPM and ASE [4] Xiang Zhou, IEEE Signal Process. Mag. 31 (2014) 35–45.
[5] H.Louchet, K.Kuzmin, A.Richter, European Conference on Optical Commu-
noise distortions. nication, Brussels, Belgium, Sept. 2008, paper.
[6] A. Bononi N. Rossi P. Serena Optical Fiber Communication 2011 Los Angeles CA.
paper OWO7.
[7] K. Toyoda, Y. Koizumi, T. Omiya, M. Yoshida, Opt. Express 20 (2012)
Acknowledgments 19815–19821.
[8] A. Diaz A. Napoli S. Adhikari Z. Maalej A. Lobato M. Kuschnerov J. Prat Optical
This work is supported by the National High Technology Re- Fiber Communication Conference 2012 Los Angeles California.paper OTh3C.5.
[9] L.B. Du, A.J. Lowery, Opt. Lett. 36 (2011) 1647–1649.
search and Development Program (863 Program) of China (Grant [10] T. Kobayashi, A. Sano, A. Matsuura, Y. Miyamoto, K. Ishihara, J. Lightwave
no. 2013AA013403), the National Key Scientific Instrument and Technol. 30 (2012) 3805–3815.
Equipment Development Project of China (2013YQ160487) and [11] M.M. Osman, Q. Zhuge, L.R. Chen, Opt. Express 19 (2011) 329–336.
[12] X. Zhang, X. Pang, L. Deng, D. Zibar, I.T. Monroy, Opt. Express 20 (2012)
the National Natural Science Foundation of China (Grant no. 19990–19995.
61475053). [13] T. Pfau, S. Hoffmann, R. Noé., J. Lightwave Technol. 27 (2009) 989–999.
[14] I. Fatadin, D. Ives, S.J. Savory, IEEE Photonics Technol. Lett. 22 (2010) 176–179.
[15] X. Zhou, IEEE Photonics Technol. Lett. 22 (2010) 1051–1053.
[16] J. Cheng, M. Tang, S. Fu, P.P. Shum, D.M. Liu, Opt. Express 22 (2014) 1257–1266.
Appendix A. Supplementary Information [17] R. Kudo, T. Kobayashi, K. Ishihara, Y. Takatori, J. Lightwave Technol. 27 (2009)
3721–3728.
[18] I. Fatadin I, S.J. Savory, IEEE Photonics Technol. Lett. 23 (2011) 1246–1248.
Supplementary data associated with this article can be found in [19] Xingwen William Shieh, Yiran Yi, Ma, Qi Yang, J. Opt. Netw. 7 (2008) 234–255.
the online version at http://dx.doi.org/10.1016/j.optcom.2015.06.030.