frequency estimation for optical coherent mqam system without removing modulated data phase
Post on 11Sep2016
213 views
Embed Size (px)
TRANSCRIPT
nt
W
ive
y q
by a
da
4, t
16
igital signal processing (DSP)irment compensation of thever, bed froment DSdth tolpproacfrequenO) laseon ofent. T
symbols is introduced in this MQAM scheme. In addition, it is to
to counteract the noise. Simulations of 10Gbaud 16QAM and
d Df
fd,n yd,n2pDfnTs2pm
Contents lists available at
els
Optics Comm
Optics Communications 285 (2012) 36923696/fd,nS and the residual y0d,n.Email address: yusong@bupt.edu.cn (S. Yu).256QAM coherent systems demonstrate that the range of the where m is an integer to ensure fd,n to fall into the range ofp,p. For simplicity, laser phase noise and ASE noise areneglected. It is worth to mention that in Eq. (2) the randomvariable of fd,n has been decomposed into its average value
00304018/$  see front matter & 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.optcom.2012.05.027
n Corresponding author.derive that specic constellation points should be rotated by p=4, /fd,nSy0d,n, 2extended to Mary quadrature amplitude modulation (MQAM)format, which is more likely to be employed in the next generation optical coherent communications system. Different from ourformer MPSK scheme [8], partition according to the amplitude of
l,n
(Df l), Ts is the symbol duration, Nn represents ASE noise, andenotes the frequency offset.
The argument of the signal is of the formestimator is necessary to ensure reliable performance of PE.We have presented a frequency estimation algorithm for
Mary phaseshift keying (MPSK) optical coherent system [8].Compared with previous algorithm, the modulated data phase isnot required to be removed. In this paper, the algorithm is
The nth received MQAM symbol can be presented as
Sn An expjyd,nyl,n2pDfnTsNn, 1
where An, yd,n represent modulated amplitude and phase ofMQAM format, y is the phase noise induced by laser linewidthIn optical coherent detection, dshows its competence for the impalonghaul transmission [13]. Moreoof phase recovery can be transferresoftware domain [1]. Among differtions (PEs), considering laser linewitation complexity, feedforward aHowever, this approach needs thetransmit and the local oscillator (LMeanwhile, the frequency resoluticannot satisfy the above requiremcause of DSP, the stagehardware domain to
Pbased phase estimaerance and implemenh [47] is preferred.cy offset between ther to be small enough.a typical tunable laserherefore, a frequency
2. Principle
For MQAM coherent system, the modulated data phase israndomly chosen from the constellation states. Thus, the corresponding phase can be treated as a discrete noise. On the otherhand, the frequency offset is comparatively stable and its resulting argument change would be periodical. Therefore, after Fouriertransform, the frequency offset can be extracted directly [8].1. Introduction frequency offset estimation can cover [1.2 GHz, 1.2 GHz],while the maximum estimation error is below 1 MHz.Frequency estimation for optical coheremodulated data phase
Song Yu n, Yinwen Cao, Haijun Leng, Guohua Wu,
State Key Laboratory of Information Photonics and Optical Communications, Beijing Un
a r t i c l e i n f o
Article history:
Received 17 January 2012
Received in revised form
15 May 2012
Accepted 16 May 2012Available online 31 May 2012
Keywords:
Coherent detection
Fast Fourier Transform (FFT)
Mary quadrature amplitude modulation
(MQAM)
a b s t r a c t
For optical coherent Mar
can be extracted directly
removing the modulated
constellation points by p=Numerical simulations of
algorithm.
journal homepage: www.MQAM system without removing
anyi Gu
rsity of Posts and Telecommunications, Beijing 100876, PR China
uadrature amplitude modulation (MQAM) system, the frequency offset
pplying Fast Fourier Transform (FFT) to the signals argument, without
ta phase. By categorizing the constellation points and rotating some
his algorithm is robust to extract the frequency offset against the noise.
QAM and 256QAM coherent systems are presented to demonstrate this
& 2012 Elsevier B.V. All rights reserved.
SciVerse ScienceDirect
evier.com/locate/optcom
unications
pDfn
S. Yu et al. / Optics Communications 285 (2012) 36923696 3693From Eq. (2), the average term is governed by
Fig. 1. The division of constellation for 16QAM when 2
Table 1The derivation of /mS for range I.
i m Possibility /mS
4k k 13=16 i=43=16k1 3/16
4k1 k 9=16 i=43=16k1 7/16
4k2 k 5=16 i=43=16k1 11/16
4k3 k 1=16 i=43=16k1 15/16/fd,nS/yd,nS/2pDfnTsS/2pmS 2pDfnTs2p/mS: 3Without loss of generality, 16QAM is demonstrated to illus
trate the algorithm. In order to ensure that m is only dependenton yd,n, 2pDfnTs is divided into three ranges, range I:p=4 ip=2,p=2a ip=2; range II: p=2a ip=2,p=2a ip=2;range III: p=2a ip=2;3p=4 ip=2, where a arctan1=3 and iis an integer which is denoted as 4kr r 0;1,2;3.
As for range I, all four cases are discussed in Fig. 1 and Table 1.Fig. 1 shows that the constellation is divided into two classes(open symbols and lled symbols). If the modulated data belongsto open symbols, m k1. If not, mk. The explanation andanalysis will be given in detail later.
As for range I, if r1, which is corresponding to Fig. 1(a), lledsymbols exhibit argument less than 3p=4 and 2pDfnTsAp=42kp,p=2a2kp. Therefore, for lled symbols, mk isrequired to keep fd,nA p,p and for the rest symbols (opensymbols), one additional 2p is required. The requirement fordifferent m for different modulate data accounts for the two setsof symbols in Fig. 1. The possibility of open symbols is 3=16, whilethe possibility of lled symbols is 13=16. Therefore, /mSi=43=16. Considering Eq. (3) and Table 1 together, if 2pDfnTsis in range I, Eq. (3) can be rewritten as
/fd,nS 2pDfnTs3p=8ip=2: 4Similarly, for range II and range III, we have
/fd,nS 2pDfnTsp=2ip=2; 5or
/fd,nS 2pDfnTs5p=8ip=2: 6In order to express Eqs. (4)(6) in a more general form, Gx isintroduced
Gx 1, xry1,0, y1oxry2,1, x4y2:
8>: 7
Therefore
/fd,nS 2pDfnTspG2pDfnTs=8y3ip=2, 8
where y3 p=2.The trajectory of /fd,nS for 16QAM (black solid line) is different
from QPSK (black dashed line), as shown in Fig. 2. The argumentchange induced by frequency offset (argexpj2pDfnTs) is alsoplotted (gray solid line). It can be seen that the period of /fd,nSfor 16QAM is the same as that of QPSK, a quarter ofargexpj2pDfnTs. Therefore, after applying FFT to /fd,nS, the rstharmonic frequency represents the absolute value of 4Df , as shownin Fig. 3.
However, compared with the spectrum of /fd,nS for QPSK and16QAM in Fig. 3(a) and (b), the amplitude of the rst harmonicfor 16QAM is much lower, making it difcult to be extracted.
In order to increase the amplitude of rst harmonic, rstly, wepartition different rings into two classes, CI and CII. In CI, the ringsconsist of constellation points that the shorter distance from axis Iand II is less than that from axis X and Y in Fig. 4(a), while CII
includes the rest rings. Specically, CI is the middle ring and CII
includes the inner and outer rings. We rotate CI by y, as shown inFig. 4(a). Considering the symmetry of the constellation, rotationof 0,p=2 is sufcient. The derivation of /mS is similar, and its
Ts is in range I. (a)(d) denote r 1;2,3;0, respectively.expression is the same as Eq. (8). The only differences are y1, y2and b, which is shown in Table 2.
To facilitate the expression of the rst harmonic F1, we usecontinuous variable t to substitute nTs, yielding
F1 1
T
Z t0 Tt0
/fd,nSejo1t dt, 9
where T 1=4Df ando1 2p 4Df . Here, we assume Df40. Theintegral range is p=4;3p=4 and Eq. (9) can be further simplied
F1 j
16ej4y1 ej4y22: 10
The amplitude of F1 is
9F19 18cos2y1y2fcos2y1y22 cos2y1y2g1
p: 11
It is worth to mention that all the constellation points best lineup when the CI points are rotated by y p=4. And this can befurther conrmed by Figs. 3 and 4(b). It can be mathematically
S. Yu et al. / Optics Communications 285 (2012) 369236963694
3/4calculated from Eq. (10) that the increase in 9F19 is 0.777, which isconsistent with Fig. 3.
Fig. 5(a) is presented to illustrate the modied frequencyestimator. Firstly, partition the rings including CI by the amplitude;
T/4 T/2
Pha
se (r
ad)
/4
0
/4
/2
/2
3/4
period of
Fig. 2. /fd,nS and argexpj2pDfnTs, T 1=Df : The pe
1 3 5 70
0.05
0.10
0.15
0.20
0.25
f/(4f)
FFT

QPSK
1 30
0.05
0.10
0.15
0.20
0.25
f
FFT

16QAM w
Fig. 3. Spectrum of /fd,nS. (a) /fd,nS for QPSK; (b) /fd,nS for 16QA
0.09
0.10
0.11
0.12
0.13
0.14
0.15
0.16
F1
Y
X
III
Fig. 4. (a) Rotation of CI by y. Black symbols: CI before rotation. Gray s
then, rotate these rings by p=4; next, obtain the argument ofall the symbols; after that, use FFT to get the correspondingfrequency of the rst harmonic frequency; nally, divide it by 4 toget 9Df 9.
3T/4 Tt
Arg[exp(j2fnTs)]
riod of /fd,nS is T/4 for both 16QAM and QPSK.
5 7/(4f)
/o Rotation
1 3 5 70
0.05
0.10
0.15
0.20
0.25
f/(4f)
FFT

16QAM w. Rotation
M before p=4 rotation; (c) /fd,nS for 16QAM after p=4 rotation.
0 /8 /4 3/8 /2
0.09
0.16
Recommended