Paper Title (use style: paper title)

6 downloads 46 Views 581KB Size Report
"almost unlimited" bandwidth offered by the optical fiber. Recent experimental reports show the generation, transmission and detection of bit rates up to 100 ...
Evaluation and Compensation of Interchannel Interference Effects in a 16-QAM Nyquist-WDM System with LMS Equalization Jhon James Granada Torres1, Ana María Cárdenas Soto1, Neil Guerrero González2 1

Universidad de Antioquia, Electronic Engineering - Graduated Programs. Colombia. 2 Centro de pesquisa e desenvolvimento em telecomunicações (CPqD). Brazil. [email protected], [email protected], [email protected]

Abstract— Nyquist-WDM systems have been proposed as a promising solution in multicarrier transmissions to be a technology which could face the elastic networks challenges. In the spectrum, Nyquist-WDM requires an ideal rectangle, characteristic that is impossible in practice. With a channel spacing near or equal to baud rate, the roll-off factor of the pulse shaping filter has influence in the channel spreading, generating spectral overlapping between adjacent channels. This phenomenon is called Interchannel Interference (ICI). This research evaluates the impact of the ICI effects in a 5 x 32 Gbaud Nyquist-WDM system, with 16-QAM modulation. The impact of the Roll-off and channel spacing value is analyzed for distances up to 5 km. Error Vector Magnitude in constellation and bit error rate are estimated according to channel spacing penalty, roll-off factor, and transmission distance variation. ICI effects minimization using LMS algorithm in the digital coherent receiver is proposed. LMS adapts the coefficients of the adaptive filter making equalization by a training sequence, superchannels signals are received in independent coherent receivers in a joint equalization among channels. The performance of the systems is improved with LMS equalization, reducing the log(BER) from -1 to -11 for 5 km in a channel spacing equal to the symbol rate, and from ~-2 to ~-20 for a channel spacing penalty of 1.5 GHz, with roll-off factor Nyquist filter of 0.15, which is feasible in practice in the pulse shaping filter at the transmitter. Keywords—Interchannel Interference Square (LMS) equalization; Nyquist-WDM;

(ICI);

Least-Mean-

I. INTRODUCTION In the last years, traffic on data optical networks has had an exponential growth due to the increasing demand of new telecommunication services, such as video on demand, fullhigh definition video transmission, cloud, and grid computing, among others. Thus, an upgrade in access, metro, and longhaul optical networks is expected to optimize the use of the "almost unlimited" bandwidth offered by the optical fiber. Recent experimental reports show the generation, transmission and detection of bit rates up to 100 Gbps over the current optical transport network [1, 2]. These experimental tests have new technological concepts related to the commercial 100-G systems such as: designing of multichannel transmitters, digital coherent receivers, advanced modulation

formats from 16QAM to 1024-QAM and hybrid modulations [3]. Nevertheless, high order modulation formats do not allow long reach transmission due to optical signal to noise ratio (OSNR) [4,5]. Therefore, maintaining a fixed-grid spectrum of 50 GHz in Wavelength Division Multiplexing (WDM) networks, according to ITU standard, will not be possible to reach bit rates up to 100 Gbps for each channel with m-ary modulation and for long distances [6, 7]. Thereby, a new concept known as elastic networks emerges, proposing a gridless or flexible grid spectrum, where the spectral width changes according to bandwidth demand. Flexible grid optimizes the spectral efficiency with a minimum separation between adjacent channels seen as multiple carriers. They are called as superchannels [5, 8-10]. Nyquist-WDM (N-WDM) has been proposed as a promising technology to lead the elastic optical networks. This technology allows transmitting multiple channels with a spectral spacing equal to baud rate, making an efficient and optimum use of the optical spectrum. However, N-WDM requires a nyquist optical pulse represented in the spectrum by an ideal rectangle, a characteristic that is impossible in practice, due to the finite length of pulse-shaping filter that not allow a perfect spectrum slicing required for conforming such rectangular pulse shape. Hence, eventual spectral overlapping between channels leads to generate a linear and nonlinear crosstalk between adjacent channels. Both effects are called as interchannel interference (ICI). This phenomenon is due to inherent linear and nonlinear optical fiber impairments [11]. In this paper, a 5 x 32Gbaud N-WDM system with coherent detection was modeled in Virtual Photonics Inc (VPI) software. ICI effects according to the channel spacing and the bit error rate (BER) were evaluated. We propose a equalization using Least-Mean-Square (LMS). LMS algorithm has been widely used in communication systems to mitigate noises and compensate chromatic dispersion in optical links. In this work, LMS equalization reduced the ICI effects with regard to BER, distance and channel spacing. This paper is organized as follows: in section II, N-WDM systems are explained. In section III, the experimental set up is described.

978-1-4799-7162-6/14/$31.00 ©2014 IEEE

The results and discussions are showed in Section IV, and finally, conclusions are presented in section V.

In Nyquist-WDM, the optical pulses have a rectangular spectrum, which occupy a bandwidth close or equal to the baud rate. The channel spacing can then be ideally as small as the symbol rate [5, 6]. Maximum spectral efficiency can be achieved, at zero inter-symbol interference (lSI) under ideal conditions, and avoiding ICI effects [12]. However, N-WDM suffers distortions in its pulse generation, due to hardware implementation limitations such as: the finite length of the pulse shaping digital filter, the digital-to-analog conversion (DAC) pulse shaping, the non-ideal timing in the sampling of analog-to-digital conversor (ADC) [13], and the finite amplitude resolution of DAC and ADC [3]. These constraints do not allow transmitting channels as close as baud rate spacing. Therefore, the roll-factor of the digital filter affects the system performance in terms of channel spacing as well as bit error rate. In fig. 1, spectrum of optical nyquist pulse is showed by two different roll-off factor: 0.01 and 0.15. This last value can be obtained in practice with 17 tap filter [14].

A multichannel N-WDM system is simulated using VPI. Each channel is shaped by a raised-cosine optical filter. The roll-off factor used is 0.15. Laser source has a linewidth of 1 MHz. Symbol rate is 32 Gbaud for 16-QAM modulation. Transmitter is configured with a pair of single drive Mach Zender Modulators (MZM). Each channel is multiplexed and transmitted through standard single mode fiber (SMF). The experimental distances are from Back-to-Back (BTB) to 5 km.

Tx 1 Tx 2

SSMF

Tx 3 Tx 4

Coherent Receiver

II. NYQUIST-WDM SYSTEM

III. SIMULATION SET UP

Tx 5 Figure 2. Layout of Nyquist-WDM system

Superchannel signals are detected by independent coherent receivers using a local oscillator and synchronous sampling. In-phase and quadrature component signal is jointly processed to enable ICI mitigation, using LMS algorithm. LMS is an iterative algorithm which finds the coefficients of adaptative filter by an error minimization between training sequence and received signal [15-16]. It can be seen as follows:

y  n  w

Figure 1. Nyquist pulse spectrum for Roll-off factor of 0.01 and 0.15

Equation 1 shows the relation of the roll-off factor (roffwith the channel spacing (CHS):  ChS

 (1  roff ) Rs 

WhereRS is the symbol rate. Roll-off factor cannot be equal to zero in the practice, in this sense, the expression of channel spacing could be written as:

 n 1 u  n  e  n  d  n  y  n w  n   w  n  1  f  u  n  , e  n  ,  

 Rs  ChsP 

ChsP  Rs roff  Where CHsP is the channel spacing penalty of the Nyquist pulse shaping. If the condition (3) is not met, a crosstalk between adjacent channels is induced.

(5) (6)

Weights update function is given by:

f  u  n  , e  n  ,    µe  n  u*  n  Where:

 ChS For

(4)

n : The current time index. w(n) : weight vector. d (n) : desired signal (training sequence) y(n) : output equalizer signal. u (n) : input equalizer signal. u* (n) : conjugated of u (n) . e : estimated error.

978-1-4799-7162-6/14/$31.00 ©2014 IEEE

(7)

µ : step size. In figure 3, a coherent receiver with LMS equalization in electronic domain is showed. LMS modifies the equalizer’s coefficients adaptatively.

Table I. EVM and BER estimation for a 16-QAM transmission with a channel spacing penalty of 1 GHz and roll-off of 0.15 in 1 km

LMS

Analog-to-Digital Converter

LO

Optical Hybrid

Photodetectors

channels, as well as, BER values. It can be verified with their standard deviation (. For 1 km EVM = 0.014 and log(BER) = 0.015, for 3 km EVM = 0.0097 and log(BER) = 0.0054. The following information is the result from only one channel, assuming the similar performance of the other channels, how we can demonstrate at the end of the section.

EVM Log(BER)

Ch1 11.424 -3.262

Ch2 11.443 -3.253

Ch3 11.436 -3.257

Ch4 11.444 -3.258

Ch5 11.462 -3.245

Table II. EVM and BER estimation for a 16-QAM transmission with a channel spacing penalty of 1 GHz and roll-off of 0.15 in 3 km

EVM Log(BER)

Figure 3. Coherent Receiver + LMS.

Ch1 14.211 -2.340

Ch2 14.208 -2.341

Ch3 14.205 -2.342

Ch4 14.226 -2.337

Ch5 14.224 -2.338

Error vector magnitude (EVM) is calculated over digital signal as follows: 1

EVM RMS

2 2 2 1 T  T  t 1 I t  I 0,t  Qt  Q0,t    (8) N 1  I  2   Q  2     0, n 0, n    N n 1 

Where It is the normalized in-phase signal for measured symbols, and I0,t is the normalized in-phase signal for the transmitted symbols in the constellation. Qt is the normalized in-phase signal for measured symbols, and Q0,t is the quadrature signal of the transmitted symbols. N is the number of ideal constellation points and T the number of the measured constellation points. According to the equation 8, when It = I0,t and Qt = Q0,t EVM value is 0%; it means an ideal transmission. Therefore, EVM’s value increases with regard to the symbols distortion increasing. According to the EVM values, the bit-error-rate (BER) is probabilistically calculated by the estimated expression in [17] as follows:

PBER

 1 2 1     2 L  3log L    Q  2 2   (9) 2 log 2 L   L  1  EVM RMS log 2 M 

Where M is the number of symbols, L is the number of levels in each dimension of the M-ary modulation system, and Q[.] is the Gaussian complementary error function (erfc).

IV. SIMULATION RESULTS AND DISCUSSION A. ICI effects evaluation Tables I and II show the EVM and BER for transmission of 1 km and 3 km respectively. These values measure the distortion of the signal received. EVM values are similar in the 5

Figure 4. Log(BER) vs Roll Off factor for different channel spacing penalty in a) 5 km, b) 3 km and c) Back-to-Back.

In figure 4, an evaluation of the roll-off impact is done for 3 different channel spacing penalties of 0.1, 0.5 and 1 GHz in a 32 Gbaud 16-QAM transmission. Roll-off value has a high impact in the ICI effects measured in terms of BER. It can be noticed in fig 4.c, that even in a Back-to-Back (BTB) transmission, roll-off factor with a value change from 0 to 0.1 represents a log(BER) increase from -30 to -10, for a channel spacing penalty of 1 GHz. Nevertheless, with an ideal Nyquistoptical pulse, optical fiber impairments even in short distance considerably affect the BER. For roff the log(BER) is ~-4 in 3 km (fig. 4.b.) and ~-2 in a 5 km transmission, while, in BTB case, log(BER) is equal to ~-30.

978-1-4799-7162-6/14/$31.00 ©2014 IEEE

Besides of roll-off factor, optical fiber length and channel spacing penalty increase the ICI effects. In figure 5, BER vs. Channel Spacing Penalty for different transmission distances, with 16-QAM and QPSK modulation for a roff is shown. QPSK is more tolerant to the ICI effects than 16QAM at the same symbol rate. QPSK reaches log(BER) of ~-12 for a 1 GHz channel spacing penalty in 3 km; however, in 16-QAM, the log(BER) is ~-2 for the same case. Depending on the distance, BER can increase considerably. For example, in 16QAM transmission, log(BER) increases from -9 to -4 in only 2 km, having a log(BER) of ~-2 for 5 km distance. Channel spacing penalty variation from 1 to 2 GHz, decreases the log(BER) from ~-4 to ~-9, in a 1 km transmission. For 5 km, the BER variation is minimum for different channel spacing penalty, reaching only a log(BER) of ~-2 in 3 GHz.

Figure 7 shows log(BER) vs. channel spacing penalty. LMS reduces the log(BER) from ~-1 to ~-11 for an ideal Nyquist spacing. BTB transmission with LMS can have a log(BER) of -10 for a channel spacing penalty of 0, equal to BTB transmission without LMS but with a channel spacing penalty of 3 GHz. Table III. Measured EVM in 16-QAM transmission over 3 km, before and after LMS equalization. %EVM %EVM Channel Spacing Before LMS After LMS Penalty 0 GHz 21.82 5.41 0.4 GHz

16.35

5.49

0.7 GHz

14.52

4.59

1 GHz

12.78

3.84

1.5 GHz

11.12

3.37

1.8 GHz

9.72

2.99

2 GHz

8.47

2.78

Figure 5. log(BER) vs Channel Spacing Penalty for different transmission distances for QPSK and 16-QAM Modulation at 32 Gbaud with roll-off of 0.15.

According to above, 5 km is a highly limited transmission in terms of BER due to the ICI effects. LMS equalization technique is performed in the coherent receiver as a technique to reduce the BER, improving the system performance. B. LMS equalization performance In figure 6, constellation diagrams are showed, for a 3 km transmission with different channel spacing penalty, before LMS and after LMS equalization. In each constellation obtained before LMS equalization of figure 6, when channel spacing penalty increases, symbol dispersion in each of the constellations decreases. After LMS, each constellation qualitatively presents similar dispersion symbols for different channel spacing penalty. For the constellation of figure 6, EVM is calculated (see table III). Measured EVM in the constellations is reduced after LMS equalization. Maximum value of EVM before LMS equalization is 21.82%, and after LMS its value is 5.41%. It means a reduction of the ICI effects impact.

Figure 6. 16-QAM constellations over 3 km, for channel spacing penalty of: a) 0 Ghz before and b) after LMS, c) 0.4 GHz before and d) after LMS, e) 0.7 GHz before and f) after LMS, g) 1 GHz before and h) after LMS, i) 1.5 GHz before and j) after LMS.

Figure 8 shows the log(BER) values with regard to the channel number, for 4 transmission cases, before LMS (green

978-1-4799-7162-6/14/$31.00 ©2014 IEEE

dots) and after LMS (blue dots) equalization. At the beginning of this section was analyzed that BER variation among channels is minimum before LMS processing. After channel equalization, the log(BER) also has a small variation among channels. The penalty of the channel location is minimum with regard to transmission parameters. The BER can have minimum variations according to the training sequence, and the parameters used to adapt the coefficients by the LMS algorithm.

V. CONCLUSION The impact of the ICI effects in a 16-QAM Nyquist-WDM systems at 32 Gbaud was investigated. The roll-off factor induced an important penalty in terms of BER even in a BTB transmission. The optical fiber impairments do not allow transmissions with an ideal Nyquist channel spacing with a roll-off factor feasible in practice without equalization techniques. The high transmission limitation by the optical fiber impairments can be reduce using LMS algorithm in the digital coherent receiver, without using pre-filtering techniques at the transmitter. Log(BER) of -10 is achieved with a roll-off of 0.15 over 5 km in an ideal Nyquist channel spacing, and an error-free transmission over 3 km with a channel spacing penalty of 1.5 GHz. This system transmits over 540 Gbps, regardless of non-ideal spectral generation, providing flexibility in the channel spacing, with a low computational cost algorithm.  ACKNOWLEDGEMENT This work has been supported by the Grant 567 of Colciencias and Sostenibilidad 2014 project. References [1]

Figure 7. log(BER) vs Channel Spacing Penalty for 16-QAM transmission, before and after LMS equalization

Figure 8. Log(BER) for each channel in a transmission: a) of 1km with channel spacing penalty = 1 GHz before LMS and b) after LMS. c) of 3km with channel spacing penalty = 1 GHz before LMS and d) after LMS. e) of 1 km with channel spacing penalty = 2 GHz before LMS and f) after LMS. g) of 3 km with channel spacing penalty = 2 GHz before LMS and h) after LMS.

Cvijetic, N.; Cvijetic, M.; Ming-Fang Huang; Ip, E.; Yue-Kai Huang; Ting Wang, "Terabit Optical Access Networks Based on WDMOFDMA-PON," Lightwave Technology, Journal of , vol.30, no.4, pp.493,503, Feb.15, 2012. [2] Wei Wei; Chonggang Wang; Jianjun Yu, "Cognitive optical networks: key drivers, enabling techniques, and adaptive bandwidth services," Communications Magazine, IEEE , vol.50, no.1, pp.106,113, January 2012. [3] Fickers, J.; Ghazisaeidi, A.; Salsi, M.; Charlet, G.; Horlin, F.; Emplit, P.; Bigo, S., "Design rules for pulse shaping in PDM-QPSK and PDM16QAM Nyquist-WDM coherent optical transmission systems," Optical Communications (ECOC), 2012 38th European Conference and Exhibition on , vol., no., pp.1,3, 16-20 Sept. 2012.. [4] Ze Dong; Hung Chang Chien; Zhensheng Jia; Xinying Li, "Joint Digital Preequalization for Spectrally Efficient Super Nyquist-WDM Signal," Lightwave Technology, Journal of , vol.31, no.20, pp.3237,3242, Oct.15, 2013 [5] Tomkos, I.; Palkopoulou, E.; Angelou, M., "A survey of recent developments on flexible/elastic optical networking," Transparent Optical Networks (ICTON), 2012 14th International Conference on , vol., no., pp.1,6, 2-5 July 2012. [6] F. Cugini et al., "Demonstration of flexible optical network based on path computation element," J. Lightwave Technol., 2012. [7] Rosa, A.; Cavdar, C.; Carvalho, S.; Costa, J.; Wosinska, L., "Spectrum allocation policy modeling for elastic optical networks," High Capacity Optical Networks and Enabling Technologies (HONET), 2012 9th International Conference on , vol., no., pp.242,246, 12-14 Dec. 2012. [8] G. Shen and Q. Yang, “From coarse grid to mini-grid to gridless: How much can gridless help contentionless?” inProc. Optical Fiber Communication Conference, 2011 [9] Shuqiang Zhang; Mukherjee, B., "Energy-efficient dynamic provisioning for spectrum elastic optical networks," Communications (ICC), 2012 IEEE International Conference on , vol., no., pp.3031,3035, 10-15 June 2012. [10] Angelou, M.; Christodoulopoulos, K.; Klonidis, D.; Klekamp, A.; Buchali, F.; Varvarigos, E.; Tomkos, I., "Spectrum, cost and energy efficiency in fixed-grid and flex-grid networks," Optical Fiber Communication Conference and Exposition (OFC/NFOEC), 2012 and the National Fiber Optic Engineers Conference , vol., no., pp.1,3, 4-8 March 2012.

978-1-4799-7162-6/14/$31.00 ©2014 IEEE

[11] Mingchia Wu; Way, W.I., "Fiber nonlinearity limitations in ultra-dense WDM systems," Lightwave Technology, Journal of , vol.22, no.6, pp.1483,1498, June 2004. [12] J. Cai, et al., “20 Tbit/s transmission over 6,860 km with sub Nyquist channel spacing”, J. Lightwave Technology, accepted for publication, available in pre-print on IEEE Xplore. [13] A. Khilo et aI., ""Photonic ADC: overcoming the bottleneck of electronic jitter,""Opt. Express 4, 4454 (2012). [14] Wang, J.; Xie, C.; Pan, Z., "Optimization of DSP to Generate Spectrally Efficient 16QAM Nyquist-WDM Signals," Photonics Technology Letters, IEEE , vol.25, no.8, pp.772,775, April15, 2013.

[15] A. D. Poularikas, Z. M. Ramadan, “Adaptive Filtering Primer with MATLAB,” Taylor & Francis Group, 2006 [16] Paulo S. Ramirez, “Adaptive Filtering: Algorithms and Practical Implementation”, Springer, Ed 2. 2002. [17] R. Shafik, Md. Rahman, R. Islam, "On the Extended Relationships Among EVM, BER and SNR as Performance Metrics", 4th International Conference on Electrical and Computer Engineering, ICECE 2006, 1921 December 2006, Dhaka, Bangladesh.

978-1-4799-7162-6/14/$31.00 ©2014 IEEE