Open Access
Add to BibSonomyAbstract
We propose to employ 8B10B line coding as a simple technique to mitigate the impact of non-uniform laser diode FM response on the system performance of directly-modulated CPFSK. Experimental results show its effectiveness in decreasing signal distortion and penalty.
©2008 Optical Society of America
1. Introduction
Within the TOSCA (Transmission of Optical Signals exploiting Competitive Amplification techniques) project, funded by the Italian Ministry of Education, Research and University, we investigate the application of Semiconductor Optical Amplifier (SOA) in multi-channel Wavelength Division Multiplexed (WDM) systems to allow a new and cost-effective generation of metropolitan area networks. The limiting factor comes from SOA non-linearities, mainly Cross Gain Modulation (XGM) [1], so that the standard Intensity Modulation Direct Detection (IMDD) technique cannot be employed. The use of a Constant Envelope (CE) modulation format is a promising solution that we analyzed in the framework of TOSCA.
Continuous Phase Frequency Shift Keying) (CPFSK) is a well known CE modulation format, extensively studied in the past, both with coherent [2] and incoherent detection [3]. CPFSK, compared to other CE formats, such as Differential Phase Shift Keying (DPSK) or Polarization Shift Keying (PolSK), is economically attractive due to the possibility of implementing a low-cost transmitter based on a directly-modulated laser diode. In recent years the availability of reliable and stable Delay Interferometers (DI) and wideband Balanced Photodetectors (BPD) has revamped the attention on this format.
The major drawbacks of the directly-modulated transmitter scheme, highlighted in earlier studies [4], is its non-uniform Frequency Modulation (FM) response at low frequencies (less than 1 MHz), which can cause severe signal distortion. This problem is due to the transition between the FM thermal response and the FM carrier density response in the range 100 kHz–1 MHz.
In the past, several solutions have been proposed to combat this problem: for example, the use of a preamplifier with a compensation network, that enhances the signal at the laser FM response dip [5], or the use of line-codes, such as Alternate Mark Inversion (AMI) [6]–[7] or Manchester [7], that deplete the low-frequency spectral content of the signal. However, AMI and Manchester coding have drawbacks. Both require a much larger modulation bandwidth with respect to a standard uncoded signal, which would in turn require a laser supporting such increased modulation bandwidth. Moreover, AMI coding generates a three-level (ternary) signal that requires a doubled FM laser modulation deviation. To obtain it, a doubled driving-current modulation would have to be supplied to the laser, resulting in a doubled spurious amplitude modulation, which would cause a non-negligible sensitivity penalty. In addition, the optical spectrum would be substantially broadened, too.
Here we present for the first time a more flexible solution based on the use of an 8B10B line coding. It depletes the low-frequency spectral content of the signal, too. This method has several advantages: the most important is its complete independence from the laser type and the operating conditions (bias point), together with easy implementation. It requires a moderate increase in modulation bandwidth (25%) and does not need a three-level signal.
2. 8B10B coding scheme
In order to shape the spectrum of the signal, we selected the well-known 8B10B line code [8]. This proven scheme is already used in optical communications, e.g., for Gigabit Ethernet and Fibre Channel.
Fig. 1. Spectra of an uncoded PRBS (solid line) and of a 8B10B coded PRBS (dashed line): the 8B10B coded sequence shows reduced levels at low frequencies.
8B10B provides a low DC content, as shown in Fig. 1, thanks to a maximum run length of 5 and a maximum digital sum variation of 6. In addition, the hardware complexity of the encoder and the decoder are low, less than that of a Forward Error Correction (FEC) code or an electronic equalizer. This is obtained with an implementation based on two smaller and simpler sub-codes (a 5B6B and a 3B4B) that can be easily integrated in a 10 Gbit/s transceiver.
The main disadvantage is represented by the 25% data rate overhead, but a positive feature of using a digital coding scheme is the availability of an error detecting signal that can be used for instance as a feedback signal for DI real-time tuning. Unfortunately, 8B10B is not an error correcting code. Research in the field of codes capable of spectrally shaping the signal and simultaneously providing error correction is still ongoing. It is possible that in the future more powerful line-codes will make more efficient use of the overhead by providing error correction as well.
3. Experimental and simulation results
In order to demonstrate the advantages of using the 8B10B coding, we performed a set of sensitivity measurement at 10.7 Gbit/s, according to the experimental set-up visible in Fig. 2. We carried out all our experiments at a physical bit-rate of 10.7 Gbit/s, both with 8B10B coded and uncoded sequences. We define as sensitivity the Optical Signal to Noise Ratio (OSNR) value needed to get a 10-6 bit error rate. Moreover, in order to present a general result, note that our definition of OSNR assumes an Amplified Spontaneous Emission (ASE) noise bandwidth equal to the bit-rate (sometimes called “OSNR per bit”).
At the transmitter we used a commercial laser diode intended for 10 Gbit/s direct intensity modulation. It is not a CPFSK optimized component. Its modulation bandwidth is 8 GHz bandwidth. We bias it at 100 mA, obtaining an average power of +9.5 dBm, and directly modulate it with a signal taken from the pattern generator, whose peak-to-peak range is Vpp. The Vpp voltage was optimized sweeping between 0.8 V and 1.8 V in order to reach the best value of sensitivity for this set-up, which corresponds to a modulation index h of 1/2, with a phase rotation between the two symbols of π/2 (Δf=RB/2). This value corresponds to the so-called Minimum Shift-Keying (MSK) format.
The sensitivity measurement is performed using the noise loading technique, implemented as shown in Fig. 2. A 50/50 coupler mixes the generated signal with ASE noise. Thanks to a pair of variable optical attenuators, we can finely select the OSNR.
We also evaluate resilience to fiber dispersion of CPFSK using G.652 fiber installed through the city of Turin by Fastweb, Italy’s second largest telecom operator. In order to achieve linear propagation and avoid non-linearity, we launched a low power in the link and we added noise after propagation, just before the receiver.
The receiver is composed of a 40 GHz optical filter, a DI followed by a BPD and a linear RF amplifier. The DI can be finely frequency tuned thanks to a thermal controller. Note that CPFSK with modulation index h=1/2 requires the DI to be detuned by RB/4 with respect to the standard DPSK working point, which consists of one DI port set at a transmission maximum and the other at a minimum.
The experimental Bit Error Rate (BER) vs. OSNR results in back-to-back, obtained both using Pseudo Random Bit Sequences (PRBS) with different lengths and a specific 220-1 PRBS coded using 8B10B, are presented in Fig. 3. The measurement was taken using Vpp=1.5 V, the optimum value for Vpp as shown in Fig. 4. We can see that the smaller the PRBS degree, the better the sensitivity is. This is due to the fact that the lower the PRBS degree, the more depleted are the low-frequency components in the signal spectrum. The use of the 8B10B code allows to improve the sensitivity because it effectively reduces the low frequency components: such behavior is independent of the sequence length. We proved its effectiveness up to a 220-1 bits long PRBS. Performances was completely independent of sequence length, confirming that 8B10B can solve the laser FM-response frequency-dip problem. The best sensitivity achieved with 8B10B coding was OSNR=14.35 dB (Vpp of 1.5 V).
The Vpp value corresponds to a peak-to-peak modulation current of 28 mA. From this number and knowing that the modulation index was h=1/2, we could derive the FM efficiency of the laser source: 178 MHz/mA. This value is not very high, because the laser is intended to be used for intensity modulation with reduced chirp. In case of laser sources with larger FM efficiency, a smaller peak-to-peak modulation current would be needed, which would also reduce the spurious laser AM (Amplitude Modulation).
Fig. 3. Experimental results: BER vs. OSNR in back-to-back condition using Vpp=1.5 V. Comparison between PRBSs of different length and a 8B10B 220-1 coded sequence.
Fig. 4. Sensitivity vs. modulation signal amplitude (Vpp). Comparison between measurements, carried out with a 27-1 PRBS sequence and a 8B10B 220-1 coded one, and simulation.
In [3] the authors obtained similar results, but they did not deal with the non-uniformity of FM response: measurements were restricted to the use of a single sequence. The aim of our paper is to demonstrate that 8B10B coding solves the problem regardless of the sequence length.
In Fig. 5 we show results of transmission experiments in the installed G.652 fiber with D=17 ps/nm/km. They were performed with the optimum driving voltage Vpp=1.5 V. We set a low transmission power in order to avoid non-linearity. We wanted to measure only the impact of linear dispersion. Using an 8B10B code the OSNR penalty after 35 km was about 1 dB. After 63 km we measured an OSNR penalty of about 3 dB. Figure 6 shows eye diagrams in back-to-back and after 63 km of propagation in the G.652 fiber.
Fig. 5. Sensitivity vs. link length, using SMF fiber with D=17 ps/nm/km. Comparison between experimental results obtained using the 8B10B 220-1 coding and simulation.
Fig. 6. Eye diagrams measured at BER=10-6 when using the 8B10B coding with optimum driving voltage Vpp=1.5V. Top: L=0 km. Bottom: L=63 km. Time base scale: 20 ps/div. Amplitude scale: 35 mV/div.
We wanted to compare to compare the experimental results on chromatic dispersion resilience with simulations, to make sure that the system behaved as expected. To carry out such comparison, we needed to implement a model of the laser source used as a transmitter and of the receiver. For the receiver, detailed and proven models are available, because they have been developed for the simulation of DPSK systems. Laser models are also widely available, but they are typically physical-level, such as rate-equation-based models. These models are powerful but parameter identification with actual devices is difficult. In our case a much simpler model of the directly-modulated laser was enough to provide a sufficiently precise description of the system behavior. Such simplified model requires only the threshold, the optical power vs. injection current (P-I) slope and the FM efficiency (chirp) of the laser source. While the first two parameters were measured, the latter was estimated from the experiment, as described above. In our model the non-uniformity of laser FM response was not implemented: optical systems simulations take place in too narrow time windows (thousands of bit, i.e. hundreds of nanoseconds) to consider this slow effect. Therefore our model is good for simulation only if the non-uniform FM response problem has been overcome, for example by employing the proposed 8B10B coding scheme. In Fig. 4 and Fig. 5 we also plotted simulation results: there is good agreement with measurements, even after propagation in linear regime along a dispersive fiber, showing that our simplified model was adequate.
7. Conclusions
In this paper we demonstrate for the first time that using the 8B10B coding scheme together with CPFSK directly-modulated optical systems mitigates the penalty due to the laser non-uniform FM response in a cost-effective way. This technique is completely independent of laser type, operating conditions and bit-rate. Moreover, we demonstrate that a simple model considering only the P-I characteristic and a linear chirp is precise enough to accurately simulate the behavior of CPFSK systems when the 8B10B is used to solve the non-uniform FM response problem.
References and links
1. D. T. Schaafsma, E. Miles, and E. M. Bradley, “Comparison of conventional and gain-clamped semiconductor optical amplifiers for wavelength-division-multiplexed transmission systems,” J. Lightwave Technol. 18, 922–925 (2000). [CrossRef]
2. R. Wyatt, D. W. Smith, T. G. Hodgkinson, R. A. Harmon, and W. J. Devlin, “140 Mbit/s optical FSK fibre heterodyne experiment at 1.54 µm,” Electron. Lett. 20, 912–913 (1984). [CrossRef]
3. B. Wedding, R. Jung, C. Haslach, and H. Söhnle, “10.7 Gbit/s FSK transmission with 61 dB power budget,” Proceedings of ECOC2003, Vol. 4, paper Th1.5.5 (21–25 Sep. 2003).
4. K. Emura, T. Ono, A. Noda, and S. Yamazaki, “Requirements for high-frequency LD FM response in an optical CPFSK heterodyne delay demodulation system,” J. Lightwave Technol. 9, 1602–1608 (1991). [CrossRef]
5. K. Iwashita and T. Matsumoto, “Modulation and detection characteristics of optical continuous phase FSK,” J. Lightwave Technol. 5, 452–460 (1987). [CrossRef]
6. S. P. Mazunder, R. Gangopadhay, E. Forestieri, and G. Prati, “Sensitivity penalty for AMI-coded,” IEEE Photonics Technol. Lett. 7, 1207–1209 (1995).
7. E. Forestieri and G. Prati, “Analysis of delay-and-multiply optical FSK receivers with line coding and nonflat laser FM response,” IEEE J. Sel. Areas Commun. 13, 543–556 (1995). [CrossRef]
8. A. X. Widmer and P. A. Franaszek, “A DC-Balanced, partitioned-block, 8B/10B transmission code,” IBM J. Res. Dev. 27, 440–451 (1983). [CrossRef]
