The major drawback of frequency modulation (FM)-based directly modulated laser (DML) is its non-uniform FM response at low frequency range which gives rise to a severe pattern-dependent performance degradation. In this paper, we investigate the use of line coding to deplete the low-frequency spectral contents of the signal and thus to alleviate the degradation. We examine various line codes (8B/10B, 5B/6B, 7B/8B, 9B/10B, and 64B/66B) with continuous-phase frequency-shift keying/ amplitude-shift keying (CPFSK/ASK) signals generated using a DML and a delay interferometer. Experimental demonstrations are performed with a long pseudorandom bit sequence length of 220-1 and the bandwidth expansion by each code is taken into consideration. The results show that among the five codes we tested, 9B/10B code outperforms the other codes in terms of receiver sensitivity an dispersion tolerance. We demonstrate successful transmission of 10-Gb/s CPFSK-ASK signals over 65-km standard single-mode fiber with a bandwidth expansion of only 11.1%.
© 2010 OSA
Directly modulated laser (DML)-based optical transmitters offer many advantages over external modulator-based counterparts, including small footprint, cost-effectiveness, high output power, and low driving voltage . However, directly intensity-modulated lasers have two major drawbacks: small extinction ratio (ER) and large frequency chirp. The current modulation of the DML is always accompanied by frequency modulation (FM) and consequently induces large frequency chirp. When the DML is biased with a high bias current to minimize the transient chirp and to increase the modulation bandwidth, the ER should be reduced, otherwise the large current modulation required to achieve high ER would lead to a large frequency chirp . Nevertheless, even with a limited ER, the adverse effects of the frequency chirp severely limit the transmission distance over standard single-mode fiber (SSMF) at the 1550-nm window. For example, at 10 Gb/s, the transmission distance that can be achieved with DMLs is typically limited to <10 km without dispersion compensation [1,3]. Several schemes have been proposed to mitigate the chirp problem, such as the use of negative-dispersion fibers  and dispersion compensation by electrical or optical means [4,5]. They, however, either require costly electrical/optical devices or installing of new type of optical fibers.
Frequency-shift keying (FSK) modulation of DML takes advantage of the frequency chirp and can improve the dispersion-limited transmission distance of DML . It requires an FM to intensity-modulation (IM) converter either at the transmitter or the receiver, which can be a narrow optical filter or a delay interferometer (DI). However, the main problem in this approach is that DMLs exhibit highly non-uniform FM response at low frequencies, which in turn causes severe pattern dependency . The laser FM response is mainly governed by two effects, the thermal effect which is dominant at low frequencies (<~10 MHz) and the carrier density-induced FM that exhibits a flat response from DC to several GHz. The overall FM response of the laser diode then becomes non-uniform at low frequencies with a ‘frequency dip’ at ~1 MHz . Different approaches have been reported to combat this problem such as pre-equalization of the laser driving signal  and the use of multi-electrode laser diodes . Nevertheless, these methods require either a complicated implementation or replacement of the conventional lasers with those having modified structures in order to produce a highly flat FM response. Recently, Baroni et al. have proposed the use of 8B/10B line coding to deplete the low-frequency spectral content of the signal to utilize the uniform FM response region only . However, 8B/10B coding has a large overhead of 25%, which requires 25% larger bandwidth for the transmission devices (e.g., drivers, DML, receiver) and makes the system 56% ( = 1.252-1) more vulnerable to fiber dispersion.
In this paper, we experimentally investigate the use of various line codes, including 8B/10B, 5B/6B, 7B/8B, 9B/10B, and 64B/66B, to find out a code most suitable for 10-Gb/s directly modulated systems. Unlike in , we take into account the bandwidth expansion of each code to accurately assess the performance. Among the five examined codes, we demonstrate that the best performance is achieved using 9B/10B (the overhead is 11.1%). We transmit 10-Gb/s signals over 65-km SSMF with a dispersion penalty of 2.5 dB.
2. Line coding
The 8B/10B, 5B/6B, 7B/8B, and 9B/10B codes are based on mapping blocks of data into predefined codewords that maintain the run length and DC-balance constraints. For example, 8B/10B code translates each source byte into a predefined 10-bit codeword with a maximum run length of 5 and a disparity of either 0 or ± 2 in each 10-bit codeword. The run length is defined as the number of identical contiguous ‘1’s or ‘0’s which appear in the signal stream and the disparity is the difference between the number of 1’s and 0’s in the codeword. Since the disparity is bounded, the 8B/10B code is DC-balanced, i.e., provides a low DC content . Similarly, 9B/10B, 5B/6B and 7B/8B codes map 9-bit, 5-bit and 7-bit blocks into predefined 10-bit, 6-bit and 8-bit codewords, respectively [11–13]. Unlike the codes described above, in 64B/66B code each 64-bit block is scrambled using the polynomial x58 + x39 + 1, and then 2-bit preamble is added . The 64B/66B code has an overhead of only 3.125% but with poorer transition density and longer run length compared to the other codes. Table 1 summarizes the basic characteristics of the line codes used in this paper. The line rate is defined as the total transmission rate including the overhead.
3. Experiment and results
The experimental setup is depicted in Fig. 1 . A 220-1 pseudorandom bit sequence (PRBS) was used as an input to the line encoder. The generated signals were fed to a commercial DML operating at 1549.2 nm. The threshold current was 10 mA at 20°C, and the DML was biased at 60 mA obtaining an output power of 5.3 dBm and modulation bandwidth of 24 GHz. The FSK-modulated DML output [Fig. 1(a)] was sent to a DI to be converted into continuous-phase FSK/amplitude-shift keying (CPFSK/ASK) signals [Fig. 1(b)]. The free spectral range (FSR) of the DI was 10.7 GHz and the DI phase was adjusted to locate its spectral null at the mark wavelength. The CPFSK-ASK signals with a 3-dB bandwidth of 7.7 GHz and signal power of 0 dBm were launched into an SSMF link and detected with a PIN detector followed by a clock and decision recovery. There was no optical amplifier to boost the signal power.
In our experiment, the DI is located at the transmitter to simplify the wavelength alignment between the laser and DI. By placing the DI together with the DML at the transmitter side, simple wavelength monitoring and locking circuitry as shown, for example, in Fig. 8 can be utilized and thus simplifying the system.
We first optimized the DML driving voltage. Figure 2 shows the measured receiver sensitivity at a bit-error rate (BER) of 10−9 as a function of peak-to-peak driving voltage. In this measurement, we utilized 27-1 PRBS at 9.95 Gb/s. The optimum driving voltage was found to be 1.28 Vpp. Since the FM efficiency of the DML is 0.184 GHz/mA, the frequency deviation between the marks and spaces of the FSK signals at the DML output is 4.7 GHz, which is approximately half the data rate. This optimum driving voltage remains unchanged for different PRBS lengths and line codes running at their corresponding line rates.
When we utilize different line codes, the line rate varies with the overhead. For fair comparison between the line codes having different line rates, we should optimize the bandwidth of electrical and optical devices for each line coding. Due to the limitation of experimental setup, however, we utilized the same devices for all the line codes. Thus, we measured the receiver sensitivity penalty incurred by the limitation of the experimental setup. To this end, we measured the sensitivity penalty for 27-1 PRBS as we increase the data rate from 9.95 to 12.44 Gb/s. Figure 3 shows the results. The penalties are plotted with reference to the receiver sensitivity for the 9.95-Gb/s signals. The results show that the penalty increases as the data rate increases. For example, a penalty of 3.1 dB is incurred at a data rate of 12.4 Gb/s. This penalty should be attributed to (1) the data-rate increase, (2) the delay-to-line rate mismatch of the DI, and (3) implementation imperfection (e.g., caused by using a fixed bandwidth amplifier). To see the contribution of those effects, we analyze the penalty caused by the data-rate increase and the delay-to-line rate mismatch using a theoretical calculation and computer simulation, respectively. As the data rate increases, the received signal power should increase with the data rate to maintain the signal-to-noise ratio. Thus, the penalty by the data-rate increase (blue curve in Fig. 3) is proportional to the data rate. The relative delay between the two arms of the DI we used in the experiment is 93.5 ps, which best matches to 10.7-Gb/s signals. When the line rate deviates from this number, a sensitivity penalty is incurred. The computer simulation (VPI TransmissionMaker) shows that a penalty of 0.4 dB is expected when 12.44-Gb/s signals are used with the 93.5-ps DI. Thus, Fig. 3 shows that the penalty of 3.1 dB for 12.44-Gb/s signal can be broken down into 1.0 dB by the data-rate increase, 0.4 by the delay-to-line rate mismatch, and 1.7 dB by implementation imperfection.
We next measure the back-to-back BER curve for each line code operating at its corresponding line rate and depict in Fig. 4 . Also plotted are the BER curves for PRBS with lengths of 27-1, 215-1, and 220-1 for comparison. The measured receiver sensitivities for the line-coded signals range from −19.3 to −16.9 dBm except for 64B/66B-coded signals where an error floor is observed at around >10−8. This should be attributed to the little depletion of low-frequency signal content by 64B/66B line coding.
Figure 5 shows the measured RF spectra of the coded signals. Unlike other line codes, 64B/66B induces little change in the power level of low-frequency spectral content of the signals compared to 220-1 PRBS, as shown in Fig. 5(g). Nevertheless, 64B/66B coding exhibits in Fig. 4 more than an order-of-magnitude improvement in BER over 220-1 PRBS. Considering that the line rate of 64B/66B is only 3.1% higher than that of PRBS, the use of 64B/66B coding has a slight effectiveness in depleting the low-frequency spectral content of the signals. Figure 4 shows that 9B/10B exhibits the best receiver sensitivity among the five line codes. This is because the 9B/10B coding greatly depletes the low-frequency spectral content of the signal, yet having a small overhead of 11.1%. The receiver sensitivity is measured to be −19.3 dBm, which is only ~0.7 dB worse than 27-1 PRBS. As explained earlier, this penalty should be attributed to line-rate increase, delay-to-line rate mismatch, and other implementation imperfection.
It is interesting to note that in Fig. 4 the sensitivity penalties of the line-coded signals with respect to the sensitivity of 27-1 PRBS are smaller than the results of Fig. 3. For example, the sensitivity penalty of 12.44-Gb/s 8B/10B-coded signal is measured to be 2.1 dB in Fig. 4 but the sensitivity penalty of 12.44-Gb/s 27-1 PRBS is 3.1 dB in Fig. 3. This implies that even 27-1 PRBS is a little affected by the FM dip of the DML and the line coding is very effective in utilizing the flat FM response region only. This can be also explained by the sensitivity improvement of the line-coded signals over 27-1 PRBS when the line coding is applied at 9.953 Gb/s (i.e., without taking the overhead into account). For example, 9.953-Gb/s 8B/10B signals give us a sensitivity improvement of 0.2 dB with respect to 9.953-Gb/s 27-1 PRBS.
Figure 6 shows the eye diagrams measured at the DI output. The eye diagrams for 27-1 PRBS, 5B/6B, 7B/8B, 8B/10B and 9B/10B show clear eye opening, but 220-1 PRBS and 64B/66B show degraded eye diagrams with scattered dots in the middle of the eyes. The ER is measured to be 9~10 dB for all the cases.
The dispersion tolerance of the line-coded signals is measured with SSMF. Figure 7(a) shows the receiver sensitivity versus the transmission distance. The results for 64B/66B are not included because we are not able to achieve a BER of <10−9 for all distances. The results show a couple of decibels improvement in receiver sensitivity after 15-km transmission. This is because of duobinary-like phase characteristics of the generated CPFSK-ASK signals. Since the frequency deviation between the mark and space at the DML output is equal to half the line rate, the phase slide of the signals at the space becomes π, just like duobinary signals . The difference between duobinary and CPFSK-ASK signals is that duobinary has an abrupt π phase jump at the center of the space whereas CPFSK-ASK has a smooth phase shift over the bit duration of the space . Only with 9B/10B and 7B/8B coding, we successfully transmit 10-Gb/s signals over 65-km SSMF. The dispersion penalties are measured to be 2.5 and 3.5 dB for 9B/10B and 7B/8B, respectively. Figure 7(b) shows the BER curves measured after 65-km transmission. There is no evidence of error floor for 5B/6B and 8B/10B. Due to the large bandwidth expansion by the overhead, however, there are large dispersion-induced penalties and we are not able to achieve a BER of 10−9 at the maximum received signal power of −14 dBm. It is interesting to note that in the case of the line coded signals we have a synchronization loss at the error detector when the BER is poorer than 10−6. We believe this is because, unlike PRBS, the line-coded signals do not have a distinct spike-like profile for the autocorrelation function, which can be used to facilitate automatic synchronization at the error detector. For 64B/66B coding, we have a synchronization loss even at high received signal power of −14 dBm. Thus, we expect that the BER is poorer than 10−6 after 65-km transmission.
It should be noted that the laser wavelength should be precisely aligned to the delay interferometer. This could be achieved by applying a low-frequency electrical tone signal to the DML and then utilizing the Fabry-Perot (FP) etalon-based monitoring module . By monitoring the wavelength drift between the DML and DI, we can control the laser wavelength to be locked to the DI. It can be readily applied to wavelength-division-multiplexed systems without additional DI or monitoring modules. Figure 8 shows the schematic diagram of such a system. To discriminate between the channels, each DML is applied with a tone signal having a unique frequency. Since the line coding depletes the low-frequency spectral contents of the signal, the tone signal would not interfere with the data signal, provided that the tone frequency is lower than tens of MHz. For monitoring of multi-channels, the FSR of the FP etalon should be equal to the channel spacing of the laser diode, which is also same as the FSR of the DI.
We have experimentally investigated the application of various line codes (8B/10B, 5B/6B, 7B/8B, 9B/10B, and 64B/66B) to enhance the performance of 10-Gb/s CPFSK-ASK signals generated using DML and delay interferometer. We take into consideration the bandwidth expansion by the overhead of each line code for fair comparison. A long PRBS length of 220-1 is used for all the line codes to accurately examine the pattern dependency. Among the five codes we tested, 9B/10B exhibits the best performance due to its efficiency in depleting the low-frequency spectral contents of the signals, yet it expands the bandwidth by only 11.1%. Thus, we achieve 65-km transmission of directly modulated 10-Gb/s signals without dispersion compensation. The dispersion-induced penalty is measured to be 2.5 dB. Therefore, we believe that the use of 9B/10B for DMLs can be beneficial for implementing cost-sensitive metro/access transmission systems.
The authors would like to thank F. Effenberger (Futurewei Technologies) and A. Carena (Politecnico di Torino) for their help in implementing the 8B/10B, 9B/10B and 64B/66B codes. This work was supported by Singapore Ministry of Education Academic Research Fund Tier 1.
References and links
1. S. C. Lin, S. L. Lee, and C. L. Yang, “Spectral filtering of multiple directly modulated channels for WDM access networks by using an FP etalon,” J. Opt. Netw. 8(3), 306–316 (2009). [CrossRef]
2. H. S. Chung, Y. G. Jang, and Y. Chung, “Directly modulated 10-Gb/s signal transmission over 320 km of negative dispersion fiber for regional metro network,” IEEE Photon. Technol. Lett. 15(9), 1306–1308 (2003). [CrossRef]
3. I. Tomkos, B. Hallock, I. Roudas, R. Hesse, A. Boskovic, J. Nakano, and R. Vodhanel, “10-Gb/s transmission of 1.55-μm directly modulated signal over 100 km of negative dispersion fiber,” IEEE Photon. Technol. Lett. 13(7), 735–737 (2001). [CrossRef]
4. M. D. Feuer, S. L. Sun-Yuan Huang, O. Woodward, Coskun, and M. Boroditsky, “Electronic dispersion compensation for a 10-gb/s link using a directly modulated laser,” IEEE Photon. Technol. Lett. 15(12), 1788–1790 (2003). [CrossRef]
5. M. Du, L. G-. Nielsen, C. G. Jorgensen, and D. DiGiovanni, “Dispersion Compensated 10-Gb/s Directly Modulated Lasers for 6x80km, DWDM Metro Network Applications,” in Proceedings of ECOC2006, pp. 1–2.
6. W. Idler, A. Klekamp, R. Dischler, and B. Wedding, “Advantages of frequency shift keying in 10 Gb/s systems,” in Proceedings of IEEE/LEOS Workshop on Advanced Modulation Formats, 2004, pp. 51–52.
7. P. Baroni, V. Miot, A. Carena, and P. Poggiolini, “8B10B line coding to mitigate the non-uniform FM laser response of direct modulated CPFSK transmitter,” Opt. Express 16(10), 7279–7284 (2008). [CrossRef] [PubMed]
8. E. Forestieri and G. Prati, “Analysis of delay-and-multiply optical FSK receivers with line-coding and non-flat laser FM response,” IEEE J. Sel. Areas Comm. 13(3), 543–556 (1995). [CrossRef]
9. S. Saito, Y. Yamamoto, and T. Kimura, “S/N and error rate evaluation for an optical FSK-heterodyne detection system using semiconductor lasers,” IEEE J. Quantum Electron. 19(2), 180–193 (1983). [CrossRef]
10. S. Ogita, Y. Kotaki, M. Matsuda, Y. Kuwahara, H. Onaka, H. Miyata, and H. Ishikawa, “FM response of narrow-linewidth, multielectrode λ/4 shift DFB laser,” IEEE Photon. Technol. Lett. 2(3), 165–166 (1990). [CrossRef]
11. X. Widmer and P. A. Franaszek, “A DC-balanced, partitioned-block, 8B/10B transmission code,” IBM J. Res. Develop. 27(5), 440–451 (1983). [CrossRef]
12. F. Effenberger, F. Yu, Z. Wang, and J. Gao, “A 9b10b line code for 2.5Gb/s upstream PONs,” in Proceedings of Optical Fiber Communication, 2009, pp. 1–3.
13. X. Widmer, “DC balanced 7B/8B, 9B/10B, and partitioned DC balanced 12B/14B, 17B/20B, and 16B/18B transmission codes,” U.S. Patent 6614369 B1, Sept. 2, 2003.
14. R. Walker, and R. Dugan, “64B/66B low-overhead coding proposal for serial links,” IEEE 802.3ah (10 GE) Task Force, 2000.
15. H. Kim, S. K. Kim, H. Lee, S. Hwang, and Y. Oh, “A novel way to improve the dispersion-limited transmission distance of electroabsorption modulated lasers,” IEEE Photon. Technol. Lett. 18(8), 947–949 (2006). [CrossRef]
16. K. J. Park, S. K. Shin, and Y. C. Chung, “Simple monitoring technique for WDM networks,” Electron. Lett. 35(5), 415–417 (1999). [CrossRef]