Open Access
Add to BibSonomyAbstract
High data-rate upstream transmission in wavelength division multiplexed passive optical network (WDM-PON) based on reflective semiconductor optical amplifier (RSOA) is limited by severe inter-symbol interference (ISI), owing to low-bandwidth transmitter and fiber dispersion. To overcome the limitations in RSOA-based WDM-PON, we propose a novel receiver based on partial response maximum likelihood (PRML) equalization which combines the use of partial response (PR) signaling with maximum likelihood sequence estimation (MLSE). MLSE has been long considered as the optimal reception technique to overcome various types of impairment in optical transmission such as dispersions. It is demonstrated in this paper that PRML surpasses standard MLSE in ISI and reflection suppression with reduced complexity for RSOA-based WDM-PON system. 150 km unidirectional distance is demonstrated for 10 Gb/s uplink, while bidirectional uplinks up to 50 km and 20 km are achieved for data rate of 10 Gb/s and 20 Gb/s respectively, in WDM-PON using PRML. Furthermore, the impacts of discrete reflections on various equalization techniques are investigated, where PRML also shows superiority over MLSE.
©2011 Optical Society of America
1. Introduction
Bandwidth demand of access network is rapidly growing due to the emerging high definition video/teleconference services. To meet future capacity requirement, wavelength-division-multiplexed passive optical network (WDM-PON) has been considered as an ultimate solution for optical access. However, high expenditure delays its mass commercialization. The employment of reflective semiconductor optical amplifier (RSOA) opens a way to inexpensive colorless optical network unit (ONU) in WDM-PON [1]. To further bring down the cost per customer and per bandwidth, it is vital to develop cost-effective methods to increase the capacity and reach of WDM PON up to and beyond 10 Gb/s and 100 km [2]. However, upstream transmission based on RSOA is limited by intersymbol interference (ISI) resulted from low electrical bandwidth (1~2 GHz) of RSOA and high chirp (4-10) of RSOA coupled with fiber dispersion. To mitigate these problems, electrical equalization has been applied at the receiver in the optical line terminal (OLT). Reported equalizers include feed-forward equalizer (FFE) [3], decision feedback equalizer (DFE) [4] and maximum likelihood sequence estimation (MLSE) [5]. Among all those equalizers, MLSE exhibits the strongest capability in compensating ISI. However, its main problem is the complexity that increases exponentially with the memory length of the channel. Besides ISI, another serious issue with single-feeder RSOA-based WDM-PON is the vulnerability to reflections. Although it is economical to deploy single fiber for both up- and downlink, the in-band crosstalk induced by unexpected reflections can seriously deteriorate the signal quality and limit the maximum reach. For this problem, several techniques have been reported. They are mainly focused on broadening signal spectrum by frequency dithering or phase modulation [6, 7]. Nevertheless, those techniques either work for only low data rate (< 3 Gb/s) or require complex transceivers. Electrical equalizers are also proposed and studied as an effective means to overcome reflection problems [8].
Partial-response maximum likelihood (PRML) is a signal processing technique that augments MLSE with effective equalization based on partial response (PR) signaling [9]. It combines an equalizer based on finite-impulse-response (FIR) filter with MLSE to perform two-step equalization. As a multi-stage equalizer, PRML has reduced complexity and superior performance compared with MLSE-only approach. We previously demonstrated 50 km uplink for 10 Gb/s and 10 km uplink for 20 Gb/s in two-fiber RSOA-based WDM-PON, by PR equalizer with symbol-by-symbol (SBS) detector [10, 11]. It was shown that the upstream channels based on RSOA share similarity with PR channels, and performance can be greatly improved by employing PR equalizer instead of conventional equalizers. In this paper, we propose to implement the upstream receiver with an equalizer based on the principles of PRML for WDM-PON with 1 GHz RSOA. This new scheme replaces the SBS detection in previous works by MLSE. We experimentally demonstrate 150 km upstream transmission in 10 Gb/s unidirectional (two-fiber) WDM-PON by PRML. Moreover, bidirectional (one-fiber) distances up to 50 km and 20 km are enabled by PRML for data rate of 10 Gb/s and 20 Gb/s, respectively. Further, PRML is compared with other standard equalizers to verify its superior performance in RSOA-based uplink. In addition, the reflection tolerance of PRML method in 20 Gb/s system is also experimentally evaluated and proven to outperform other equalizers.
2. PRML equalization
MLSE is widely recognized as the optimal detection technique in transmission systems affected by ISI and intra-channel deterministic non-linearity. In the context of optical communication systems, a collection of scientific studies have shown that MLSE is indeed capable of effectively detecting signals corrupted by all typical fiber propagation impairments, albeit with residual penalties [12, 13]. On the other hand, the well known shortcoming of MLSE is its exponentially increased complexity in respect to the length of the channel memory, i.e., the number of bits (or symbols) affecting any given bit (or symbol) through ISI. Therefore, partial response signaling has been combined with MLSE in magnetic storage applications for the purposes of complexity reduction and performance improvement [14]. In optical communication, PRML has been investigated as electronic dispersion compensation (EDC) for standard single mode fiber (SSMF) [15]. Figures 1(a), (b) illustrate the typical systems employing MLSE and PRML, respectively. Standard MLSE requires good knowledge of the channel, and therefore channel estimation and training sequence are compulsory, as shown in Fig. 1(a). Furthermore, high complexity is the major hurdle to the real-time implementation of MLSE in systems impaired by large memory, e.g. bandwidth-limited long-distance optical link like the extended RSOA-based transmission. Unlike conventional equalizers such as MLSE, trying to eliminate ISI in one stage, PRML performs equalization in two steps, shown by Fig. 1(b). Firstly, the received signal is equalized to a predetermined PR signal which is called target impulse response (TIR). In PR signal, the ISI is intentionally introduced by letting a certain number of adjacent symbols interfere with each other. So the PR channel can be expressed by a universal polynomial:
Where D is 1-bit delay and fn is coefficient. The response of the TIR generation channel is supposed to be as close to that of the transmission channel as possible. PR equalizer allows a predetermined amount of ISI in its output, hence zero forcing can be avoided so that noise enhancement is reduced. Secondly, the MLSE designed for this TIR eliminates the residual ISI contained in the TIR and recover the transmitted binary signal. Since the input to the MLSE is known, channel estimation that is essential in standard MLSE is not required in PRML. Moreover, the PR equalizer shortens the memory length of the received signal to the desired PR response, and hence makes the realization of low-complexity MLSE feasible by reducing the number of states. Before equalization, the signal is at first applied to the receiver filter that can be set to optimize the detected signal-to-noise ratio (SNR) for the TIR of the equalizer. This analog filter also can adjust the ISI of different received signals to similar amount desired for the chosen TIR. A proper selection of TIR can maintain the noise enhancement of the receiver at low level even if the distortion is severe. By matching PR signals with received signals and proper equalizer design, the PRML has better performance-complexity trade-off than MLSE for long-memory channels.It is found that RSOA-based transmission shares similarity with the PR channels. All of their spectra have frequency dips inside the signal bandwidth. The measured channel responses of the RSOA-based uplink with distances from 0 km to 150 km are shown in Fig. 2(a) . Those resonant frequency dips on the channels are resulted from high chirp of RSOA coupled with fiber chromatic dispersion, and can cause severe noise enhancement if conventional equalizers are used. The chirp parameter of the employed RSOA is measured to be 4.4. When distance increases, the dips move closer to the baseband. The first dip is the most important in determining the shape of the received signal spectrum. Figure 2(b) plots three examples of PR channel responses with F(D) of (1) as 1 + D (duobinary), 1 + D + D2 (triobinary) and 1 + D2. It is evident that the frequency dip positions vary in respect to the ISI pattern of the PR signals. The optimal TIR can be found out by searching the frequency dips’ locations on PR signal spectra to match with the channel. The similarity between transmission channel and target PR channel gives PR equalizer significant improvement and reduced complexity over conventional equalizers.
Fig. 2 (a) Frequency responses of the uplink channel and (b) Channel responses of 10 Gb/s PR signals.
3. Receiver design
Figure 3 displays the structure of the proposed PRML equalizer working as a blind equalizer. As shown in Fig. 3, the captured signal samples are firstly passed through a 3rd-order Bessel low pass filter (LPF). This LPF can expand the working range of the receiver by adjusting received signals from different distances to similar conditions so that the same setup of equalizer can be applied on different channels. The PR equalizer is built by a DFE with a symbol-spaced feed-forward filter (FFF), a feedback filter (FBF) and a quantizer that has multiple thresholds according to the selected TIR. The signal is equalized to the TIR by this DFE. After that, the MLSE detects the equalized PR signal by searching the most likely path on the trellis of the TIR using Viterbi algorithm [16]. Given that the proposed equalizer is blind, training sequence and channel estimation are not required here.
The selection of TIR depends on the channel characteristics analysis. For upstream in RSOA-based WDM-PON with two feeder fibers, 10 Gb/s transmission can be extended up to 150 km in our proposed system. Figure 2(a) shows that the 50 km channel has null near 3.4 GHz, which is generally matched with triobinary channel, and 150 km channel has two nulls near 1.9 GHz and 7.5 GHz, which is generally matched with (1 + D2) channel. It is found that the received signal after 75 km can be equalized to either triobinary or (1 + D2) signal. Thanks to the LPF and adaptive equalization, the same equalizer can be applied to a range of distances. Triobinary signal is chosen as the TIR for transmission distance from 0 to 75 km, while (1 + D2) signal is chosen as the TIR for 75 to 150 km transmission. In the 10 Gb/s unidirectional experiments, for distances ≤ 75 km the PR equalizer is implemented by a 3 GHz LPF and a (13, 2) DFE of 13-tap FFF and 2-tap FBF, while for distances ≥ 75 km, it is composed of a 3.5 GHz LPF and a (18, 5) DFE.
For uplinks using bidirectional architecture, another technique called optical filter detuning (OFD) is applied to convert the phase modulation from frequency chirp to useful amplitude modulation by blue-shifting the central wavelength of the optical bandpass filter (OBPF) [11]. To take advantage of OFD, a Fabry-Perot filter (FPF) detuned by −0.1 nm is used before photodiode (PD) for both 10 Gb/s and 20 Gb/s bidirectional transmission. We measured the RF spectrum of the 10 Gb/s signal after 50 km and 20 Gb/s signal after 20 km with and without OFD. Results are shown in Figs. 4(a), (b) and compared with PR signals’ spectra. When single fiber is used, there is considerable amount of RBS. Although the signals have similar amount of ISI with the received signal after unidirectional transmission, the dip disappears for 10 Gb/s signal after 50 km fiber since the SNR degrades dramatically due to reflection noise. It is obvious from Figs. 4(a), (b) that the channel bandwidth is improved by OFD. Then the spectrum of received 10 Gb/s signal generally matches with the triobinary spectrum from 0 to 2 GHz. The equalizer for 10 Gb/s bidirectional uplink is designed as a 2 GHz 3rd-order Bessel LPF, a (17, 3) DFE and an MLSE for triobinary signal. At data rate of 20 Gb/s, the frequency spectrum around 5 GHz is suppressed after 20 km, but after using OFD it is generally matched with duobinary signal from 0 to 7 GHz. The equalizer here is composed of a 7 GHz 3rd-order Bessel LPF, a (7, 2) DFE and an MLSE for duobinary signal.
Duobinary sequence dn, expressed by original binary bits bn as dn = bn + bn-1, has 1-bit memory and three levels. Triobinary sequence tn, represented by tn = bn + bn-1 + bn-2, has memory length of 2 bits and four levels. From these relationships, we can obtain the trellises of duobinary and triobinary signal for Viterbi algorithm used in MLSE, as shown in Figs. 5(a), (b) . Hence 4-state and 2-state MLSEs are required for TIR as triobinary and duobinary signals, respectively. (1 + D2) signal wn represented by wn = bn + bn-2 is actually the interleaved form of two duobinary signal (1 + D). Thus the MLSE for (1 + D2) signal can be realized by two parallel duobinary MLSE with demultiplexer (DEMUX) and multiplexer (MUX), shown by Fig. 6 . Thanks to the known relationship between TIR and the transmitted binary signal, the implementation of MLSE here is much simpler than standard MLSE for unknown channels.
4. Experiments and results
Figure 7 depicts the generic setup for all experiments. The seed light at 1547 nm with 1 dBm power was generated in OLT, passed through a variable attenuator and L2 km fiber. The injected power on RSOA was adjusted to −10 dBm in all cases. The RSOA was biased at 80 mA to maximize its bandwidth for 20 Gb/s bidirectional system and 10 Gb/s unidirectional system, but at 55 mA to minimize the reflection noise for 10 Gb/s bidirectional system. The continuous-wave (cw) signal was modulated with a 3.2-Vpp 211-1 pseudo-random binary sequence (PRBS). Figure 8(a) shows that the RSOA has an amplified spontaneous emission (ASE) spectrum centered at 1547 nm and an optical bandwidth of 55 nm. The frequency responses in Fig. 8(b) confirm that the RSOA has 3 dB electrical bandwidth of 1 GHz in all experiments. The uplink signal was reflected back to the L2 km fiber followed by L1 km fiber and received by a 15 GHz photodiode (PD). L1 = 0 for unidirectional experiments, while for bidirectional experiments L2 = 0. The OBPF was realized by a tunable FPF with 0.19 nm bandwidth and 32 nm free-spectral range (FSR). The detected signal was sampled by a 15 GHz storage oscilloscope at 40 Gsamples/s with a length of 1.6 × 107 bits and processed offline for equalization and bit error rate (BER) calculation. For 10 Gb/s experiments focused on long reach, an amplifier is added, and the optical SNR (OSNR) was monitored by an optical spectrum analyzer (OSA) before PD.
4.1. Unidirectional upstream transmission
For unidirectional systems, the BER curves against transmission distances are measured separately from 0 km to 75 km using triobinary TIR and from 75 km to 150 km using (1 + D2) TIR, shown in Figs. 9 and 10 , respectively. To evaluate the superior performance of PRML method compared with other equalization techniques, we measured the BER of the uplink using the proposed PRML, PR-DFE (where a SBS detector is used after PR equalizer [10,11]), conventional MLSE (4 or 8 states), DFE and FFE. For shorter distances, MLSE with 8 states is not included, since the memory length of the channel is relatively short. For longer distances, FFE is not included because both FFE and DFE have similar BER results that are above 0.1. From the measured results, it is evident that PRML presents significant advantages over FFE and DFE, and outperforms PR-DFE and MLSE for all distances. The equalized eyes at the output of DFE are clear, shown by the insets of Figs. 9 and 10, indicating good performance. Since hard decision is used in MLSE, its output is ideal binary signal. Limited by storage length, there is no error at 0 km for MLSE, PR-DFE and PRML approaches. The results prove that only two types of PRML equalizer can cover the transmission distance from 0 km to 150 km. The achieved BERs for up to 150 km are well below the threshold of forward error correction (FEC) with 7% overhead (2.3 x 10−3) by PRML that surpasses other equalizers.
4.2. Bidirectional upstream transmission
For 10 Gb/s bidirectional uplink, the required OSNR (ROSNR) results for BER of 5 × 10−4 are measured for variable fiber lengths from back-to-back to 50 km and given in Fig. 11 . The BER results for different distances are calculated using the same triobinary PRML equalizer. Eye diagrams of the received signal, equalized triobinary signal and detected binary signal are displayed by the insets of Fig. 11. The equalized four-level eye is clear, indicating good performance of the proposed technique. It is demonstrated that error-free transmission from 0 km to 50 km can be achieved at reasonable OSNR by the same triobinary PRML together with FEC.
For 20 Gb/s bidirectional uplink, the eye diagrams of the received signal after 20 km fiber, the equalized duobinary signal and the recovered binary signal are shown in Fig. 12 . The clear three-level eye at the output of the DFE indicates good match between the TIR and the channel. The measured BER values against transmission distances from 0 to 20 km at the received power of −6 dBm and the BERs vs. the received optical power for 20 km transmission are displayed in Figs. 13(a) and (b) , respectively. Other electrical equalization techniques under comparison are PR-DFE, MLSE (4 states), (7, 2) DFE and (7) FFE. The BER results verify the superior performance of PRML compared with other techniques and same PRML equalizer can be applied for all distances (0~20 km). 20 Gb/s transmission over 20 km fiber can be achieved by PRML with the BER below 5 × 10−4.
4.3. Reflection tolerance
The performance of RSOA-based WDM-PON system implemented in a bidirectional single-fiber configuration is vulnerable to the unwanted external reflections caused by imperfect devices with high reflectivity. To evaluate the reflection tolerance in upstream transmission that employs PRML receiver, we measured the BERs for various reflectivity levels and compared them with the results of other equalizers. There are two kinds of interferers from discrete reflection: the reflected cw light (I) and the reflected upstream signal (II). The second interferer is the dominant noise source because of reamplification at RSOA. In-band crosstalk incurred by both reflections I and II is investigated using the setups illustrated in Figs. 14(a), (b) . Both of the feedback loops in two setups comprise a variable optical attenuator (VOA), polarization controller (PC) and a 5 m SSMF, simulating the reflected cw light and uplink signal, respectively. PC is used to co-polarize the interferer and signal for worst-case analysis. The cw-induced reflection is evaluated by the crosstalk-to-signal ratio (CSR), which is the ratio between the power at point b and power at point a in Fig. 14(a). The signal-induced reflection is evaluated by the reflectivity defined as the ratio between the power at point b and power at point a (when point b is disconnected) with additional adjustment due to the circulator and coupler losses, shown in Fig. 14(b). Figures 15(a), (b) plot the BER curves against CSR for interferer I and against reflectivity for interferer II, respectively, via tuning the VOA in the loop. The same equalizers used in 20 Gb/s bidirectional experiments are applied here. It is evident that PRML outperforms all other schemes. BER below FEC limit of 2.3 × 10−3 can be achieved at maximal CSR of −25 dB and reflectivity of −23.5 dB for cw-induced and signal-induced reflections respectively. Therefore PRML is able to enhance the reflection tolerance and extend the reach of single-feeder WDM-PON.
Fig. 14 Experiment setups to evaluate effects of reflection noises induced by (a) cw light and (b) uplink signal.
5. Conclusions
Impairments caused by ISI and noises can be alleviated by electrical equalization utilizing different degrees of complexity with various levels of performance. It is shown in this paper that the proposed PRML equalization technique enhances the performance of standard MLSE with reduced complexity and is an adequate solution for ISI and noise mitigation in RSOA-based WDM-PON. Error-free 10 Gb/s unidirectional upstream transmission over 150 km using PRML and sufficient FEC is demonstrated in WDM-PON based on 1 GHz RSOA. Furthermore, 50 km and 20 km bidirectional single-fiber WDM-PONs are enabled for 10 Gb/s and 20 Gb/s by PRML with 1 GHz RSOA. As the combination of PR equalizer and MLSE, PRML achieves better BER results than MLSE and other compared equalizers in all experiments. The performance of PRML at the presence of both cw- and signal-induced reflections is also evaluated for 20 Gb/s uplink and shown to surpass other equalizers. The blind equalization setup of the proposed PRML allows for simple hardware implementation and fast signal processing. The PRML equalization technique proposed in this paper potentially offers a cost–efficient solution to the implementation of high-capacity long-reach WDM-PON.
References and links
1. S. J. Park, G. Y. Kim, and T. S. Park, “WDM-PON system based on the laser light injected reflective semiconductor optical amplifier,” Opt. Fiber Technol. 12(2), 162–169 (2006). [CrossRef]
2. Y. C. Chung, “Recent Advancement in WDM PON Technology,” Proc. ECOC, paper Th.11.C.4, Geneva, Switzerland, (2011).
3. A. Agata and Y. Horiuchi, “Data rate enhancement of RSOA-based WDM PON systems using feed-forward equalizer and forward error correction,” Proc. ECOC, paper P.6.11, Brussels, Belgium, (2008).
4. K. Y. Cho, Y. Takushima, and Y. C. Chung, “10-Gb/s operation of RSOA for WDM PON,” IEEE Photon. Technol. Lett. 20(18), 1533–1535 (2008). [CrossRef]
5. I. Cano, M. Omella, J. Prat, and P. Poggiolini, “Colorless 10Gb/s extended reach WDM PON with low BW RSOA using MLSE,” Proc. OFC, paper OWG2, San Diego, CA (2010).
6. P. J. Urban, A. M. J. Koonen, G. D. Khoe, and H. de Waardt, “Interferometric crosstalk reduction in an RSOA- Based WDM Passive Optical Network,” IEEE Photon. Technol. Lett. 27(22), 4943–4953 (2009).
7. C. W. Chow, G. Talli, and P. D. Townsend, “Rayleigh Noise Reduction in 10-Gb/s DWDM-PONs by Wavelength Detuning and Phase-Modulation-Induced Spectral Broadening,” IEEE Photon. Technol. Lett. 19(6), 423–425 (2007). [CrossRef]
8. J. Prat, “Rayleigh back-scattering reduction by means of Quantized Feedback Equalization in WDM-PONs,” Proc. ECOC, paper Th.10.B.3, Torino, Italy, (2010).
9. R. D. Cideciyan, F. Dolivo, R. Hermann, W. Hirt, and W. Schott, “A PRML System for Digital Magnetic Recording,” IEEE J. Sel. Areas Comm. 10(1), 38–56 (1992). [CrossRef]
10. Q. Guo, A. V. Tran, and C. J. Chae, “Extended-reach 10 Gb/s RSOA-based WDM-PON using partial response equalization”, Proc. Photonics Society Annual Meeting, paper WA2, Denver, CO, (2010).
11. Q. Guo, A. V. Tran, and C. J. Chae, “20 Gb/s WDM-PON System with 1 GHz RSOA using Partial Response Equalization and Optical Filter Detuning”, Proc. OFC, paper NTuB5, Los Angeles, CA, (2011).
12. P. Poggiolini, “MLSE receivers: Application scenarios, fundamental limits and experimental validations,” Proc. ECOC, paper Tu.1.D.1, Brussels, Belgium, (2008).
13. A. Faerbert, “Application of Digital Equalization in Optical Transmission Systems”, Proc. OFC, paper OTuE5, Anaheim, CA, (2006).
14. K. Han and R. R. Spencer, “Performance and implementation of adaptive partial response maximum likelihood detection,” IEEE Trans. Magn. 34(5), 3806–3815 (1998). [CrossRef]
15. S. Elahmadi, M. Bussman, J. Edwards, D. Baranauskas, D. Zelenin, K. Tran, C. Gill, L. Linder, D. Ng, H. Tan, M. Srintah, and D. Rajan, “A monolithic one-sample/bit partial-response maximum likelihood SiGe receiver for electronic dispersion compensation of 10.7Gb/s fiber links,” Proc. OFC, paper JWA34, San Diego, CA, (2009).
16. G. D. Forney, “Maximum likelihood sequence estimation of digital sequences in the presence of intersymbol interference,” IEEE Trans. Inf. Theory 18(3), 363–378 (1972). [CrossRef]
