Open Access
Add to BibSonomy
Abstract
Directly modulated lasers (DMLs) and electro-absorption modulated lasers (EMLs) are key transmitter options in future short-haul networks. However, both of them suffer from frequency chirp, which incurs nonlinear distortions, especially to high order modulation signals. In this paper, we investigate their application in PAM4-based digital mobile fronthaul and propose a scheme to remarkably improve the fidelity of radio signal. We first give a detailed study of the BER distribution of DML/EML based PAM4 signals and find that the BER of the second bit is much higher than that of the first bit in both systems. Accordingly, we propose to adopt sample bit interleaving to reduce the radio signal distortions caused by sample bit errors. Experimental results of 56Gbps I/Q data transmission reveal that, in a DML-based transmission system, the proposed scheme respectively leads to up to 8dB and 13dB EVM reduction to accommodate 33 × 100MHz 1024QAM OFDM signals and 64QAM OFDM signals in 10km and 20km cases. As well as in an EML-based transmission system, 14dB EVM reduction is achieved in 10km to finally accommodate 33 × 100MHz 256QAM OFDM signal.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Cloud radio access network (C-RAN) has been considered as both cost-effective and energy-saving architecture for future mobile communication [1]. In C-RAN, the baseband processing units are consolidated in central office to offer centralized control (BBU pool), and remote radio units are left in cell sites (RRUs) to provide access points (APs) for terminals. The data transport between BBU pool and RRUs is via optical fiber link, which is referred to as optical mobile fronthaul (MFH). For fronthaul system, both analog and digital radio over fiber (RoF) schemes have been actively studied [2–5]. Analog RoF directly transmits continuous waveform; whereas the digital RoF converts radio signals into sample bits before transmission. Compared with analog scheme, the digital RoF is more robust to various impairments in fiber link and, is able to support high-order QAM signals with sufficient distortion margin for wireless channel [4, 5].
However, to meet the continuously growing mobile traffic demand, capacity upgrade of digital MFH is highly desirable. In the meantime, the enabling transmission technologies are expected to be low-cost, otherwise the increased Capital Expenditure (CAPEX) will deviate the initial purpose of C-RAN architecture [6, 7]. As for the modulation format, a basic option is to use high order modulation format to replace the NRZ-OOK [8, 9]. 4-level pulse amplitude modulation (PAM-4) is a relatively low-cost candidate, due to its simpler digital signal processing (DSP) and relaxed linearity requirement compared with other popular formats such as discrete multi-tone modulation (DMT) and carrier-less amplitude/phase modulation (CAP) [10]. On the other hand, it is highly desired to use low-cost devices, among which, directly modulated lasers (DMLs) and electro-absorption modulated lasers (EMLs) are most commonly used because of their compact footprint to save energy and cost. However, both of them suffer from frequency chirp, which will incur severe nonlinear distortions, especially to high order modulation signals [11–13]. Although chirp effect is not an issue in O-band with ultra-low dispersion, it is beneficial to investigate their application in C-band considering DWDM deployment or full-duplex bidirectional transmission using 1.3/1.55-μm windows [3, 14]. In C-band, such distortion is difficult to be mitigated by cost-effective linear equalizers such as feed forward equalizer (FFE) and decision feedback equalizer (DFE). Finally, the increased BER will cause severe distortion on radio signals in fronthaul system [11].
In this paper, we study the implementation of DMLs and EMLs in PAM4-based digital mobile fronthaul and propose a scheme to remarkably enhance the performance of radio signals. We first give a detailed study of the BER distribution of PAM4 signals. In DML-based system, the adiabatic chirp will produce eye skewed PAM4 signal, whereas in EML-based system, the intensity-dependent transient chirp will cause uneven level distribution in PAM4 signal. Attributed by these features, the BER of the second bit (2ndb) is much higher than that of the first bit (1stb) in both systems. On the other hand, the digital fronthaul system has an important feature that, high order sample bits have higher weight than low order ones; thus, errors of high order sample bits will induce much severer distortions on radio signal [15]. Accordingly, we propose to adopt sample bits interleaving to reduce the radio signal distortions, where high order sample bits are mapped to more accurate 1stb of PAM4 symbols for transmission, and low order sample bits are transmitted by less accurate 2ndb. Experimental results of 56Gbps I/Q data transmission reveal that, in DML-based system, the proposed scheme respectively leads to up to 8dB and 13dB EVM reduction to accommodate 1024QAM-OFDM signals and 64QAM-OFDM signals in 10km and 20km cases. As well as in EML-based transmission system, 14dB EVM reduction is achieved in 10km to accommodate 256QAM-OFDM signal.
Note that this work is an extension of [16]. In addition to more detailed analysis and results on DML compared with [16], we also study EML in this manuscript. For the first time, we study the influence of EML’s chirp effect on bit error distribution in a high baud rate system, which was not considered in low-baud-rate systems [15]. The remainder of this work is organized as follows. In Section 2, we give a detailed study on the non-uniform bit error distribution incurred by the frequency chirp in DML/EML based systems. Section 3 presents the principles of sample bits interleaving. In Section 4, the experimental setup is introduced. In Section 5, we show the experimental results and the performance improvement; whereas, this work is concluded in Section 6.
2. Analysis of bit error distribution in DML/EML-based PAM4 system
In this section, qualitative analyses on bit error distribution of DML/EML-based PAM4 signals are given. We will show that, the nonlinear distortion of DML-based transmission mainly comes from adiabatic chirp, causing eye skewing effect in the receiver side. On the other hand, the distortion of EML-based transmission originates from the non-uniform transient chirp, which results in non-uniform eye height in the receiver side. Our analysis indicates that, in spite of different origins of nonlinear distortions in the two systems, the PAM4 signals in receiver has a common feature that, BER of 2ndb is quite larger than 1stb.
2.1 Analysis of DML-based PAM4 transmission
In DML-based transmissions, the electrical signal is directly applied to the gain section of laser cavity, so the electrical signal simultaneously modulates optical intensity and frequency, causing adiabatic chirp expressed as
where P(t) is laser output power, C is a positive constant related to the laser parameters [11]. Thus, in DML-based C-band transmissions, higher level of PAM4 signal has larger optical frequency and arrives at the receiver earlier due to the chromatic dispersion (CD), which is expected to result in an eye skewing effect as shown in Fig. 1.Figures 2(a) and 2(b) compare the eye diagrams of 28G Baud PAM4 signals before and after 20km standard single mode fiber (SSMF) transmission with an accumulated dispersion of 340ps/nm. The skewing effect shown in 20km case confirms the above analysis. Such distortion is expected to be more obvious in a longer distance with a larger CD. This eye skewing is in fact a nonlinear effect and cannot be mitigated with commonly used linear DFE/FFE. To analyze the BER distribution feature, the histogram of received amplitude distribution of the 4 symbols (noted as S0, S1, S2 and S3) is plotted in Fig. 2(c). The overlap of adjacent symbols causes symbol error (SE), and we define them as Type 1 SE, Type 2 SE, and Type 3 SE for distinction, as marked in this figure. We calculate the symbol error ratio (SER) of the three types of SE versus decision point with a resolution of unit interval (UI)/32, as plotted in Fig. 2(d). It is observed that, optimal decision time for different eyes are different because of the eye skewing effect. For example, at the decision time of T = 0, Type 1 and Type 3 SER is much worse than that of Type 2.
Fig. 2 Analysis of received PAM4 signal in DML-based transmission. Eye diagrams of 28G Baud PAM4: (a) 0km and (b) 20km. (c) Amplitude distribution of received symbols after 20km transmission. (d) Different types of symbol error ratio (SER) versus decision point. (e) Total BER as well as 1stb and 2ndb versus decision point.
According to the Gray mapping rule, BER of 1stb equals Type 2 SER, and BER of 2ndb is the sum of Type 1 SER and Type 3 SER. The total PAM4 BER as well as 1stb and 2ndb BER are plotted in Fig. 2(e). As per the figure, at the decision time of T = 0, the 1stb BER is 1 × 10−3 and the 2ndb BER is 8.5 × 10−3. That is to say, the total BER is mainly limited by the high 2ndb BER; whereas, 1stb BER is several times lower than BER of 2ndb.
2.2 Analysis of EML-based PAM4 transmission
In EML-based transmissions, intensity modulation is always associated with phase modulation due to the variation of complex refractive index, resulting in drift of optical frequency [17]. However, the drift occurs when intensity changes, which means that, only transient chirp exists in EML. After fiber transmission, such chirp will change the frequencies of fading dips caused by CD, which can be described in the channel response by small-signal approximation
where α is the chirp factor, L, β2 and f are the fiber length, dispersion parameter and frequency of the signal, respectively [18]. The cosine term determines the frequency of the first fading dip, which should satisfy. Accordingly, at a fixed fiber length with the same CD distortion, the frequency of the first dip becomes smaller with a positive α, while becomes larger with a negative α.One important feature of EML is that, α is related to the applied electrical voltage of electro-absorption modulator (EAM) [17]. To clearly illustrate the relationship, we measure the channel responses of 40km fiber under different applied voltages (bias). 40km is selected here, because the frequencies of fading dips are smaller than the bandwidth of optical devices and the dips can be captured. Three normalized channel responses of C-band SSMF transmission at bias of −0.5V, −1.2V and −1.9V are depicted in Fig. 3(a), and theoretical response without chirp effect (α = 0) is also attached as a reference. One can observe that, the bias influences the frequency of the first fading dip: as the bias voltage decreases, the dip shifts to the high-frequency end. Figure 3(b) shows the first fading dips’ frequencies of 40km fiber channel as well as the calculated chirp factor α according to Eq. (2). As for the PAM4 transmission, we optimize the electrical levels to obtain equally-spaced optical PAM4 signals by considering both the P-I curve and extinction ratio. For example in Fig. 3(b), the four levels of PAM4 signal are respectively selected as −2.3V, −1.5V, −0.9V and −0.1V. Since the open-close degree of the three eyes in PAM4 signals directly influence the BER, only intensity changes between adjacent levels are considered here for simplicity. The quality of three eyes can be analyzed by considering three small-signal responses respectively biased at −1.9V, −1.2V and −0.5V, which are exactly centers of the three eyes. As per Fig. 3(a), the three bias voltages result in different fading dips. In other words, the three eyes in PAM4 signals experience different transient chirp, which will cause different transmission performance. Generally, the lower eye is expected to be larger due to the relatively larger main lobe bandwidth in Fig. 3(a), while the upper eye is expected to be smaller. To verify this, we plot the eye diagrams of EML-based PAM4 after 0km and 10km transmission and FFE equalization in Figs. 4(a) and 4(b). For 0km case, three eyes are with approximately the same size. For 10km case, the three eyes are with quite different sizes and the upper eye obviously shrinks. Figure 4(c) depicts the histograms of received amplitude distribution of the four symbols in the 10km case with 170ps/nm accumulated dispersion. One can clearly observe that the heights of the three eyes are unequal, and symbol error (SE) increases from Type 1 to Type 3 as shown by the three overlap regions among the four symbols. According to the Gray mapping rule, BER of 2nd b is dominated by Type 3 SER; therefore, BER of 2ndb is quite larger than 1stb in EML-based PAM4 transmissions, which is similar to DML as discussed above.
Fig. 3 (a) Channel response of 40km fiber under different bias voltages, (b) Frequencies of first dip at 40km, the calculated chirp factors and transfer function.
Fig. 4 Analysis of received PAM4 signal in EML-based transmission. Eye diagrams of 28G Baud PAM4: (a) 0km and (b) 10km. (c) Amplitude distribution of received symbols after 10km transmission.
3. Principles of sample bit interleaving
In a typical digital mobile fronthaul system, each sample of baseband signal is represented by an n-bit codeword [20], where a sign bit represents the polarity, and the other bits represent the amplitude. The basic idea of the scheme proposed in this work is based on two facts:
- (1) In digital fronthaul, higher order sample bits contribute more to the quality of wireless signal, thus it is beneficial to guarantee the quality of high order sample bits with high priority [15].
- (2) In DML/EML based PAM4 system, the BER is non-uniformly distributed between 1stb and 2ndb as discussed in Section 2, which inherently provide two priorities to respectively accommodate high order and low order sample bits.
Therefore, we adopt sample bits interleaving [15] to cope with the 2ndb BER problem. The detailed principles are illustrated in Fig. 5. In this paper, 8-bit non-uniform quantization with μ law signal compression is adopted in order to reduce the data rate [19]. Note that the proposed scheme should also be effective for other kinds of quantization methods, such as the standard uniform quantization [16], as sample bits in these quantization methods also have different priorities. After digitization, the sample bits in a codeword are divided into high order and low order halves, which are respectively marked with blue and magenta colors. After respective line coding, the high order half and low order half are interleaved and Gray-mapped to PAM4 symbols. In this way, high order sample bits are mapped to 1stb of PAM4 symbol; whereas low order sample bits are mapped to 2ndb. As a result, bit errors caused by the chirp-induced distortions mostly concentrate on low order sample bits half, rather than uniformly distribute in the two halves. Consequently, EVM distortion of wireless signal induced by bit error can be greatly reduced.
Fig. 5 Principle of sample bits interleaving and corresponding line coding. Gray mapping rule is adopted for PAM4 modulation and demodulation.
Considering the practical implementation, line coding is always applied for direct current (DC) balance or clock recovery. For the proposed system, line coding is introduced between sample bits interleaving and PAM4 modulation. Detailed design is depicted in Fig. 5. The interleaved bit stream is firstly divided into two branches, which respectively contain high order bits and low order bits. Then the two branches of bit stream are independently encoded. After that, the two bit streams are combined again and modulated into PAM4 signals. At receiver side, reverse procedures are adopted. It should be noted that, encoding or decoding of high order bits and low order bits can be operated in parallel. Hence, almost no extra latency is induced compared with existing line coding methods.
4. Experimental setup
The experimental setup is shown as Fig. 6. Baseband OFDM signal is generated according to 3GPP specifications [21], and then 5/6 down-sampled to remove frequency redundancy, and clipped with a peak to average power ratio (PAPR) of 10dB to suppress quantization noise. Since signals above 10 times of mean value are of low probability, the nonlinearity effect caused by clipping can be neglected. Then the clipped signal is quantized and sample bits are obtained. High order and low order sample bits are interleaved and then modulated onto PAM4 signals. The PAM4 signal is pre-emphasized with 1 sample per symbol to compensate the bandwidth insufficiency of the arbitrary waveform generator (AWG) and electrical amplifier. The derived sequence is up-sampled and loaded into AWG of 70G sampling rate, generating 56Gbps PAM4 signal. Noting that 56Gbps PAM4 is adopted here, because as a baseline data rate for future short-haul transmission, the cost of related transceivers can be greatly reduced with the large-scale commercialization. The number of supported 100MHz OFDM signals can be calculated as: 56Gbps/(122.88MSa*5/6)/2/8bit/(66/64)≈33 [21]. Then, the output electrical signal is used to drive the DML or EML with parameters listed in the table of Fig. 6. For DML, a high optical output power (10dBm) is adopted to help suppress the potential skewing effect due to signal dependent rise/fall time [22, 23]. The peak-to-peak voltage (Vpp) of driving signal is optimized considering both extinction ratio and adiabatic chirp. For EML, the bias voltage and Vpp are jointly optimized under the tradeoff among extinction ratio and nonlinearity distortion. The output optical signal is launched into a standard single mode fiber (SSMF). At the receiver side, a variable optical attenuator (VOA) is used to adjust the received optical power (ROP). Then the optical signal is retrieved by a PIN-TIA and sent to a digital sampling oscilloscope (DSO) at 80GSa/s for offline processing. The offline processing includes FFE equalization, binary bits recovery, de-interleaving of sample bits and reconstruction of OFDM signal.
Fig. 6 Experimental setup of DML/EML-based digital mobile fronthaul system. AWG (arbitrary waveform generator), DML (directly modulated laser), EML (Electro-absorption Modulated Lasers), VOA (variable optical attenuator), TIA (trans-impedance amplifier), DSO (digital storage oscilloscope), FFE (feed forward equalizer).
5. Results and discussions
5.1 Performance of DML-based transmission
We first evaluate the BER performance of 1stb and 2ndb of DML-based PAM4 signals. In back-to-back (BtB) case, the distortion is almost completely compensated after FFE equalization. 1stb and 2ndb have similar performance, and the slight difference between 1stb BER and 2ndb BER is attributed to the Gray coding rule. After fiber transmission, due to the interaction between chirp and dispersion, both overshooting and eye skewing effect occur as shown in the inset eye diagram. This phenomenon gets severer in 20km case. Although the FFE equalizer can help filter out the overshooting noise and reopen the eye, the eye skewing effect cannot be mitigated since it is a nonlinear distortion. From the eye diagrams of 10km and 20km, the eye height of the middle eye is larger than that of upper/lower eyes at the decision point, which means that the BER is non-uniformly distributed between 1stb and 2ndb. Total BER of PAM4 before and after equalization are also presented in Fig. 7. As the figure shows, although total BER of PAM4 after 10km and 20km fiber transmission are inferior to BER of BtB, BER of 1stb is relatively good.
Then we reconstruct the OFDM signal based on the PAM4 signal after FFE, of which the EVM are calculated and plotted in Fig. 8. Two conditions are compared: (i) conventional scheme (without interleaving) and (ii) scheme with sample bits interleaving. In BtB case, sample bits interleaving can give a slight improvement. The EVM is reduced by 1~2 dB regardless of ROP, which is mainly attributed that 1stb BER is half that of 2ndb according to the Gray mapping rule. In 10km case, sample bits interleaving can lead to a more prominent performance improvement. This improvement gets larger under a higher ROP. Particularly when ROP is −2dBm, EVM of −38dB is finally achieved with 8dB EVM reduction, which enables the transmission link to accommodate 1024-QAM format (EVM of −35.5dB required), leaving 2.5dB margin to bear distortions from wireless channel [24]. The constellations of 1024-QAM OFDM signal are inserted in Fig. 8(b). In 20km case, for conventional scheme, the EVM performance has a floor of −15dB because of the poor total BER; which even cannot satisfy the requirement of 16-QAM signal (EVM of −18dB required). In contrast, with bit interleaving scheme, EVM distortion is significantly reduced. It can be seen that at −2dBm ROP, up to 13dB EVM reduction is realized and EVM of −28dB is achieved to support 64QAM (EVM of −22dB required), which leaves up to 6dB margin to bear signal distortion caused by wireless channel. Interestingly, it is shown that higher ROP shows more EVM reduction. This is because of the logarithm relationship between SNR (dB unit) and BER, and also because the middle eye is not only tolerant to eye skewing but also the clipping effect caused by saturation of PIN-TIA.
Fig. 8 EVM performance of recovered wireless signal after DML transmission: (a) BtB, (b) 10km, (c) 20km.
5.2 Performance of EML-based transmission
Then, we evaluate the BER performance of EML-based PAM4 signals. The results are shown in Fig. 9. For BtB case, only very slight inter symbol interference (ISI) distortion of the received PAM4 signal is observed. This is because the EML adopted here has a larger bandwidth than the DML in Section 5.1. After FFE equalization, the residual ISI distortion is removed and the eye diagram is further improved. In this case, the 1stb shows slight superiority over 2ndb due to the Gray mapping rule. For 10km case, the eye diagram is severely distorted and even 7-level partial response signaling appears because of the power fading effect as above-mentioned in Section 2. After FFE equalization, the 4 levels of the signal are recovered however are distributed quite unequally. In particular, the upper eye is almost closed which is the main cause of the poor performance of 2ndb as well as total BER; both of which has a floor at ~10−3. In contrast, performance of the middle eye is much better, yielding a more accurate 1stb than 2ndb. When ROP is −1dBm, the BER of 1stb is reduced to 10−4 level. In addition, similar to DML-based system, the BER difference between 1stb and 2ndb gets larger under a higher ROP.
We then reconstructed the OFDM signal, and the EVM results are shown in Fig. 10. In BtB case, sample bits interleaving gives 2~3dB performance improvement compared with conventional scheme. When ROP is −9dBm, EVM of −34dB can be achieved to accommodate 256-QAM OFDM signal (EVM of −29dB required), where 5dB margin is left to bear the distortion from wireless channel. In 10km case, the achievable EVM is limited within −19dB by using conventional scheme, which can only support 16QAM-OFDM signals (EVM of −18dB required). On the other hand, with sample bits interleaving, the EVM is greatly reduced. Particularly at −1dBm ROP, EVM of −32dB is realized, which enables 256-QAM OFDM signal transmission. Here, we also attempt to extend the fiber distance to 20km, but find that the performance is too poor that PAM4 signal cannot be recovered. Thus, the results are not given here. It is interesting that, DML can support a longer distance for 56Gbps PAM4 transmission than EML. This is because that, the adiabatic chirp caused frequency modulation in DML can be converted to amplitude modulation after CD effect, thus enhancing the channel response [25].
6. Conclusions
In this paper, we study the implementation of DML and EML in PAM4-based digital mobile fronthaul system. According to the qualitative analysis of bit error distribution after fiber transmission, 1stb is more accurate than 2ndb in both systems. By taking advantage of the non-uniform BER distribution, we adopt sample bit interleaving to counteract the chirp induced performance degradation. More accurate 1stb of PAM4 is used to deliver high order sample bits and less accurate 2ndb to transmit low order sample bits. 56Gbps digitized OFDM signal delivery, which is capable to accommodate 33 × 100MHz OFDM signals, is experimentally conducted. We show that EVM is greatly reduced by using sample bits interleaving. For DML, the proposed scheme respectively leads to up to 8dB and 13dB EVM reduction to accommodate 1024QAM OFDM signals and 64QAM OFDM signals in 10km and 20km cases. For EML, 14dB EVM reduction is achieved in 10km to finally accommodate 256QAM OFDM signals.
Funding
National Natural Science Foundation of China (NSFC) (61431009 and 61521062).
References and links
1. I. C. Lin, C. Rowell, S. Han, Z. Xu, G. Li, and Z. Pan, “Toward green and soft: A 5G perspective,” IEEE Commun. Mag. 52(2), 66–73 (2014). [CrossRef]
2. D. Che, F. Yuan, and W. Shieh, “High-fidelity angle-modulated analog optical link,” Opt. Express 24(15), 16320–16328 (2016). [CrossRef] [PubMed]
3. B. G. Kim, S. H. Bae, H. Kim, and Y. C. Chung, “RoF-based mobile fronthaul networks implemented by using DML and EML for 5G wireless communication systems,” J. Lightwave Technol. 36(14), 2874–2881 (2018). [CrossRef]
4. H. Li, R. Hu, Q. Yang, M. Luo, Z. He, P. Jiang, Y. Liu, X. Li, and S. Yu, “Improving performance of mobile fronthaul architecture employing high order delta-sigma modulator with PAM-4 format,” Opt. Express 25(1), 1–9 (2017). [CrossRef] [PubMed]
5. M. Xu, F. Lu, J. Wang, L. Cheng, D. Guidotti, and G. K. Chang, “Key technologies for next-generation digital RoF mobile fronthaul with statistical data compression and multiband modulation,” J. Lightwave Technol. 35(17), 3671–3679 (2017). [CrossRef]
6. X. Liu and F. Effenberger, “Emerging optical access network technologies for 5G Wireless,” J. Opt. Commun. Netw. 8(12), B70–B79 (2016). [CrossRef]
7. “5G network architectures: a high level perspective”, (Huawei Technologies, 2016). https://www.huawei.com/minisite/5g/img/5G_Network_Architecture_A_High-Level_Perspective_en.pdf
8. S. H. Bae, H. K. Shim, U. H. Hong, H. Kim, A. Agata, K. Tanaka, M. Suzuki, and Y. C. Chung, “25-Gb/s TDM optical link using EMLs for mobile fronthaul network of LTE-A system,” IEEE Photonics Technol. Lett. 27(17), 1825–1828 (2015). [CrossRef]
9. J. Wang, Z. Yu, K. Ying, J. Zhang, F. Lu, M. Xu, L. Cheng, X. Ma, and G. K. Chang, “Digital mobile fronthaul based on delta-sigma modulation for 32 LTE carrier aggregation and FBMC signals,” J. Opt. Commun. Netw. 9(2), A233–A244 (2017). [CrossRef]
10. K. Zhong, X. Zhou, T. Gui, L. Tao, Y. Gao, W. Chen, J. Man, L. Zeng, A. P. T. Lau, and C. Lu, “Experimental study of PAM-4, CAP-16, and DMT for 100 Gb/s short reach optical transmission systems,” Opt. Express 23(2), 1176–1189 (2015). [CrossRef] [PubMed]
11. K. Zhang, Q. Zhuge, H. Xin, M. Osman, E. E. Fiky, L. Yi, W. Hu, and D. V. Plant, “Intensity directed equalizer for the mitigation of DML chirp induced distortion in dispersion-unmanaged C-band PAM transmission,” Opt. Express 25(23), 28123–28135 (2017). [CrossRef]
12. C. Sun, S. H. Bae, and H. Kim, “Transmission of 28-Gb/s Duobinary and PAM-4 signals using DML for optical access network,” IEEE Photonics Technol. Lett. 29(1), 130–133 (2017). [CrossRef]
13. H. Ji, L. Yi, L. Xue, and W. Hu, “Upstream dispersion management of 25 Gb/s duobinary and PAM-4 signals to support 0–40 km differential reach,” Chin. Opt. Lett. 15(2), 022502 (2017). [CrossRef]
14. K. Zhang, Q. Zhuge, H. Xin, H. He, W. Hu, and D. V. Plant, “Low-Cost WDM fronthaul enabled by partitioned asymmetric AWGR with simultaneous flexible transceiver assignment and chirp management,” J. Opt. Commun. Netw. 9(10), 876–888 (2017). [CrossRef]
15. H. Xin, K. Zhang, H. He, W. Hu, and M. Zhang, “Fidelity enhancement in high-data-rate digital mobile fronthaul with sample bits interleaving and unequally-spaced PAM4,” Opt. Express 25(5), 5559–5570 (2017). [CrossRef] [PubMed]
16. H. Xin, K. Zhang, H. He, M. Zhang, and W. Hu, “High tolerance against chirp induced PAM4 eye skewing in DML-based digital mobile fronthaul with 11dB EVM Reduction,” in Proceedings of European Conference on Optical Communication (ECOC) (2017), paper M1B.3. [CrossRef]
17. J. C. Cartledge and B. Christensen, “Optimum operating points for electro-absorption modulators in 10 Gb/s transmission systems using non-dispersion shifted Fiber,” J. Lightwave Technol. 16(3), 349–357 (1998). [CrossRef]
18. H. Y. Chen, N. Kaneda, J. Lee, J. Chen, and Y. K. Chen, “Optical filter requirements in an EML-based single-sideband PAM4 intensity-modulation and direct-detection transmission system,” Opt. Express 25(6), 5852–5860 (2017). [CrossRef] [PubMed]
19. B. Guo, W. Cao, A. Tao, and D. Samardzija, “CPRI compression transport for LTE and LTE-A signal in C-RAN,” in Proceedings of Int. Conf. on Comm. and Networking in China. (IEEE, 2012) pp. 843–849.
20. CPRI specification V6.1 (2014–7-1). (2014). [Online]. Available: http://www.cpri.info/spec.html
21. 3GPP TS 36.104 version 14.3.0, “Base station (BS) radio transmission and reception,” 2017.
22. N. Kikuchi, R. Hirai, and T. Fukui, “Non-linearity compensation of high speed PAM4 signals from directly modulated laser at high extinction ratio,” in Proceedings of European Conference on Optical Communication (ECOC) (2016), paper M2B.4.
23. J. M. Castro, R. J. Pimpinella, B. Kose, Y. Huang, A. Novick, and B. Lane, “Eye skew modeling measurements and mitigation methods for VCSEL PAM-4 channels at data rates over 66 Gb/s,” in Optical Fiber Communications Conference (OFC) (2017), paper W3G.3. [CrossRef]
24. P. T. Ferrera, S. Straullu, S. Abrate, and R. Gaudino, “Upstream and downstream analysis of an optical fronthaul system based on DSP-assisted channel aggregation,” J. Opt. Commun. Netw. 9(12), 1191–1201 (2017). [CrossRef]
25. B. Wedding, B. Franz, and B. Junginger, “10-Gb/s optical transmission up to 253 km via standard single-mode fiber using the method of dispersion supported transmission,” J. Lightwave Technol. 12(10), 1720–1727 (1994). [CrossRef]
