Open Access
Add to BibSonomy
Abstract
This paper investigates solutions for mitigating multipath interference (MPI) impact for intensity modulation direct detection systems, particularly for PAM4 systems. We propose a scheme that uses a high pass filter (HPF) at the receiver to remove the MPI induced carrier-carrier beat noise and therefore improve the MPI tolerance. The scheme has been verified via theoretical analysis, numerical simulation, and experiment. For a typical optical transmitter based on Mach-Zehnder modulator (MZM) without frequency chirp, over 5dB MPI tolerance can be improved for 106.4Gb/s PAM4 by using the proposed scheme. Investigation was also extended to electro-absorption modulated laser (EML) and directly modulated laser (DML) based optical transmitters. Results show that transient chirping in EML reduces the effectiveness of proposed MPI mitigation scheme. For EML with chirp factor smaller than 1.5, MPI tolerance gain of 2dB is still achievable. However, the scheme is ineffective for typical DML where the adiabatic chirp significantly broadened the laser linewidth.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
The fast growth of data center business is driving operators, enterprises, and service providers to rely on Ethernet solutions that provide high speed network communications. Intensity modulation direction detection (IMDD) has been a solution for short reach networks due to its low complexity and low power consumption [1–7]. The previously deployed 100Gbit/s Ethernet (100GbE) used 4 lanes of 25Gbaud non-return-to-zero (NRZ) modulation formats, either through space division multiplexing with 4 parallel single mode (PSM4) fiber for 500 m, or wavelength division multiplexing (WDM) with 4 channels of LAN (local area network) WDM (LWDM-4) or coarse WDM (CWDM-4) for 2 km, 10 km and 40 km reaches. For the 400GbE currently being deployed, 4 level pulse amplitude modulation (PAM4) was adopted as the optical modulation format. In IEEE standards, 50Gbaud PAM4 PSM4 was used for 500 m, and 8 channel 25Gbuad (LWDM-8) was used for 2 km and 10 km reaches. In 100G Lambda multi-source agreement (MSA) industry consortium, 4 channels of 50Gbaud CWDM-4 PAM4 have also been adopted for 2 km and 10 km to reduce the cost. Now the IEEE standards are setting up a Call for Interest (CFI) consensus and going forward for the request to form the new study group on “Beyond 400GbE” for the next generation of Ethernet, like 800GbE or 1.6TbE. PAM4 (either single polarization or polarization multiplexed) may be still promising format for next generation Ethernet due to its low complexity and ease to upgrade from existing 400GbE.
However, it is well known that PAM4 is very susceptible to multipath interference (MPI) caused by multiple reflections originating from fiber connectors, transmitters, and receivers. In practical deployments, low MPI tolerance is a major limitation of directly detected PAM4 applications. Numerous works have been done on evaluating the impact of MPI [8–11]; particularly, [10,11] reported experimental measurement of the MPI impact. To mitigate MPI, some schemes using digital filtering have been proposed in patents [12–14]. [12] uses a tunable notch filter in digital signal processing (DSP) for filtering the received optical signal to remove the interference or uses a low-pass filter to extract the MPI which is then subtracted for mitigation. In [13], the MPI-mitigation circuit has an error generator that estimates a PAM level of samples received from MPI impaired signal and generates a corresponding error signal. The error signal is filtered with a low-pass filter to produce the estimate of MPI, which is then subtracted. In [14], a mean MPI signal representing a mean amplitude of the MPI in an input signal is generated using a mean MPI feedback loop or using an iterative feedforward process, then subtracted for MPI compensation. [12–14] may require a large tap number or large FFT (Fast Fourier Transform) size because of the low frequency of notch filter or low pass filter, which result in high DSP power consumption. Furthermore, patents [12–14] only proposed the concepts and to our knowledge, no experiment demonstration has been reported.
Meanwhile, due to the lack of practical MPI mitigation solution, the standards bodies like IEEE802.3 and 100G Lambda MSA tightened the fiber connector return loss (RL) specification from −26 dB to −35 dB in 2016, in order to accommodate PAM4 applications. But in real systems, return loss from “dirty connectors” (e.g., standard/subscriber connector (SC) or ferrule connector (FC)) can be still well above −26 dB, making it hard to use PAM4 format. In fact, MPI has still been a major limitation in performance in short reach fiber transmission systems like data centers.
Different from [12–14], this paper investigates solutions in analog domain for mitigating the impact of MPI effect and therefore increase MPI tolerance for IMDD systems, particularly PAMn systems. We propose a scheme that uses a high pass filter (HPF) at the receiver to suppress the MPI induced carrier-carrier beat noise, therefore, to mitigate the MPI effect. The scheme only involves passive analog filter and introduces no additional power consumption. The paper is organized as follows. Section 2 introduces our proposed scheme for mitigating MPI effect. Section 3 explains the operation principle where theoretical analysis is provided. Section 4 presents simulations which show the effectiveness of the scheme. Section 5 provides the experiment verification results. Section 6 investigates the impact of frequency chirp on the effectiveness of the proposed MPI mitigation scheme, where the chirp from electro-absorption modulated laser (EML) and directly modulated laser (DML) are discussed. Section 7 summarizes the paper.
2. Proposed MPI mitigation scheme
It is noted that the intensity modulated optical signal has a strong optical carrier in addition to broadband data. The carrier spectral width is determined by the laser linewidth, which is very narrow compared to data spectral width. MPI is the accumulation of reflected signals in the fiber link, which is combined with signal via superposition in optical field when detected at the optical receiver. After photo-detection, the beating of the optical signal carrier and MPI carrier generates low frequency components, here referred to as carrier-carrier beat noise, which is undesired crosstalk. The spectrum of the carrier-carrier beat noise is the convolution between the signal carrier and MPI carrier. Its spectral profile is determined by the laser linewidth and the delay between the signal and MPI reflections. For a typical laser linewidth of a few MHz, the spectral width of fully de-correlated carrier-carrier beat noise is about twice the laser linewidth, which is still very narrow compared to the data spectrum (tens of GHz). In principle, this low frequency carrier-carrier beat noise can be removed by an HPF in the optical receiver. Therefore, the impact of MPI can be greatly reduced.
The proposed MPI mitigation scheme using HPF for IMDD system is shown in Fig. 1, which consists of an intensity modulated optical transmitter, a fiber link that experiences multipath interferences, and an optical receiver with HPF.
In the optical transmitter, the intensity modulated optical signal can be generated from an intensity modulator like Mach-Zehnder modulator (MZM) following a continuous wave (CW) laser, which may be a distributed feedback (DFB) laser or an external cavity laser (ECL). The intensity modulated optical signal can also be generated by an EML or a DML. The optical signal may be in baseband format like multi-level pulse amplitude modulation PAMn, including PAM2 (NRZ), PAM4, or higher-level PAM.
The fiber link transmits an optical signal from the optical transmitter to the receiver. During the signal transmission, there may be multiple reflectance points, such as end-facets of transmitter and receiver or fiber connectors. Double reflections of a forward traveled signal generate a forward traveled reflected light. MPI is the accumulation of all forward traveled reflected lights. The intensity of MPI depends on the return loss of connectors, transmitter and receiver, as well as the polarizations orientation and delay between signal and each double reflected light, and most likely exhibits a statistic behavior.
The optical receiver consists of a photodetector (PD) followed by a transimpedance amplifier (TIA) and a proposed HPF. The HPF is used to mitigate the carrier-carrier beat noise caused by the beating between the signal carrier and MPI carrier. The electrical output signal of the optical receiver is fed to either a digital signal processor after analog to digital conversion or an analog decision circuit. The HPF is usually located after TIA. It also could be located between the PD and TIA. The latter configuration may experience parasitics and limit the receiver bandwidth.
It should be emphasized that the HPF proposed in this paper is different from a DC block in a conventional optical receiver. It is well known that the DC block exhibits a high pass filtering aspect, but mostly with a cut-off frequency in range of ∼10kHz to 100kHz, well below MHz. As demonstrated later, such a DC block has little or no effect on MPI effect mitigation. The HPF proposed in this paper has a cut-off frequency in order of 10MHz. Since forward error correction (FEC) like KP4 FEC is used, the conventional concern of DC wandering from HPF is not an issue. The wavelength drift between MPI and signal is also not an issue for short reach of a few kilometers.
A simple configuration of the HPF may be an RC high pass filter. A surface mount capacitor may be introduced on the receiver micro-stripe transmission line. The frequency transfer function of the HPF is
where ${f_c} = 1/({2\pi RC} )$ is the cut-off frequency with the load impedance R and the capacitance C. For example, for a cut-off frequency (fC) of 20MHz, the required capacitance is around 160pF for 50 Ohm load impedance.3. Theoretical analysis
For an optical fiber link with MPI, if the number of reflection points, including those from transmitter and receiver, is n, the total number of forward traveled reflections (each by two reflectance points) is ${\boldsymbol N} = {\boldsymbol n}({{\boldsymbol n} - \mathbf{1}} )/\mathbf{2}$. The optical field with signal and MPI can be expressed as
For intensity modulated signal, the optical power can be approximated as ${\ P}({\ t} )= {{\ P}_0}({1 + {{\ V}_{\ S}}({\ t} )} )$, which consists of a DC part and an AC part. Here VS is the normalized modulation signal and P0 is the average power. The MPI can also be expressed as DC part and AC part, so Eq. (3) can be rewritten as,
It is worthwhile to point out that the delay τ plays an important role in MPI effect mitigation. When we say the signal and MPI phase noise are fully de-correlated, it means the delay is sufficiently long such as over the coherence length. In such a scenario, the carrier-carrier beating noise has wider spectrum and is relatively more difficult to separate from data spectrum, and changing the delay length would not result in an obvious change in performance. However, when the delay is less than the coherence length, reducing the delay would improve the performance, where the coherence can reduce carrier-carrier beat noise to some extent with reduced noise spectral width. When τ = 0, the term $2\sqrt {\varepsilon} \; cos[\phi{(t )- \phi({t - \tau} )} ]$ becomes pure DC.
For intensity modulated optical signal with finite extinction ratio, the DC part is larger than AC part, so the carrier-carrier beat noise is more dominant than the broadband crosstalk. Removing the carrier-carrier beat noise is expected to significantly improve the system performance.
4. Simulation verification
To verify the proposed scheme, simulations have been carried out with simulation parameters listed in Table 1. We first look at the RF spectra to better understand the principle. Figure 2(a) shows the RF spectrum of detected signal with −26.5dB equivalent MPI and Fig. 2(b) is the RF spectrum after an RC HPF with cut-off frequency of 20MHz applied. Strong low frequency components with a spectral profile close to Lorentz shape are observed in Fig. 2(a), which is due to carrier-carrier beat noise. After the HPF is applied to the MPI distorted signal, these low frequency noises are removed, as shown in Fig. 2(b).
Fig. 2. RF spectra of signal with −26.5 dB MPI. (a), no HPF; and (b), with HPF with a cut-off frequency of 20 MHz.
Figure 3 shows the eye diagram of a PAM4 signal with −26.5dB MPI, and without HPF (a) and with HPF (b). It can be seen that the high level eyes are almost closed due to the MPI induced crosstalk. Once an HPF with a cut-off frequency of 20MHz is applied to the distorted signal, the eyes have been significantly opened due to the removing of the carrier-carrier beat noise.
Figure 4 shows the simulated BER versus received optical power (ROP), with −26.5dB MPI, for various HPF cut-off frequencies. Figure 4(a) shows the result for a DFB + MZM transmitter with a laser linewidth of 1MHz. There is almost no performance improvement for HPF cut-off frequency of 0.1MHz, so a typical DC block has no effect in removing carrier-carrier beat noise for a laser linewidth of 1MHz. Results also show that the optimal cut-off frequency is around 20MHz, which is the tradeoff between sufficiently suppression of the carrier-carrier beat noise and HPF introduced distortion. These simulation results clearly show that the BER performance has been significantly improved by the proposed method. Figure 4(b) shows the results for a transmitter with a laser linewidth of 10MHz. The optimal cut-off frequency for this case is about 50MHz. It shows that the BER performance is also improved by using HPF for 10MHz linewidth, even though it is not as effective as for the case of 1MHz linewidth.
Fig. 4. BER vs. ROP for various HPF cut-off frequencies. (a) 1 MHz linewidth; and (b) 10 MHz linewidth.
MPI tolerance is another parameter to show the advantage of the proposed scheme. Figure 5 shows the MPI power penalty comparison between scenarios without and with HPF at 20MHz fc for various laser linewidths. Here the power penalty is referenced to the case of no MPI and no HPF @2e-4 BER by KP4 FEC. Figure 5(a) is for 1MHz laser linewidth. With very low level MPI, such as −40dB, the performance with HPF is slightly worse than that of without HPF due to the HPF introduced filtering penalty. HPF improves performance when MPI is higher than −37dB. For 1dB power penalty, HPF provides over 6dB gain in MPI tolerance, corresponding to over 3dB requirement reduction in connector return loss.
For 2MHz laser linewidth, the gain in MPI tolerance is reduced compared with that of 1MHz linewidth, while it is still above 5dB, as shown in Fig. 5(b). Therefore, for a typical DFB + MZM transmitter, MPI tolerance gain of 5dB can be achieved. Figure 5(c) is for 10MHz laser linewidth. Due to a significantly broader laser linewidth, the gain in MPI tolerance at 1dB power penalty is reduced to only about 1dB, and HPF improves performance only when MPI is higher than −35dB.
As discussed above, the effectiveness of HPF for MPI mitigation strongly depends on laser linewidth. Figure 6(a) shows the optimal HPF cut-off frequency as a function of laser linewidth with MPI of −26.5dB. The optimal HPF cut-off frequency increases as laser linewidth increases. For a typical laser linewidth of 1∼2MHz like for DFB + MZM, the optimal HPF cut-off frequency is around 20∼30MHz. For an optical transmitter using ECL with linewidth of 50kHz, the optimal HFP cut-off frequency is around 4MHz. For a transmitter with linewidth of 10MHz, the optimal HPF cut-off frequency is around 50MHz. Figure 6(b) shows the ratio of the optimal HPF cut-off frequency to laser linewidth as a function of laser linewidth. The ratio decreases as laser linewidth increases. For ECL with linewidth of 50kHz, the ratio is around 80; for a typical DFB + MZM with linewidth of 1∼2MHz, the ratio is around 20∼13; for transmitter with linewidth of 10MHz, the ratio is around 5. Figure 6(c) is the sensitivity gain dependence on linewidth. Here the gain is defined as the sensitivity difference between those of no HPF and with HPF at optimal cut-off frequency, @2e-4 BER. Figure 6(c) shows that sensitivity gain decreases as linewidth increases.
Fig. 6. Impact of laser linewidth, evaluated at MPI = −26.5 dB. (a) Optimal HPF cut-off frequency; (b), Ratio of HPF cut-off frequency to laser linewidth; (c), Sensitivity gain.
Above discussion shows that the optimal HPF cut-off frequency is strongly dependent on laser linewidth. HPF also introduces filtering penalty in addition to removing carrier-carrier beat noise. This penalty is dependent on baud rate; therefore, the optimal cut-off frequency is also related to baud rate. Figure 7 shows the HPF filtering itself (without MPI) induced power penalty as a function of cut-off frequency for 26.6Gbuad, 53.2Gbuad and 106.4Gbaud PAM4. The penalty is again defined relative to that of no HPF at BER@2e-4. Penalty induced by filtering increases with the HPF cut-off frequency. For 20MHz cut-off frequency, the penalty induced by filtering is about 0.2dB for 53.2Gbuad PAM4 and about 0.1dB for 106.4Gbaud PAM4. Higher baud rate PAM4 suffers from lower HPF filtering penalty at the same HPF cut-off frequency.
Although the discussion so far is focused on PAM4, it should be noted that the proposed scheme can also be extended to other orders of PAM. In the final part of this section, we evaluate the effectiveness of the scheme via simulation for two other popular PAMs: PAM2 and PAM8. 2MHz laser linewidth, 6dB ER and 53.2G baud rate were used in the simulation. We first compare the HPF filtering itself (without MPI) induced power penalty as a function of cut-off frequency among various PAMs, as shown in Fig. 8. The results show that PAM2 is much more tolerant to HPF filtering than PAM4, as the penalty is still less than 0.4dB even for up to 160MHz cut-off frequency. However, PAM8 is much more sensitive to the HPF filtering than PAM4, as the penalty is over 1.0dB even for 15MHz cut-off frequency.
Figure 9 shows the power penalty as a function of MPI with or without HPF for different PAMs. As expected, PAM2 is more tolerant to MPI than PAM4, and PAM8 is more sensitive to MPI than PAM4. Without HPF, for 1dB penalty defined at BER = 2e-4 relative to the case of no MPI and no HPF, the allowed MPI is around 41.3dB, 32.2dB and 22.4dB for PAM8, PAM4 and PAM2 respectively. As discussed in Fig. 8, different PAMs suffer from different HPF filtering penalty, therefore HPFs with fc of 5MHz, 20MHz and 100MHz were applied to PAM8, PAM4 and PAM2 respectively in the simulations, for MPI mitigation. This provides MPI tolerance gain of about 1.4dB, 5.2dB and 6.1dB for PAM8, PAM4 and PAM2, respectively. Even though higher order PAM benefits less MPI tolerance gain from high pass filtering, all three PAMs show effectiveness of the proposed scheme.
5. Experimental verification
Experiments were also carried out to further verify the proposed scheme, with the setup shown in Fig. 10. The optical signal with PAM4 format was generated by an MZM following a DFB laser. In the experiment, all the connectors were well cleaned and well connected, so the intrinsic MPI of the system due to the connectors was negligible. The equivalent MPI was introduced by delaying a portion of the signal via introducing a 2.2km single mode fiber (SMF). The polarization of MPI was adjusted by a polarization controller (PC), and the intensity of MPI was controlled by a variable optical attenuator (VOA). The combined optical signal and MPI are detected at an optical receiver which has a capacitor on the microstrip of receiver board for DC blocking. The capacitor was changed with different capacitance values to get the desired cut-off frequencies of the HPF. The filtered electrical signal was digitized by a Keysight digital storage oscilloscope (DSO) with an 80GSa/s sampling rate and was processed by DSP which compensates for any residual S21 and adjusts for timing offset. The receiver with the capability of capacitor replacement was a 25G device, so the experiment was limited to 26.6Gbaud for verification purpose.
In the experiment, three surface mount capacitors with capacitance of 150pF, 47pF and 27pF were used to replace the original DC blocking capacitor of the receiver, respectively. These capacitances resulted measured HPF cut-off frequencies of 6MHz, 23MHz and 37MHz respectively, as shown in the HPF frequency response of Fig. 11.
Figure 12(a) shows the measured BER versus received optical power for 26.6Gbaud PAM4 with −26.5dB MPI, with three HPF cut-off frequencies: 6MHz, 23MHZ and 37MHz. All HPFs provide improved BER performance compared to the case without HPF, indicating the effectiveness of the proposed scheme. 23MHz HPF delivers better performance than those of 6MHz and 37MHz. Simulations at 26.6Gbaud PAM4 with parameters similar to those of the lab setup were also run for comparison, and results are shown in Fig. 12(b). The simulation result exhibits relatively more effectiveness in MPI suppression by HPF than experiment, which might be due to the TIA nonlinearity in experiment. Simulations show that 23MHz HPF provides better performance than those with 6.3MHz and 37MHz cut-off frequencies, which agree well with experiments.
Fig. 12. BER vs. ROP for 26.6Gbaud PAM4 with different HPF cut-off frequencies. (a), experiment; (b) simulation.
Figure 13(a) shows the measured BER versus ROP with no MPI for different HPF cut-off frequencies. At fC = 23MHz, the HPF filtering power penalty defined at 2e-4 BER relative to no HPF is close to 0.5dB; and at fC = 37MHz, the HPF filtering power penalty is close to 1dB. These test results are consistent with those of the simulation result on 26.6Gbuad in Fig. 7. Figure 13(b) shows the measured BER versus ROP with −29.5dB MPI for different HPF cut-off frequencies. There is still performance improvement even though it is less than the case of −26.5dB MPI, indicating the effectiveness of the proposed scheme. 23MHz cut-off frequency still shows better performance than those of 6MHz and 37MHz cut-off frequencies.
Fig. 13. Measured BER vs. ROP for 26.6Gbaud PAM4 with different HPF cut-off frequencies. (a), no MPI; (b) −29.5 dB MPI.
6. Impact of frequency chirp
The above discussion only considered optical transmitters based on MZM where there is no frequency chirp. Other types of intensity modulated transmitters like EML and DML exhibit frequency chirp, therefore we need to investigate HPF effectiveness for those kinds of transmitters. We first look at the EML case where only transient chirp is involved which can described by the following equations [15].
Here Δν is the transient frequency chirp, α is the chirp factor or linewidth enhancement factor, ϕ(t) the laser phase noise, φ(t) the phase shift introduced by transient chirping, P(t) the optical power and E(t) the optical field. The simulation uses the same parameters as in Table 1, with 53.2G baud rate and 2 MHz laser linewidth, where chirp is introduced via Eqs. (7–9) with different chirp factors.
Table 1. Simulation Parameters
Figure 14 shows the power penalty versus equivalent MPI for various chirp factors. Again, the power penalty is defined at 2e-4 BER relative to the case of no MPI and no HPF. For zero chirp, the MPI tolerance defined at 1 dB power penalty is improved over 5 dB when an HPF with 20 MHz cut-off frequency is applied. Once there is a chirp, the MPI tolerance is reduced, and increasing the chirp factor further reduces the MPI tolerance. When chirp factor is increased to 3.5, the MPI tolerance gain from HPF is totally removed. While if chirp factor is no more than 1.5, the MPI tolerance gain over 2 dB is still achievable.
The degradation in effectiveness of MPI mitigation by transient chirping is likely due to the broadened spectrum as shown in Fig. 15, which shows the optical spectra without and with transient chirping at 2 MHz resolution. With transient frequency chirping, the intensity in low frequency part near DC is reduced while the spectrum in high frequency part is increased. This is similar to the impact of increased linewidth, which degrades the effectiveness in MPI mitigation as discussed in Section 4.
Now we look at the DML case where both transient chirp and adiabatic chirp are involved which can described by the equation below [15]:
Here κ is the adiabatic chirp constant, and Pave is the average optical power inside laser cavity.
Figure 16 shows the power penalty versus equivalent MPI for various adiabatic chirp constants. α and Pave were set to 1.0 and 10dBm in the simulation. For κ = 0, the MPI tolerance gain with HPF at fc=20 MHz over no HPF is still around 4 dB for α = 1, which is similar to that of Fig. 14. With adiabatic chirp, the MPI tolerance gain is reduced. For κ=5 GHz/mW, which is a typical value for an InP DFB laser at 1550 nm waveband, the MPI tolerance gain is totally removed. Therefore, the proposed MPI mitigation scheme is not effective for DML.
The lack of effectiveness in MPI mitigation for DML is due to the significantly broadened laser linewidth by adiabatic chirp. Figure 17(a) shows the optical spectra of DML with various adiabatic chirp constants for α = 1.0. For a typical adiabatic chirp constant like 5 GHz/mW, the laser linewidth is significantly broadened as shown in the pink curve of Fig. 17(a). The broadened linewidth induces significantly broadened auto-correlation trace as shown in Fig. 17(b) which has a 1 km delay. Auto-correlation trace is equivalent to RF spectrum of signal combined with MPI. For the case of κ=1 GHz/mW, the RF spectrum is significantly broadened and for the case of κ=5 GHz/mW, the spectrum is totally flattened, both making the high pass filtering for MPI mitigation ineffective.
Fig. 17. Spectra with various adiabatic chirp constants at 2 MHz resolution. (a) Optical spectra; (b) auto-correlation traces.
7. Summaries
We have proposed and demonstrated a scheme using HPF at receiver to remove the MPI induced carrier-carrier beat noise and improve MPI tolerance for IMDD systems, particularly PAM4 systems. The solution has been verified via theoretical analysis, numerical simulation and experiment demonstration. For a typical optical transmitter based on MZM without frequency chirp and at 53.2Gbaud PAM4, the MPI tolerance can be improved by 5dB via using the proposed scheme without introducing any additional power consumption. Investigation was also extended to EML and DML based optical transmitters. Results show that transient chirping in EML reduces the effectiveness of MPI mitigation using HPF, while for EML with chirp factor no larger than 1.5, 2dB MPI tolerance gain is still achievable. However, this scheme is ineffective for typical DML where the adiabatic chirp significantly broadened the laser linewidth.
Acknowledgment
The authors would like to thank Dr. Andy Shen for his helpful discussion.
Disclosures
The authors declare no conflicts of interest.
References
1. J. Lee, S. Shahramian, N. Kaneda, Y. Baeyens, J. Sinsky, L. Buhl, J. Weiner, U. Koc, A. Konczykowska, J. Dupuy, F. Jorge, R. Aroca, T. Pfau, and Y. Chen, “Demonstration of 112-Gbit/s optical transmission using 56 GBaud PAM-4 driver and clock-and-data recovery ICs,” European Conference on Optical Communication (ECOC) 2015, paper Mo.4.5.4.
2. J. Man, W. Chen, X. Song, and L. Zeng, “A low-cost 100GE optical transceiver module for 2 km SMF interconnect with PAM4 modulation,” in Optical Fiber Communication Conference (OFC) 2014, paper M2E.7.
3. T. Wettlin, S. Ohlendorf, T. Rahman, J. Wei, S. Calabro, N. Stojanovic, and S. Pachnicke, “Beyond 200 Gb/s PAM4 Transmission using Tomlinson-Harashima Precoding,” European Conference on Optical Communication (ECOC) 2019, Paper Tu2B.6, Sept 2019.
4. J. Verbist, M. Vanhoecke, M. Lillieholm, S. A. Srinivasan, P. D. Heyn, J. V. Campenhou, M. Galili, L. K. Oxenløwe, X. Yin, J. Bauwelinck, and G. Roelkens, “4:1 Silicon Photonic Serializer for Data Center Interconnects Demonstrating 104 Gbaud OOK and PAM4 Transmission,” J. Lightwave Technol. 37(5), 1498–1503 (2019). [CrossRef]
5. M. Jacques, Z. Xing, A. Samani, X. Li, E. El-Fiky, S. Alam, O. Carpentier, P.-C. Koh, and D. V. Plant, “Net 212.5 Gbit/s Transmission in O-band With a SiP MZM, One Driver and Linear Equalization,” Optical Fiber Communication Conference (OFC) 2020, Post deadline paper, Th4A.3.
6. S. Kanazawa, H. Yamazaki, Y. Nakanishi, Y. Ueda, W. Kobayashi, Y. Muramoto, H. Ishii, and H. Sanjoh, “214- Gb/s 4-PAM Operation of Flip-Chip Interconnection EADFB Laser Module,” J. Lightwave Technol. 35(3), 418–422 (2017). [CrossRef]
7. S. Yamaoka, N. P. Diamantopoulos, H. Nishi, R. Nakao, T. Fujii, K. Takeda, T. Hiraki, S. Kanazawa, H. Tanobe, T. Kakitsuka, T. Tsuchizawa, F. Koyama, and S. Matsuo, “239.3-Gbit/s net rate PAM-4 transmission using directly modulated membrane lasers on highthermal-conductivity SiC,” the 45th European Conference on Optical Communication (ECOC2019), 2019, pp. 1–4, doi: 10.1049/cp.2019.1022.
8. G. Nicholl, C. Fludger, M. Mazzini, T. Kupfer, and M. Traverso, “Analytic estimation of MPI penalty with supporting experimental measurements,” IEEE802.3bs, 400GE Task Force Meeting, Nov. 2014, presentation nicholl_3bs_01_1114.
9. F. Zhu, Y. J. Wen, Y. J. Zhu, and Y. S. Bai, “Technical Feasibility of Single Wavelength 400GbE 2km &10km PMD,” IEEE802.3bs, 400GE Task Force Meeting, May 2014, presentation zhu_3bs_01a_0514.
10. W. Way and T. Chan, “Revisit MPI penalties for 400GBASE-FR/LR8 Links,” IEEE802.3bs, 400GE Task Force Meeting, Jan 2016, presentation way_3bs_01a_0116.
11. C. R. S. Fludger, M. Mazzini, T. Kupfer, and M. Traverso, “Experimental measurements of the impact of multi-path interference on PAM signals,” Optical Fiber Communication Conference (OFC) 2014, paper W1F.6.
12. X. Zhou, R. Urata, E. Mao, H. Liu, and C. L. Johnson, “In-band optical interference mitigation for direct-detection optical communication systems,” U.S. Patent, US9998235B2.
13. B. P. Smith, J. Riani, S. Bhoja, A. Farhoodfar, and V. Bhatt, “Circuit for multi-path interference mitigation in an optical communication system,” U.S. Patent, US9876581B2.
14. W. Bliss and J. Wang, “Compensation for optical multi-path interference,” U.S. Patent, US9246587B2.
15. K. Zhang, Q. Zhuge, H. Xin, W. Hu, and D. Plant, “Performance comparison of DML, EML and MZM in dispersion-unmanaged short reach transmissions with digital signal processing,” Opt. Express 26(26), 34288–34304 (2018). [CrossRef]
