Coherent optical reception promises performance gains for a wide range of telecom applications and photonic sensing. However, the practical implementation and the particular realization of homodyne detection is by no means straight-forward. Local oscillator requirements and polarization management need to be cost-effectively supported for accurate signal detection at high sensitivity, preferably without relying on digital processing resources. Towards this direction we propose a conceptually simple, laser-based homodyne receiver. We exploit the injection locking of a pair of externally modulated lasers that simultaneously serve as optically synchronized local oscillators and photodetectors in a polarization-diversity analogue coherent receiver arrangement. We demonstrate signal detection at 2.5 Gb/s over an optical budget of 35 dB and a dynamic range of >20 dB. Long-term measurements over field-installed fiber confirm the correct operation independent of the polarization state of light. Stability considerations for the injection locking process are drawn in view of even higher loss budgets.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Transceivers for optical telecommunications are characterized by a trinity of merits: cost, energy and performance density. Among these, cost still prevails as the primary criterion when it comes to practical deployment. The vast performance benefits offered by coherent systems with respect to their direct detection counterparts are therefore often disregarded. Especially in cost-sensitive segments such as optical access or datacom, leaps are still to be made to break the pre-dominance of direct detection, including, amongst others, tunable optical references, polarization management and real-time signal processing. Research work has proposed a series of complexity-performance trade-offs to accelerate the commercial up-take. These include bandwidth-inefficient yet phase-agnostic heterodyne receivers , transmitter-side polarization management [2,3] or non-linear analogue processing of on-off keying through envelope detection after phase-diversity reception assisted by an optical hybrid [4,5]. Self-homodyne reception with an embedded local oscillator (LO) at any physical multiplexing dimension such as polarization, time , frequency or space  avoids a local reference but makes its characteristics dependent on the transmission channel. Homodyne reception can be further enabled by phase locked loops [8,9], which again raise the receiver-side complexity significantly while further requiring a high loop bandwidth to accommodate larger frequency offsets. In view of lean component technology, we have recently proposed to exploit an externally modulated laser (EML) as coherent homodyne detector  and have further validated the ability of this low-cost laser element to simultaneously serve as transmitter and coherent receiver . Although this fits very well within a low-cost envelope as required for mass applications, polarization management has been neglected so far.
In this work, we demonstrate polarization-independent operation through a tandem-EML design. We experimentally validate coherent data reception at 2.5 Gb/s over 35 dB loss budget in presence of polarization drift over field-deployed fiber. We further prove stable coherent homodyne reception in virtue of the high dynamic range for injection-locking of the employed EMLs and present a mitigation scheme for strong polarization-selective fading to retain an analogue coherent receiver design even with just partially locked LO pair.
The paper is organized as follows. Section 2 discusses the receiver concept and its operation regimes. Section 3 introduces the experimental setup. Section 4 investigates the polarization diversity functionality and the locking stability before Section 5 proves its operation for data transmission. Section 6 evaluates the transmission performance over field-deployed fiber. The impact of an unlocked receiver branch is investigated in Section 7, for which Section 8 provides a mitigation scheme. A conclusion is drawn in Section 9.
2. Polarization-independent coherent homodyne receiver
Coherent receivers based on integrated LO and photodetector can be as simple as an EML . In such a receiver implementation the distributed feedback (DFB) laser is injection-locked to the signal that is to be detected and the electro-absorption modulator (EAM) serves as fast photodetector. Since the coherent detection process is polarization-selective, a diversity scheme based on a tandem-EML has been recently introduced in brevity . A performance that is independent of the received polarization state after an arbitrary polarization transform along an optical transmission channel is obtained through detection of the transverse-electric (TE) and -magnetic (TM) polarizations of the data signal. A possible implementation is shown in Fig. 1(a) where a polarization beam splitter (PBS) connects to an EML-based receiver in each of its output branches. Although the inclusion of polarization optics indicates a substantial increase of complexity and assembly cost at first, the monolithic integration of InP EML technology with waveguide-based polarization management has been recently demonstrated .
Although this polarization diversity scheme is known for coherent receivers that split a single LO and the incoming signal in orthogonal polarization components to then add the detected electrical baseband signals [14,15], the EMLs in Fig. 1(a) serve as independent receivers. In particular, two LOs provide their optical reference ΛTE, ΛTM (Fig. 1(b)) at the spectral vicinity of the incident data signal λS. This leads to two intermediate frequencies IF = ΔFTE,TM after photodetection, for which the frequency detuning ΔF is determined by the wavelength mismatch |ΛTE,TM – λS|. By choosing this mismatch between LO and data signal small enough through tuning of the LO bias β,β*, homodyne reception is obtained for a mismatch lower than the locking range LR of the DFB laser (Fig. 1(c)). Under this condition the DFB laser is injection-locked to the data signal and takes the optical emission frequency of the injected signal so that IF = 0 , while the optical phase of the LO is determined by the frequency deviation between LO and the optical input signal within the locking range . This guarantees a correct analogue electrical summing of the detected signals of both polarizations.
However, the range LR is a function of the optical injection power Pin (Fig. 1(c)), which together with a polarization-dependent transmission ε and μ at the PBS for its TE and TM branches can lead to an EML injection level that is much lower than that of the coherent receiver input, i.e., the PBS input. Injection-locked operation therefore greatly depends not only on the received optical power Pin but also on the incident state of polarization (SOP) and may not always apply to both polarization tributaries. In this work we will first prove correct operation for the case that both tributaries are stably locked, corresponding to region ξ in Fig. 1(c). The extent of region ξ is determined by the ability of the EMLs to reside within the locking range to the data signal. Previous findings have shown that an integrated micro-cooler of a transistor-outline (TO) packaged EML enables a stable emission frequency. Even at a low optical input of −30 dBm the locking range is typically 20-times larger than the emission wavelength deviation caused by remaining bias drifts . Stable locking can be therefore guaranteed for the ideal SOP, while the polarization extinction of the PBS needs to be considered for the proposed polarization-diversity receiver. We will therefore particularly investigate the implications of an unlocked polarization tributary (region Ξ), which incurs a summing error during analogue post-processing of the detected photocurrents for TE and TM polarizations. Compared to our previous work , an analogue monitoring stage will be incorporated to detect an unlocked polarization tributary for its sub-sequent blanking before the summing point, as it will be dynamically evaluated for a drop in optical input power to the modified receiver.
It should be noted that a spectral carrier component is required to enable injection locking, which prevents the use of phase-modulated signals unless a carrier-pilot is ensured, for example, through non-rigorous antipodal phase shift keying. Phase diversity reception would then be required additionally. At the same time, injection locking at very low power levels can be supported through additional phase locked loop configurations .
3. Experimental evaluation setup and methodology
Figure 2 shows the experimental evaluation setup. Two TO-can EMLs followed by low noise amplifiers (LNA) build the core of the polarization-diversity homodyne receiver. The choice for 50Ω-LNAs rather than transimpedance amplifiers (TIA) was made due to the use of TO-can packaged EAM-based photodetectors, though LNAs are considered sub-optimal in terms of noise performance and, consequently, sensitivity. The optical transmitter includes an optical seed source at λS = 1548.51 nm and an in-phase/quadrature (I/Q) modulator used for optical single-sideband (SSB) modulation. SSB has been chosen for data modulation in order to reduce dispersive effects in case of fiber transmission, thus yielding the intrinsic receiver performance in absence of additional penalties while retaining a realistic network scenario. The optical carrier λS has been suppressed by ~10 dB due to its sole use for injection locking, which does not require high optical power levels .
Data transmission was based on an orthogonally frequency division multiplexed (OFDM) bandpass signal. The OFDM bandwidth and was center frequency were 2.5 and 5 GHz, respectively. While the center frequency is arbitrary, the same settings as for a previous single-polarization experiment  have been applied to allow for a comparison. The 128 OFDM sub-carriers were individually bit-loaded. The data signal was launched with 6 dBm from the transmitter. A lab fiber span of 27.5 km was inserted as optical channel in case of data transmission and a variable optical attenuator (Att) was included to investigate the compatible loss budget.
In the following Sections IV to VII the coherent receiver is first characterized statically. Data transmission is then evaluated over lab and field-installed fiber. Reception penalties will be evaluated as function of the input SOP but also for the unfavorable condition that a LO is not locked to the input signal anymore as the optical injection power fades.
4. Characterization of the polarization-diversity receiver and locking stability
The partitioning of optical power among the branches of the tandem-EML receiver depends on the input SOP, which can be described by its azimuth α on the Poincaré sphere. When aligned with the axes of the PBS, either the TE- or TM-EML are fed at full optical power. Figure 3(a) shows these ideal conditions for 2α = 0 and 90°, respectively. In the worst case the polarization states fall along the meridian of the Poincaré sphere that spans through diagonal and circular states, at an azimuth offset of 45°. In this case both branches receive half the optical power, leading to a drop in received radio frequency (RF) power of 3 dB after summing the TE and TM polarizations in the electrical domain (Fig. 3(a)). If the receiver branches show an optical loss imbalance δR, an extra penalty is incurring. This is applied in Fig. 3(a) to the TM branch at an imbalance of δR = 0.5 dB (dashed) and 1 dB (dotted line).
To prove the static polarization-diversity operation of the proposed coherent receiver in experiment, an unmodulated RF carrier at fRF = 5 GHz was transmitted for characterization purposes while sweeping the input SOP to the receiver. Figure 3(b) shows the received RF power level at fRF. For reception at either the TE- or the TM-EML branch, the power level spreads (ΔP) due to polarization-selective fading due to extinction at the PBS but does not drop to a low background noise level due to retention of coherent homodyne reception for which fRF is preserved. The summed TE + TM output shows a reduced fading of ΔPTE + TM = 3.5 dB (×). This proves the correct polarization-diversity reception. The remaining excess penalty of 0.5 dB is attributed to the imbalance in fiber-to-EML coupling efficiency at both branches.
An important feature of the EML-based coherent receiver is its ability to retain coherent homodyne reception to ensure that there is no frequency deviation between transmitted and received signal components. However, the PBS as polarization-selective device may cause a locking instability as the injection level fades. This would result in a beat note at fΔ that deviates from the transmitted RF carrier at fRF. Figure 3(c) presents the power-dependent locking stability in terms of frequency deviation from the expected carrier frequency fRF for various input power levels resulting from polarization-swept light. The optical power Pin delivered to the PBS was −20 dBm and the frequency deviation ΔF of the DFB lasers towards the seeding signal λS has been minimized. The reference for the power fading shown in Fig. 3(c) is made to maximum PBS transmission for each of the EML branches and their summed output. Stable injection locking is obtained for the vast majority of input SOP, as it is evidenced by the non-existent frequency deviation fΔ – fRF. Coherent homodyne reception applies to all SOP states for the TE-EML. The TM-EML remains locked (ℓ) even for a polarization-induced RF fading of 33.2 dB (●), corresponding to operation in the preferred region ξ of Fig. 1(c). For slightly higher values first unlocking effects appear (υ) only for the TM-EML, which then falls back to a coherent intradyne reception mode. This corresponds to region Ξ with partially locked tandem-EML. Nevertheless, the high dynamic power range of >30 dB proves the robustness of injection-locking for the independent LOs of the constituent EMLs. Nevertheless, it is required to assess the dynamic range in relation to the supported optical loss budget for data transmission in order to provide a more practical measure.
5. Performance for coherent homodyne data reception
Polarization-independent operation was further evaluated for data transmission, for which the error vector magnitude (EVM) of the single-sideband, OFDM data signal has been assessed. The received OFDM signal is shown in Fig. 4(a) for an optical budget of 16 dB. It is correctly translated from the optical to the electrical domain in virtue of coherent homodyne detection process at both EMLs. There is no frequency shift since the LOs are locked to the data signal and the signal spectrum is not smeared out as it would apply to intradyne reception. With this, the OFDM signal can be demodulated without modification of the digital signal processing (DSP) stack and without additional correction functions.
Figure 4(b) shows the EVM performance (◆) for all OFDM sub-carriers at an azimuth of 2α = 45° and 0° for 27.5 km lab fiber transmission at an optical loss budget of 22 dB between transmitter and receiver. An average modulation efficiency (▲) of η = 2.55 bits/symbol is obtained and leads to a data rate of R = 5.56 Gb/s after removal of overheads due to OFDM modulation and hard-decision forward error correction (FEC). For the co-polarized case at α = 0° a similar efficiency of 2.57 bits/symbol and post-FEC data rate of 5.61 Gb/s have been found. This confirms the correct operation of the polarization-diversity receiver under data transmission.
Figure 4(c) shows the supported OFDM post-FEC data rate for 27.5 km lab fiber transmission as function of the optical loss budget. Results are presented for signals that are received co-polarized to the LOs (●) and under 2α = 45° (▲). An overlap in performance is given for a wide range of optical loss budget. A post-FEC data rate of 2.5 Gb/s can still be obtained at a budget of 35.2 dB, for which the noise floor given by the LNA front-end leads to a roll-off. For high input power levels an onset of overload can be noticed (ς). This is attributed to cross-talk arising from direct-detection (DD) terms at the EAM photodetector (Fig. 4(a)). Taking into consideration this saturation effect and the LNA noise floor, the dynamic range found for the assessed range of delivered optical power is more than 20 dB. This value addresses well the specifications sought for demanding applications such as passive optical networks (PON) .
It has to be stressed that the absence of a TIA as electrical front-end leads to a sensitivity that is below that found in state-of-the-art works on coherent signal reception. Nevertheless, the work proves the concept of a greatly simplified, polarisation-independent coherent reception engine that performs homodyne detection without the need for additional DSP functions, while enabling at the same time simultaneous data transmission as demonstrated elsewhere .
6. Evaluation over field-deployed fiber link
In order to investigate the receiver performance under a more realistic polarization transform, the lab fiber span has been replaced by a 40 km field-installed fiber link in the city of Vienna. The evolution of the SOP has been acquired by a polarimeter and is shown in Fig. 5(a) for a measurement period of ~200 min during a winter afternoon. The trajectory on the Poincaré sphere shows an average drift of 9°/min and features an evolution in azimuth and inclination.
Figure 5(b) presents the EVM over time, in reference to an 8-ary quadrature amplitude modulated (QAM) sub-carrier of the OFDM signal. The EVM (▲) remains stable as the polarization transform evolves and is well below the EVM limit. This confirms the correct operation for a realistic transmission channel. The degradation during the second half of the measurement is explained by the partial loss of carrier-suppression at the transmitter: the power in the received modulation sideband (■) fades as the optical carrier recovers due to a drift in the bias points of the nested Mach-Zehnder modulator that serves the optical I/Q modulation. The fading modulation index leads to the observed EVM degradation.
7. Reception penalty under partially locked operation
In the actual experimental investigation, the electrical receiver front-end is a LNA and is therefore not optimized in terms of noise performance. When the bias control minimizes the detuning ΔF of the LOs, the unlocking of the LO would occur at a low optical power level for which the received electrical signal is already covered by the noise floor. This means that regions Ξ and Ψ in Fig. 1(c) are practically hardly reached – though they would apply with proper TIA-based signal conditioning. Nevertheless, in order to investigate the effect of an unlocked data tributary in this work, one LO has been artificially detuned to force unlocking of this LO even at a high delivered optical power Pin. In such a situation, corresponding to region Ξ, the incorporation of an unlocked tributary with poor signal integrity in the analogue summing process degrades the received electrical signal. This effect is quantified in Fig. 6(a,b) in terms of EVM performance for an 8-QAM loaded OFDM and a loss budget of 20 dB. For this purpose, both outputs of the receiver branches have been acquired and eventually summed in the digital domain. Two scenarios have been investigated in terms of input SOP.
In the first the received signal is co-polarized with the TE-axis of the EML tandem (α ≈0°), with tributary-selective power extinctions of ε ≈0 and µ > 16 dB for the TE- and TM-polarizations. Figure 6(a) shows the corresponding EVM performance. For locked LOs in both receiver branches there is no significant difference in EVM for summing either a single (▲) or both polarizations (◆). Although correct TE + TM summing is performed for locked LOs, the average EVM degrades by 0.8% since the TM-tributary is mainly contributing through noise as its signal is highly extinct by a large µ value. When the LO in the TM-branch is artificially detuned by ~500 MHz (■), the EVM performance of the summed signal worsens by 3.2%. This penalty derives from the frequency-shifted TM-tributary that is perceived as crosstalk. The difference in EVM performances for various cases can be also seen in the constellations, which are appended to Fig. 6 for the central OFDM sub-carrier.
In the second scenario the SOP of the input signal is adjusted for an azimuth of 2α = 45°, yielding ε = µ = 3 dB. As it can be seen in Fig. 6(b), the summing process with locked LOs in both polarizations (◇) now improves the performance by 0.3% when compared to that of single polarization (△). However, since the TM-tributary is now associated to approximately the same power as the TE-tributary, an unlocked LO in the TM-branch causes the EVM to deteriorate by 5.1% (□). Since the input SOP cannot be controlled, it is therefore necessary to suppress such crosstalk if operation at the region Ξ (Fig. 1(c)) is desired.
8. Homodyne receiver robust to strong polarization-selective power fading
Figure 6(c) presents the modified coherent receiver that avoids this source of erroneous operation. The unlocked data tributary is continuously detected and instantaneously switched off in the RF domain. In this way an analogue receiver design is retained. In particular, a portion of the detected signal is split off with a 10% RF coupler (CPL). Measurement of the spectral IF component of the residual optical carrier is performed in the low-frequency range up to 1 GHz through a low-pass filter (LPF) and a RF detector (DET). The magnitude of spectral components in this IF range indicates whether the respective LO is locked or not. Next, a thresholder (THR) with a reference adjusted to the integrated power of the noise floor delivers the control signal σ for the RF switch (SWI), which either feeds the homodyne signal (IF = 0) to the summing point or drops the intradyne signal (IF > 0). The threshold for decision making whether or not the tributary is unlocked relies on the fact that an the unlocked EML receiver results in a beat note that can be distinguished in terms of RF power from the background noise due to EML receiver without optical input and LNA. In this work the threshold was manually chosen according to the integrated noise power. A realistic deployment scenario would require an additional calibration procedure and additional low-frequency electronics and control circuitry.
Figure 7(a,b) shows the instantaneous spectrum of the received TM tributary R, which was tapped with a 50/50 RF splitter just before the RF switch during swept optical input power Pin. The transition between locked and unlocked LO can be noticed for decreasing and increasing optical power. A beat note between optical carrier and LO at an IF > 0 appears as the power reaches a value that does not support injection locking. Consequently, the originally clear OFDM spectrum washes out. It shall be stressed that the response of the locking process does not show clear boundaries and a bouncing behavior is visible for the IF beat noise. This is due to the finite sweeping speed of the variable optical attenuator used to emulate the power drop, leading to a window of partially stable locking in the order of 200 to 600 µs.
Figure 7(c) presents the dynamic operation of the switched receiver architecture for a continuous drop in delivered optical power. As the analogue monitor signal for the optical input power Pin drops, the locking process becomes unstable. This can be noticed by the increased power for the received TM data tributary R, which results from the additional IF beat note of the residual optical carrier. For the intended homodyne case this beat note would be suppressed as a result of the typical low-frequency cut-off in the RF chain. At the same time as the LO unlocks, the thresholder eventually provides its trigger signal σ to the RF switch, which blanks the undesired intradyne TM data tributary S that is delivered to the summing point. In this way detrimental crosstalk but also optical receiver noise is mitigated.
A polarization-independent coherent receiver based on a tandem of low-cost TO-can EMLs has been demonstrated. Homodyne detection is obtained through stable injection-locking of the EML-based LO and photodetector with high dynamic input power range. Correct analogue RF signal processing of the detected electrical signals of both polarization tributaries has been shown without use of digital correction functions. Data reception at 2.5 Gb/s was achieved over a loss budget and dynamic range of 35 dB and >20 dB, respectively. Measurements over a 40 km field-deployed fiber link have confirmed the polarization-independent operation. Moreover, the implications of polarization-selective power fading on the locking stability have been assessed in view of extreme loss budgets. Continuous monitoring of the LO state and an electrical drop of an erroneous tributary deriving from an unlocked LO ensures correct analogue operation, even for the worst case in which a strongly quenched injection level at one receiver branch results in a frequency-shifted electrical tributary. Co-integration with a TIA is required to obtain a better sensitivitiy.
H2020 European Research Council (804769).
1. J. A. Altabas, G. Silva Valdecasa, M. Didriksen, J. A. Lazaro, I. Garces, I. Tafur Monroy, and J. B. Jensen, “Real-time 10Gbps Polarization Independent Quasicoherent Receiver for NG-PON2 Access Networks,” in Proc. Opt. Fiber Comm. Conf., San Diego, United States, Mar. 2018, Th1A.3. [CrossRef]
2. M. S. Erkilinc, D. Lavery, K. Shi, B. C. Thomsen, R. I. Killey, S. J. Savory, and P. Bayvel, “Comparison of Low Complexity Coherent Receivers for UDWDM-PONs (λ-to-the-User),” J. Lightwave Technol. 36(16), 3453–3464 (2018). [CrossRef]
3. S. Narikawa, H. Sanjoh, and N. Sakurai, “Coherent WDM-PON using heterodyne detection with transmitter-side polarization diversity,” IEICE Electron. Express 7(16), 1195–1200 (2010). [CrossRef]
4. A. W. Davis, M. J. Pettitt, J. P. King, and S. Wright, “Phase Diversity Techniques for Coherent Optical Receivers,” J. Lightwave Technol. 5(4), 561–572 (1987). [CrossRef]
5. M. Artiglia, M. Presi, F. Bottoni, M. Rannello, and E. Ciaramella, “Polarization-Independent Coherent Real-Time Analog Receiver for PON Access Systems,” J. Lightwave Technol. 34(8), 2027–2033 (2016). [CrossRef]
7. Z. Feng, L. Xu, Q. Wu, M. Tang, S. Fu, W. Tong, and D. Liu, “Large-Capacity Optical Access Network Utilizing Multicore Fibre and Self-Homodyne Coherent Detection,” in Proc. Opt. Fiber Comm. Conf., Los Angeles, United States, Mar. 2017, Th1K.2. [CrossRef]
8. S. Ristic, A. Bhardwaj, M. Rodwell, L. Coldren, and L. Johansson, “An Optical Phase-Locked Loop Photonic Integrated Circuit,” J. Lightwave Technol. 28(4), 526–538 (2010). [CrossRef]
9. K. Balakier, L. Ponnampalam, M. J. Fice, C. C. Renaud, and A. J. Seeds, “Integrated Semiconductor Laser Optical Phase Lock Loops,” IEEE J. Sel. Top. Quantum Electron. 24(1), 1 (2018). [CrossRef]
10. B. Schrenk, “Injection-Locked Coherent Reception Through Externally Modulated Laser,” IEEE J. Sel. Top. Quantum Electron. 24(2), 1–12 (2018). [CrossRef]
11. B. Schrenk and F. Karinou, “A Coherent Homodyne TO-Can Transceiver as Simple as an EML,” J. Lightwave Technol. 37(2), 555–561 (2019). [CrossRef]
12. B. Schrenk and F. Karinou, “Polarization-Immune Coherent Homodyne Receiver Enabled by a Tandem of TO-can EMLs,” in Proc. Europ. Conf. Opt. Comm., Rome, Italy, Sep. 2018, We4G.3. [CrossRef]
13. M. Baier, F. M. Soares, T. Gaertner, A. Schoenau, M. Moehrle, and M. Schell, “New Polarization Multiplexed Externally Modulated Laser PIC,” in Proc. Europ. Conf. Opt. Comm., Rome, Italy, Sep. 2018, Mo4C.2. [CrossRef]
14. R. Noe, H. Rodler, A. Ebberg, G. Gaukel, B. Noll, J. Wittmann, and F. Auracher, “Comparison of Polarization Handling Methods in Coherent Optical Systems,” J. Lightwave Technol. 9(10), 1353–1366 (1991). [CrossRef]
15. B. Glance, “Polarization Independent Coherent Optical Receiver,” J. Lightwave Technol. 5(2), 274–276 (1987). [CrossRef]
16. F. Morgensen, H. Olesen, and G. Jacobsen, “Locking Conditions and Stability Properties for a Semiconductor Laser with External Light Injection,” IEEE J. Quantum Electron. 21(7), 784–793 (1985). [CrossRef]
17. D. S. Wu, R. Slavik, G. Marra, and D. J. Richardson, “Direct Selection and Amplification of Individual Narrowly Spaced Optical Comb Modes Via Injection Locking: Design and Characterization,” J. Lightwave Technol. 31(14), 2287–2295 (2013). [CrossRef]
18. R. Hui, A. D’Ottavi, A. Mecozzi, and P. Spano, “Injection Locking in Distributed Feedback Semiconductor Lasers,” J. Quantum Electron. 27(6), 1688–1695 (1991). [CrossRef]
19. “40-Gigabit-capable passive optical networks 2 (NG-PON2): Physical media dependent (PMD) layer specification”, Recommendation ITU-T G.989.2 (2014).