Open Access
Add to BibSonomy
Abstract
The increasing demand of real-time applications poses a huge challenge to building next-generation radio access network (NG-RAN) with higher stability and lower system complexity. Parallel signal detection (PSD), which aggregates signals of different intermediate frequencies (IFs) on different wavelengths with a single photodiode (PD), becomes a promising candidate for uplink mobile fronthaul with the advantage of low-latency. However, high requirements on the transmitters inhibit the large-scale deployment of radio units (RU). In this paper, we propose an economical, low-latency, multipoint-to-point (MP2P) uplink fronthaul architecture capable of aggregating four end-users with commercial 25G-class optical modules and a single PD. With delta-sigma modulation (DSM), commercial off-the-shelf optical modules can replace analog transmitters in traditional systems. As a demonstration, we aggregated 4 × 380.16-MHz 5 G new radio (NR) orthogonal frequency division multiplexing (OFDM) signals in an IF band with a fixed interval of 400 MHz over 20 km fiber with 4 users.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Nowadays, emerging latency-sensitive applications (such as automatic driving, remote surgery, and virtual reality) put forward extremely high requirements on the delay and stability of the radio access network (RAN), especially to its uplink transmission link [1]. A typical next-generation radio access network (NG-RAN) divides the entire function of a radio base station (BS) into three key functional blocks: a central unit (CU), a distributed unit (DU), and a radio unit (RU) [1]. To reduce latency and achieve real-time services, DUs take on more data processing tasks. Therefore, a low-latency link between DU and RU can significantly improve network response time by reducing network delay. Moreover, due to the dense deployment of RUs, it is necessary to apply multipoint-to-point (MP2P) uplink transmission with large-scale access and low-latency connectivity to provide ∼1 Gb/s capacity for each uplink RAN node at low cost [2].
In the existing standardized time division multiplexing (TDM)-based uplink scheme, the uplink frame latency is high because the RUs transmit their uplink frames in separate non-overlapping time slots to avoid collision (Fig. 1(a)). To settle this problem, the latency can be readily reduced by having a short cycling period in recent 50G-PON (passive optical network) systems [3]. In addition, all the TDM RUs need to share the full rate, resulting in limited per-user uplink capacity [4]. In contrast, wavelength division multiplexing (WDM) is a better choice to provide dedicated low-latency services to all end-users (Fig. 1(b)). Recently, a novel coarse wavelength division multiplexing (CWDM) and circulator-based system is proposed [5]. The architecture utilizes two sets of CWDM six-wavelength optical modules to achieve non-interfering 6-channel, 25 Gbit/s low-cost bidirectional fronthaul application. However, some existing networks deploy colorless optical couplers. In WDM systems, if the number of users and wavelengths are to be increased, additional de-multiplexer (DMUX) and its port mapping relationship need to be added, which makes it difficult to the deployment and upgrade of the fronthaul network for multi-user access. Thus, upgrades of fronthaul indicators such as delay and capacity with coupler-based optical fiber network is essential and challenging.
Different from other MP2P WDM uplink schemes, the parallel signal detection (PSD) scheme aggregates signals of different intermediate frequencies (IFs) on different wavelengths with a single photodiode (PD) (Fig. 1(c)). 12 × 100 MHz 4G + orthogonal frequency division multiplexing (OFDM) signal from four users is successfully aggregated with negligible interference [6]. Different from the traditional WDM scheme, it can use only one PD instead of DMUX and equal number of receivers to the users. In the case of large-scale access, the complexity of DU will be greatly decreased. However, it was demonstrated mostly by analog transmitters consisting of expensive Mach-Zehnder modulators (MZMs) in each RU to generate and modulate multi-IF signals, which greatly increases the cost and deteriorates maintainability of RUs [2,6,7]. Therefore, it is essential to lower the requirements of the transmitter and utilize more cost-effective transmitters such as integrated optical modules to simplify the RU structure. Nevertheless, optical modules will generate wide-spectrum non-return-to-zero (NRZ) signal, which is contradictory to the principle of traditional PSD scheme. Compared to the traditional one, we proposed an economical, low-latency, MP2P aggregation uplink fronthaul architecture with widely used commercial 25G-class optical modules and a single PD, as is shown in Fig. 1(d) [8]. Delta-sigma modulation (DSM) is suitable for fronthaul and wireless applications due to its simplified and cost-efficient structure [9]. With DSM, analog IF signals can be converted into NRZ signals, which are delivered by commercial off-the-shelf optical modules. Combined with PSD scheme of low-latency and support for multi-user, it can be well adopted to access scenarios such as Fiber to the Room (FTTR) [10]. However, since signal received by the PD is a linear combination of the electrical signals sent by each user, the noise from different users is cumulated. Thus, when using transmitters with high noise levels, the performance of each OFDM band degrades slightly as the number of access users increases. According to Fig. 1(e), the proposed PSD scheme features low-latency and low-complexity. It utilizes 25G-class WDM optical modules, which are widely deployed in commercial network and possess mature industrial support chain and obvious cost advantages [5]. Furthermore, in the proposed scheme, using different wavelengths for different RUs is to avoid optical channel-channel beating interference at lower frequencies. It is more expensive compared to TDM optical modules with the same wavelength. However, to enhance the latency and stability of existing TDM-PON systems, retrofitting them into traditional WDM architectures would result in prohibitively high deployment costs. In contrast, the proposed approach offers a more cost-effective solution by replacing devices.
In this paper, we demonstrated the uplink aggregation of 4 × 380.16 MHz 5 G new radio (NR) OFDM signal over 20-km single mode fiber (SMF) by utilizing four WDM optical modules. To verify the flexibility of the proposed architecture, the application scenarios under different numbers of users are also discussed to emulate heavy traffic and light traffic. The results indicate that PSD aggregation technology can support low-latency, cost-effective, and power-efficient uplink mobile fronthaul networks by exploiting existing integrated optical transceivers.
2. Principle
Considering an uplink network of n users, center frequencies of the narrowband OFDM signals are IF1, IF2, … IFn. In current existing schemes, a complete set of analog transmitters is required for e-o conversion in each RU. To adapt to the existing commercial fronthaul system, we introduce DSM to convert the OFDM signals into 1-bit sequences, which can be delivered by optical modules [11].
The conversion is realized through oversampling, noise-shaping and 1-bit quantization. The OFDM signal is first oversampled to expand its Nyquist zone. Then, it is fed to the delta-sigma modulator, where noise-shaping pushes the quantization noise N out of the signal band, leaving a frequency range with trivial noise for settling OFDM bands as shown in Fig. 2(a). Thus, analog OFDM signal can be converted to on-off keying (OOK) with only one quantization bit.
To obtain a flat quantization noise distribution, a 6th-order delta-sigma modulator is designed. The structure and parameters are shown in Fig. 3 and Table 1. ${a_n}$ and ${g_n}$ are weight factors and feedback factors, respectively. Thus, the noise transfer function (NTF) is given by [11]:
Table 1. Parameters of the designed delta-sigma modulator
Figure 4 shows the magnitude-frequency response of quantization noise for the designed delta-sigma modulator. The curve, which represents the distribution of quantization noise in the frequency domain, is calculated and plotted by NTF with fs = 28.05 GSa/s. A frequency range of around 1.6 GHz is provided for settling signals. Different from lower-order delta-sigma modulators, there are three notches (corresponding to zeros of NTF) in the frequency domain. Spreading the notches (zeros) over the signal band (unit circle) ensures a low and flat in-band noise distribution, which greatly benefits multi-IF transmission.
Fig. 4. Distribution of quantization noise and zero-poles plots of the designed delta-sigma modulator.
After delta-sigma modulation, the 1-bit sequences are sent by WDM optical modules with wavelengths of ${\lambda _i}({i = 1\sim n} )$. Then optical signals are combined by a coupler. After single-end PD detection, quantization noise accumulates at the two side-bands and IF signals are placed consecutively in the multi-band range (Fig. 2(b)). Optical channel-channel beating interference is automatically filtered out because the wavelength interval is beyond receiver bandwidth [2]. In this way, only a PD is needed to receive multi-IF signals from different users. The signal received by the PD is a linear combination of the electrical signals sent by each user. However, the noise in the multi-band range is cumulated, thus, the performance of each OFDM band slightly degrades.
3. Simulation
In order to verify the feasibility of the proposed scheme, we first conduct a simulation on the proposed MP2P aggregation architecture with four end-users using VPI Transmission Maker. Figure 5 shows the setup of the simulation system.
At the transmitter side, we first generate four 380.16-MHz OFDM signals. Key parameters are listed in Table 2. Each signal has a subcarrier spacing of 120 kHz, and the fast Fourier transform (FFT) size and the number of subcarriers are set to be 4096 and 3168, respectively. All these parameters follow the specifications defined for the 5 G frequency range 2 (FR2) [12]. The gap between the center frequency of each signal is set at 400 MHz. Therefore, the total bandwidth of the aggregated signal is 1.6 GHz. Then, OFDM signals are independently converted into 1-bit sequences by delta-sigma modulators working at 28.05 GSa/s. Additional noise is added to emulate the fixed SNR of practical electrical ports. Modulated by four electro-absorption modulated lasers (EMLs) and coupled, the optical signal is transmitted over 20 km standard single-mode fiber (SSMF). The wavelengths of the EMLs are 1547.72, 1548.51, 1550.12, and 1551.72 nm, which follow the dense wavelength division multiplexing (DWDM) wavelength grid. At the receiver side, the optical signal was detected by a single PD. The low-pass filter is used to emulate the effective bandwidth of the PD.
Table 2. Parameters of numerical analysis of the four-user aggregation architecture
Figure 6 shows the average BERs under different received optical power (RoP). In order to explore the impact of receiving signals from four different users simultaneously, the simulation is divided into “four-channel” mode and “single-channel” mode by controlling the switches of four EMLs. Four-channel mode (square) means that four EMLs work at the same time and send four signals modulated by delta-sigma modulators from four users, respectively. In single-channel mode (triangle), performance of four users is individually investigated while other three EMLs are turned off.
In the four-channel scenario, optical sensitivity with and without fiber transmission is -12.3 dBm and -14.2 dBm, respectively. Sensitivity increases around 11.38 dB when the number of detected wavelengths increases to 4. This is partly due to the 6-dB increase caused by quadrupling the number of wavelengths. The other part comes from SNR penalty caused by the linear addition of noise of all wavelengths during square-law detection.
The above experimental results are measured when the thermal noise of the PD is set to 4e-12 A/Hz1/2. By adjusting the value of “Thermal Noise”, we also studied the impact of PDs with different noise levels on system performance. Table 3 lists the difference in sensitivity between “single-channel” and “four-channel” modes in the back-to-back (B2B) case. The simulation results show that as the noise level of receiver increases, the difference in sensitivity decreases. This is because receiving signal from other three users quadruples the noise from the transmitter ${N_{tx}}$, while the noise from the receiver ${N_{rx}}$ remains at the same magnitude. When the noise of the receiver is negligible, the noise received in the “four-channel” scenario simply increases by a factor of 4 compared to the original noise. However, when the noise level of the receiver increases and becomes dominant, the SNR penalty of the aggregated signal gradually decreases. This means that the multi-user aggregation penalty is related to the noise level of the receiver.
Table 3. Sensitivity difference between “four-channel” and “single-channel” mode
The above simulation sets the same power of the four channels, resulting in similar BER performance. Furthermore, since different RUs may have different distances from the DU, the aggregated power of the four channels after coupling may vary. To study the impact of distance induced power variation on other channels during aggregation, we adjust the launch power of EML1 while keeping the same launch power of the other three channels, resulting in a power offset from -4 dBm to 4 dBm compared to the other three channels. When the power decreases, the offset is positive. Figure 7 illustrates the SNR penalty (gain) brought to the other three channels. The SNR penalty (gain) refers to the difference compared to the case where the power remains unchanged.
Fig. 7. Impact of power offset of channel 1 on the other three channels: simulation result (black) and fitting curve (red).
The black line (square) in Fig. 7 represents the simulation results from VPI. As the power of channel 1 decreases, the noise (contributed by channel 1) in the aggregated signal decreases, leading to an improvement in SNR for the other three signals.
Assuming the power of channel 1 before adjustment is ${P_1}$, the SNR of the other three OFDM signals are represented as $SN{R_1} = \frac{S}{{{n_0} + a{P_1}}}$, where S represents the power of the other three OFDM signals, a represents the coefficient of noise caused by channel 1, and ${n_0}$ represents the power of other noise. When the power of channel 1 is adjusted to ${P_2}$, $SN{R_2} = \frac{S}{{{n_0} + a{P_2}}}$. The SNR penalty/gain can be expressed as:
4. Setup
The experimental setup for MP2P uplink transmission is shown in Fig. 8. To verify the performance of the system with different numbers of users, the experimental procedure is divided into “one-user” and “four-user” scenarios. It corresponds to the number of working optical modules (wavelengths). Transmitter and IF assignment of these two cases is listed in Table 4. Case I simulates heavy traffic scenario. In this scenario, transmitter-2 occupies the entire 1.6 GHz spectrum range and transmits all four OFDM bands located at different frequencies. The other three transmitters are turned off. In contrast, Case II simulates light traffic scenario where each transmitter sends non-overlapping OFDM bands with center frequencies from IF1 to IF4.
Table 4. Transmitter and IF assignment of two cases
In the transmitter, corresponding to 5 G specifications, four 16-QAM OFDM signals with 380.16 MHz bandwidth are generated offline. To avoid spectral overlap after PSD, the four bands are respectively located at 2.2 GHz, 2.6 GHz, 3 GHz and 3.4 GHz with the channel spacing of 400 MHz. OFDM signals are individually fed to the delta-sigma modulators, where the analog inputs are digitized to one/four 28.05-Gb/s NRZ with 1-bit quantization. Then the delta-sigma modulated optical NRZ signals are separately transmitted by one/four commercial WDM optical modules embedded in the transmitter shown in inset (i) of Fig. 8, and combined by a 1 × 4 coupler. To prove the flexibility of the proposed scheme, four WDM optical modules are randomly selected with wavelengths of 1533.58 nm, 1548.52 nm, 1550.15 nm and 1551.74 nm, respectively. The optical module model used in the experiment is RTXM330-9332 produced by Accelink. The power of transmitters is approximately 4 dBm. The wavelength spacing is over 1.6 nm (200 GHz). It means that the beat noise between wavelengths will be filtered by the analog bandwidth of the PD. The number of working wavelengths depends on the above two scenarios. The eye diagram of transmitter-4 is also shown and SNR is around 22 dB, which is much lower than that of analog external modulator with MZM and DAC (around 35 dB). Thus, the modulation format is limited to 16-QAM in exchange for utilization of low cost and commercial optical modules. To support 5 G formats such as 256-QAM, there are several improvement methods: (1) compared with integrated optical modules, using MZM and DAC can achieve higher ER and linearity [13,14]; (2) improving the SNR of the optical module, for example, enhancing clock data recovery (CDR) performance; (3) increasing the number of bits in the digital transmitter [15]; (4) smaller signal bandwidth can improve SNR, thus enabling support for higher-order modulation formats.
After transmission over 20 km fiber, optical signal is detected by a PD and sent to a digital sampling oscilloscope (DSO) of 4 GHz bandwidth at 20 GSa/s for offline processing. Since the signals are at different center frequencies, the four bands can be independently filtered and demodulated. Finally, we measure the BER performance of each signal band one by one.
The experimental results of both cases including electrical spectra and BER performance are shown in Fig. 9 and 10. In case I (one-user scenario), only one optical module (transmitter-2) is working. It means that transmitter-2 can occupy the entire frequency range without sharing it with other users. As shown in Fig. 9(a), it transmits the four bands IF1-IF4 at the same time. The capacity of transmitter-2 reaches 4 (four bands) ${\times} $ 380.16 MHz ${\times} $ 4 (spectral efficiency) = 6.08 Gb/s. It can be helpful for RU under heavy traffic time.
Fig. 9. Results of case I: electrical spectrum in the transmitter (a) and receiver (b), (c) Average BER vs. RoP.
Fig. 10. Results of case II: (a) electrical spectra in the four transmitters, (b) electrical spectrum in the receiver, (c) Average BER vs. RoP of all four wavelengths in the single-channel and four-channel scenarios.
When the capacity requirement of a single user decreases, the number of aggregated users increases. Figure 10 shows the system performance in the four-user scenario (case II). The electrical spectra of the transmitters in different colors are shown in Fig. 10(a). The colors indicate the different source transmitters of the IFs. The uneven noise floor is caused by noise-shaping. The electrical spectrum at the receiver side is shown in Fig. 10(b). It shows that the four bands remain non-interfering.
Then the performance in the single-channel and multi-channel scenarios are evaluated. Figure 9(c) and Fig. 10(c) illustrate the average BER versus RoP of all wavelengths (In single-channel case, the performance of the transmitters is individually investigated and other channels are turned off). The curves show that the proposed MP2P network can support all four bands without exceeding the 7% forward error correction (FEC) threshold (BER = 3.8e-3) in both cases.
In the four-channel situation of case II, optical sensitivity with and without fiber transmission is -4 dBm and -6 dBm, respectively. However, the sensitivity in simulation is -12.3 dBm and -14.2 dBm, respectively. The large sensitivity penalty between simulation and experiment mainly comes from the extinction ratio (ER) of the transmitters. The ER of the electro-absorption modulator (EAM) used in simulation is about 20 dB, while the ER of the optical modules used in the experiment is about 8 dB. The difference in ER will lead to the difference in AC signal power under the same RoP, thus affecting the sensitivity. In addition, the difference in PD noise level between simulation and experiment is also a minor factor leading to the difference in sensitivity.
Since the power of transmitters is about 4 dBm, the link budget is about 8 dB in this 4-channel 20-km system. Semi-conductor optical amplifiers (SOA) and erbium-doped fiber amplifiers (EDFA) for signal amplification can be inserted to significantly improve the budget. Furthermore, using optical modules with higher ER and lower noise can also achieve a larger power budget. As shown in Fig. 10(c), sensitivity increases around 11.7 dB when the number of detected wavelengths increases to 4. It is consistent with the previous simulation results. Apart from 6 dB increase caused by quadrupling the number of wavelengths, the other part comes from the SNR penalty due to the linear addition of noise of all wavelengths during square-law detection. Compared with quantization noise, most of the noise comes from modulation of the optical module itself. SNR of the NRZ signal sent by the optical module, which is about 22 dB shown in inset of Fig. 8, is lower than that of analog transmitter with high-resolution DAC and external modulation. We believe that it is possible for uplink fronthaul networks to upgrade modulation format and capacity using transmitters with high linearity and low noise, which support SNR of around 35 dB and the modulation format up to 256-QAM [16].
In this work, the aggregation of four OFDM signals with 400 MHz bandwidth each is demonstrated. To aggregate more users, the signal bandwidth for each user can be adjusted, for instance, by sending eight signals with 200 MHz bandwidth. Alternatively, digital transmitters with higher bit rate and higher order modulation can be used, such as the 100GBASE-LR SFPDD single Lambda 1310 nm 10 km [15]. This optical module can transmit 53.125 GBd PAM4 signals. Due to the doubled sampling rate, the number of supported users can be increased to eight while keeping the individual signal bandwidth unchanged. To meet the link budget requirements, inserting SOA/EDFA for signal amplification can solve the problem of insufficient link budget.
To explore the impact of channels with different powers on other channels during aggregation, transmitter-1 is additionally attenuated by 1.5 dB. Figure 11 depicts BER performance of all four channels when the power of channel 1 is individually attenuated. For comparison purposes, the red line illustrates the performance when all four channels have the same power. It can be observed that the performance of channel 1 deteriorates significantly, while the BER of the other three channels show a slight reduction. This is consistent with the simulation results: channels with lower power have less impact on signals from other channels in the proposed aggregation architecture.
5. Conclusion
In this work, we demonstrate a simple, low-latency and compatible uplink fronthaul with four users. Applying DSM and PSD in uplink fronthaul, four RUs are implemented with commercial pluggable 25G-class WDM optical modules without any analog devices. In the receiver, a single PD is utilized to receive 4 × 380.16 MHz OFDM bands (6.08 Gb/s) at the same time over 20 km fiber transmission without WDM devices. It solves the contradiction between commercial digital devices and multi-channel analog detection schemes, which may facilitate the development of integrated RAN. This study shows the feasibility of radio-over-fiber (RoF) technology utilizing existing integrated optical transceivers. Applying next-generation PAM4 optical modules with higher data rate (e.g. 100GBASE-LR SFPDD single Lambda 1310 nm 10 km with 50 G PAM4 optical interfaces [15]) can further upgrade the modulation format and number of users in the proposed architecture. Moreover, the experimental results with different number of aggregated DUs proves superior functionality and flexibility of the proposed architecture, which brings cost-effective scalability for MP2P connection in the future.
Funding
Innovation Fund of WNLO; National Natural Science Foundation of China (62275091); Science and Technology Planning Project of Shenzhen Municipality (JCYJ20200109142010888).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. J. Li, L. Chen, and J. Chen, “Enabling technologies for low-latency service migration in 5 G transport networks [Invited],” J. Opt. Commun. Netw. 13(2), A200–A210 (2021). [CrossRef]
2. M. Xu, J. Yan, J. Zhang, F. Lu, J. Wang, L. Cheng, D. Guidotti, and G. Chang, “Bidirectional fiber-wireless access technology for 5 G mobile spectral aggregation and cell densification,” J. Opt. Commun. Netw. 8(12), B104–B110 (2016). [CrossRef]
3. X. Liu, “Enabling Optical Network Technologies for 5 G and Beyond,” J. Lightwave Technol. 40(2), 358–367 (2022). [CrossRef]
4. ITU-T G.989.2, 2014.
5. D. Zhang, Z. Du, M. Cheng, M. Jiang, and X. Liu, “Innovation and Demonstration of a New CWDM and Circulator Integrated Semi-Active System for 5 G Fronthaul,” J. Lightwave Technol. 41(4), 1223–1229 (2023). [CrossRef]
6. H. Xin, H. He, K. Zhang, S. B. Hussain, and W. Hu, “Flexible baseband-unit aggregation enabled by reconfigurable multi-IF over WDM fronthaul,” IEEE Photonics J. 10(1), 1–10, Art no. 7200210, (2018). [CrossRef]
7. W. Du, H. Xin, H. He, and W. Hu, “A resource sharing C-RAN architecture with wavelength selective switching and parallel uplink signal detection,” in Asia Communications and Photonics Conference (2015), paper ASu3E.3.
8. L. Zhong, W. Gao, X. Dai, M. Cheng, L. Deng, Q. Yang, and D. Liu, “A simple and low-latency MP2P aggregation architecture for uplink fronthaul,”, in Conference on Lasers and Electro-Optics (2023), paper SF2M.2.
9. J. Wang, Z. Jia, L. A. Campos, and C. Knittle, “Delta-sigma modulation for next generation fronthaul interface,” J. Lightwave Technol. 37(12), 2838–2850 (2019). [CrossRef]
10. F. Effenberger and Y. Luo, “PON for fiber to the room,” in Asia Communications and Photonics Conference (2021), paper T3C.1.
11. L. Zhong, Y. Zou, S. Zhang, X. Dai, J. Zhang, M. Cheng, L. Deng, Q. Yang, and D. Liu, “An SNR-improved transmitter of delta-sigma modulation supported ultra-high-order QAM signal for gronthaul/WiFi spplications,” J. Lightwave Technol. 40(9), 2780–2790 (2022). [CrossRef]
12. 3GPP TS 138.104 v 15.2.0 release 15, “Base station (BS) radio transmission and reception,” 2018.
13. S. Ishimura, H. -Y. Kao, K. Tanaka, K. Nishimura, and M. Suzuki, “IF-over-Fiber Transmission of 40×400-MHz 256QAM OFDM Signals Using Commercial 100Gb/s EML TOSA for Analog Mobile Fronthaul Networks,” Opto-Electronics and Communications Conference (2020), pp. 1–3.
14. M. A. Fernandes, B. T. Brandão, A. Lorences-Riesgo, P. P. Monteiro, and F. P. Guiomar, “5G-compatible IF-over-fiber transmission using a low-cost SFP-class transceiver,” IEEE Access 10, 24601–24610 (2022). [CrossRef]
15. https://resource.fs.com/mall/doc/20230706175155q3dx0 t.pdf
16. N. Shibata, T. Murakami, K. Ishihara, T. Kobayashi, J-i Kani, J. Terada, M. Mizoguchi, Y. Miyamoto, and N. Yoshimoto, “256-QAM 8 wireless signal transmission with DSP-assisted analog RoF for mobile front-haul in LTE-B,” in OptoElectronics and Communication Conference and Australian Conference on Optical Fibre Technology (2014), pp. 129–131.
