目录
1.PON标准中单波50G（ITU-T） 相比于 2*25G（IEEE）的优势
2. 50G PON 使用NRZ 不使用PAM4 因为功率代价
3.突发接收机的问题 接收到的信号功率不同，需要大的动态范围和收敛快的均衡器
4.线性FIR滤波器的缺点 增加高频部分噪声，不能均衡窄带限
6.DFE比FFE性能好 因为，DFE是非线性的，而且DFE能够去掉当前符号中由之前符号引起的ISI
7.FEC-threshold 400G-PSM4 2e-4
18 通过改变调制格式改变系统速率的缺点，速率不连续，因此要用概率整形
19 batch normalization layer 的作用
1.PON标准中单波50G（ITU-T） 相比于 2*25G（IEEE）的优势
Compared with the IEEE’s wavelength stacked 2∗25G solution, the ITU-T’s single carrier non-return-to-zerobased 50G TDM-PON has the advantages of low cost, easy operation, and convenient management. Although the IEEE task force’s objective also includes 50G [5], it is based on a solution that stacks two fixed wavelengths, which will add more cost to the system and more maintenance difficulties. As a counterpart, ITU-T also began to standardize 50G TDM-PON, but with a single carrier in early 2018
DSP enabled next generation 50G TDM-PON
2. 50G PON 使用NRZ 不使用PAM4 因为功率代价
当速率增加时，要维持大的功率预算成了问题，对于NRZ信号，50Gbps比25Gbps的灵敏度降3db（理想情况），25GBaud的NRZ和PAM4，灵敏度代价是4-6dB，所以50Gbpsd的PAM4信号比50Gbps的NRZ灵敏度代价为1-3dB，因此说50G NRZ wins against PAM4
When the bit rate increases, the large power budget is an obvious issue. That is why 50G non-return-to-zero (NRZ) wins against 25 GBaud four-level pulse amplitude modulation (PAM4) as the adopted modulation format.
DSP enabled next generation 50G TDM-PON
3.突发接收机的问题 接收到的信号功率不同，需要大的动态范围和收敛快的均衡器
The BM receiver in the OL T receives optical burst signals from different optical network units (ONUs) that have different optical power levels and initial phases. As a result, not only is a BM-TIA with high dynamic range essential to overcome the different power levels, but also fast convergence in both the BM-TIA and the equalizer is required to reduce the BM overhead and improve the US traffic efficiency
4.线性FIR滤波器的缺点 增加高频部分噪声，不能均衡窄带限
While linear feed-forward equalization is an effective approach to mitigate the distortions caused by low-bandwidth devices, it also enhances the high-frequency noise, which limits its application in systems with severe bandwidth-narrowing effect.
5. 色散 + PD =非线性失真 为什么
Chromatic dispersion can be easily compensated for under coherent detection with an analytical formula, but it becomes nonlinear distortion with an intensity modulation and direct detection system.
6.在带限系统中MLSE比FFE要好
Figure 4 is the simulated bit error rate (BER) performance when typical 25G-class optoelectronics are adopted. The transceiver bandwidth is simulated by a 5thorder 18 GHz Bessel filter. It is shown that an obvious BER floor is observed when only a linear feedforward equalization (FFE) is adopted. By contrast, MLSE can remarkably improve the sensitivity by more than 2 dB at the BER threshold of 1×10−2. Therefore, MLSE is an effective approach to alleviate the noise enhancement of FFE in bandwidth-limited 50G-PON systems.
6.DFE比FFE性能好 因为，DFE是非线性的，而且DFE能够去掉当前符号中由之前符号引起的ISI
Functionally, the feedback filter is used to remove that part of the ISI from the present estimated symbol caused by previously detected symbols. Since the detector feeds hard decisions to the feedback filter, the DFE is nonlinear.
7.FEC-threshold 400G-PSM4 2e-4
7.1 2e-4
Assuming the currently standardized KP4 (RS(544,514,10)) FEC-threshold with a pre-FEC BER-limit of 2E-4 [5], (visualized as a solid black line in the plots) only for the optical back-to-back (b2b) case BERs below the threshold are obtained.
[5] “IEEE P802.3bs/D1.2 draft standard for ethernet amendment: Media access control parameters, physical layers and management parameters for 400 Gb/s operation,” 2016. [Online]. Available: http://www.ieee802.org/3/bs/
Experimental Demonstration of 84 Gb/s PAM-4 Over up to 1.6 km SSMF Using a 20-GHz VCSEL at 1525 nm
7.2 3.8e-3 6.7%FEC
Z. Wang, “Super-FEC codes for 40/100 Gbps networking”, IEEE Communications Letters, vol. 16, no. 12, pp. 2056–2059, 2012.
ITU -T Recommendation G.975.1, 2004, Appendix I.9.
7.3 15% FEC
《ommunication: A Comparison of Deep Neural Networks and Volterra Series》但悲伤的是，文中没有给出误码率阈值
“Open ROADM MSA 3.01 W-Port Digital Specification (200G-400G).” [Online]. Available: http://openroadm.org/download.html
7.4 20% FEC
F. Pittala, I. Slim, A. Mezghani, and J. A. Nossek, “Training-aided frequency-domain channel estimation and equalization for single-carrier coherent optical transmission systems,” Journal of Lightwave Technology, vol. 32, no. 24, pp. 4849–4863, 2014.
7.5 5.2*10-5 KR4 FEC
First Real-Time 400G PAM-4 Demonstration for Inter-Data Center Transmission over 100 km of SSMF at 1550 nm
8. FFE训练数据的描述
For all the evaluated equalizer combinations, training symbols are used at startup. In principle, the equalizer could also work in a completely blind mode [17], however, we experienced a more stable performance of the equalizer with training symbols, especially in case of strong distortions. The first 5000 samples of the received signal are used as training symbols and afterwards the equalizer is switched to a blind adaptive mode with a hard decision in the equalizer.
Experimental Demonstration of 84 Gb/s PAM-4 Over up to 1.6 km SSMF Using a 20-GHz VCSEL at 1525 nm
9 选用FPGA部署CNN 而不是 ASIC的理由
While ASIC can allow full customization and small form factor, it leads to complex design cycle, high initial investments and lack of reconfigurability which is especially crucial given the rapid improvements in CNN architectures.
A survey of FPGA-based accelerators for convolutional neural networks
10 初期阶段
in its infancy / In the early, nascent period of development or implementation / In the initial stage of design
11 误码率描述
The system is able to push 3 bits per channel use and deliver 12.28 Gb/s at a long-accumulated counted BER of 4×10−5.
12 有公式，但是非常难解
Such systems, referred to as auto-encoders (AE), are particularly suitable for communication over channels where the optimum transceiver is unknown or its implementation is computationally prohibitive.
13 端到端网络仿真比实验差 （这句话怎么写得这么好。。）
The performance of transceivers learned on a specific model assumption is often deteriorated when applied to an actual transmission link.
(学弟说迁移学习中这种表述很常见。。fine，下次多看看计算机方面的论文)
14 神经网络拟合信道的表述
mimick some ground truth distribution of the data
15 数据中心光模块400GE接口标准
数据中心光模块400GE 的接口标准已经确定采取4 路并行56Gbaud 的PAM-4 调制信号，并且一般对系统进行性能评估也往往利用PAM 信号完成
16 夸一项技术对于光通信的贡献 提高覆盖范围 吞吐量
For fiber communications, PS was highlighted as an effective method for reach and throughput increase as well as rate adaptation.
《End-to-End Learning in Optical Fiber Communications: Experimental Demonstration and Future Trends》
17 8倍上采样的波形可以准确描述非线性 在AE中
A waveform up-sampled by a factor of 8 was considered to accurately capture the nonlinear effects.
《End-to-End Learning in Optical Fiber Communications: Experimental Demonstration and Future Trends》
18 通过改变调制格式改变系统速率的缺点，速率不连续，因此要用概率整形
However, entropy tuning using uniform distributed signals is stepwise and improvement is hampered by the large sensitivity difference between different modulation formats.
《Probabilistic and Geometric Shaping for Next-Generation 100G Flexible PON》
19 batch normalization layer 的作用
The power normalization is performed with batch normalization (BN), while disabling scale and shift operations.
《End-to-End Deep Learning for Phase Noise-Robust Multi-Dimensional Geometric Shaping》
20 相位噪声来源
In coherent optical communications, the PN may come from laser spectrum linewidth, fiber nonlinearity, imperfect phase recovery, etc.
《End-to-End Deep Learning for Phase Noise-Robust Multi-Dimensional Geometric Shaping》
21 内存与性能之间有折中
It can be observed that for each memory size an architectural trade-off between complexity and representation capacity exists.
《Soft-Demapping for Short Reach Optical Communication: A Comparison of Deep Neural Networks and V olterra Series》
22 实验与仿真参数不一样
The lower data rates used in the experimental demonstration presented here in comparison with those presented in the simulation studies in Part I (50 Gb/s) is due to the capabilities of the equipment available.
《Linearisation Method of DML-based Transmitters for Optical Communications, Part II: Experimental Demonstration and Implementation Methods》
23 通信支撑的应用
4k/8k HDTV(高清视频)，VR edge computing, blockchain technology, and internet of things
24 为什么不用监管层信息而使用码型识别 因为有可能中断
Though modulation format information can be notified via the supervisory-control layer of the EONs, accidental disruption of the supervisory channel may induce the overdue key information acquirement.
25 由直检系统转向相干系统
To meet the rapidly increasing bandwidth/data demand in long-haul communication links, coherent solutions became the standard technology, offering high capacity with high transmission margins. Moreover, they started to supersede the direct detection solutions in core networks.
26 功率预算标准
PR 10 20dB
PR 20 24dB
27 选用一个基础算法的理由
In this work, we focus on VSM, which is the foundation of many state-of-the-art IR-based bug localization approaches and has been shown to outperform many other models.
28 TDM PON难以继续提高速率
For example, with FDM it is relatively simple to increase capacity, but with TDM PON, the requirement for fast transceivers causes an issue with the demands of continuous traffic growth. It is not simple to build TDM PON with coherent technology, and it is not possible to exploit advanced DSP techniques, e.g., to improve the reach. This is an important limitation, and it is exacerbated by the burst mode transmission typical of this architecture. DSP algorithms can be severely impacted during the learning phase. For example, the tracking of polarization must be carried out for each burst.
In addition, as discussed in Section IV -B, if the leaf modules are placed at different geographic locations in the aggregation stage, a TDM approach might suffer synchronization issues during the uplink10, which cannot be compensated by a proper DSP. Moreover, burst transmission induces transients on the electrical current and on the laser output power and affects the laser stability.
Point-to-Multipoint Optical Networks Using Coherent Digital Subcarriers
29 使用DSP的理由
Advanced DSP technology is a prevalent choice to enhance performance while maintaining low costs.
30.使用低带宽器件，不使用50G APD
Although 50G APDs have been demonstrated in research, their limited obtainability and maturity necessitate the investigation of using limited receiver bandwidth approaches for 50G PON by using 25G APDs.
Strategies for economical next-generation 50G and 100G passive optical networks [Invited]
31.达到功率预算的发送和接收功率要求
Roughly speaking, such a loss budget corresponds to a receiver sensitivity of ≤ −25 dBm at a given pre-forward error correction (pre-FEC) threshold bit error rate (BER) at a launch power of 4 dBm to co-exist with the previous PON generations
PON transceiver technologies for ≥50 Gbits/s per λ: Alamouti coding and heterodyne detection [Invited]
32. RNN比LSTM好的原因
For a long time, RNN was thought to be theoretically fascinating. Unlike standard feed-forward networks, RNN can deal with arbitrary input sequences instead of static input data only. This combined with the ability to memorize relevant events over time makes RNN in principal more powerful than standard feed-forward networks.
33.RNN的问题
The extent to which this potential can be exploited, is however limited by the effectiveness of the training procedure applied. Gradient based methods share an important limitation.
34.IMDD系统中的概率整形需要加上FEC
When the PS is applied to IMDD system, some of the common techniques and assumptions used in coherent optical systems are no longer valid therefore modifications are required. While coherent system has access to phase modulation and FEC parity bits are often accommodated in the sign bits of the constellation mapping, the optimal distribution for IMDD system with optical amplifier often asymmetric, hence inserting FEC parity bits is not trivial.
《First Experimental Demonstration of Flexible Rate PON Beyond 100Gb/s with Probabilistic and Geometric Shaping》
35.单波50G PON标准
In 2018, the ITU T elecommunication Standardization Sector (ITU-T) Question 2/Study Group 15 (Q2/SG15) launched the process to standardize a single wavelength 50G-PON
“Optical systems for fibre access networks,” ITU-T Q2/SG15, https://www.itu.int/en/ITU-T/studygroups/2017-2020/15/Pages/q2. aspx.
36.DFE的优势
After adding DFE, the noise enhancement induced by FFE can be compressed so that the receiver sensitivity is improved.
《100G PAM-4 PON with 34 dB Power Budget Using Joint Nonlinear Tomlinson-Harashima Precoding and Volterra Equalization》
扫一扫
412
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
