500 Gbit / s - record speed in fiber optic networks
Engineers from Germany managed to achieve a record speed of data transmission over fiber in real, not laboratory, conditions - 500 Gbit / s in one channel.
/ Flickr / tony webster / cc by
According to the OECD , in three years the number of Internet of things devices can reach 50 billion. With the increase in the number of gadgets, the volume of traffic in mobile networks will also grow - by some estimates , about four times. Deloitte says that the existing fiber-optic infrastructure, which will become the basis for 5G networks, will not cope with such a load.
')
For this reason, more and more companies and research organizations are working on technologies that increase the throughput of "optics". One of these organizations is the Munich University of Technology (TUM). Its employees, five years ago, developed an algorithm for probabilistic formation of a signal constellation - Probabilistic Constellation Shaping, or PCS (we will tell about it in more detail later). In 2016, with its help, it was possible for the first time to achieve the terabit data transfer rate in the laboratory.
In February of this year, the same group of scientists set another record - they carried out data transfer at 500 Gbit / s, but did so in “field” conditions. For the tests, the signal processor Nokia PSE-3 was used, which was introduced into the network of the German operator M-Net.
PCS is a method that complements quadrature amplitude modulation (QAM) in fiber optic networks. In the classical case of QAM, all points (signal amplitude values) have equal weights and are used with the same frequency.
The PCS algorithm, developed by engineers from TUM, each time selects the optimal group of points that is best suited for the current state of the channel. For each of the constellation points, the probability of data distortion and the value of the energy required to send a signal are calculated. The less distortion of the message and energy consumption, the more often the specific amplitude is used . How often the constellation point is used is determined by the probability distribution functions. They are derived experimentally for each specific network based on data on the average noise level in the optical channel.
/ Wikimedia / Splash / CC BY-SA / Signal constellation for 16-QAM
Typically, a PSC less frequently uses high amplitude signal points. According to the developers, this allows to increase the stability of the signal to noise and increase the transmission speed. For example, for 16-QAM, the “increase” is from 15 to 43%.
According to the president of Nokia Bell Lab Marcus Weldon (Marcus Weldon), in the future PCS will allow fiber optic networks to transmit large amounts of data and dynamically adapt to current traffic needs (for example, in 5G networks).
The technology is already supported by the Infinera network equipment provider. The company uses probabilistic modulation in ICE digital signal processors. Infinera claims that the devices will be able to increase network bandwidth up to 800 Gbps, but their capabilities have not yet been tested. Representatives of the company say that the technology will help mobile operators and Internet providers to reduce the cost of infrastructure development and the construction of new lines.
But one drawback can affect the popularity of probabilistic modulation: it is poorly optimized to work with existing methods of forward error correction (FEC) in data transmission. FEC methods are designed to ensure that all combinations in the channel are used equally often. In the case of PCS, some constellation points are chosen more often than others, which may affect network performance. To solve this problem , more advanced FEC methods are developed , for example, “parallelize” correction schemes and conduct several checks simultaneously.
/ Flickr / Groman123 / CC BY-SA
There is another type of modulation of the signal constellation - geometric. It differs from the probabilistic in that it changes not the frequency of use of a particular point, but the shape of the constellation. To this end, the phase signal is added to the amplitude modulation of the signal, which allows you to “shift” the points relative to each other. Like probabilistic modulation, geometric helps to achieve more efficient use of the optical channel: the location of points in the constellation is chosen so that in each of them the signal-to-noise ratio (SNR) is maximum.
The advantage of the geometric form over the probabilistic one is the smaller number of possible amplitude values. This feature reduces the chance of signal distortion. However, geometric modulation has a drawback: in practice, it is less effective in reducing signal distortion than probabilistic.
Experts hope to improve geometric modulation using machine learning methods, using them to determine the optimal shape of the signal constellation. The results are not very impressive so far: in the 2018 study, a simple single-layer neural network helped increase the SNR value by one percent. However, engineers plan to continue working and experimenting with recurrent neural networks.
So far, the geometric modulation of the signal constellation loses probabilistic when working in real networks, and therefore the latter is considered the most promising method of increasing the bandwidth of Internet channels. It is expected that in the near future, probabilistic modulation will benefit Internet providers in creating high-speed fiber to the home lines, as well as cloud providers, for example, when transferring data between different data centers.
/ Flickr / tony webster / cc by
Who set the record
According to the OECD , in three years the number of Internet of things devices can reach 50 billion. With the increase in the number of gadgets, the volume of traffic in mobile networks will also grow - by some estimates , about four times. Deloitte says that the existing fiber-optic infrastructure, which will become the basis for 5G networks, will not cope with such a load.
')
For this reason, more and more companies and research organizations are working on technologies that increase the throughput of "optics". One of these organizations is the Munich University of Technology (TUM). Its employees, five years ago, developed an algorithm for probabilistic formation of a signal constellation - Probabilistic Constellation Shaping, or PCS (we will tell about it in more detail later). In 2016, with its help, it was possible for the first time to achieve the terabit data transfer rate in the laboratory.
In February of this year, the same group of scientists set another record - they carried out data transfer at 500 Gbit / s, but did so in “field” conditions. For the tests, the signal processor Nokia PSE-3 was used, which was introduced into the network of the German operator M-Net.
How does the algorithm work
PCS is a method that complements quadrature amplitude modulation (QAM) in fiber optic networks. In the classical case of QAM, all points (signal amplitude values) have equal weights and are used with the same frequency.
The PCS algorithm, developed by engineers from TUM, each time selects the optimal group of points that is best suited for the current state of the channel. For each of the constellation points, the probability of data distortion and the value of the energy required to send a signal are calculated. The less distortion of the message and energy consumption, the more often the specific amplitude is used . How often the constellation point is used is determined by the probability distribution functions. They are derived experimentally for each specific network based on data on the average noise level in the optical channel.
Typically, a PSC less frequently uses high amplitude signal points. According to the developers, this allows to increase the stability of the signal to noise and increase the transmission speed. For example, for 16-QAM, the “increase” is from 15 to 43%.
Application and technology potential
According to the president of Nokia Bell Lab Marcus Weldon (Marcus Weldon), in the future PCS will allow fiber optic networks to transmit large amounts of data and dynamically adapt to current traffic needs (for example, in 5G networks).
The technology is already supported by the Infinera network equipment provider. The company uses probabilistic modulation in ICE digital signal processors. Infinera claims that the devices will be able to increase network bandwidth up to 800 Gbps, but their capabilities have not yet been tested. Representatives of the company say that the technology will help mobile operators and Internet providers to reduce the cost of infrastructure development and the construction of new lines.
But one drawback can affect the popularity of probabilistic modulation: it is poorly optimized to work with existing methods of forward error correction (FEC) in data transmission. FEC methods are designed to ensure that all combinations in the channel are used equally often. In the case of PCS, some constellation points are chosen more often than others, which may affect network performance. To solve this problem , more advanced FEC methods are developed , for example, “parallelize” correction schemes and conduct several checks simultaneously.
What we write about in our corporate blog:
/ Flickr / Groman123 / CC BY-SA
Analogue of probability modulation
There is another type of modulation of the signal constellation - geometric. It differs from the probabilistic in that it changes not the frequency of use of a particular point, but the shape of the constellation. To this end, the phase signal is added to the amplitude modulation of the signal, which allows you to “shift” the points relative to each other. Like probabilistic modulation, geometric helps to achieve more efficient use of the optical channel: the location of points in the constellation is chosen so that in each of them the signal-to-noise ratio (SNR) is maximum.
The advantage of the geometric form over the probabilistic one is the smaller number of possible amplitude values. This feature reduces the chance of signal distortion. However, geometric modulation has a drawback: in practice, it is less effective in reducing signal distortion than probabilistic.
Experts hope to improve geometric modulation using machine learning methods, using them to determine the optimal shape of the signal constellation. The results are not very impressive so far: in the 2018 study, a simple single-layer neural network helped increase the SNR value by one percent. However, engineers plan to continue working and experimenting with recurrent neural networks.
So far, the geometric modulation of the signal constellation loses probabilistic when working in real networks, and therefore the latter is considered the most promising method of increasing the bandwidth of Internet channels. It is expected that in the near future, probabilistic modulation will benefit Internet providers in creating high-speed fiber to the home lines, as well as cloud providers, for example, when transferring data between different data centers.
Additional reading in our blog on Habré:
Source: https://habr.com/ru/post/442342/
All Articles