How Netflix and YouTube are killing fibre optic broadband as we know it; and how algorithms are here to keep us connected.

Photo by Jens Kreuter on Unsplash

Thats the clickbait title out of the way. The real title should read: How machine learning (ML) and artificial neural networks (ANN) will come to dominate the broadband market and how agile players can get there first.

Our consumption of Netflix, YouTube and almost all other sources of data-intensive online content is heavily reliant on how that data is delivered to us. This means the story of Netflix and YouTube is now fundamentally intertwined with the story of fibre optic broadband.

To understand how the internet is instantly delivered to our fingertips it helps to understand how raw data is actually transmitted in the first place. Almost all of the data borne by the Internet today – irrespective of the final leg of transmission being wireless – is transmitted over fibre optic cables. These cables consist of small glass fibres that transmit light (data) along its axis by a process called total internal reflection.

This essentially means that data signals can travel over extremely long distances at near the speed of light.

However the growing popularity and our ever growing consumption of these streaming services is resulting in sustained increases of data traffic which is putting our infrastructure (the cables themselves) under a huge amount of pressure to meet the demand. Existing fibre optic technology is currently unable to meet the projected future data needs and requirements of society. The limiting factor here is the infrastructure itself. Specifically, as cables are being laid across more and more rural terrain and across further and further distances, the actual data signals transmitted along the axis of these cables become distorted, losing actual data and therefore quality as is travels. With these ever longer cables we are finding ever more distortions.

This isn’t a new discovery and some scientists have found a solution that partially solves the problem. Their solution is to use software coding to digitally correct the distortion. The issue they have found with this though, is the that a fairly significant proportion of the distortions are completely random and therefore extremely difficult to pinpoint and address.

One approach to this has been to utilise machine learning, in particular artificial neural networks (ANN)’s which are models that simulates the complexity and processing power of the human brain. The key differentiator for the use of a neural network as opposed to a more typical machine learning model is that we do not have train the model, it learns by itself. You simply expose it to the data and et voila.

So, how can ANN help our internet networks?

Photo by Taylor Vick on Unsplash

The use of an ANN provides a complex and advanced statistical approach coupled with a powerful ability to deal with randomness (the random distorted signals in the fibre optic cables). It achieves this by training the signals itself. This is a key feature in ANNs, which are dedicated solely to making predictions based on the feeding of data into the model and permitting the ANN itself to learn more about the processed information. As with any brain it grows as it learns and whilst an ANN also functions as a digital fix to the distortion problem – programmed in electronic telecom modems, using this training process to gather and assimilate historical data about the fibre optic network – it learns about network performance impairments, building a probabilistic model.

Using this probabilistic model as a guide, its artificial neurons are then responsible for deciding if and how distorted data signals can be repaired, essentially operating as multiple ‘digital filters’. Another benefit of using an ANN is that whilst the field of machine learning and neural networks may be niche given the level of understanding required — it is fundamentally less complexity (at least in terms of integration) than other technological techniques, therefore making it a financially cost effective solution.

Whilst the commercialisation of this solution is still in its infancy, the concept itself has been proven to work in a research setting.

The potential impact of this work and further developments in ML could be far reaching, particularly in the connection of rural areas by laying longer and longer fibre optic cables without the need for additional hardware.

ML is also expected to be key to sustaining a low-latency network aka 5G. Additionally, ML will be fundamental to enabling the two-way communications needed for driverless cars.

Though ML has been implemented already in certain areas, such as in medical diagnosis, robotics, and online marketing, its application to the world of telecoms will be more challenging because of the time-critical nature of typical communications: ML has only fractions of a second to respond when an issue is found e.g. financial transactions on a market such as the London Stock Exchange. It is for this reason telecoms companies should be investing in their long term futures and not existing to remunerate shareholders on a quarterly basis.

Venture Investor // Practice Lead @ Radically Digital