Satellite operators need to transform from capacity providers to service providers, offering their customers an entire package. AI has the potential to help. Credit: Courtesy of the Satcoms Innovation Group

This op-ed originally appeared in the June 10, 2019 issue of SpaceNews magazine.

Satellite communications is under threat and in danger of losing its relevance. Fiber connections continue to roll out across the world, giving consumers fast and reliable connectivity even in some of the more rural, previously unconnected areas. With communities continuing to put pressure on governments and councils, that is likely to continue over the coming months and years. At the same time, consumers are viewing more and more content on mobile phones and pay TV subscriptions are starting to decline in many areas of the world. Given that broadcast is one of the biggest customer groups for satellite operators, it is not surprising these changes are having an impact on their bottom line. At the same time, the mobile industry continues to set its sights on gaining valuable C-band spectrum from the satellite industry.

As the world becomes more connected, many don’t see a need for satellite, yet we must remind ourselves that the global communications network cannot survive without satellite communication. How can satellite services survive? Could artificial intelligence (AI) be the answer?

The satellite challenge

The threat from other services and network complexity is where the boundaries between terrestrial and satellite become ever more blurred. This means that it is more important than ever for teleports to be well managed and ensure performance and reliability are second-to-none. We have seen a real difference in the way teleports are operated, whereby operators need to have the ability to quickly grow the number of dishes, but also be able to swap dishes out at a moment’s notice and ensure resilient redundancy switching — all of which is geared toward ensuring constant connectivity.

Not only is the satellite industry under threat from other communication methods, it is also under threat from itself. Space is congested and getting more so. We already have a colossal problem with debris, not unlike the plastic filling the seas. In fact, a collision in 2017 generated an estimated 3,800 fragments. We have had others since, adding vast quantities of debris to the equation. If there were a collision now at GEO, the entire belt would be affected within 24 hours.

However, it is about to get a whole lot worse. OneWeb plans to launch 720- plus satellites, Telesat is planning 117, and SpaceX a further 4,425 satellites. Others plan to launch hundreds each. With that many new satellites about to launch, how do we manage the RF environment? This is going to have a significant impact in the number of LEO satellites in view. Currently, we might see one satellite every four to five hours. By the time the SpaceX satellites have all launched, we might see 20 in one view at a time. That will mean the existence of interference, as the more satellites transmit at the same frequency range, the more noise will exist. This may potentially also lead to a reduction in quality of service delivery via geostationary satellites for the same frequency bands.

When we look at LEO launches, we need to bear in mind that many will be launched by nontraditional satellite companies, which may not be so well versed on the best practices followed by most satellite operators to avoid risk of collision and other issues. Also, LEO operators will only see their satellites four times a day instead of them being in constant view. So, how can we make sure they remain under control the rest of the time?

AI: Is it the answer?

Satellite operators need to transform from capacity providers to service providers, offering their customers an entire package. AI has the potential to help the satellite industry transform itself and with that perhaps save itself, too. Yet it does come with its own set of challenges that will need to be overcome if it is to be successful.

SpaceX, which deployed 60 Starlink satellites May 23, intends to put up more than 4,000, creating challenges for satellite operators managing an increasingly crowded RF environment. Credit: SpaceX via Instagram

AI is about using data and machine learning to teach a system certain automated tasks and error resolution. It is not new. In fact, it has been around since the late 1950s. However, what we have now is data, and in the satellite industry we definitely have lots of it, enough to make us the envy of pretty much every other industry. Used well, it can reduce the cost of today’s network management by enabling much faster resolution of incidents. It should be viewed as a really helpful support for satellite engineers. What it will never be is a replacement to those engineers, as many decisions will still need the benefit of their wealth of experience. AI results may often not be clear enough to let the machine decide, as getting zero percent or 100 percent probability with AI will never happen.

Use cases for AI

The potential use cases for AI in satcoms are vast, as in most industries. We have already heard of some being tested which have the potential to make a massive difference for operators. Here are just a few examples of potential use cases:

Interference Detection: Gathering data from base stations to classify interference so the system can recognize it when it happens and raise an alert. With more data, this could go one further and suggest possible ways to resolve a particular type of interference. This has been tested successfully by CTTC, a research center in Barcelona.

Interference Prediction: Gathering data from ticketing systems for past incidents that have happened and correlate it with future fixed events in the calendar to predict potential interference cases. Pilot projects are already started to find the right mathematical correlation functions for machine learning.

Anomaly detection in telemetry data: As with interference detection, this is about teaching the system to recognize and detect anomalies in the telemetry data. This is another use case being researched by CTTC.

Future flexible satcom systems: AI could allow prediction of how the radio sources will be requested. Or in the case of satellite broadcast, it could help predict what type of content might be requested so that it can be moved ready for broadcast.

Putting intelligence to VSAT measurement data: Using VSAT measurement data, AI could determine what each station costs, how well they are performing, whether they are well installed, etc. This would ultimately enable VSAT operators to increase productivity.

Cybersecurity: This could be a significant use case for AI. A system is only secure until something new comes in to exploit weaknesses. As satellite networks become virtualized, this will be even more of an area for concern. AI can enable effective monitoring of the security of satcoms systems and raise an alert should a breach happen or be likely.

Collision Avoidance: If we feed operator ephemeris data and data from public sensor networks into an AI system, could we let that calculate the probability of collision and alert operators when a close approach is imminent? Currently Space Situational Awareness is done by multiple organizations, including the Space Data Association, but it is clear this needs to improve to secure the future of the space environment. Perhaps AI can help resolve that.

Predicting Orbits: Could we use AI to predict the path of a satellite once we lose visibility, either in the case of an error or when LEO satellites are out of view and control from a gateway station?

How do we get there?

There are several things the satellite industry needs to do to get to an AI-enabled world.

1. Virtualized networks

We need to virtualize the ground station. If RF is converted into digital, it can be sent to cloud storage where software modems can be applied. There are already well over two hundred satellites being flown with digital infrastructure. And this is great, because digitizing the system and getting data from that into the cloud means more can be done with that data.

2. Data Sharing

There are companies and organizations emerging with the capability to process the data and turn that into real-life use cases to improve the efficiency of satellite communications. What they need is the data, something the satellite industry has in abundance but has historically been reluctant to share. The more data we are willing to share with these types of organizations, however, the more the industry can reap the benefits.

It is also important to decide whether to run an AI task force within the company or have it outsourced – including sharing relevant corporate data with industry partners. This is a significant dilemma between passing business-crucial data along to entities outside of any company or building up a local company infrastructure team to host AI and machine learning tasks internally. It is a decisive question to take with regards to the economic outlook for the satellite communication environment.

3. Math skills

We need math to enable AI and a whole different skill set. RF engineers know satellites and are needed to make a final decision on a course of action. But we also need data scientists and mathematicians to work with the data, run the AI and assess the data effectively. This could be in the form of employing a different type of staff with this skill set or it could be outsourcing to those companies that can process the data effectively.

4. Cybersecurity

While cybersecurity is one area that can be solved with AI, it is also a bigger threat because of AI. How secure are the algorithms? Could someone hack into them and change behavior? If a satellite is attacked, it would not be a question of simply bringing it down. As networks become more virtualized, it is important that we ensure networks are as secure as they can be but also that there are contingency plans in place ready for when an attack does happen.

Conclusion and outlook

Satellite communication companies gather millions of datagrams every day from sensors on board their satellites and in their ground networks. The reason is traditionally very clear. We need more sensors when operators are unable to look at their machines. No sensor means no data, no alarm, and no notification if something goes wrong.

Dealing with Big Data is something common to finding the correct way forward for operational and commercial decision making. So it is the next logical step to go for machine learning and AI as it can make use of the big data repository satellite communication operators already possess.

It is not a question when it will be integrated industry-wide — because it will be done naturally in the development of each different company. The question is rather how and in how much depth it will be done — and that is a decision that each company needs to answer for itself according to its needs and requirements.

At the end, the machine will give a proposition that is either a ‘YES,’ a ‘NO’ or a ‘CAN’T DECIDE.’ This is not dependent on the algorithm used, but rather on the questions asked and the answers recorded, and – most important – on the data quality used as input. The lower source data quality is, the more ‘CAN’T DECIDE’ events will occur – potentially erasing efficiency gains.

Hence, it is not only important that AI/machine learning design instances, sources, result quality management and its impact is understood by technicians and engineers who will work with AI systems daily, but also by their managers and the company’s senior decision makers.

Martin Coleman is the executive director of the Satcoms Innovation Group.