People Story

AI sees what you don't

We've probably all experienced it: We're in the middle of a voice call far from home and suddenly we hear the familiar beeps. The other person is gone and the call is disconnected. To keep these annoyances to a minimum and ensure that mobile phone users can benefit from high-performance networks, they need to be thoroughly tested. The classic approach: Conduct measurements in fully developed networks and analyze the quota of dropped calls. But that is not only time-consuming. It is also costly.

To find a more effective way to analyze the reliability of wireless connections, Rohde & Schwarz utilizes artificial intelligence (AI) methods. In 2018 we established the Data Intelligence Lab which explores ways of simplifying the optimization of mobile networks, among other issues. One of the people behind this Lab is Miguel Angel Román. Miguel is a software development manager and has been working at our Alicante office for over three years. In this interview he explains the advantage of using AI methods and the skill he considers most important in AI programming.

Miguel Román talking about AI

"Deep learning has the potential to change every sector of the economy."

Miguel Angel Román

Miguel, you spend every day looking at how AI can be used to optimize mobile networks. What is special about AI methods that makes them a good fit for Rohde & Schwarz?

Our machine learning methods target what we call "guided optimization" of mobile networks. By taking into account the temporal dimension of testing data, they expose strange behavior in the network, for example sudden drops of KPIs or glitches that affect network performance. This would be very difficult to detect using other standard statistical methods. Moreover, our methods also help to benchmark different operators by scoring their tests based on a learning process that examines hundreds of thousands of other tests performed under multiple network conditions. These scores obtained through machine learning measure test performance under existing network conditions, which is unique and provides a new way to compare test results.

What is the advantage of AI compared with human analysts?

Mainly cost saving, as it would reduce the long time it would take for a human analyst to revise thousands of hours of mobile testing data in order to extract the rules that would either score a test or detect an abnormal network condition. With machine learning, these rules are automatically learned from data. The human analyst can then focus solely on the exposed tests for further analysis in order to find a good solution.

We're interested in your career path. How did your first encounter with AI come about?

A few years ago, I started to see AI based applications capable of doing things that I had no clue of how to make happen using conventional software. Soon I realized I would become obsolete as a software engineer if I did not learn more about AI. I enrolled in a nanodegree from Udacity, attended a deep learning summer school, and started my Ph.D. in AI applied to music.

What aspects of AI do you find most fascinating?

I am particularly fascinated by the AI subfield of deep learning, with which we can create software applications that were previously unthinkable. Deep learning can be seen as a huge leap in the level of automation that we can achieve with software. Since software and data are everywhere, deep learning has the potential to change every sector of the economy.

It's part of our company philosophy to give our employees as much freedom as possible to make their ideas a reality. What do you personally like the most about Rohde & Schwarz?

I am a huge advocate of this culture, which shows trust in employees and creates a healthy working environment in return. I particularly appreciate the opportunity to work from home, which helps me to concentrate better at work and keep up with new emerging technologies.

And finally, we'd like to ask for your advice for everyone who would like to develop AI methods themselves. What skills would they definitely need?

Understanding what is possible with deep learning and what is not, is crucial for building AI applications that are robust and useful.

Thank you for the very informative interview, Miguel!