True Autonomy Through Artificial Evolution

Our ultimate research goal is to achieve an almost miraculous feat, to build an artificial engineer who can autonomously solve tasks posed by current and future networks, with little or no human supervision. But what is autonomy? Autonomy is the independence to reflect about and adapt a behavior. In other words, to go beyond programming and tackle a problem that is considered the most appropriate: this could be choosing an approach that is most likely to succeed and/or offering the highest reward.

Various machine learning technologies are essential in solving problems today. Deep learning is often used to find anomalies, recognize people, or translate language; however, when AI is applied to optimize a system, it does not necessarily mean that the system is acting autonomously as development and maintenance relies on (human) engineers. Autonomous systems, on the other hand, possess the ability to go beyond their predefined program and learn on their own. Learning, in this sense, must be wider in scope and more powerful in execution than what is currently being deployed in our networks.

Nature has shown how an artificial engineer can come into existence: Evolution. Through semi-arbitrary recombination and modification of small building blocks (genomes) and feedback through survival of the fittest, better mating chances for stronger individuals and so on. Drawing inspiration from this real-life trial and error experimentation of self-evolving systems, evolution is, thus, one of the employed strategies in creating our artificial engineers.

Radio Capacity and Coverage Optimization

The radio access network (RAN) is responsible for transmission, reception, and management of the radio and its traffic and is a key feature of telecommunication networks. With the increasing number of subscribers, RAN is also facing pressures of ever-increasing diversity of system elements, complexity of wireless environment dynamics and the scattering of radio resources. We are exploring the application of machine leaning (ML) techniques to the key and classic challenge of tuning antenna parameters. We target the most challenging dynamic and time varying version of the problem but also apply our approach to other RAN challenges, such as signal detection and channel estimation in massive MIMO, spectrum sensing in cognitive radio, load balancing, coverage and capacity optimization, intercell interference coordination, and cell outage detection.

Mobile Edge Computing

Edge computing, geographically distributed ‘little’ groups of computers, are being embraced by telco operators as an opportunity to reduce service latency embrace data sovereignty. We are taking a two-pronged approach on the research necessary to create effective edge computing platforms. Firstly, though collaboration with universities, we are engaged in research projects that explore the necessary technologies to enable future edge. This includes new transport protocol interaction, programming approaches for both network applications and maintenance, marketplaces for data and service, and adaptive-performance intelligence services. Secondly, as edge computing exhibits many of the issues found in a full network, it is a meaningful use case domain that our overall research efforts on autonomous networks can be applied.