Optimizing Networks with Machine Learning

With the adoption of software systems and with the increase in data processing, the new 5G system can use existing and new machine learning algorithms to optimize its functioning. Specifically, of interest are three main directions:

1) Optimizing the usage of the system by the end devices – in a 5G system the end devices will not be any more equally treated, as requested by the consumer as well as by the vertical markets.

Machine learning enables to determine and optimize the subscriber mobility patterns and their QoS usage as well as to predict network congestions at specific locations.

With this, a better scheduling of the control plane procedures and of the allocation of the resources is possible, ultimately providing a better network service with a significantly lower resource consumption.
This image has no alt text.
Fraunhofer FOKUS
2) Optimizing the network management – machine learning is providing additional insight into the functioning of the mobile network, providing additional capabilities for fault, performance and security management.

By monitoring a large number of parameters across a longer duration of time, the following optimizations may be considered.
    • Fault discovery and mitigation solutions in highly complex systems
    • Performance degradation and anomaly detection
    • Simplification of the security policies
    • Determining network elasticity and scaling patterns
This image has no alt text.
Fraunhofer FOKUS

3) Reduced human interaction – albeit a certain human interaction will still be needed to overlook the network and for setting the governance policies, there is a stringent need that these operations will be executed without requiring very specialized, highly educated experts (e.g. you don’t need to be able to repair a gear box to be able to get a driver license).

Machine learning is one of the main technologies which enable to expose the most appropriate parameters (even by considering multiple administrative perspectives), in order to be able to enable the current network administrators to easily manage even more complex and more automated networks.