Contact Persons
Mehdi Amirijoo +46 73 0430445
mehdi.amirijoo@ericsson.com
Master Thesis – Machine Learning for 5G Networks
Background
Machine learning constitutes a set of algorithms that can learn from and make predictions on data. Such algorithms operate by building a model from inputs in order to make predictions or decisions, rather than following strictly static program instructions. These set of algorithms have successfully been applied to various applications such as computer security, bioinformatics, computer vision, medical diagnosis and search engines. Common to all these fields is the need to automatically process large sets of data in order to generate useful insights and take appropriate decisions.
Mobile networks are complex by nature and it is expected that the next-generation 5G mobile communication systems will be even more complex. They will need to handle a broader set of scenarios, not fully addressed by current cellular systems, such as massive deployment of ultra-low power sensors, intelligent traffic systems, critical low-latency communications, enterprise networks, etc.
To handle this complexity there is a need to deploy intelligent methods for analysing 5G network data.
Such methods need to reduce efforts for network management, i.e. offloading human effort needed to operate these networks, be able to draw new insights, and predict future network and user behaviour in order to make smarter decisions. This will result in higher network performance, better reliability and more adaptive systems.
Thesis Description
This thesis work will investigate machine learning methods for improving 5G network performance by automatically tuning key network parameters to their optimal values. In particular for 5G systems, the usage of beam forming is a central technology component for increasing data rates and coverage. Beam forming consist of each base stations ability to communicate with users through a set of defined beams pointing in different directions. Not all beams are active in the network, e.g. in order to reduce energy consumption and interference. In this context, there is a need to decide which beams should be (de)activated as the user moves around the network. We would like to evaluate the application of machine learning techniques for addressing this problem using supervised and unsupervised learning as well as reinforcement techniques. Due to limited computational capacity, it is preferred that machine learning techniques with low complexity are used. Evaluation will be conducted using a 5G simulator.
Contact Persons
Mehdi Amirijoo +46 73 0430445
mehdi.amirijoo@ericsson.com
Qualifications
This project aims at masters students in computer science, computer engineering, or electrical engineering, with strong background in mathematics/statistics. Excellent Java and Python/Matlab programming skills is a must. Background in machine learning is preferred. Successful candidate has average grade above B/4.0.
Extent
1 student, 30hp
Location
Linköping, Östergötland, SE
Preferred Starting Date
January 2016
Keywords
Machine Learning, Data Mining, Data Analysis, Statistics, 5G, Beam Forming
