The main objective of the project is to propose a novel approach of distributed, scalable, dynamic and energy efficient algorithms for managing resources in a mobile network. This new approach relies on the design of an orchestration mechanism of a portfolio of algorithms. The ultimate goal of the proposed mechanism is to enhance the user experience, while at the same time to better utilize the operator resources.

You will find here and there two presentations of the project done resp. for the Institut Mines-Telecom workshop on 5G and for the NETLEARN workshop.

More specifically we address in this project the following technical bottlenecks:

  1. Excessive interference is a crucial aspect of todays (LTE) and future (LTE-A) radio access networks (RAN) that prevents mobile users to have a homogeneous quality of service whatever their location and along their movements. This excessive interference is due to the chosen frequency reuse 1 and to the heterogeneity of the network (composed of macro, small, femto cells, and relays). Adopting Inter-Cell Interference Coordination (ICIC) and Coordinated MultiPoint (CoMP) techniques can help reducing interference but radio resource management schemes are required to coordinate these techniques.
  2. Access delay to popular contents may be excessive for users in mobility. Content Delivery Networks (CDN) are in charge of placing contents in surrogate servers and controlling the content distribution. In a mobile environment, classical prefetching techniques may become obsolete because of the user mobility, the variations of its radio channels and of the traffic demand.

The NETLEARN project intends to tackle these two technical bottlenecks by considering distributed learning algorithms with the aim of designing scalable, dynamic and energy efficient solutions. There are however several scientific bottlenecks in this field:

  1. Several such algorithms exist in the literature, each one with its own characteristics in terms of reached equilibrium, convergence speed, stability, etc. Each one can also be tuned by one or several parameters.
  2. Environments we are considering are characterized by stochastic variations (due to variations of the traffic demand, of the radio channels, due to mobility), non-stationary situations (sudden changes/discontinuities of the system).

The NETLEARN project intends to adopt a novel approach to address these issues: First building a portfolio of distributed learning algorithms; Second, proposing an orchestration of this portfolio based on a ‘learn to learn’ approach. Our goal will thus be to devise adaptive learning schemes that select dynamically between different learning schemes so that their long-term learning power exceed the regret of any individual algorithm. Based on this framework, we will propose distributed, scalable, dynamic and energy efficient learning algorithms for managing interference in a RAN, contents and cache servers in a mobile CDN. We will propose possibly architecture and protocols enhancements to existing mobile networks in order to allow the implementation of above proposed algorithms. We will demonstrate the effectiveness of the approach through extensive simulations and a demonstrator (testbed).