Mobile ad-hoc networks (MANETs) are comprised of nodes that can configure themselves autonomously to provide communication services without relying on a pre-existing infrastructure. In the absence of fixed infrastructure, MANETs must adapt quickly to topology changes resulting from node mobility and to channel condition changes due to signal attenuation (path loss and interference effects). Traditional MANETs use omnidirectional antennas which ease the implementation as they cover a 360-degree area. The downside is that omnidirectional antennas inherently create interference which decrease the capacity of the system. There is therefore a strong interest to utilize directional antennas to form P2P wireless links between nodes. In addition to significant capacity gains and extended ranges, smart antennas also reduce undesired interferences, effectively increasing spatial diversity between transmitters.

Topology control is a fundamental problem in MANETs with directional antennas and is especially useful for constructing and maintaining networks with desirable properties. In essence, the problem amounts to selecting the P2P links that should be established and configuring them so as to maintain a connected topology in spite of node mobility while minimizing interferences and optimizing network performance (e.g., by minimizing the diameter of the network or maximizing its total capacity). The configuration of the selected links does not only require coordination of antenna steering at both the receiver and transmitter end of each one, but also the determination of the appropriate frequency and transmission power for each of them.

This internship will explore the potential of machine learning techniques for improved topology control in directional antennas-equipped MANETs. The design of learning-based algorithms for topology control will be based on an exact formulation of the problem as a convex optimization problem, using the concept of conflict graph to identify interfering links. The research avenue we intend to explore amounts to devising an efficient heuristic to solve the problem, and then to use a Deep Feedforward Neural Network for learning the input-output relationship of this heuristic. Network simulations in representative scenarios will be used to validate the proposed schemes and to study their impact on network performance.

Host laboratory: the internship will be carried out within the SARA team of LAAS-CNRS, Toulouse, France. It will be supervised by Olivier BRUN and Balakrishna PRABHU.

Starting date: 03 February 2025.

Application material: should include a CV, a motivation letter, and the names of one or two persons who can write a recommendation letter for the candidate.

Contacts (for sending the application material or for queries): Olivier BRUN (brun@laas.fr)