Workshop on Wireless Communications with limited feedback

Optimization and control when shifting from full to bandit feedback

Positioning

This workshop provides an opportunity to discuss how the tools provided by learning theory could be used in deriving optimal adaptive communication algorithms, which also reduce the amount of control messages exchange. Therefore, this workshop aims at bringing together the wireless communication community with that part of the mathematical and computer science communities working on learning algorithms.

Scope

Recently, learning theory and algorithms have been considerably developed by the scientific community in order to exploit the information within the abundance of available data. At the same time, the increasing number of wireless devices connected to the network presents to service providers several challenges spanning from increasing network capacity to reducing the energy consumption. However, such increasing densification of wireless devices could be exploited in learning the communication conditions and make adaptive communication algorithms less needy of control signalling than those currently used, thus making available more resources to the users. The scope is then to discuss the performance gain that could be achieved under different scenarios of temporal and spatial correlation of both traffic and propagation conditions. Contributions on adaptive communication algorithms minimizing feedback at physical layer (e.g., Zero Forcing precoding in Multi-User MIMO in the large array regime) and at upper layers (opportunistic scheduling, context-aware video streaming, efficient caching) are encouraged for submission.

List of topics

  • Spatial and temporal correlation models for wireless communications in MU-MIMO
  • Adaptive algorithms for CSI acquisition in the large array regime
  • Stochastic bandit optimization for opportunistic scheduling
  • Distributed learning algorithms for optimal resource allocation accounting for the cost of feedback
  • Video streaming optimization for predictable feedback
  • Efficient caching, minimizing the amount of control signaling