Cognitive Networks


Nowadays, wireless networks are becoming pervasive, highly populated and increasingly complex. Under these conditions exploiting rich interactions among mobile devices are better fulfilled. These trends are giving rise to new communications paradigms which are based on cooperation and cognition as the main underlying principles. The symbiosis between these two principles, confer to the wireless networks some degrees of consciousness or understanding about their own existence, such as internal structure, capabilities, relationships to the outside world, limitations, current use of radio resources and many more.

A cognitive process involving observing, planning, reacting and learning from experience can be applied to wireless networks in order to adapt the system to the highly dynamic wireless ecosystem. The ultimate goals are to enhance the efficiency in the use of radio resources as well as to improve both link and network performance.
Recently, some research effort has been addressed to the issues regarding Cross Layer Optimization. With reference to the protocol stack of a wireless standard, the concept of vertical integration (Cross Layer Design) refers to the joint optimization of techniques crossing different, adjacent or even non-adjacent, layers of the stack. Cross Layer techniques adapt the link/network/transport parameters to the channel status, or the application instantaneous requirements (channel or application aware protocols) and to the current state of the algorithms running at the other layers. However, Cross Layer Optimization (CLO) issues go beyond the concept of channel or application aware protocol design. In fact, CLO requires mutual adaptation of the parameters of separate layers, based upon the channel and/or application characteristics. In this context, the activities of the SIGNET group will include the following.

  1. Cross layer approach Definition and analysis of channel-aware MAC/scheduling mechanisms, the understanding of the trade-offs between overhead and error-protection provided by hybrid FEC/ARQ error correction mechanisms. Definition of guidelines for the joint design of physical and MAC layer mechanisms;
  2. Machine learning techniques are applied in order to improve wireless networks performance in several networks scenarios. Neural networks and Reinforcement Learning have shown to be flexible algorithms with low complexity (adequate for practical implementations) . We are looking towards the migration of these techniques in the wireless world;
  1. Network coding : Current research in this field show that network coding has some characteristics that bring significant improvements on the cognitive cycle. The integration of a network coding module in the cognitive architecture and the evaluation of its performance are still under investigation;

People involved

Alfred Asterjadhi, Nicola Baldo, Andrea Zanella, Michele Zorzi