Back Wireless networks with cognitive topology

Wireless networks with cognitive topology

Wireless networks with cognitive topology

Cellular systems have undergone four generational transitions, beginning with the 1st Generation (1G) in the 1980s. We are now undergoing the transition from 3G to 4G. Despite substantial differences, all generations have conformed to the same architecture that underpinned 1G, namely the cellular architecture. The defining notion in this architecture is a small geographical region termed cell. A cellular network tessellates a territory into cells, each centered on a site where a so-called base station houses all the necessary equipment to radio-communicate with users. Wireline infrastructure then backhauls all these base stations. A user communicates only with the base station in its cell.

The cellular architecture has served us well, yet the time is approaching when it will become exhausted. The first step to construct something new is to deconstruct the old. At the heart of every cell there has always been a base station with comprehensive functionalities. Now, some of these functionalities should be pushed towards the antennas while others should be pulled away. On one hand, as antennas multiply, radios should also multiply and each should attach to one antenna yielding antenna-radio units that become the basic building blocks of the infrastructure. On the other hand, the signal processing needs to move up to a higher plane such that all these units can work together and the network can be smarter.

The upwards shift of signal processing leads to the much-touted concept of cloud radio access. Borrowing from the cloud computing paradigm, this would entail centralizing the signal processing for the entire radio access on the cloud, where the intelligence would concentrate leaving at the network edge only a massive number of inexpensive antenna-radio units. This opens the door to virtualization, whereby base stations could cease to exist physically rather becoming software processes. Such virtual base stations could be created at will and the structure of the network would then be programmable. This would offer huge elasticity and reduced energy consumption as no physical resources would ever be idle.

The objective of the present project is to devise, based on machine learning, algorithms that allow the network to automatically and dynamically define its base stations on the basis of the conditions experienced by terminals at each given place and time. Specifically, the algorithms would operate on the basis of data supplied by the mobile devices to the infrastructure. This data would provide a sample map of the quality and condition of the channel over the network, refreshed with fast periodicity.

The work requires a combination of knowledge and skills in wireless networks (supplied by the Wireless Communications group) and in machine learning (supplied by the Artificial Intelligence group). The project may also benefit from the Wireless Communication group’s ongoing collaboration with Vodafone, which has supplied field data from one of their UK deployments. Although limited in size, this data could be used to test algorithms devised on larger computer-generated data sets. Moreover, it is possible that further and more extensive amounts of field data are made available by Vodafone.

Principal researchers

Àngel Lozano


Rasoul Nikbakht