Achievable regions of bursty multi-user wireless communications – ARburst

The exponential growth of connected devices has lead to a new kind of networks called machine to machine (M2M) and Internet of Things (IoT) networks. Future scenarios in mobile communications envisage a huge number of wireless devices located in physical proximity, generating multiple information flows with different requirements and sharing the same radio resources. To face the development of highly connected devices in the next telecommunication roadmap, new theoretical tools are needed to characterize these networks. Indeed, the classical tools we have at disposition such as Shannon theory and stochastic geometry are limited in the context of M2M and future 5G services due to spurious and bursty nature of information flows. While the small packet sizes invalidate the use of the asymptotic Shannon capacity as a performance indicator, the impulsive nature of interference also invalidates the Gaussian assumption.

This project aims at addressing the fundamental limits of this kind of networks. The first objective is to define a proper space in which the performances of these networks could be evaluated. The capacity is certainly not the main challenge and we propose to work in a three-dimensional space energy-efficiency/delay/reliability to characterize the achievable region of dense and bursty netwoks. To this end, we will first extend the non-asymptotic channel coding theory developed by Polyanskiy to large and dense IoT networks in multiple access channel (MAC) and broadcast channel (BC) scenarios. In a same time, suitable models for non-Gaussian mixture of interference for bursty communications able to take into account rare but strong events will be sought. The large number of nodes and random deployment will be taken into account and the stochastic geometry tool will play an important role in the interference characterization. In a second time, the non-Gaussian interference models and the achievable and converse results with packet non-asymptotic regime will be merged in order to define the achievable region of our common framework; the non-asymptotic channel capacity is able to capture the energy-efficiency, delay and error probability in its formalism and hence would allow characterizing the performances of M2M and IoT networks in the three-dimensional space energy-efficiency/delay/reliability. One of the key issue will be to derive the channel dispersion quantity for non Gaussian channels.

The second objective is to evaluate the margin of improvement of actual techniques in the theoretical framework developed in the first objective. Techniques related to the PHY and layer-2 will be considered here. The up-to-date channel coding, e.g. LDPC, fountain codes and retransmission mechanisms, e.g. HARQ-IR, will be characterized in our three-dimensional space, which implies to be able to characterize these techniques with the three objectives mentioned earlier which is not a so simple task due to the complexity of the analysis of these schemes. Hybrid methods, involving numerical and analytical analysis will be performed. It would be an unprecedented try and would allow, finally, stating what is the margin of improvement of such schemes in the particular environment of bursty networks.

The third objective is to identify the most promising way of researches to effectively achieve the pareto-front described in the first objective. In particular, non Gaussian signal processing will be investigated. Superposition coding and interference alignment in impulsive conditions will be tackled in order to decide if they are suitable solutions for the development of future networks.

This project is unprecedented research and finally gives a try on the comprehension of fundamental limits of dense and bursty networks.