Lecturer : Daniela TUNINETTI
The bottleneck of today’s wireless networks is interference. The solution of commercial available networks is to avoid interference through division of resources (such as time, spectrum, space, waveform, etc.) among the competing users. This approach results in simple network architectures. It is also a theoretically optimal solution in the up-link and down-link of single-cell cellular systems with single antenna nodes in presence of fading when throughput maximization is the design goal. Perfect orthogonalization of users is however not possible in practice, in which case the residual interference is usually treated it as noise. ;It has become apparent however that interference avoidance and treating interference as noise are highly suboptimal in interference networks.
For example, when the power imbalance between the useful signal and the interfering signal at a destination is large enough, a receiver can first decode the interfering signal by treating its own signal as noise, and then decode its signal as in an interference-free channel.
More generally, it is possible to consolidate the aggregate interference into roughly half of the receiver available degrees of freedom and thus each user achieves half the rate it would get if it were alone on the network, irrespectively of the number of interferers. Moreover, node cooperation further improve on the overall network interference management problem. With these observations in mind, much progress has been made recently in understanding the ultimate performance limits of interference networks.
This talk summarizes the last decade of information theoretic studies on cooperative interference networks. We will start by describing commonly adopted cooperation models (in-band vs. out-of-band, noisy vs. rate limited noiseless links, generalized feedback, causal vs. cognitive, etc.) and then recap the fundamental lessons learnt from exact and approximate capacity results available in the literature. An overview of current challenges and open problems will conclude the talk.
This work was supported by NSF. Part of this work has been done in collaboration with Dr. Echo Yang, Dr. Stefano Rini, Martina Cardone, Alex Dytso, and Dr. Natasha Devroye.