Information Theory provides the theoretical basis for many communication systems. We aim to obtain a genuine understanding of the fundamental trade-offs in point-to-point communication as well as in self organizing networks. Here below we list some of the projects we are currently involved.
Wideband Low-Power Communication
Recent information theoretic results point to the inefficiency of current communication methodologies in extreme wideband or low-power applications through fading channels. While finding capacity for real cases of interest is a daunting task, we focus on constructing useful bounds and identify good signalling schemes. Thus channel effects like delay and Doppler spread, multi-paths, estimation of channel coefficients, timing etc and their effects on capacity are analysed. We further investigate the issues associated with multi-access over fading channels. Our constructions try to incorporate the so far neglected component of communication delay into information theoretic framework. Lack of user synchronisation is modelled by imposing higher moment constraints on the input in certain situations.
Random Matrices in Communication
The study of fundamental limits to various modern communication systems relies on the study of large random matrices. Our focus is on multi-antenna channels and ad hoc networks. – Multi-antenna channels: We have established invariance principles giving the optimal input covariance, required for the channels. More has to be said in the Ad-hoc networks: We have established precise scaling laws for the capacity of arbitrary one-dimensional networks and regular two-dimensional networks. These scaling laws rely on refined estimates of eigenvalues of random large matrices. The extension to the arbitrary two-dimensional case is in progress.
Network Information Theory
Network information theory is the theory about transmitting or storing information when more than two devices are involved. Within the last decade networks have evolved from static and centrally-controllable to dynamic and self-organised. How do we deal with the transmission and the storage of information when a large number of unreliable devices are involved? We are currently examining three aspects of this question: Ad-hoc detection: How does one detect an event (like an earthquake or a burglar) using a large number of imprecise devices with very limited communication capabilities? – Multi terminal source coding: Suppose several devices can provide some information about an event. How many bits of information should every device send in order to provide enough information so as to achieve a target criterion? – Spectral properties of random graphs: A network can be modelled by an undirected graph. How does the spectrum of the adjacency matrix relate to the network structural properties, like connectivity or the speed of information diffusion in the network.
Combining Networking and Information Theory
In a multi-user communication system the transmitters are generally driven by independent sources generating bursty streams of packets. Information theory views multi-user systems in terms of noise and interference, the issue of bursty packet arrival is dealt with appropriated source coding and no consideration for end-to-end delay is made. On the other hand, network theory directly treats the bursty arrival of packets via resource allocation and distributed scheduling, but ignores the issues of noise and interference. Is there a unified view? In this project, we handle this question by exploring approaches in which both aspects of multi-user communications are dealt with.
Universal Channel Coding and Feedback
A well known (and early) result of Information Theory establishes that feedback does not improve the capacity of memoryless channels. Even though this result does not apply to networks, nor to channels with memory, it has a damping effect on research on the use of feedback. Furthermore, just because it does not improve capacity it does not mean that feedback cannot simplify the design of communication systems. To that end we investigate the use of feedback in communication over unknown channels. As in any network the the communication channels can dramatically change over time, this is a natural question. Recent results show that in some cases, even if the channel is revealed neither to the transmitter nor to the receiver, there exists coding strategies that perform as well as the best coding strategies tuned for the channel under use. In these cases feedback enlarges the cooperation capabilities between the transmitter and the receiver, and ultimately frees them from the knowledge of the channel.