NCD Seminar: Toward a Many-User Information Theory

Toward a Many-User Information Theory 

Classical multiuser information theory studies the fundamental limits of models with a fixed (often small) number of users as the coding blocklength goes to infinity. In this talk, I introduce a new regime, where the number of users and the blocklength tend to infinity simultaneously. This paradigm is motivated by systems in which the number of devices is comparable or far exceeds the blocklength, such as in large machine-to-machine communication systems and sensor networks. The focus is on the Gaussian many-access channel, which consists of a single receiver and a massive number of transmitters, where a subset of users transmit in a given block and need to be identified. Since the conventional notion of capacity in bits per channel use is ill-suited for the task, a new notion of capacity is introduced and characterized. Also discussed are many-broadcast channels, lossless many-source coding, and an outlook on a general many user information theory. Parts of the work were done in collaboration with Xu Chen and Tsung-Yi Chen at Northwestern University, and with Gregory W. Wornell at MIT.