Can finite be more than infinite in distributed information coding?

With the deployment of communication network infrastructures
such as packet-switched wireline networks, mobile cellular wireless networks, and distributed
sensor networks, it has become important to have a deeper
understanding of how information needs to be stored, processed and
transmitted efficiently to harness the full potential of these networks.

In this talk we look at distributed information processing and coding
and report a new coding phenomenon.
In particular, we show that for distributed source coding (data compression)
involving two or more information sources, as the block-length of the code is increased,
the performance improves initially, but plateaues, and then
decreases. The best performance is achieved for some finite
block-length that depends on the source distribution. To achieve the
same performance, the standard Berger-Tung approach requires
multi-letterization. This explains why for distributed source coding,
Berger-Tung coding scheme is not optimal which was
recently established using an alternate argument based on
continuity. The new approach leads to new computable characterization
of performance limits as well as a new coding strategy.
We explore application of this approach in other network communication problems.