Information Theory for Dynamic Networks

Lizhong Zheng
Communications and Networking Group
Research Laboratory of Electronics

Information theory has been a main driving force for the development of the digital communication industry. In the past two decades, several information theoretic results, including the use of multiple antennas, opportunistic transmissions, and interference management, have been adopted in most commercial wireless data systems, resulting in orders of magnitude improvements in system efficiency and many successful wireless products on the market today. The new era of industrial development is, however, focused on applications over dynamic wireless networks, which challenges the foundation of information theory, and calls for refocusing of previous theoretical studies.

In our vision, the new generations of wireless data networks should operate in a way depicted in the figure below. Here, information source is often broken into multiple pieces; passed along different paths in the network; further divided and/or merged as it travels through the network; and eventually different pieces get collected at destination nodes, with possibly different levels of fidelity and reliability. The advantage of such architecture is several fold: it allows nodes in wireless networks to cooperate in forwarding messages, and allows easy local reconfiguration when network topology changes. The data redundancy over the network can also be easily controlled to combat data loss.

Depiction of operation for new generations of wireless data networks.

Depiction of operation for new generations of wireless data networks.

The main theoretical challenge in implementing such architecture is that it fundamentally deviates from conventional theory, which is based on a homogenous view of information: all information bits are treated the same way, and are transmitted with idealized end-to-end reliability.

Supported by DARPA and the AirForce, our work addresses several aspects of the dynamic network communication problem. We have made progress in developing coding techniques that encode multiple streams of data on a single channel, while efficiently offering different levels of error protections to different streams. This approach helps us to develop a heterogeneous view of information, prioritizing network controls and cooperative transmissions under a unified framework. Under this framework, we particularly focused on “dynamic coding” problems, where the coding contents and techniques can change dynamically according to real-time observations of wireless nodes. In such problems, conventional concepts such as long-term average throughput and error rates are no longer appropriate to measure the system performance. Instead, we have developed the new notion of “dynamic information progress”, which is suitable for communication over time varying networks.

Conceptually, our approach fundamentally differs from the conventional theory in that we allow the exchange of “soft information”, i.e., information without perfect reliability. By quantifying and efficiently using soft information, in the form of suggestive network protocols and interior cooperative signaling, we can greatly improve the performance of dynamic wireless networks.

Tags: , , , , , , , , , , ,

Leave a Reply

Replies which add to the content of this article in the EECS Newsletter are welcome. The Department reserves the right to moderate all comments. If you would like to provide any updated information please send an email to newsletter@eecs.mit.edu.