Low-Complexity Methods for Provably Good Information Transmission and Network Anomaly Detection via Packet Timings In Networks

摘要

My interest in understanding fundamental limits and practical algorithms/codes to instantiate such limits -- within the context of communication based upon packet timings -- was inspired by my understanding how neurons signal to one another with the timings of their spikes. Remarkably, using this modality to communicate over packet networks has not been explored in nearly as much depth as other communication modalities. Applications of this approach are abundant: including unequal error protection, covert communications, and computer network security. Although an 1990s paper titled 'Bits through queues' characterized fundamental limits of communication with packet timings over queuing channels, my curiosity led to a very different -- and in some sense simpler -- way to arrive at the fundamental limits of such complicated queuing channels (that are nonlinear and have memory). By considering the queuing system as a stochastic dynamical system with feedback coupled to a memoryless channel, I was able to provide a much simpler explanation of the fundamental limits of such a system. Also, this new proof methodology can be extended to many other information-theoretic contexts, which I have also been actively pursuing. Lastly, this new proof technique has led to the development of low-complexity, provably good, error-correcting codes that can be used over computer networks. Extensions of this approach and its industrial potential have led to us winning UIUC's Grainger Award in Emerging Technologies in 2008.