Information Theory of Real-Time Communication
Event details
| Date | 11.03.2014 |
| Hour | 16:15 › 17:15 |
| Speaker | Prof. Ashish Khisti, University of Toronto http://www.comm.utoronto.ca/~akhisti/ |
| Location | |
| Category | Conferences - Seminars |
An increasing number of multimedia applications require real-time communication i.e., information must be transmitted sequentially with strict delay constraints. Both the fundamental limits and optimal codes for such systems can be very different from classical models in information theory. In this talk I will present our recent work on source and channel coding problems under such real-time constraints.
The first part of the talk will focus on communication over packet-erasure channels and introduce a class of "streaming codes" . The encoder must process a stream of source packets in a causal fashion, and the decoder must reconstruct each source packet with a fixed delay. For a class of channels that introduce both burst and isolated erasures, we will present near-optimal streaming codes and discuss their algebraic properties. Simulations over the Gilbert-Elliott channel indicate that our proposed codes achieve significant gains over baseline erasure codes for certain channel parameters.
In the second part of this talk we will consider real-time transmission of a Markov source over a burst-erasure channel. The encoder observes a sequence of source-vectors, which must be compressed in a causal fashion. The decoder must reconstruct each source-vector with a fixed delay, except those sequences that belong to an outage period, associated with the burst-erasure. We will present an information theoretic tradeoff between the compression-rate and the outage-duration. We will discuss how traditional separation based architectures are sub-optimal and insights from some multi-terminal source coding are relevant to the streaming setup.
The first part of the talk will focus on communication over packet-erasure channels and introduce a class of "streaming codes" . The encoder must process a stream of source packets in a causal fashion, and the decoder must reconstruct each source packet with a fixed delay. For a class of channels that introduce both burst and isolated erasures, we will present near-optimal streaming codes and discuss their algebraic properties. Simulations over the Gilbert-Elliott channel indicate that our proposed codes achieve significant gains over baseline erasure codes for certain channel parameters.
In the second part of this talk we will consider real-time transmission of a Markov source over a burst-erasure channel. The encoder observes a sequence of source-vectors, which must be compressed in a causal fashion. The decoder must reconstruct each source-vector with a fixed delay, except those sequences that belong to an outage period, associated with the burst-erasure. We will present an information theoretic tradeoff between the compression-rate and the outage-duration. We will discuss how traditional separation based architectures are sub-optimal and insights from some multi-terminal source coding are relevant to the streaming setup.
Practical information
- General public
- Free
Organizer
- IPG - Prof. Khisti is hosted by Rudiger Urbanke