PhD Defense: "Complexity and Delay Constrained Compression and Transmission of Information"

Emrah Akyol

March 10th (Thursday), 12:00pm
ESB Room 2003

This talk is concerned with optimal strategies for delay and complexity constrained communications. The first part of the talk is focused on optimal joint source-channel coding for zero-delay communications. This problem, originally posed by Shannon in his seminal paper, received recent attention due the increasing need for low delay and low complexity communications. The necessary conditions for optimality of encoding and decoding functions (mappings) are derived and a corresponding numerical algorithm is proposed which is shown to discover locally optimal mappings that outperform all prior results. The approach is then extended to provide optimality conditions and design algorithms for distributed source channel coding. The second part is concerned with the conditions for linearity of optimal decoding mappings, and of optimal estimators in general, in terms of the channel constraint, source and noise densities and the distortion measure. Specifically, the necessary and sufficient conditions for linearity of the optimal estimator along with existence and uniqueness of source and noise densities that satisfy such conditions, are derived. The remainder of the talk is focused on low delay source coding methods including contributions to dithered quantization and transform coding. Dithered (randomized) quantization which has traditionally been considered in its natural setting of uniform quantization, is extended to encompass nonuniform quantizers by dithering in the companded domain. Finally, a long standing theoretical problem of transform coding is solved. The necessary and sufficient condition for optimality of a transform, in conjunction with variable rate quantization at high resolution is derived. This condition not only determines when the Karhunen-Loeve transform (KLT) is optimal, but also leads to an algorithm that obtains the optimal (non-KLT) transform. The optimal transform is also derived for the setting of transform coding in conjunction with dithered quantization, resulting in a universally optimal fixed rate source coding scheme.

About Emrah Akyol:

Emrah Akyol is a Ph.D. candidate at UC Santa Barbara, working under supervision of Prof. Kenneth Rose. He obtained his BS. degree from Bilkent University, Ankara, Turkey and MS. degree from Koc University, Istanbul, Turkey. He worked at Hewlett-Packard Research Labs and NTT DoCoMo Research Labs, Palo Alto, CA. His research interests span information and estimation theory, signal processing, and multimedia compression and transmission. He is co-author of over 20 refereed conference and journal papers and one patent, and recipient of the 2010 ECE Dissertation Fellowship.

Hosted by: Professor Kenneth Rose