Vinay Melkote (University of California, Santa Barbara), “Optimal Delayed Decisions in Encoding and Decoding of Audio Signals and General Sources”, Advisor: Kenneth Rose (2010)
The dissertation commences with the exploration of delayed-decision approaches to optimize the encoding operation over the entire signal, a concept that is of considerable benefit to applications that employ off-line coded content. In the representative audio-compression framework, a two-layered trellis effectively optimizes both intra- and inter-frame encoding decisions while minimizing a psycho-acoustically relevant distortion measure under a prescribed bit-rate constraint. Motivated by this rate-distortion optimization paradigm modifications are proposed to the audio distortion metric itself that endeavor to enable subjectively optimal decisions.
The focus then gravitates towards optimally exploiting delay at the decoder end: specifically, can a delayed decoder exploit information from future frames to improve the reconstruction of the current frame, while retaining the existing encoder structure? In the relevant scenario of predictive coding applications, this dissertation proposes an estimation-theoretic optimal delayed decoder that recursively calculates the required conditional probability densities, given both past and available future information, and computes the optimal reconstruction via conditional expectation.