University Relations
Towards fundamental bounds for physical-layer power consumption.
Anant Sahai, Univ. of California Berkeley, Dept of EECS
Traditional information-theoretic analysis uses explicit models for the communication channel to study the power spent in transmission. However, practitioners have long observed that the decoder complexity, and hence the total power consumption, goes up when attempting to use sophisticated codes that operate close to the Shannon waterfall curve.This is especially true when short-range communication is considered. This talk asks the question of what "capacity-achieving" should mean if we have to care about total power rather than just transmitter power.
Power consumption is thus considered to be the natural metric for complexity. Before we can build a full bridge to the semiconductor roadmap and the circuit side of things, we have to start with the basics. Classical results are reinterpreted in this context and are shown to imply that neither classical dense linear codes with ML decoding nor convolutional codes can be capacity achieving in any reasonable sense. An explicit model is given for the power consumption of an idealized decoder that allows for extreme parallelism in implementation. This decoder architecture is in the spirit of message passing and iterative decoding for sparse-graph codes.
Generalized sphere-packing arguments are used to derive lower bounds on the decoding power needed for any possible code given only the gap from the Shannon limit and the desired average probability of bit error. As the gap goes to zero, the energy per bit spent in decoding is shown to go to infinity. This suggests that to optimize total power, the transmitter should operate at a power that is strictly above the minimum demanded by the Shannon capacity. The lower bound is also plotted to show an unavoidable tradeoff between the average bit-error probability and the total power used in transmission and decoding. In the spirit of conventional waterfall curves, we call these `waterslide' curves.
-
Excerpt from the academic lecture at Qualcomm on Sept 24, 2008
Digital image capture, processing, and sharing has become pervasive in our society. This has had significant impact on how we create novel scenes, how we share our experiences, and how we interact with images and videos. In this talk, I will present an overview of series of ongoing efforts in the analysis of images and videos for rendering novel scenes. First I will discuss (in brief) our work on Video Textures, where repeating information is extracted to generate extended sequences of videos. I will also describe some our extensions to this approach that allows for controlled generation of animations of video sprites. We have developed various learning and optimization techniques that allow for video-based animations of photo-realistic characters. Using these sets of approaches as a foundation, then I will show how new images and videos can be generated.
