|1- Bit Quantization for Parameter Estimation: Performance Limits and Optimal Designs|
|陈豪 Hao Chen|
|  Dr. Hao Chen received the B.S. and M.S. Degree in electronic engineering from University of Science and Technology of China in 1999 and 2002, respectively, and Ph.D. degree in electrical engineering from Syracuse University, Syracuse, NY, USA in 2007.
He is currently an Associate Professor with the Department of Electrical and Computer Engineering and Computer Science, Boise State University, Boise, ID, USA. His research interests include statistical signal and image processing, and communications.
|  Quantization is an essential component in digital signal processing, e.g., analog-to-digital converter. In 1-bit quantization, every input data sample is mapped to a binary symbol of either 0 or 1. Despite the seemly massive information loss due to the compression, 1-bit quantization techniques have been employed in many applications with surprisingly good performances. Due to the virtually infinity possibilities of quantizer designs, the performance bounds and performance limits of 1-bit quantizers are not well established, except for few cases.
In this talk, we discuss the applications of 1-bit quantization in parameter estimation where the goal is not to recover the data itself, but to estimate the parameter of interest. In particular, we address problem of designing the optimal or near-optimal distributed estimation systems with many distributed agents or sensors. In such systems, the observations are compressed to bits using independent and identical 1-bit quantizers and transmitted to a central node to estimate the parameter. To evaluate the system performance, we establish an asymptotic performance limit for all possible quantizer designs. Aided by the performance limit, we derive a set of the optimal quantizer and observation models that achieve the PL by exploring the relationship between the probabilistic quantization scheme and the dithered quantizers. We show that the performance limits can be nearly achieved for the cases of Gaussian and Laplacian parameters, which in turn, establishes the optimality of the respective quantizers