随机性标量参数分布式量化的最优比特分配

Optimal bit allocation for stochastic scalar parameter distributed quantization

中文关键词  分布式量化估计  最优比特分配  无偏均匀量化器  均方误差  贝叶斯Cramer-Rao下界

英文关键词  distributed quantization estimation  optimal bit allocation  unbiased uniform quantizer  mean-square error  Bayesian Cramer-Rao lower bound

中文摘要

      本文研究了在总比特率设定的情况下, 改良并给出表现更优的量化器, 以及如何实现基于网络的随机标量 参数分布式量化估计, 重点讨论传感器比特数最优分配. 与常规给定各传感器的量化比特率不同的是, 本文将结合 估计器算法使用和不同量化器的构建, 来研究固定总比特率下的分配. 文中的观测模型噪声服从高斯分布, 并且以 此模型为对象通过均匀量化探讨基于一般类型与线性估计器的最理想比特分配方式. 前者均方误差上限与后者对 应下限在高精度处理方案下结果几乎相同, 都表现出网络中观测噪声误差反比于量化级数这一特性. 此外还借用 交替序列比特分配算法以确保求解出的数值解恒非负. 最后从Matlab仿真结果可以看到, 本文给出的最优比特分配 估计器较传统方案的表现更优.

英文摘要

      Given the total bit rate, this paper studies the improved quantizer, the distributed quantization estimation for random scalar parameters, and focuses on the optimal allocation of sensor bit numbers. Different from the existing results in which the quantization bit rate of each sensor is given, this paper will combine the estimator algorithm and the construction of different quantizers to study the optimal allocation under the fixed total bit rate. The noise in the observation model obeys the Gaussian distribution, and this observation model is used to discuss the optimal allocation based on linear and general types estimators through uniform Quantization. The former mean square error upper limit and the latter corresponding lower limit are almost the same under the high-precision quantization scheme, and all show that the observation noise error is inversely proportional to the quantization series. In addition, an alternating sequence bit allocation algorithm is borrowed to ensure that the solved numerical solution is constant and non-negative. Finally, it can be seen from the Matlab simulation that the optimal bit allocation estimator given in this paper is better than the traditional scheme.