基于神经网络的一种改进的向量量化方法

用LBG算法产生的码书,其码向量在码书中的排列是无序的.用此序号作为向量量化器编码输出时,对信道误码特别敏感.为了控制由于信道误码而导致整个向量量化通信系统性能严重下降,基于Kohonen网络的自组织特征映射(SOFM)算法进行向量量化分析,并针对SOFM算法性能上的缺陷,提出了一种改进的自组织特征映射算法.新算法引入失真敏感参数,对网络参数进行优化,通过调整码字的部分失真来指导神经网络的学习.通过仿真试验,从峰值信噪比的提高验证了算法的优越性.

An Improved Vector Quantization Algorithm Based on Neural Network

LBG vector quantization algorithm is used to create codebook traditionally, but the sequence is not in order. If the sequence is used to be the coding output of vector quantization, it is sensitive to channel error coding. To avoid this problem, this paper uses self -organizing feature mapping algorithm applying to vector quantization based on Kohonen network. Aimed to get over the defects, an improved SOFM algorithm is presented. New method cites a distortion-sensitivity parameter, and optimizes learning parameters in the network. Adjust part distortion of code-words to instruct the neural network learning. Simulation result shows that the new algorithm has good performance in the view of PSNR.,