噪声信道下最佳矢量量化器的人工神经网络实现方案

噪声信道影响下的矢量量化器设计,实际上是一个信源、信道联合编码(joint soruce/channel encoding)问题。本文提出了一种利用人工神经网络ANN(Artificial Neural Network)技术解决联合编码的组合优化问题的方案,该方案较好地解决了有噪声信道条件下的最佳矢量量化器的设计问题。由于将信道传输特性直接引入到神经网络的构造中,因此,对网络进行训练后,最终得到对噪声信道影响具有一定程度的抑制作用的矢量量化器。我们针对BSC信道对所提方案进行了分析和模拟,结果表明,在量化网络设计中,考虑噪声信道影响的因素,可使训练得到的量化网络的量化特性对噪声信道的影响表现出明显的韧性。

Artificial Neural Network Based Optimum Vector Quantizer Design for Noisy Channels

Designing vector quantizer for noisy channels is a joint source and channel encoding problem. In this paper, we introduce an approach to solve the joint source/channel vector quantization problem by applying ANN(Artificial Neural Network). we provide the architecture and learning algorithm of the quantizing net. Simulation results with different sources and channel properties show that the robustness of vector quantizer designed with the method described in this paper has been improved.