Neural Network Quantizers for Discrete-valued Input Control

The performance deterioration of the system due to quantization is a serious problem that affects the networked control systems. Appropriately designed dynamic quantizers can reduce this performance degradation. However, most of the existing dynamic quantizers are based on a model of the plant and in many situations this model cannot be used. Therefore, this study proposes a type of quantizer implemented with a neural network and a design method based on a series of input and outputs of the plant. Then, this quantizer does not need a plant model, and could be applied to any type of system. The effectiveness of the proposed quantizer and its design method are verified with several numerical examples.