Non-Binary Analog-to-Digital Converter Based on Amoeba-Inspired Neural Network

Abstract

An analog-to-digital converter (ADC) based on neural networks is proposed, and the feasibility of using no binary coding is discussed with circuit simulation. An amoeba-inspired computing technique is used to construct the present ADC, where switched-capacitor circuits are used as unit neurons. Dummy units are also added to improve the stability of circuit operation. For an ADC with a radix of 2, large quantization errors were observed due to the local minima. It was found that introducing a radix smaller than 2 effectively reduced the quantization error. Low-power operation can be expected by using a dynamic analog circuit technique in the present neuro-ADC.