A new design strategy for the Tank and Hopfield neural analog-to-digital converter

Tank and Hopfield (1986) have described the implementation of a neural-based 4-bit analog-to-digital converter and discussed several of the problems they encountered such as missing output codes. High-gain amplifiers were used for the artificial neurons in their implementation. System noise and non-ideal amplifier characteristics such as offset voltages are multiplied by gain which results in a loss of conversion accuracy. The new approach proposed here use low-gain analog amplifiers for the artificial neurons where the gain is more like that of a biological neutron. Simulation and experimental results are presented.<
>