The performance of traditional chaotic optical communications depends on the matching parameters of the chaotic transmitter and receiver. The chaotic receiver based on the neural network (NN) can solve the dependence on matching parameters, but the analog-to-digital converter (ADC) has an effect on the decryption performance. In this paper, the effects of sampling rate and digitalizing bit for ADC on NN-based chaotic optical communication systems with different feedback strengths at different bit rates are numerically studied. The results show that the requirements for ADC will be higher if the feedback becomes stronger. In order to achieve higher speed chaotic optical communications, the ADC with higher performance is needed. These results can give guidance for the applications of NN-based high-speed chaotic optical communication systems.