In this paper, we analyze neural spike dynamics of a double feedback neural unit (DFNU) and its networks. An essential emphasis of the DFNU is not only on its simple formulations that can provide quantitative analytic results, but also on physiological plausibility of the dynamics that is comparable to that of Hodgkin-Huxley's model. The results suggest that a proportional coding of neural information on firing frequency may not be always reasonable due to sensitivity to noisy inputs especially for low-frequency firings. On the other hand, high-frequency firings are relatively appropriate for a neural informational carrier due to the reliability and robustness to noisy inputs. It is also demonstrated by simulation studies that use of noisy inputs can enhance the dynamic neural performances.