Study On Stochastic Resonance And Sensory Information Processing Using Model And Experiment

In daily life, noise is normally considered as a harmful factor for information detection and transmission. In last two decades, however, the discovery of stochastic resonance (SR) reveals the defect of this assertion: indeed, addition of appropriate amount of noise can enhance signal to facilitate signal detection or transmission in noisy environments. The stochastic resonance we mention is a generalized concept. The term "resonance " here refers to an effect of noise-enhanced signal detection or transmission that can be characterized by a certain measure of the detection or transmission displaying a non-monotonic variation with the noise level and peaking at a maximum value for an appropriate noise level.Firstly we introduced the phase plane method in nonlinear dynamics to analyze qualitatively the threshold dynamical behavior of neurons in this thesis. The results demonstrate that a single neuron with threshold behavior can be considered as a simple threshold detector. A typical characteristic of sensory systems is that the signal from peripheral receptors converges to nerve center, thus a summing parallel network can be used to simulate this process.Secondly, we investigated a summing parallel neuronal network constituted by the Hodgkin-Huxley (HH) model, which is used to simulate the peripheral part of sensory systems. In comparison with the single HH model, SR in the network has a wider range of optimal noise intensities for sub-threshold input signals. We obtained interesting results that noises do not deteriorate the capability of the detection of the suprathreshold input signals. These results prove that sensory systems may utilize SR to transfer diverse signal in a circumstance with relatively invariant noise.From the viewpoint of signal transmission, we studied the SR in summing network constructed by threshold neurons by measuring mutual information and correlation coefficient. In addition to conventional sub-threshold SR, we also observe the suprathreshold stochastic resonance (SSR) from the results of simulation when many threshold neurons are stimulated with independent noise sources, which means the information transmission and the match between input and output signal match can also be enhanced by the noise even if information-bearing signal is suprathreshold.Furthermore, we qualitatively verified our modeling results experimentally. According to the signal detection theory (SDT) of psychophysics, the receiver and the classifier of detector may be an adequate model for sensory systems constructed by the peripheral receptors and the nerve center (e.g. the cortical cells). We only consider SR in nerve center initially, which is considered as the classifier of detector. The neuronal Poisson model was introduced to model the probability distribution instead of normal distribution. Based on this model, different SR phenomena are observed. Comparing with conventional models regarding the receiver part of sensory system as a linear or single non-linear system, we employed a summing network constructed by MacCulloch-Pitts neurons in our model to simulate the receiver. Simulation results show that the relevant indices of both percent correct measure and detectability of signal exhibit the stochastic resonance behaviors.Finally, We then carried out the psychophysical experiments using both 1IFC (one interval forced choice) and 2IFC (two interval two alternative forced choice) methods. The experimental results qualitatively verify the conclusion in accordance with the theoretical model. These works give a proof that stochastic resonance is not only epiphenomenon in sensory systems.Our works suggests that noise-based technique can be used to improve and restore sensory functions in human. Furthermore application of the SR effect of signal transmission in sensory systems on engineering might provide some novel bionic methods for information processing.