Synchronous Discharge Characteristics Of Small-scale Neural Network By Hardware Implementation Based On FPGA

The phenomenon of synchronous discharge of neurons is widely present in the nervous system.It is the key to the normal functioning of advanced brain functions such as cognition,emotion,and behavior,as well as an important mechanism for neural information processing.Synchronous discharge can not only suppress noise interference,but also improve the accuracy of neural information processing.Researching it helps to decipher the working mechanism of the nervous system.In addition to using computer software to simulate the synchronous discharge of neurons,it can also be achieved by hardware with flexibility and real-time response.Starting from the action potential of biological neurons,this thesis proposes a simulation method and hardware modeling method for the synchronization characteristics of small-scale neural networks with biological attributes.Using the Hodgkin-Huxley neuron model with biological properties and Leonid chemical synapses and electrical synapses,small-scale neural networks with different parameters are modeled by MATLAB&Simulink to research the synchronous discharge characteristics of the 10 kinds of structures including regular topology and complex topology,and the networks with different coupling strengths.Use the standardized coherence coefficient method to quantitatively analyze the synchronization characteristics of the neural networks with different parameters.And on this basis,the anti-interference characteristics of the neural network in a noisy environment are studied.Using the method of joint compilation of DSP Builder and Quartus II to establish the circuit model of the network with different parameters according the simulation.,then implement the synchronous discharge process of the neural networks on FPGA.Modular modeling not only intuitively reflects the structure and coupling methods of biological neural network,but also facilitates the functional partitioning of the nervous system.It lays the foundation for the subsequent realization of large-scale neural nuclei and clusters.The results show that the coupling strength of the synapse and the topology are the key to the synchronous discharge of the neuron network.Keeping the network coupling strength constant,the synchronization index of the chain neural network with the lowest degree of connectivity is 0.169,and the synchronization index of the global coupling neural network with the highest degree of connectivity is 0.96.Whatever the network is regular or complex,as the degree of connectivity in the network increasing,the synchronization index of the network gradually approaches 1,at the same time,neural networks are getting more and more synchronized.In a certain topology,the stronger the coupling strength,the better the synchronization.When the coupling strength of the simple loop network increases from0.05m S/cm~2 to 7m S/cm~2,the synchronization index of the neural network also increases from0.08 to 0.998.After introducing a certain intensity of noise into the network,it is found that the action potential of a single neuron is distorted with interference.The synchronous discharge of the neural networks can eliminate the noise and restore the normal discharge of tneurons,and suppress interference effectively.When the standard deviation of Gaussian white noise is 20,the loop ring network with a synchronization index of 0.187 can effectively suppress interference.The standard deviation increases to 40,the loop networks with synchronization indexes of 0.187 and 0.964 can still effectively suppress interference,but the anti-interference effect of the complex topology 7 network whose synchronization index is0.809 has declined.Through comparison,it is found that the anti-interference effect obtained by changing the coupling strength between neurons is better than the effect obtained by changing the network topology.The hardware implementation results in this thesis are basically consistent with the simulation results.It proves that the hardware method can be used to research the synchronous discharge characteristics of the small-scale neural network.