Simulation And Implementation Of Information Coding Based On Biological Neural Network

Neural information encoding is the basic information transfer mechanism in the nervous system,in which neurons change their own membrane potential and firing rate to encode different information.Neurons form biological neural networks through synaptic connections,which enable the transmission and processing of information.Frequency encoding and temporal encoding are the main ways of encoding neural information.Neural information encoding has been widely used in neuroscience fields such as sensory information processing,motor control,memory and learning.Therefore,studying the information encoding properties of neural networks is crucial for understanding the operating mechanisms of neural systems.Neural network information encoding can be implemented not only by numerical simulation with computer software,but also by flexible and real-time hardware.In this thesis,we propose a numerical simulation and hardware implementation method of information encoding properties of biological neural networks based on chemical synapses using Hodgkin-Huxley neurons with biological properties as neural network nodes.Simple biological neural networks(e.g.,ring networks)and complex biological neural networks(e.g.,globally coupled networks,small-world networks)were constructed by MATLAB&Simulink simulation platforms,respectively.The average frequency coding method and the Inter-Spike Intervals(ISIs)coding method were used to study the information coding characteristics of the biological neural networks under the applied stimuli,and the specificity of the network coding patterns and the influence of the network topology on the information coding under different applied stimuli were analyzed on this basis.In this thesis,a neural network circuit model corresponding to the simulation is established by modular modeling method,and the information encoding process is implemented in hardware on FPGA by using DSP Builder and QuartusⅡfor joint compilation.The method can visually reflect the network structure and coupling mode,facilitate functionalized partitioning of the system,and provide an experimental method for exploring the information transmission and processing mechanisms between neurons and studying the functions of biological neural networks.The information encoding characteristics of the networks were simulated and analyzed under sinusoidal signal stimulation with an amplitude of 10μA/cm~2and frequencies of 20 Hz,50 Hz,and 200 Hz,respectively.The results show that under sinusoidal signal stimulation at a frequency of 20 Hz,the neurons in the simple ring network and the globally coupled network exhibit a cluster firing pattern,and the neurons in the small-world network exhibit a rhythmic single firing.Under the sinusoidal signal stimulation with frequencies of 50 Hz and 200 Hz,respectively,the higher the stimulation signal frequency,the higher the firing rate and the denser the firing sequence of the biological neural network with the same topology;at the constant stimulation signal frequency,the more complex the network topology,the higher the firing rate and the denser the firing sequence of the biological neural network.Simulation analysis of the information encoding characteristics of the networks under the stimulation of random audio signals showed that the firing patterns of neurons in the three networks exhibited high synchronization,and the greater the intensity of the random audio stimulation signal,the greater the degree of network response to the stimulation and the more dense the firing of neurons in the networks.From the temporal structure of the discharge sequences,it was derived that the temporal structure of the discharge sequences of the neuronal networks with different topologies differed within the time window of higher stimulus signal strength:in the time window of 0-300 ms,the simple ring network discharge sequences were denser;in the time window of 300-500 ms,the small-world network discharge sequences were denser;in the time window of 500-800 ms,the global coupled network discharge sequences were dense.The comparison concludes that the encoding patterns of biological neural networks are specific to different types of stimulus signals,and the topology of the network affects the temporal structure of the discharge sequences.the ISI encoding method is more accurate and contains more information,and the encoding method combined with the average frequency encoding can effectively express the dynamic changes of information encoding patterns of neuronal networks under stimulus signals.In this thesis,the simulation results are verified using a hardware implementation method,and it is demonstrated that the adopted hardware method can accomplish the study of information encoding characteristics of biological neuronal networks.