Self-repairing Neural Network Modeling And FPGA Hardware Implementation

The Central Pattern Generator(CPG)is the basis for the generation and regulation of rhythmic movements in vertebrates and invertebrates,with the ability to spontaneously produce rhythmic output in a real external environment where interference and noise prevail.As a biological neural network,the robustness and self-repair ability of CPG provide a new idea for the motion control of bionic robot based on brain-like computing.Although many types of CPG models are established to reproduce rhythmic signals,these existing models do not reflect the robustness and self-repairing ability of rhythmic output.The underlying mechanism needs to be further studied.Recent studies have shown that astrocytes play an important role in the self-repairing function of the nervous system.Therefore,this paper proposes a novel CPG model based on astrocyte-neuron self-repair mechanism,which captures the above-mentioned CPG rhythmic output behavior by introducing factors such as random synaptic input and the presence of "wrong" synapses.Two cases are used to evaluate self-repairing properties,one of which is when two neurons in the CPG experience an external synapse failure with the same failure rate,and the second is when one neuron is "healthy" and the other neuron are damaged.The repair function is evaluated by comparing the output characteristics of the CPG network with the dynamic behavior of the external synapse in the presence or absence of astrocytes.Firstly,in order to evaluate the robustness of the self-repairing network,this paper introduces a random environment based on the probability-based external synapse model to compare the output characteristics of CPG networks with or without astrocytes under external random stimulation.The study found that indirect feedback signals from astrocytes can minimize the adverse effects of random input by adjusting synaptic transmission probabilities,thereby increasing the ability of CPG to resist external stimuli.Secondly,in order to study the self-repairing effect of astrocytes on CPG rhythm,this paper sets the external synapses connected to CPG neurons to different failure rates,and then analyzes the output behavior of CPG and external synaptic dynamics under different failure rates.Studies have found that astrocytes reduce CPG system loss due to synaptic failure and can repair network output rhythms at low failure rates.Finally,a novel CPG hardware simulation platform based on astrocyte-neuron self-repair mechanism is proposed based on FPGA,and the improved CPG model is realized.The system provides a visualization platform closer to real neural networks for the study of mechanisms with self-repairing motion control systems.In the design,the multiplier is replaced by logical shift and addition and subtraction,which significantly reduces the hardware resource consumption and computational complexity,and provides a good idea for extending the more close-to-physical large-scale self-repairing CPG network.The platform has important application value for implementing brain-like bionic motion controllers.