Research On DTN Routing Approach Based On Spiking Neural Network

The DTN(Delay-Tolerant Network)protocol is the key to complete the construction of a long-latency,intermittently connected environment network for the deep-space network in the future.However,with the increase in the number of deep space network nodes,the computation and storage costs of the DTN protocol,especially the Dijkstra-based CGR(Contact Graph Routing)routing protocol,have increased significantly,which may not be able to adapt to the resource-constrained satellite nodes.For such challenges,this paper proposes a routing method based on Spike Neural Network(SNN),which converts the routing computation process of each bundle into a classification problem based on SNN and reduces the amount of computation.The SNN-based routing module uses the link information as the feature vector and the CGR's result as the labeled training data to train the SNN model to obtain the weight parameters.Spike neurons that store a large amount of time domain information are used as the computation unit of the neural network,and a variant of the STDP weight update algorithm adapted to the spike neurons is proposed.Since only time information is considered,the SNN-based DTN routing method may have under-fitting problems.This paper tries to improve the routing performance by expanding the available link capacity,queuing time,and delay information as feature vectors and selecting appropriate parameters to optimize the machine learning model.This paper builds an ION-DTN-based international space station data backhaul scenario,verifies that adding the available link capacity as feature vector can improve the routing accuracy,and uses a randomly generated DTN network to verify the complexity characteristics of the SNN-based DTN routing algorithm.A 48-node DTN network simulation scenario was built based on DTNSim software.The experimental results show that the routing results of the improved model are equivalent to the CGR's results under the premise of greatly reducing the computational cost.It is concluded that the routing method based on machine learning is superior in computational performance.The research in this paper proves the advantages and feasibility of DTN routing method based on SNN,and has certain reference value for the design and development of DTN routing.