Research On Detection Technology For Shortwave And Ultrashortwave Hybrid Signal

With the development of communication technology,the type and number of signals increase,and the signal confusion and complexity also increase,which puts forward higher requirements for signal detection technology.Traditional signal detection and recognition techniques have limitations in modeling complex channels and extracting high-dimensional signal features,which leads to low detection and recognition accuracy in the absence of relevant expert prior knowledge.The rapid development of deep learning technology,its adaptability to large-scale data and automatic feature learning capabilities make it an important solution in the field of signal detection.In this context,the signal detection algorithm based on deep learning studied in this paper has important significance and value.The main work of this paper is as follows:First,for the task of single-signal detection,based on the classic deep learning models VGG network,residual network and convolution-long-short-time-deep neural network,starting from the basic concepts and challenges of signal detection technology,this paper discusses the methods based on the above three models A method for single-signal detection,and experimentally verified on the RML2018.01a dataset.The experimental results show that these three models can achieve better accuracy when the signal-to-noise ratio is high,which verifies their effectiveness in signal detection tasks and provides model reference for subsequent mixed signal detection tasks.Second,this paper constructs a mixed-signal dataset of shortwave and ultrashortwave scenarios.Starting from the characteristics of mixed signals,this paper simulates 6 mixed scenarios based on the characteristics of shortwave and ultrashort wave band signals,and generates a mixed signal dataset based on the RML2018.01a dataset that has undergone spectrum shifting,and completes the construction of the dataset.Third,aiming at the nature of short-term classification in mixed signal detection tasks,this paper proposes a network model based on convolutional recurrent neural network,which can simultaneously learn the spatial and temporal features of signal samples to complete mixed signal detection Task.At the same time,this paper also proposes some indicators and evaluation methods for evaluating the performance of mixed signal detection systems,and introduces a weakly supervised learning method based on multi-instance pooling to deal with the practical problem that mixed signals are difficult to label.In terms of experimental verification,this paper uses the generated shortwave and ultrashort wave scene mixed signal data set,and performs experimental verification on the mixed signal detection system according to the mixed signal detection performance index evaluation standard.In addition,this paper also conducts comparative experiments with the convolutional neural network model and the recurrent neural network model,which proves the effectiveness of the convolutional recurrent neural network model.Finally,this paper explores the optimal model structure and parameters by modifying the network parameter structure,including the structure part of the cyclic neural network,convolution kernel,activation function,and learning strategy.Through a large number of experiments,this paper obtains the optimal model scheme in the mixed signal detection task scenario,and effectively completes the mixed signal signal detection task.