| With the rapid development of wireless communication,artificial intelligence(AI),the Internet of Things(IoT),and other technologies,the current electromagnetic environment is becoming more and more complex,and the estimation of environmental signal-to-noise ratio(SNR)is facing new challenges.The traditional mathematical model-driven SNR estimation algorithms are no longer suitable for the current complex environment SNR estimation due to the huge computational overhead.It is urgent to explore a SNR estimation method suitable for complex electromagnetic environments.Emerging artificial intelligence technologies have brought new ideas for the rapid and accurate assessment of SNR.As a representative technology of AI,deep learning can learn highorder hidden features from massive data and can use high-order features to make decisions or predictions,which has attracted widespread attention.However,the SNR estimation based on deep learning faces the problems of lack of complex electromagnetic environment datasets and high model complexity.To this end,this topic focuses on the intelligent estimation technology of SNR in complex environments,constructs a dataset of unmanned aerial vehicles(UAV)remote control signals,and proposes an intelligent estimation algorithm of SNR based on a single-pathway convolutional neural network(CNN)and an intelligent estimation algorithm based on two-pathway CNN.The proposed algorithms take into account model complexity and recognition accuracy,which greatly improves the accuracy of SNR estimation,and provide a reference for accurate SNR estimation in complex environments.The main innovations of this topic are summarized as follows:(1)A single-pathway CNN-based SNR intelligent estimation algorithm called 2D-CNN is proposed,which combines the single-pathway CNN with the SNR estimation.Firstly,a two-dimensional dataset of UAV remote control signals is constructed,and then the influence of model depth and convolution kernel size on the estimation of SNR is discussed,to determine a SNR estimation model that takes into account the performance and complexity.Experiments showed that the proposed model outperforms two known classical algorithms on the constructed dataset,the average relative error is reduced by about 0.3.(2)An intelligent estimation algorithm of SNR based on two-pathway CNN called TP-CNN is proposed.First,the constructed dataset is augmented to increase the accuracy of the SNR classification.Then,2DCNN is improved,and the features extracted by two different neural networks of structures are fused to form a new network architecture for SNR estimation.The experimental results showed that the proposed model can further improve the estimation accuracy of the SNR,the average relative error which is about 0.01 less than 2D-CNN. |
