| With the development of wireless communication technology and the large-scale deployment of5 G network,the electromagnetic spectrum environment becomes more and more complex,which poses a more severe challenge to the signal detection technology at the receiver.Traditional communication signal detection algorithms and most current deep learning-based signal detection algorithms adopt a “feature-driven” approach,and their detection performance is strongly dependent on external feature extraction algorithms.Deep learning itself has a powerful feature extraction capability,which relies on the multi-layer nonlinear processing unit(deep architecture)to directly extract the known and possibly hidden input signal features.Therefore,with the help of deep learning technology,in this paper we focus on the efficient detection of direct sequence spread spectrum(DSSS)signals and orthogonal frequency division multiplexing(OFDM)signals by means of “data-driven” approach and use universal software radio peripheral(USRP)to collect corresponding real-world signals to verify the performance of the proposed detection algorithms.DSSS is one of the basic modes of spread spectrum communication.It has the advantages of strong anti-jamming ability and low probability of interception,which plays an essential role in both civil and military communications.The detection of DSSS signals becomes very difficult because of its low power spectral density.In this paper,the detection of DSSS signals under noncooperative conditions is carried out based on the classification technology of deep learning.Firstly,a detection method based on residual neural network(ResNet)is proposed,in which the neural network is used to learn the features of DSSS signal and noise automatically without extracting the features in advance.Secondly,in order to reduce the computational complexity,a detection method based on ResNet and autocorrelation is also proposed.In this method,the autocorrelation of the received signal is truncated and used as the input of the neural network for training and inference.Finally,the detection results of the DSSS signals generated by computer simulation and the actual DSSS signals collected show that the detection performance of the proposed two methods is significantly better than the traditional autocorrelation-based detection method.In addition,the computational complexity and storage complexity of the ResNet-autocorrelation-based detection method are lower than those of the ResNet-based detection method.As a multi carrier modulation multiplexing technology,OFDM has been widely used in modern communication systems.Detection of OFDM signals has received paramount interest especially for spectrum sensing in cognitive radios.Most of the current deep learning-based OFDM signal detection methods treat the feature extraction and classification separately which may cause performance loss due to the immature feature extraction or learning method.In this paper,firstly,an OFDM signal detection method based on deep learning is proposed,namely OFDM-DetNet,which combines feature learning and classifier in a designed densely connected network(DenseNet)structure.OFDM-DetNet completes the detection in an end-to-end manner,with the in-phase(I)and quadrature(Q)components of the received signal as the input and the detection result as the output.Secondly,when the length of the input signal is larger than the length of the network input data,a confidence-based fusion method is proposed,namely OFDM-DetNet-F.The received signal is segmented,and each segment of the signal is input into the trained network to obtain the confidences which are then fused to judge whether an OFDM signal is present.This can improve the detection performance under low signal-to-noise ratio(SNR)scenarios.Finally,three datasets with simulations and three datasets with real-world signals are constructed to evaluate the performance of the proposed methods.The detection results of the computer simulation generated OFDM signals and the three collected actual OFDM signals show that the detection performance of the proposed OFDM-DetNet is better than that of other two deep learning-based OFDM detection methods,i.e.,stacked autoencoder(SAE)-based method and covariance matrix(CM)-based detection method,in terms of probability of detection.Furthermore,the detection performance can be further improved with the OFDM-DetNet-F method in low SNR region.In terms of complexity,the computational complexity of OFDM-DetNet method is less than that of OFDM-DetNet-F method,and the storage complexity is the same. |
