| With the rapid development of communication technology and the upgrading of communication equipment,communication signal processing is facing severe challenges,and it is urgent to study new technologies to meet the era's demands.As the most popular branch of machine learning,neural network technology is the closest learning algorithm to human intelligence,which is widely used in image,speech,medicine,and other fields.In view of the mature application of neural network algorithms in other fields,in this paper,the deep neural network algorithm is applied to the field of signal processing,such as signal detection,recognition,and demodulation algorithm.The main contributions and innovations are as follows:1.For the wideband multi-signal detection,this paper designs a signal detection model based on object detection.The algorithm consists of two parts: the wideband signal preprocessing and the detection model.In the aspect of preprocessing,the special presentation of the signal on timefrequency spectrum is analyzed,and then the standard flow of the sample processing is given.In the aspect of the detection network,we mainly set the YOLO V3 detection algorithm as the prototype,and improve the adaptability of the multi-signal detection problem in this paper.Theoretical and simulation experiments show that the algorithm does not need any prior information,can realize the signal location under the condition of low SNR,and is not sensitive to time-frequency aliasing and colored noise.2.For the modulation signal recognition,the recognition set is {BPSK,QPSK,OQPSK,8PSK,16 QAM,16APSK,32 APSK,64QAM}.In this paper,a modulation recognition algorithm based on the classical convolutional neural networks is proposed.The algorithm takes the enhanced signal vector diagram as the network sample,uses several classic convolution neural networks as the recognition model.On this basis,a multi-inputs convolution neural network model is proposed,which takes the vector diagram and eye diagram as the network samples.Experimental results show that the method can recognize signals from different dimensions,and the performance is better than other algorithms.3.For specific signal recognition,a recognition algorithm based on dilated residual network is proposed.Firstly,the visual characteristics of signals under different protocols on timefrequency spectrums are theoretically deduced,and the feasibility of signal recognition based on time-frequency spectrums is summarized.Then,the neural network structure based on dilated convolution is proposed.Finally,seven kinds of typical signals with specific specifications are used to verify the performance of the algorithm.The experimental results show that the algorithm can accurately identify the signal,and maintain good robustness in the case of low SNR and timefrequency aliasing,which has a strong research value.4.For the amplitude-phase signal demodulation,a demodulation algorithm based on the convolution neural network is proposed.In this paper,the baseband complex signal oversampling data is taken as the network samples,and the convolution neural network.The simulation results show that in the Gaussian white noise environment,the demodulation performance is consistent with the best receiver performance,which lays the foundation for the subsequent PCMA signal separation.5.For the PCMA signal separation,a separation algorithm based on the convolutional neural network is proposed.In accordance with the amplitude-phase signal demodulation algorithm,the network sample adopts the complex baseband oversampling data.Based on the original demodulation network,a double-label network is proposed.Through the training of the network,the PCMA signal separation of QPSK modulation and 8PSK modulation is realized,and the separation speed meets the real-time demand.On this basis,the effects of signal-to-noise ratio,amplitude ratio,delay difference on the performance are further discussed,which provides a new idea for high-order modulation of PCMA signal separation. |
PDF Full Text Request