Research On Communication Signal Enhancement Technology Based On Deep Learning

In the process of wireless signal transmission,due to the influence of random noise and interference in the channel,the communication quality is seriously affected,thus affecting the analysis of the signal at the receiving end.Therefore,removing noise and interference in communication signals,that enhancing signal is an important technology promoting the development of wireless communication.However,the traditional signal enhancement method uses the prior information of noise or interference to map to the separable transform domain for separation,such as band-pass filtering.However,due to the random characteristics of noise and interference,it is often a strong prior to noise and interference to construct the corresponding separable transform domain manually.Based on the strong feature extraction and learning ability of deep learning,this paper proposes a new idea of end-to-end self-learning to remove the noise and interference in the signal,and realizes the purpose of signal enhancement through the separable transformation of learning signal and noise and interference.The specific work and contributions of this paper include:Firstly,an embeddable convolution adaptive filtering module based on deep learning is proposed.A typical data set is generated and an end-to-end signal detection network is constructed.The filter enhancement module is embedded in the network to improve the detection accuracy and then evaluate its performance.After the adaptive filter layer,attention mechanism is introduced and Squeeze-and-Excitation-block layer is added to describe the importance of each filter and enhance the ability of filter to extract signal features.Finally,by comparing the signal detection accuracy of the network with the baseline,observing the spectrum changes before and after the enhancement process and the weights assigned by the SE-block to different filters,the adaptability and robustness of the adaptive filter module are analyzed.In order to solve the problem that the convolution-based enhancement module must establish an end-to-end network to supervise its generation,and it is not ideal to eliminate the same frequency interference and random noise,we design an end-to-end wireless signal enhancement network WSEGAN in the framework of generation countermeasure.We analyze the improvement process of the objective function,give the design structure,and prove the importance of each component of the objective function by using the method of control variables.Through a large number of experiments to verify the adaptability,robustness and versatility of the network,it is proved that WSEGAN can achieve the state of the art enhancement effect.In order to verify the effect of enhancement model based on deep learning in application network,this paper designs modulation recognition as the application network.The end-to-end modulation recognition network without signal enhancement module,the modulation recognition network based on embeddable convolutional layer-based filter and the modulation recognition network based on generative adversarial networks are constructed respectively.After the network training,the test set is input,and the modulation recognition effect before and after the signal enhancement is compared.The effect of the two enhancement modules is analyzed by time-frequency analysis and error rate.