Research On Modulation Recognition Technology Of Weak Communication Signal Based On Stochastic Resonance And Deep Learning

In the acquisition of communication signals,how to overcome the adverse effects of noise on the signal has always been an important research content in the field of communication.Due to increasingly diverse communication methods and increasingly saturated spectrum utilization,how to identify weak signals in strong interference and strong background noise has always been a hot issue in current research.Most researchers believe that extracting signal features by designing noise reduction algorithms can facilitate the detection and recognition of signals,and the various signal detection methods formed thereby have been widely used in the field of communications.However,with the elimination of noise,the characteristics of the signal will also be damaged to varying degrees due to the detection algorithm.As a special signal enhancement method,the stochastic resonance method can realize the mutual conversion of noise energy and signal energy to a certain extent,and realize the enhancement of signal energy.This subject is based on the research of stochastic resonance theory and applied it to the field of communication signal processing,with the intention of expanding the application range of the theory in weak signal enhancement and detection and recognition.The main work and research results include:1.Aiming at the bistable stochastic resonance mainly applied to binary signals,and less application on multi-ary signals,the enhancement effect of stochastic resonance on the multi-ary phase modulation signal is discussed,and the MPSK signal power expression after resonance is deduced The formula also gives different signal-to-noise enhancement ranges when the signal strength is determined.Starting from the theory of bistable stochastic resonance,the power density expression and power expression of MPSK signal after stochastic resonance enhancement are obtained,which proves that the power spectral density and power expression of MPSK signal are independent of phase;the signal-to-noise ratio gain is defined,The relationship between the signal-to-noise ratio gain and the original signal power and noise variance is calculated,and the noise variance range when the signal gains the forward gain.2.Aiming at the calculation method of the average power of the MQAM signal of the rectangular M^1/2×M^1/2 constellation diagram,it is only valid whenM=2^2k,k?^N+and is not suitable for other M values.This paper proposes a calculation of the average power of the MQAM signal and the corresponding maximum amplitude that holds for anyM=2^k+1,k?^N+method.And for the case where bistable stochastic resonance is less used in MQAM signals,the power expression of MQAM signal after stochastic resonance enhancement is derived,and the relationship between the signal-to-noise ratio gain and the average power and noise variance of the original signal is obtained and the positive result is obtained.To gain the signal-to-noise ratio interval,a bistable system parameter adjustment method is finally proposed,so that the signal within a certain signal-to-noise ratio range can obtain the specified gain.3.Aiming at the problem that it is difficult to extract features of digital modulation signals when the signal-to-noise ratio is low,and the enhancement of the signal through stochastic resonance will lead to the destruction of the signal structure and weaken the performance of traditional digital features,a modulation recognition based on stochastic resonance and Res Net is proposed.method.Input the original MPSK,MQAM,MASK,MFSK signals into the bistable stochastic resonance system to generate stochastic resonance signals,and then make the stochastic resonance signal samples and the original signal samples into experimental sample sets and control sample sets respectively,and use Res Net to analyze the above sample sets respectively Perform classification recognition and comparison.The simulation experiment results show that compared with the direct recognition using Res Net,stochastic resonance can increase the success rate of modulation recognition classification by an average of 4.07% at each signal-to-noise ratio.Experiments also show that the algorithm has less influence on the recognition rate of signals that have undergone stochastic resonance than that of signals that have not been resonance when the number of samples with high signal-to-noise ratio is reduced.