Research On Parameter Estimation And Identification Technology For Non-cooperative Radio Stations

As an effective anti-interception and anti-jamming communication method,frequencyhopping communication is widely used in military communication.Therefore,the reconnaissance and jamming of frequency-hopping signals have become the focus of research in the field of communication and countermeasures.This paper studies the signal parameter estimation and identification technology in the receiving process of noncooperative radio stations,including the reception and separation of non-cooperative frequency hopping signals,the parameter estimation of non-cooperative frequency hopping signals and the open-set identification of modulated signals.For the reception and separation of non-cooperative frequency hopping signals,this paper uses the linear array structure to capture the spatial characteristics of frequency hopping signals through the correlation and difference between the received signals of multiple channels,and then according to these spatial characteristics,alias hopping frequency signal separation.Specifically,this paper firstly adopts the multiple signal classification(MUSIC)algorithm based on the spatial spectrum estimation theory to estimate the direction of arrival of the received signal by distinguishing the characteristic subspace and the signal subspace of the received signal.Then the minimum variance distortionless response(MVDR)beamforming algorithm is used to keep the signal in the desired direction while suppressing the interference and noise signals to separate the aliased frequency hopping signals.For the separated frequency hopping signal,its frequency hopping parameters need to be estimated.This paper firstly analyzes the traditional frequency hopping frequency and frequency hopping time estimation algorithm based on time-frequency distribution,and finds that its estimation performance is affected by the frequency resolution and frequency interval.Therefore,this paper proposes a data-driven and model-driven frequency hopping parameter estimation algorithm.Specifically,this paper adopts a long short-term memory network(LSTM)to estimate the frequency hopping moment by using the energy near the signal hopping moment as a feature.At the same time,the sparse basis matrix is constructed according to the sparsity of the frequency domain of the frequency hopping signal,and then the frequency hopping frequency estimation is transformed into a sparse reconstruction problem.Finally,a sparse Bayesian algorithm is used to solve the above problem.Experiments show that,compared with the traditional algorithm,the data-driven and modeldriven method can perform high-resolution estimation of frequency hopping parameters under the condition of low signal-to-noise ratio and small frequency interval.Finally,in order to further decipher the information carried by the de-frequency hopping signal,the identification of the signal modulation scheme is studied in this paper.Aiming at the existence of low signal-to-noise ratio signals and unknown types of modulated signals in complex environments,an open-set recognition network based on signal enhancement is proposed.First,modulated signals with low signal-to-noise ratio are enhanced by generative adversarial network to eliminate the influence of noise on the recognition accuracy.Then,by designing a deep residual network and introducing a center error loss function,the signals of different modulation methods are guided to the center of their respective classes.This makes it possible to distinguish known and unknown classes of modulated signals by Euclidean distance in a high-dimensional space.Experiments show that the method proposed in this paper has better robustness and higher recognition accuracy for signals with different signal-to-noise ratios.