Research On Integral Recognition Of Radar And Communication Signals With Feature Learning

Radar and communication signals,of which numbers are increasing and modulations become more complex,are overlapped in spectrum both of battlefield and civil environment.Recognition of radar and communication signals,inevitably,makes significant sense in precise electronic warfare reconnaissance,unmanned battle platform reconnaissance and spectrum sensing and sharing of cognitive radio.Set on the background of complex electromagnetic environment,the thesis focuses on integral reconnaissance and recognition technique depending on feature learning.Major innovations contained are illustrated as follow.1.A method that tells radar and communication signals apart,including frequency-hopping signals,is proposed,concentrating on the problem that existing methods with duty ratio,requiring high SNR levels,are of poor robustness and incapable to frequency-hopping signals.First,the duty ratio judgement threshold with different SNR levels is derived based on Neyman-Pearson criterion,in order to distinguish communication signals(without frequency-hopping signals)and narrow-band pulse signals.Second,we present an algorithm using PRI entropy and multi-frequency-point duty ratio judgement to distinguish communication frequency-hopping signals and radar pulse signals.Simulation results demonstrate that the derived duty ratio threshold is effective when SNR is higher than 5dB,and the proposed algorithm with PRI entropy and multi-frequency-point duty ratio judgement is capable of telling communication frequency-hopping signals and radar pulse signals with different measure precision.2.The paper presented the radar modulation recognition method depending on ambiguity features and fuzzy clustering learning.The proposed method concentrates on the low SNR environment of instantaneous wide-band reconnaissance of wide-band signals,the low calculating complexity demand on real-time processing and low power limits,and practical situations where prior knowledge of the number of radar emitters is unavailable.First,an algorithm,which extracts the max energy angle of the ambiguity function using coordinate transforming and rounding function,is proposed.Second,smoothing algorithm based on derivative constraint is presented so that the slice extracted on the max energy angle is denoised.The Second-order Cone Programming form of the algorithm is derived to decrease calculating complexity.We also present an algorithm utilizing Symmetry Holder coefficient to quantize the slice waveform.Third,depending on Davies-Bouldin Index(DBI),an ergodic optimizing algorithm is proposed to determine the numbers of radar emitters.At last,feature vectors consisting of the max energy angle and slice is clustered through fuzzy C-means.Simulation results show that the proposed method achieves a correct recognition rate of 98% when SNR?0dB,the feature extracting algorithm is of lower calculation complexity when compared with similar algorithm in high sampling rate,the ergodic optimizing algorithm is effective to estimate the emitter numbers when SNR?0Db,and proposed recognition method is applicable to practical situations where four different LFM emitters exist together with other modulated emitters in low SNR levels.3.The paper proposes a communication modulation recognition method based on cyclic spectrum and Deep Belief Net(DBN).This method focuses on low SNR environment and demand for low calculating complexity.First,FFT Accumulation Method(FAM)is utilized to extract the cyclic spectrum,a preprocessing algorithm based on grey-scaled map and dimensionality reduction is also presented.Second,DBN is designed especially for recognition,and the format and generating approach of training data is presented as well.Finally,the communication signal recognition framework with DBN is put forward.Simulation results show that methods in the paper have better performance towards FSK and PSK signals in low SNR levels,the feature extraction algorithm is of lower calculating complexity.Supplement simulations also verify that the proposed method towards communication modulation recognition can be applied to radar coding modulation recognition.