Research On Key Techniques Of Target Recognition For Multiple Signals Based On Sparse Recovery Algorithm

As the rapid development of communication technology and the increasing number of radiation targets,signals captured by communication receivers tend to be overlapped and are difficult to be distinguished.The target recognition technology can estimate signal parameters and assess the difference between different signals.Then it can evaluate the status of spectrum and analysis the distributing patterns of received signals.By discriminating the specific signal from complicated and disorderly resources,the target recognition technology ultimately provides important support for the following signal processing.However,as the electromagnetic environment becomes more complicated,the non-ideal characteristics,such as low signal-to-noise ratio(SNR),few samples and spectrum collision,impose severe challenge on traditional target recognition method.Consequently a more efficient and accurate target recognition mechanism which can be applied to highly complicated electromagnetic environment is of critical concern.In recent years,the signal processing method based on sparse characteristics shows impressive advantages in non-ideal electromagnetic environment.Besides,the multi-antenna signals reception technology gradually becomes essential for signal detection,parameter estimation and signal sorting.Consequently,this thesis offers emphatic discussions on two types of target recognition problem based on sparse reconstruction and multi-antenna signals reception technology: frequency-hopping(FH)signal sorting and spectrum sensing of overlapped signal.The main work of this thesis can be summarized as the following four points:1.The existing FH signal sorting algorithm cannot meet the need of real-time processing or handle the FH signal with high hopping speed.In order to sort FH signals with few samples,a real-time tracking and parameter estimation method is proposed.According to the sparsity in frequency domain,sparse bayesian learning(SBL)is introduced to reconstruct multiple measurement vector(MMV).By constructing new statistic parameter,a hop timing detecting method with constant false alarm probability is derived.Then the proposed method estimates the carrier frequency and direction-of-arrival(DOA)by gravity of geometric center and least square method respectively.According to the hopping pattern of asynchronous and synchronous FH network,the parameters can provide sufficient reference for sorting FH signals.Experiments show that the new method has lower false alarm probability under low SNR,and improves the accuracy of parameter estimation remarkably.2.The FH signal sorting method based on standard array manifold presents higher requirement for antenna layout and channel conformance.To solve these problems a novel method in randomly distributed antenna system is proposed.SBL is introduced to reconstruct single measurement vector(SMV).The reconstruction result reveals the number of signals,carrier frequency and time-delay vector.Considering of the separability of time-delay vector between different FH signals,an improved K-means clustering algorithm which can accelerate the calculation and enhance the convergence accuracy is presented.Experiments demonstrate that the proposed method has obvious improvement in the accuracy of parameter estimation and signal sorting.3.For signals overlapped in both time domain and frequency domain,existing spectrum sensing algorithm fails to take full advantage of the angle seperation,therefore cannot make impressive assessment about the spectrum.In order to effectively distinguish the overlapped signals,a decision test combined with adaptive threshold is derived by integrating the binary probability hypothesis into iterative procedure of SBL.The proposed pruning step can accept the active components of the model and transform the sparse reconstruction into a detection problem.Therefore,the algorithm can sense the spectrum blindly with constant false-alarm rate as well as estimate the accurate angle of each incident signal.Furthermore,the adaptive threshold can be extended to other sparse reconstruction with accuracy improved and computational complexity reduced.4.After studying the stationary target recognition problem,this thesis finally focusses on the recognition of multiple moving targets.A fast DOA tracking algorithm for multiple targets is proposed based on Kalman filtering and sparse signal reconstruction.By modeling the spatial-temporal dynamics of sources with autoregressive(AR)processes,the Kalman filter equations can be obtained.Then state vectors can be efficiently estimated by a fast algorithm with lower complexity,and the coefficients of Kalman filter can be estimated in a principled manner via bayesian inference.The state vectors and model coefficients are recursively estimated to improve the estimation accuracy,which is capable of enhancing the concentration of spatial-temporal distribution.The simulation results verify that the proposed algorithm remarkably refines the tracking performance around the intersection of trajectories.Moreover,the proposed algorithm can improve the estimation accuracy of DOA with lower computing cost than previous algorithms.