Compressive Sensing-based Radar Signal Processing Techniques

To improve the detection accuracy,modern radar systems tend to utilize high bandwidth and multichannel signals.These signal processing methods would add the complexity of the radar signal acquisition system.Meanwhile,it will bring massive data.In contrast,target parameters are highly sparse distributed in the detection background.As a new theory for sparsity signal acquisition and processing,compressive sensing or compressive sampling(CS)provides a new ideal to reduce the complexity of the radar system.CS radar have potential advantages in many areas,such as data acquisition,transmission and processing.With the continuous deepening in CS radar,many of the challenges are emerged,among which the low signal-to-noise ratio(SNR)problem is the most obvious one.When faced with low SNR scene,the existing analog-to-information converter(AIC)may be invalid,the design of a measurement matrix becomes difficult,and the performances of the recovery algorithms decline significantly.As a result,the performance loss would appear in a CS radar system.This dissertation focus on the above problems in the low SNR scene based CS radar.Some works are carried out from the perspective of low complexity AIC research,measurement matrix design,recovery algorithm optimization,and sparse representation.In summary,the main contents of this thesis are listed as follows 1.The effect of noise on the recovery performance of CS is studied.The joint model for both the signal noise and the measurement noise in the CS is constructed.The restricted isometry property(RIP)of approximate sparsity is analyzed.In order to explain the noise folding phenomenon,the impacts of the signal noise and the measurement noise to the recovery performance are analyzed by comparing the SNR before and after the recovery,respectively.The result that the effect of signal noise on the recovery performance is consistent with the conclusion that obtained from the aspect of whitening.The reason that why it is hard for sparse recovery in the low SNR scene is analyzed based on the above results,which provides theoretical basis for the researches in the low SNR condition.2.The problem of measurement matrix design with low SNR condition is discussed.In view of the noise folding phenomenon in CS,a methodology for measurement matrix design using subspace method-based adaptive selective compressive sampling(ASCS)is presented in this thesis.By employing the estimated noise prior information in the CS front end,which is fed back by the processing center in the back end,the measurement matrix in the ASCS scheme can be adapted in order to selectively sense the sparse signal.The proposed ASCS method has an inherent characteristic of noise suppression,thus provides fewer signal noisy measurements;Furthermore,an ASCS-based one dimensional angle estimation scheme is proposed for CS-MIMO radar.The RIP of the CS-MIMO radar is analyzed.Taking the multidimensional structure into consideration,the tensor model for the received data in formulated.The high order singular value decomposition method is utilized for more accurate noise information estimation.Due to the fact that the proposed ASCS scheme can suppress the noise,the DOA estimation accuracy is therefore improved.3.The Bayesian framework-based sparse recovery algorithm is investigated in the low SNR situation.A block sparse Bayesian learning based recovery algorithm is proposed for the multitask compressive sensing(MCS)problem.Exploiting the vectorization method,the MCS problem is linked to a block Bayesian-based single measurement vector model.The statistical CS model is constructed.To enhance the recovery performance,the block structure characteristics is introduced in the model.The parameters iteration principles are derived using Bayesian rule;The MCS model-based angle estimation algorithm is developed for CS-MIMO radar.The temporal structure of the echo coefficients is considered.The sparse representation model is formulated for angle estimation.The problem of angle estimation for CS-MIMO radar is linked to parameters learning problem with Bayesian framework.A fast iteration learning algorithm is derived in the case of the same measurement matrixes in the MCS model.Compared to the classical Bayesian algorithm,the proposed algorithm has less computation load as well as more estimation accuracy.The proposed algorithm,which exploits intra-signal correlation,is capable of applying to limited data support and low SNR scene.4.The multidimensional sparse representation problem is also studied.The prolem of sparse representation and the algorithm for sparse recovery in two dimensional space and higher dimensional space are investigated.We fouced on tensor-based three-way CS model and overcomplete dictionary-based recovery method;A three-way compressive sensing based algorithm is developed for angle estimation in bistatic multiple-input multiple-output radar.Exploiting the multidimensional structure inherent in the received data,a third-order tensor signal model is formulated.To lower the storage and computing complexity,the high-order singular value decomposition method is applied to compressive the tensor data.The kernel tensor is linked to the trilinear model thus the compressed direction matrixes are obtained.Thereafter,the sparsity of the target in the background is utilized and two overcomplete dictionaries are constructed for angle estimation.Taking advantage of the inherent multidimensional structure of the received data,the proposed algorithm achieves better estimation accuracy than traditional algorithms.In addition,the proposed algorithm does not require further pairing computing.Furthermore,it could achieve the Doppler frequencies of the targets.Simulation results verify the effectiveness of the proposed algorithm.5.The problem of sampling system design for a CS front-end in the radar-communication integration background is studied.A multi-Chirp based methodology for modulated wideband converter(MWC)is presented.Multi-Chirp signal are used to sense the wideband sparse signal.The problem of sense matrix optimization is converted into design of multiple Chirp signals.Compared with the existing MWC schemes,the proposed scheme is easier to implicate.The experimental results demonstrate the proposed scheme improves the MWC performance under low SNR scene;A sinusoidal frequency modulation(SFM)multiplier-based MWC architecture is presented.The measurement matrix is made up by the coefficients of the frequency spectrum of multi-SFM signals.The recovery accuracy of proposed architecture is very close to the existing pseudorandom binary sequences based MWC,and the proposed architecture is more flexible than the pseudorandom binary sequences based MWC for sparse spectrum sensing.