Research On Theories And Techniques Of Adaptive Subspace Signal Detection

Adaptive signal(target)detection is one of the main issues in the field of radar sig-nal processing,which is an important prerequisite for the procedures of target focusing,imaging,recognition and tracking.However,noise and/or interference are inevitable dur-ing detection.And it is always an urgent need to enlarge the signal-to-noise ratio and reduce the influence of the interference.This dissertation considers adaptive detection of subspace signals in Gaussian noise,and carries out the researches on whether there is an interference,whether the interference is random or deterministic.The outlines and contributions of this dissertation are briefly summarized as follows:(a)When the training data is insufficient to accurately estimate the noise covariance matrix(NCM),a reduced-dimensional(RD)method is introduced.Namely,the test and training data are projected onto the signal subspace which has a lower rank,to reduce the rank of the NCM to be estimated,as well as the demand for the training data volume.For point-like target detection,two RD detectors with constant false alarm rate(CFAR)properties are designed via the one-step and two-step generalized likelihood ratio tests(GLRTs)? when the signal steering vector is partially known,the GLRT,Rao and Wald tests are used,correspondingly,the same new CFAR direction detector is proposed in ho-mogeneous environment(HE),three new CFAR direction detectors are devised in partially HE(PHE)? in the case where a rank-one random interference exists in the test data but not the training data,a RD detector is derived.Compared with existing detectors,these new detectors ensure better detection performance when the training data volume is small.(b)If no training data are available,the strategy of multiple-snapshots-within-a-short-time is adopted.This “short-time”ensures that the target signatures maintain constant during detection,this“multiple-snapshots”guarantees that sufficient test data are sampled to accurately estimate the NCM.Four new CFAR detectors are designed via the GLRT,Rao test,Wald test and the aforementioned RD method? they achieve reliable detection performance based on only the test data.(c)For the detection problem involving random interference and thermal noise,the independence between these two components are assumed and three cases are discussed:the signal and random interference are completely overlapping,the signal is embedded in the random interference,the signal and random interference are partially overlapping.And,a new CFAR GLRT detector is designed for the first case? the GLRT-,Rao- and Wald-based CFAR detectors are designed for both the second and third cases.Those new detectors have lower demands for the training data volume and achieve better detection performance,when compared with their counterparts which deal with the random inter-ference and thermal noise as a whole.(d)For the detection problem where the signal and deterministic interference are lin-early independent,a modified Rao test whose relative parameter contains both the signal-and interference-related parameters is applied to design a CFAR detector.This modified Rao detector outperforms its counterparts when the test data volume is large and the train-ing data volume is small.(e)If the deterministic interference is linearly independent from the signal and its steering vector is partially known,this steering vector is assumed to lie in a known inter-ference subspace.Two new CFAR direction detectors are designed via the one-step and two-step GLRTs? they are more effective than the existing detectors when dealing with the problem of partially known interference steering vector.(f)When the signal and deterministic interference are linearly independent,the train-ing data volume is small and the number of unknown parameters is large,the RD method is adopted to filter out the deterministic interference,to reduce the number of unknown parameters and the dimension of the detection problem.When the signal steering matrix is known,three new CFAR detectors are designed via the GLRT,Rao and Wald tests?when the signal steering vector is partially known,the same new CFAR direction detector is derived via different design techniques.These new detectors are capable to deal with the detection problems involving large number of unknown parameters? they achieve bet-ter detection performance when the training data volume is small,when compared with existing detectors.(g)When the signal and deterministic interference are partially related,the singular value decomposition is employed to recast the signal and deterministic interference as one.By resorting to the one-step and two-step GLRTs,two new CFAR detectors are derived in both HE and PHE.These new detectors outperform their counterparts? they reduce to the existing GLRT-based ones if the signal and interference are linearly independent.