This paper presents a synthesis of the subspace-based methods for direction of arrival or frequency estimation which do not requireudthe eigendecomposition of the data covariance matrix . These methods, referred to as linear methods because they only use linearudoperations on the data covariance matrix, have a potential interest for real time applications because of their low complexity andudtheir possible adaptive implementation . While presenting the methods which are referred to as BEWE, the Propagator Methodud(MP) and SWEDE, we establish the relationship between the different versions of these methods . The complexity of each method isudestablished and discussed . BEWE then appears as the less costly of the linear methods . As the asymptotical performances (for anudinfinite number of data) of BEWE and SWEDE has already been obtained in the literature, we here propose the derivation of theudasymptotical performances of a particular version of the MP, referred to as the Propagator Method with noise elimination (MPEB) .udWe then show that MPEB has the best performance of the linear methods and has the same performance as MUSIC . Simulationsudare given to strengthen the theoretical results established in the paper and to illustrate the comparaison between all the differentudmethods.
展开▼
