Specific radar emitter identification using multiset canonical correlation analysis

In order to improve the performance of a specific radar emitter recognition system, a novel framework based on Multiset Canonical Correlation Analysis (MCCA) is proposed. It extracts the Doppler cuts of the ambiguity function (AF) of each radar signal as the initial feature set and employs MCCA to perform feature fusion and redundancy reduction in such a set. By using label information, the further developed Multiset Discriminant Canonical Correlation Analysis (MDCCA) achieves competitive performance while retaining the low order of canonical vectors. Thanks to the direct fusion strategy, the proposed scheme not only avoids the uncertainty in determining the optimal cut of AF in previous methods, but also extends the conventional CCA, which can only deal with two sets of feature vectors, to the multiset version. Experiments on real radar emitter data demonstrate the effectiveness of the proposed methods.