Structured sparsity preserving projections for radio transmitter recognition

Bispectrum is an effective measurement to the intrinsic stray feature of radio transmitters, however, it is not feasible to directly exploit bispectra for radio transmitter recognition because of the high dimensionality. In this paper, a novel dimensionality reduction method named structured sparsity preserving projections (SSPP) is proposed for feature extraction. SSPP captures the structured sparse reconstructive relationship among data samples based on a structured regularization and then projects the data samples to a low-dimensional subspace with the relationship best preserved. SSPP naturally uses the class label information and sub-block information of the samples to improve the generalization capability. Experimental results on 10 MSK modulation radios, including closed-set test (target identification) and open set test (target verification), show that SSPP outperforms integral bispectra+PCA, LPP, SPP for radio transmitter feature extraction.