Through-the-wall radar imaging is an electromagnetic wave sensing technology capable of detecting targets behind walls, doors, and opaque obstacles. Identification of stationary targets is often achieved by first forming an image of the scene, and then segmenting and classifying the targets of interest. In order to provide prompt and reliable situational awareness, this paper proposes a radar signal classification approach that does not rely on image formation. Here, a dictionary learning based method is employed to classify targets behind a wall using the signals received from individual antennas. The cepstrum coefficients of the high resolution range profile are first extracted as features. Then, the latent consistent K-SVD algorithm is used to learn a discriminative dictionary and a linear classifier simultaneously. Experimental results show that the proposed method can classify individual radar signals with high accuracy, without having recourse to image formation.