Aimed at the specific emitter identification problem in existence of unintentional modulation on pulse, the specific emitter model is established on timedomain and the effect of the innovation component in feature space to individual identification is hence analyzed. First, singular values in the timefrequency domain are extracted to represent features. Then, a joint collaborative representation model is constructed and the Tikhonov matrix is introduced to enhance betweenclass discrimination. Representation coefficients of the unique component are acquired. Finally, the supervised specific identification is executed according to the lowest classification residual. Three types of emitters with unintentional modulation on pulse are adopted in simulations. Experimental results show that the proposed method can effectively identify unintentional amplitude and phase modulation features, and the accuracy can be over 80% when the signaltonoise ratio is 10 dB, which is of significance in generalization modeling.