Many existing signal recognition methods for steady-state visual evoked potential (SSVEP) do not pay sufficient attention to the importance of the phase features. In this paper, an improved extended canonical correlation analysis (eCCA) method is proposed for SSVEP signal identification. The phase parameter in the stimulus paradigm of joint frequency-phase modulation coding is added to the reference signal constructed from subjects′ training data as a way to achieve phase constraints on eCCA, thus improving the recognition performance of the eCCA method for SSVEP signals. Thus the eCCA-based SSVEP signal recognition performance is improved. To verify the effectiveness of the proposed method, SSVEP signal recognition experiments are conducted on a publicly available dataset and compared with the existing signal recognition methods. The experimental results show that the average recognition rate of the proposed method is improved to 82.76%, and the information transmission rate is reached to 116.18 bits/min with better stability.