Methods based on instantaneous statistical features and high-order cumulants are widely used in digital modulation recognition. However, the signal models used by the above two methods are different from each other, which lies in the difference of code elements. Specifically, for one method the code element ranges from 0 to M-1, while for the other method the code element ranges from 1 to $M$, where $M$ denotes the modulation order. The former one is usually used in method based on instantaneous, while the latter one is usually used in method based on high-order cumulants. Unfortunately, the signal model is usually unknown for the receiver and demodulator, hence the digital modulation recognition method based on specific signal model may lead to high error in all probability, this may become obvious for MASK signals. This paper analyzed the two signal models with different code element ranges and high-order cumulants of the two signal models, and proposed a digital modulation recognition method suitable for above two signal models based on mixed features, including two features based on high-order cumulants, spectrum feature based on FFT and signal instantaneous feature. Simulation results showed that the method proposed in this paper can recognize six classical digital modulation patterns efficiently with satisfactory recognition rate.