Abstract:

       For non-stationary digitally modulated signals, a high definition feature, high order cyclic cross cumulant(HOCCC) for is proposed to suppress interference and noise. A novel modulation classifier based on nonlinear dynamic fuzzy neural networks (FNN) is presented.  According to the general distribution of the feature samples, we establish a fuzzy inference system with initial experiences, embed the structure and self-adaptive training of the neural network to adjust and optimize the fuzzy system parameter, and complete approximation of the fuzzy neural network modeling. For MASK, MPSK, MFSK and MQAM, simulation results show better adaptability and fault-tolerance of the system at a variety of environment parameters such as SNR. The system with initial experiences possesses a short modeling phase, and can improve average probability of correct classification and efficiency compared to neural network classifiers.

Key words: modulation recognition, high order cyclic cross cumulant, fuzzy neural networks, fuzzy inference,
adaptation