The nonlinearity of weapon-radiated noise and its application for the classification of passive acoustic signal

The non-gaussianity and nonlinearity of the weapon-radiated noise (WRN) are firstly tested using bispectral analysis, the results show that the model of WRN signals should be nonlinear. Then, a bilinear model (BLM) is presented, which is used to preprocess the WRN signals, and the parameter estimation algorithm is also given. Finally, a product hidden markov model (PHMM ) classifier is builded using the BLM coefficients to classify the acoustic signals. The results show that the BLMPHHM classifier are superior to traditional AR-HMM classifier and have great potentials in the field of passive acoustic signal classification.