Abstract: Aiming at the problem that the traditional modulation signal feature extraction algorithm has low
recognition accuracy and poor classification effect in noisy environment，based on the existing modulation signal
processing method，a new feature extraction algorithm for modulation signals in radio monitoring is proposed． The
mathematical models of all kinds of modulation signals in radio monitoring are constructed firstly，based on which
the characteristics of instantaneous amplitude，instantaneous phase and instantaneous frequency are obtained by
simulation． The advantages and disadvantages of various algorithms are analyzed for the recognition of the current
signal modulation mode，using the wavelet transform to complete the noise reduction processing of the modulation
signal and the design of the algorithm for the feature extraction of the abrupt boundary，use the algorithm for the
maximum feature extraction of the spectral density of the zero center normalized instantaneous amplitude and the
kernel discriminant analysis algorithm to extract all kinds of modulation signals layer by layer． The complete
classification and improvement of all kinds of modulation signals are realized． The extraction accuracy of the
feature signal under the influence of noise and the classification effect are improved，which provides a favorable
scientific basis for the feature extraction of modulation signals in radio monitoring．

Key words: radio, modulating signal, wavelet transform, wavelet denoising, boundary feature extraction,
nuclear discriminant analysis