With the popularization of low-orbit satellite communication system, the impact of various types of interference signals on the quality of satellite communication has become increasingly serious. In this paper, for the automatic detection of various types of interference signals in low-orbit satellite communication system, an automatic interference signal detection method based on improved depth belief network is proposed. The study firstly establishes seven typical interference signal models, including single-tone interference, multi-tone interference, narrowband noise interference, broadband noise interference, comb spectrum interference, broadband linear sweep interference and impulse interference; and then extracts the signal characteristics from both time and frequency domains, focusing on the analysis of three characteristic parameters: 3dB bandwidth of normalized spectrum, peak-to-average ratio of the frequency domain, and moment skewness of the frequency domain; Then an improved deep belief network model introducing radial basis function activation function, combining -step contrast dispersion (CD)- and adaptive moment estimation algorithms is constructed to realize accurate classification and detection of interference signals. The experimental results show that the proposed method has good recognition effect in the range of dry noise ratio from -6dB to 13dB, and the AUC values for the seven kinds of interference signals as unknown signals are more than 0.7, among which the best detection effect is achieved for the comb spectral interference, with an AUC value of 0.806. The method has a false alarm probability of only 0.3 when the detection probability of the unknown interference is 0.8, which verifies the improvement of the depth belief network effectiveness and accuracy in the automatic detection of low-orbit satellite signal interference.
