Analysis Method Of Multi-Component Signal Based On Empirical Mode Decomposition

Recently, electromagnetic environment has becoming more complicated. By intercepting and analyzing techniques, critical intra pulse modulation parameters can be extracted from the radio frequency signals sent by the radar emitter. Thus, the emitter target can be identified, interfered or even attacked, which is greatly significant to obtaining the initiative of the war. A natural question that arises is “How to effectively analyze the signal forms and its feature?”. This actually influences the reliability and effectiveness of signal analysis. Signal diversity is not only depended on the diversity of signal category, but also reflects on the number of signal components which received by the radar receiver at the same time. Thus, in this thesis, I mainly focus on the analysis of multi-component signal and extract time-frequency distribution characteristic from signals, such that the key features of the signal can be extracted.First of all, I investigate the characteristic of multi-component signal. Due to the overlap of radio frequency signals received by the receiver in time domain, multi-component signal can be represented in the form of multiple single component signals. I consider Linear Frequency Modulation(LFM) and Sinusoidal Frequency Modulation(SFM) as the method of signal modulation, and the channel follows an additive Gaussian white noise(AWGN) distribution. Simulation results show that conventional signal analysis cannot effectively process multi-component signal such that the feature of signal inner modulation cannot be reflected.Moreover, Empirical Mode Decomposition(EMD) theory is usually used for adaptive time-frequency analysis. Motivated by this, I analyze its reason and influence, which provides a corresponding solution to its short comes. The main contributions include: 1) Reducing the phenomena of endpoint wing by extreme point continuation method; 2) Cutting off the false components produced during EMD by calculating their correlation between each intrinsic mode function and the original signal; 3) The thesis adopts Ensemble Empirical Mode Decomposition(EEMD) to further improve the aliasing effect of the model in a low signal-to-noise ratio(SNR) environment; 4) Independent component analysis and higher-order statistics help to remove the noise of the signal. Simulation results show that, this method can effectively reduce the noise under 5d B, which guarantees that further signal analysis will not be influenced.A novel improved Hilbert transform is proposed as a tool of processing conventional intrinsic mode function, which makes the time-frequency distribution of Hilbert-Huang spectrum(HHT) become tighter. Furthermore, the energy of time-frequency line is more concentrated with the help of Gaussian window features. Although these two methods have similar anti-noise performance, compared with the smooth pseudo Wigner distribution(SPWVD), the time-frequency concentration of the proposed method is still muc h better. Based on signals extracted from time-frequency distribution, I adopt “amplitude weighting windowed K- means clustering” to extract the time-frequency line and estimate their parameters. Therefore, the proposed method has a good time-frequency concentration and anti-noise ability. However, the computation of this method is complicated, which needs to be further improved.